dynamic traffic prediction with adaptive sampling for 5g...
TRANSCRIPT
Research ArticleDynamic Traffic Prediction with Adaptive Sampling for5G HetNet IoT Applications
Shuangli Wu 12 Wei Mao13 Cong Liu 4 and Tao Tang4
1Computer Network Information Center Chinese Academy of Sciences Beijing China2University of Chinese Academy of Sciences Beijing China3Internet Domain Name System Beijing Engineering Research Center Beijing China4Beihang University Beijing China
Correspondence should be addressed to Shuangli Wu wushuanglicniccn
Received 1 April 2019 Accepted 12 June 2019 Published 26 June 2019
Guest Editor Bo Rong
Copyright copy 2019 Shuangli Wu et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Due to the proliferation of global monitoring sensors the Internet of Things (IoT) is widely used to build smart cities and smarthomes 5GHetNets play an important role in the IoT video streamThis paper proposes an improved Call Session Control Function(CSCF) schemeThe improvedCSCF server contains additionalmodules to facilitate IoT traffic prediction and resource reservationWe highlight traffic prediction in this work and develop a compressed sensing based linear predictor to catch the traffic patternsExperimental results justify that our proposed scheme can forecast the traffic load with high accuracy but low sampling overhead
1 Introduction
The Internet of Things (IoT) makes it possible to connectvarious physical devices Vehicles buildings etc embeddedin sensors are connected through the Internet of Things anddata can be collected and exchanged via the Internet TheInternet of Things enables mutual communication betweendevices In the next decade the service targets in the InternetofThings will be improved to users in various industries andthe number of machine to machine (M2M) terminals willincrease dramatically and applications will be ubiquitous
In order to solve the problem of explosive data volumeof the IoT a heterogeneous networks (HetNets) technologyis proposed The problem of 5G communication is solvedby deploying a large number of small cells [1ndash3] Thereare several technical challenges hindering the IoT videostreaming over 5G HetNets For example the interactionsbetween HetNet and 5G optical core tremendously dependon the IoT traffic from the local HetNet which may changeperiodically for many reasons Consequently the systemmust have a forecast and reservation scheme to dynamicallybook necessary resource from optical core In addition aconnection admission control (CAC) mechanism plays anessential role in case the actually required resource overpasses
the reserved Traffic on future fifth-generation (5G) mobilenetworks is predicted to be dominated by challenging videoapplications such as mobile broadcasting remote surgeryand augmented reality demanding real-time and ultra-highquality delivery Two of themain expectations of 5Gnetworksare that they will be able to handle ultra-high-definition(UHD) video streaming and that they will deliver servicesthat meet the requirements of the end userrsquos perceived qualityby adopting quality of experience (QoE) aware networkmanagement approaches
To overcome the above challenges this paper proposesa scheme of improved CSCF server for 5G HetNets Theimproved CSCF can reside itself in a mobile edge computing(MEC) server with three extra components added to the orig-inal including IoT traffic prediction bandwidth negotiationand connection admission control [4]
The improved CSCF first forecasts the IoT traffic loadfrom a local HetNet to the 5G optical core After that itemploys Common Open Policy Service (COPS) protocol[5] to adaptively negotiate bandwidth resource Finally theconnection admission control serves to keep the system inshape and guarantee communication quality
The IoT is responsible for the communication betweendifferent devices providing better solutions and enhanced
HindawiWireless Communications and Mobile ComputingVolume 2019 Article ID 4687272 11 pageshttpsdoiorg10115520194687272
2 Wireless Communications and Mobile Computing
services for different information flows Video signal is veryimportant among multimedia services At the same time thevideo traffic on the Internet has increased dramatically anda lot of research work has been done to improve the viewingexperience of the audience
Video traffic between HetNet and optical core oftenfluctuates because of use casersquos characteristic and variationof observed object Therefore our proposed improved CSCFserver chooses a dynamic mechanism in a bid to bookbandwidth economically Apparently the traffic predictioncomponent of improved CSCF server has critical impact onthe performance of dynamic bandwidth negotiation In prac-tice sampling overhead is a real challenge as traffic predictormust talk to two other components of the improved CSCFserver during the sampling process For instance traffic pre-dictor needs the calling records from connection admissioncontrol component as the input After the prediction it hasto transfer the result to bandwidth negotiation componentwhich will further use COPS messages to negotiate with 5Gcore network The procedure above can strongly justify thesignificance of sampling rate reduction
Compressed sensing is a promising paradigm that usessignal sparsity to reduce the amount of data that needs to bemeasured [6 7] Compressed sensing (CS) theory indicatesthat the characteristics of sparse signals are not affected bythe signal sparse basis It can be obtained by a small amountof projection on another basis The reconstruction of thewideband signal can be obtained by solving the L1 normoptimization It is concluded that in the calculation processthe sparsity of the signals in the orthogonal basis determinesthe minimum projection number and sampling rate required[8]
In the literature the study of traffic prediction mainlyconcentrated on the least mean square (LMS) methods thattake sampling and prediction as two independent aspects [9ndash12] As a result the sampling often implements a constantinterval and is too prone to low efficiency in IoT environment
Other than previous work this paper focuses on theaggregated multiple streams in a HetNet instead of oneseparated video stream
Inspired by compressed sensing theory we develop anadaptive sampling rate (ASR) linear predictor to overcomethe shortcoming of traditional schemes ASR linear predictorutilizes the prediction error to control sampling interval sothat the sampling and predicting overheads are significantlyreduced
The rest of the paper is organized as the remainderSection 2 introduces 5G IoT applications in smart citiesSection 3 introduces 5G HetNets and IMS based IoT videostreaming Section 4 develops traffic prediction schemesusing adaptive sampling rate and compressed sensing theoryFinally Section 5 summarizes the advantages of our proposedimproved CSCF server and ASR linear predictor
2 5G IoT Applications in Smart Cities
21 Smart Cities The development of the IoTs and cloudcomputing technology has become an important symbol of
our modern information age especially in the city providingtechnical support for the commercial development of the cityThe comprehensive and multilevel processing of informationtechnology in the IoTs provides a service platform for datasystems Therefore in the smart city information systemthe digital information processing of the city can timelyunderstand the relevant information of the city
The use of IoT technology to improve and develop smartcity information technology shield bad information andguide and support the exchange of useful information enablesit to cover all corners of the city in a network and realizesthe management and control of the city as a whole asshown in Figure 1 At the same time with the help of theintelligent engine and the unified information service systemthe management efficiency of urbanization construction hasbeen effectively improved
For smart cities their construction and developmentare inseparable from the IoT communication technologyIt can be said that IoT communication technology is animportant component of its foundationOne of the importantapplications is intelligent building management which ismainly used in the following aspects
(a) The installation of intelligent building managementtechnology during the construction of the building willenable residents to conveniently manage and control remoteitems under the cooperation of the IoT
(b)The establishment of a large publicmonitoring systemthere are many public buildings in the city and there aremany equipment and facilities Therefore an emergencymonitoring system must be established to solve this problemquicklywith the assistance of the IoTwhenunexpected eventsoccur
(c) Adopt intelligent water coal and electricity metersThe RFID chip is built in the table of various facilities andthe data can be collected into the data processor through thetable and then transmitted to the background system throughthe 5G network to realize the function of automatic meterreading
(d) Processing of abnormal equipment remote moni-toring of air conditioners lights etc is carried out usingwireless sensors and local area network technology during theprocessing of this problem
22 IoT Intelligent Parking System in Smart City With thedevelopment of economy there are more and more vehiclesin the city and the problem of parking difficulty is followedSmart cities can bring effective solutions to this problem Inthe smart city an intelligent parking system based on InternetofThings is proposedThe framework of the system is shownin Figure 2
This system combines advanced IoT technology mobileinternet technology and big data technology As shown inthe figure the system is mainly composed of the videopile the base station the server the mobile client and themanagement cloud platform
The video pile is mainly composed of a camera aprocessor and a communication device While the systemis running the camera is used to capture images of eachparking space and monitor the status of each parking space
Wireless Communications and Mobile Computing 3
INDUSTRY
SECURITY RETAIL
SOCIETY HEALTHCARE
HOME ENERGY
MOBILITY
SMART CITY
Figure 1 5G IoT application in smart cities
BS
5G
Server
Mobile clientManagement
cloud platform
Figure 2 Smart parking system
in real time through target recognition technology Thenthe video pile transmits the information to the base stationvia the wireless network The base station forwards theinformation from different video pile to the server and thecloud management platformThe whole system adopts intel-ligent management and the big data technology is adoptedin the cloud management platform to carry out regionallevel parking space scheduling to prevent the problem ofuneven resource allocationTheuser can query the remainingnumber of parking spaces and make a parking space reser-vation in real time through the mobile phone client andif the user is not familiar with the terrain the system canrecommend the best parking space for the user and provideprecise navigation services Throughout the service process
the system performs automatic billing and supports mobileclient online payment
3 IoT Video Communication overMultilocation HetNets
31 5G Core Networks Bridged by MPLS VPN In the IoTsHetNets networks consist of eNode BS and low-power nodes(LPNs) aka access points a unit that makes it up Differentlayers are covered by the twonodes according to the transmis-sion power [13 14] This approach densifies the topology andimproves space utilization and spectrum reuseThe structureof the heterogeneous network satisfies the requirements of 5Gtechnology and greatly improves the spectral efficiency [15]
4 Wireless Communications and Mobile Computing
Cellular BS
Operator deployed smallcell
User deployed smallcell
Cloud
Macrocell link
Small cell link
Relaying
Gateway
DataCenter
Servers
Figure 3 A typical 5G IoT heterogeneous network
In order to solve this problem 5G wireless networkshave received attention and research in order to provideconnectivity and meet the requirements of different qualityof service (QoS) The key technologies of 5G enable thedeployment of heterogeneous networks (HetNets) to supporta large number of IoT data requirements by deploying a largenumber of small cells
Network density is themost effectivemeans inmanywaysto increase network capacity such as spectrum expansion andspectrum efficiency enhancement [16] By densely deploy-ing small cells indoors and outdoors network capacity isincreased and network latency is reduced Small cells include
microcells picocells and femtocells which are closer to theuser Figure 3 is a schematic diagram of deployment of a smallcell which can be widely deployed indoors and outdoorssuch as in offices intersections and plazas Among themthe data traffic of indoor users can be provided by Wi-FiThe next generation of Wi-Fi 80211ac is expected to growrapidly providing multigigabit transmission ratesThis led tothe concept of HetNets a multilayer network with multipleradio access technologies
A 5G core network could be enhanced by network slicingwhich allows different clients to bridge their private sites overa common provider-owned cloud Using MPLS or another
Wireless Communications and Mobile Computing 5
5G core networks5G HetNet
CE
PEPE
5G HetNet
5G HetNet
CE
CE
PE
PE
Figure 4 Network slicing based 5G core network connecting several local HetNets
CE PE
IMS
Improved CSCF Server
Figure 5 IMS based IoT video streaming over 5G HetNets
similar technology [17] network slicing becomes a noveltrend and has a wonderful capability to offer QoS guaran-teed services with flexibility Figure 4 shows an example ofnetwork slicing based 5G core that connects several localHetNets Using MPLS technology the core network containsthe following types of equipment
(a) customer edge (CE) router which serves to connectwith client local network directly
(b) provider edge (PE) router which serves as the inter-face between 5G HetNet and optical core
(c) provider (P) router which serves to forward the usertraffic arranged by CE and PE routers
32 IMS Based IoT Video Streaming Internet of things (IoT)is regarded recently as the most promising form of thewireless communication environments
In Internet devices a large amount of data is generatedand accumulated into a large amount of data streams overtime Therefore higher requirements are imposed on theprocessing and decomposition of online data In manyapplications a large amount of data storage is generated suchas monitoring the generated video stream data
5G HetNets transport the video streaming traffic bychanging an IoT sensor into an IP Multimedia Subsystem(IMS) terminal IMS realizes robust IMS service functionalityby working closely with Call Session Control Function(CSCF)TheCSCFnode facilitates Session Initiation Protocol(SIP) to operate session setup and teardown [18]
As a popular video streaming protocol CSCF conductssignaling and session management and enables connectioninformation to go across network boundaries
Figure 5 gives an example of IMS based IoT videostreaming over 5G HetNets The improved CSCF serveracts as an intermediate entity that receives and forwardssignaling messages Improved server offers a bunch of cru-cial functions such as authentication authorization androuting
33 Model of Improved CSCF Server By definition the mainmodule of the IMS system is call session control to allowvideovoice communication over the convergence of differentaccess networks It comprises of all the functional modulesrequired to handle all the signaling from end-user to servicesand other networks However conventional CSCF cannot
6 Wireless Communications and Mobile Computing
SIPampCOPSProtocol Stacks
Bandwidth
Negotiation
Traffic
Prediction
Connection
Admission
Control
Original CSCF Server
Improved CSCF Server
SIP Messages
COPS Messages
Enhanced Modules
Figure 6 Model of improved CSCF server
Traffic Load Sampling
Traffic Prediction Algorithm
Real Traffic Load
Predicted Traffic Load
Traffic Load Prediction Module
Figure 7 General framework of IoT traffic prediction
100 meet the requirement of IoT video streaming and bemodernized
Figure 6 shows that an improved CSCF server containsthree more modules than the original Our design has themerit that the improved CSCF server bargains with theoptical core on behalf of all IoT devices of the local HetNetin advance
For this reason the paths required by video streaming arepreconstructed before the stream even starts to reduce theunnecessary signaling delay
4 Traffic Prediction in IoT UsingCompressed Sensing
Traffic prediction in the IoT has been widely studied andapplied in recent years Sampling and prediction representusually two independent parts in the past For example mostconventional approaches keep the sampling rate unchangedwhich may result in low efficiency in IoT video streamingTo overcome this problem this paper develops a novelcompressed sensing based traffic prediction where error offorecast algorithm goes back to control the sampling intervalOur proposed scheme can significantly reduce the samplingoverhead without sacrificing the accuracy
41 Problem Formulation and Architecture Design Let 119861119899minus1119899be the sum of IoT video streaming loads from a local HetNetat time slot [119905119899minus1 119905119899](119899 = 0 1 2 ) Let improved CSCFserver be exactly aware of the value of 119861119899minus1119899 at time 119905119899minus1With this piece of information it will bargain with 5G corefor the amount of 119861119899minus1119899 bandwidth using COPS messagesAt the next time slot [119905119899 119905119899+1] IoT traffic may change to anew value of 119861119899119899+1 and the improved CSCF server will haveto rebargain with 5G core network again
It is nevertheless impossible for the improved CSCFserver to achieve the value of 119861119899minus1119899 before the time of 119905119899Thus at time 119905119899minus1 we can estimate the approximate value119861119899minus1119899 which is denoted by 1198611015840119899minus1119899 If 1198611015840119899minus1119899 lt 119861119899minus1119899 the localHetNet will suffer from the shortage of bandwidth resourceto accommodate all video streaming sessions As a solutionCAC component must reject some of connection requests inorder to guarantee overall quality of service
On the other hand if 1198611015840119899minus1119899 gt 119861119899minus1119899 part of bandwidthresource is simply overbooked and useless In this sense thetraffic prediction is critical to ensure a good performance ofour proposed scheme
Figure 7 presents the general framework of IoT trafficprediction module with the component of traffic load sam-pling and the component of traffic prediction algorithm
Wireless Communications and Mobile Computing 7
We will give more elaboration on these two componentsnext
42 IoT Traffic Sampling An IoT device initiates a videostream session by sending an INVITE message to theimproved CSCF server in a bid to indicate called URL andbandwidth requirement The improved CSCF server savesevery request in the record for IoT traffic sampling usageregardless the request is accepted by CAC or not
In this paper we use 119861119899minus1119899 to represent the averageIoT traffic interval Δ119905119899minus1119899 = 119905119899 minus 119905119899minus1 In practice thesystem achieves 119861119899minus1119899 from the record in the improved CSCFserver Conventional approach conducts traffic samplingwithconstant rate which rules
Δ11990501 = Δ11990512 = sdot sdot sdot = Δ119905119899minus1119899 = (119899 = 0 1 2 ) (1)
Let 119861(119905) be the traffic load function varying with time tand let 119861(119891) be the frequency counterpart of 119861(119905) with theupper bound of 119891max Nyquist theorem regulates the idea thatsampling ratemust be at least twice of119891max to prevent aliasing[19] In IoT video streaming 119891max could be high in somecomplicated scenarios which make constant sampling ratenot applicable at all
43 Compressed Sensing Compressed sensing theory is atheory that has been widely used in recent years and iswidely used in signal processing It is also called compressedsampling or sparse sampling CS wants to sample the signalat a sampling rate much lower than Shannons Law andfind a solution for underdetermined linear systems whilemaintaining the basic information of the signal [20ndash24]
In order to implement the compressed sensing theory thefollowing solution can be used to estimate the sparse vectorx isin C119873 using the observed measurement vector 119910 isin C119872A method based on measurement equations is applied Theformula is as follows
119910 = Φ119909 + 119908 (2)
The measurement matrix is represented by Φ isin C119872times119873119908 isin C119873 represents the unknown vector of measurementnoise and modeling error The reconstruction process canonly occur under ideal conditions ie only when the signalx is S sparse (119878 ≪ 119873 ) That is there can be at most S nonzeroitems During the operation the number of observations issignificantly smaller than the number of variables ie≫ 119872
In practice the signal we are dealing with is not a sparsesignalThe processingmethod in this time is to express Ψ119873119894=1on some basis with the corresponding sparse coefficient 120579119894Processing equation (1) yields
119910 = Φ119909 + 119908 = ΦΨ120579 + 119908 = Θ120579 + 119908 (3)
where 120579 isin 119862119871 is a L dimension sparse vector of coefficientsbased on the basismatrixΨ isin 119862119873times119871Θ = ΦΨ is a119872times119871matrix(119872 ≪ 119871 )
However because fewer measurements are applied thanthe entries in 120579 the resulting solution is uncertain In
order to recover the 119878 sparse signal 1198970 norm of the mostsparse sequence should be found from all feasible solutionsMinimization can be used for the reconstruction processTheformula is expressed as follows
120579 = min 1205790st 1003817100381710038171003817Θ120579 minus 11991010038171003817100381710038172 le Ξ (4)
where 120579 refers to the estimated vector for 120579 and w le Ξis noise tolerance However for practical operations the 1198970normminimization has huge computational complexity anda reconstruction algorithm with lower computational costmust be developed
Studies have shown that the reconstruction process of sig-nal 119909must satisfy a condition that matrix Θ cannot map twodifferent S sparse signals to the same set of samplesThereforethematrixΘmust satisfy the restricted equidistance property(RIP) [25 26]
44 Adaptive Sampling Rate (a)The definition of traditionalcompressive sensing sparse signals can be recovered from asmall number of nonadaptive linearmeasurements Let signal119909 be 119870 sparse in basisdictionary Ψ For example Ψ is DFTmatrix if the signal is frequency domain sparse Ψ can beexpressed as a matrix as shown in Figure 8(a) In Figure 9if signal 119909 is sparse then we can use compressive sensing by119910 = Φ times 119909
Φ is the measurement matrix and the theory of CS statesthat random Φ will work How to understand this Forexample if the signal is sparse in frequency domain then wemake a few random samplings in time domain In contrast ifthe signal is sparse in time domain we make a few randomsamplings in frequency domain
(b) Address the situation that the appearance time ofsparse signal can be predicted In this case Ψ is identitymatrix I as shown in Figure 8(b) Then measure matrix Φis part of I which involves the time points where the sparsesignal appears Φ is not random anymore In practice wepropose variable sampling rate scheme to predict these timepoints (or the measurement matrix)
If a fixed sampling interval applies as mentioned beforethe sampling rate must double the highest frequency of trafficcurve to achieve accurate prediction
Accordingly it is significant to develop an inconstantsampling rate scheme with much lower overhead
Variable sampling rate comes from the observation ofthe traffic curve in Figure 10 with combination of slow andfast changing periods The fluctuation of traffic significantlydepends on the timely feature of IoT devices For examplethe cameras monitor the parking carsrsquo moving and stoppingaccording to peoplersquos behavior which is obviously related tothe number of car arrivals and departures In Figure 8 fastchanging zone represents the period of rapid event varyingsuch as morning peak hour slow changing zone representsthe period of steady event numbers such as at night
With the goal of reducing the sampling overhead astraightforward adaptive rate approach is to utilize low ratein slow changing period and high rate in fast changing
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
2 Wireless Communications and Mobile Computing
services for different information flows Video signal is veryimportant among multimedia services At the same time thevideo traffic on the Internet has increased dramatically anda lot of research work has been done to improve the viewingexperience of the audience
Video traffic between HetNet and optical core oftenfluctuates because of use casersquos characteristic and variationof observed object Therefore our proposed improved CSCFserver chooses a dynamic mechanism in a bid to bookbandwidth economically Apparently the traffic predictioncomponent of improved CSCF server has critical impact onthe performance of dynamic bandwidth negotiation In prac-tice sampling overhead is a real challenge as traffic predictormust talk to two other components of the improved CSCFserver during the sampling process For instance traffic pre-dictor needs the calling records from connection admissioncontrol component as the input After the prediction it hasto transfer the result to bandwidth negotiation componentwhich will further use COPS messages to negotiate with 5Gcore network The procedure above can strongly justify thesignificance of sampling rate reduction
Compressed sensing is a promising paradigm that usessignal sparsity to reduce the amount of data that needs to bemeasured [6 7] Compressed sensing (CS) theory indicatesthat the characteristics of sparse signals are not affected bythe signal sparse basis It can be obtained by a small amountof projection on another basis The reconstruction of thewideband signal can be obtained by solving the L1 normoptimization It is concluded that in the calculation processthe sparsity of the signals in the orthogonal basis determinesthe minimum projection number and sampling rate required[8]
In the literature the study of traffic prediction mainlyconcentrated on the least mean square (LMS) methods thattake sampling and prediction as two independent aspects [9ndash12] As a result the sampling often implements a constantinterval and is too prone to low efficiency in IoT environment
Other than previous work this paper focuses on theaggregated multiple streams in a HetNet instead of oneseparated video stream
Inspired by compressed sensing theory we develop anadaptive sampling rate (ASR) linear predictor to overcomethe shortcoming of traditional schemes ASR linear predictorutilizes the prediction error to control sampling interval sothat the sampling and predicting overheads are significantlyreduced
The rest of the paper is organized as the remainderSection 2 introduces 5G IoT applications in smart citiesSection 3 introduces 5G HetNets and IMS based IoT videostreaming Section 4 develops traffic prediction schemesusing adaptive sampling rate and compressed sensing theoryFinally Section 5 summarizes the advantages of our proposedimproved CSCF server and ASR linear predictor
2 5G IoT Applications in Smart Cities
21 Smart Cities The development of the IoTs and cloudcomputing technology has become an important symbol of
our modern information age especially in the city providingtechnical support for the commercial development of the cityThe comprehensive and multilevel processing of informationtechnology in the IoTs provides a service platform for datasystems Therefore in the smart city information systemthe digital information processing of the city can timelyunderstand the relevant information of the city
The use of IoT technology to improve and develop smartcity information technology shield bad information andguide and support the exchange of useful information enablesit to cover all corners of the city in a network and realizesthe management and control of the city as a whole asshown in Figure 1 At the same time with the help of theintelligent engine and the unified information service systemthe management efficiency of urbanization construction hasbeen effectively improved
For smart cities their construction and developmentare inseparable from the IoT communication technologyIt can be said that IoT communication technology is animportant component of its foundationOne of the importantapplications is intelligent building management which ismainly used in the following aspects
(a) The installation of intelligent building managementtechnology during the construction of the building willenable residents to conveniently manage and control remoteitems under the cooperation of the IoT
(b)The establishment of a large publicmonitoring systemthere are many public buildings in the city and there aremany equipment and facilities Therefore an emergencymonitoring system must be established to solve this problemquicklywith the assistance of the IoTwhenunexpected eventsoccur
(c) Adopt intelligent water coal and electricity metersThe RFID chip is built in the table of various facilities andthe data can be collected into the data processor through thetable and then transmitted to the background system throughthe 5G network to realize the function of automatic meterreading
(d) Processing of abnormal equipment remote moni-toring of air conditioners lights etc is carried out usingwireless sensors and local area network technology during theprocessing of this problem
22 IoT Intelligent Parking System in Smart City With thedevelopment of economy there are more and more vehiclesin the city and the problem of parking difficulty is followedSmart cities can bring effective solutions to this problem Inthe smart city an intelligent parking system based on InternetofThings is proposedThe framework of the system is shownin Figure 2
This system combines advanced IoT technology mobileinternet technology and big data technology As shown inthe figure the system is mainly composed of the videopile the base station the server the mobile client and themanagement cloud platform
The video pile is mainly composed of a camera aprocessor and a communication device While the systemis running the camera is used to capture images of eachparking space and monitor the status of each parking space
Wireless Communications and Mobile Computing 3
INDUSTRY
SECURITY RETAIL
SOCIETY HEALTHCARE
HOME ENERGY
MOBILITY
SMART CITY
Figure 1 5G IoT application in smart cities
BS
5G
Server
Mobile clientManagement
cloud platform
Figure 2 Smart parking system
in real time through target recognition technology Thenthe video pile transmits the information to the base stationvia the wireless network The base station forwards theinformation from different video pile to the server and thecloud management platformThe whole system adopts intel-ligent management and the big data technology is adoptedin the cloud management platform to carry out regionallevel parking space scheduling to prevent the problem ofuneven resource allocationTheuser can query the remainingnumber of parking spaces and make a parking space reser-vation in real time through the mobile phone client andif the user is not familiar with the terrain the system canrecommend the best parking space for the user and provideprecise navigation services Throughout the service process
the system performs automatic billing and supports mobileclient online payment
3 IoT Video Communication overMultilocation HetNets
31 5G Core Networks Bridged by MPLS VPN In the IoTsHetNets networks consist of eNode BS and low-power nodes(LPNs) aka access points a unit that makes it up Differentlayers are covered by the twonodes according to the transmis-sion power [13 14] This approach densifies the topology andimproves space utilization and spectrum reuseThe structureof the heterogeneous network satisfies the requirements of 5Gtechnology and greatly improves the spectral efficiency [15]
4 Wireless Communications and Mobile Computing
Cellular BS
Operator deployed smallcell
User deployed smallcell
Cloud
Macrocell link
Small cell link
Relaying
Gateway
DataCenter
Servers
Figure 3 A typical 5G IoT heterogeneous network
In order to solve this problem 5G wireless networkshave received attention and research in order to provideconnectivity and meet the requirements of different qualityof service (QoS) The key technologies of 5G enable thedeployment of heterogeneous networks (HetNets) to supporta large number of IoT data requirements by deploying a largenumber of small cells
Network density is themost effectivemeans inmanywaysto increase network capacity such as spectrum expansion andspectrum efficiency enhancement [16] By densely deploy-ing small cells indoors and outdoors network capacity isincreased and network latency is reduced Small cells include
microcells picocells and femtocells which are closer to theuser Figure 3 is a schematic diagram of deployment of a smallcell which can be widely deployed indoors and outdoorssuch as in offices intersections and plazas Among themthe data traffic of indoor users can be provided by Wi-FiThe next generation of Wi-Fi 80211ac is expected to growrapidly providing multigigabit transmission ratesThis led tothe concept of HetNets a multilayer network with multipleradio access technologies
A 5G core network could be enhanced by network slicingwhich allows different clients to bridge their private sites overa common provider-owned cloud Using MPLS or another
Wireless Communications and Mobile Computing 5
5G core networks5G HetNet
CE
PEPE
5G HetNet
5G HetNet
CE
CE
PE
PE
Figure 4 Network slicing based 5G core network connecting several local HetNets
CE PE
IMS
Improved CSCF Server
Figure 5 IMS based IoT video streaming over 5G HetNets
similar technology [17] network slicing becomes a noveltrend and has a wonderful capability to offer QoS guaran-teed services with flexibility Figure 4 shows an example ofnetwork slicing based 5G core that connects several localHetNets Using MPLS technology the core network containsthe following types of equipment
(a) customer edge (CE) router which serves to connectwith client local network directly
(b) provider edge (PE) router which serves as the inter-face between 5G HetNet and optical core
(c) provider (P) router which serves to forward the usertraffic arranged by CE and PE routers
32 IMS Based IoT Video Streaming Internet of things (IoT)is regarded recently as the most promising form of thewireless communication environments
In Internet devices a large amount of data is generatedand accumulated into a large amount of data streams overtime Therefore higher requirements are imposed on theprocessing and decomposition of online data In manyapplications a large amount of data storage is generated suchas monitoring the generated video stream data
5G HetNets transport the video streaming traffic bychanging an IoT sensor into an IP Multimedia Subsystem(IMS) terminal IMS realizes robust IMS service functionalityby working closely with Call Session Control Function(CSCF)TheCSCFnode facilitates Session Initiation Protocol(SIP) to operate session setup and teardown [18]
As a popular video streaming protocol CSCF conductssignaling and session management and enables connectioninformation to go across network boundaries
Figure 5 gives an example of IMS based IoT videostreaming over 5G HetNets The improved CSCF serveracts as an intermediate entity that receives and forwardssignaling messages Improved server offers a bunch of cru-cial functions such as authentication authorization androuting
33 Model of Improved CSCF Server By definition the mainmodule of the IMS system is call session control to allowvideovoice communication over the convergence of differentaccess networks It comprises of all the functional modulesrequired to handle all the signaling from end-user to servicesand other networks However conventional CSCF cannot
6 Wireless Communications and Mobile Computing
SIPampCOPSProtocol Stacks
Bandwidth
Negotiation
Traffic
Prediction
Connection
Admission
Control
Original CSCF Server
Improved CSCF Server
SIP Messages
COPS Messages
Enhanced Modules
Figure 6 Model of improved CSCF server
Traffic Load Sampling
Traffic Prediction Algorithm
Real Traffic Load
Predicted Traffic Load
Traffic Load Prediction Module
Figure 7 General framework of IoT traffic prediction
100 meet the requirement of IoT video streaming and bemodernized
Figure 6 shows that an improved CSCF server containsthree more modules than the original Our design has themerit that the improved CSCF server bargains with theoptical core on behalf of all IoT devices of the local HetNetin advance
For this reason the paths required by video streaming arepreconstructed before the stream even starts to reduce theunnecessary signaling delay
4 Traffic Prediction in IoT UsingCompressed Sensing
Traffic prediction in the IoT has been widely studied andapplied in recent years Sampling and prediction representusually two independent parts in the past For example mostconventional approaches keep the sampling rate unchangedwhich may result in low efficiency in IoT video streamingTo overcome this problem this paper develops a novelcompressed sensing based traffic prediction where error offorecast algorithm goes back to control the sampling intervalOur proposed scheme can significantly reduce the samplingoverhead without sacrificing the accuracy
41 Problem Formulation and Architecture Design Let 119861119899minus1119899be the sum of IoT video streaming loads from a local HetNetat time slot [119905119899minus1 119905119899](119899 = 0 1 2 ) Let improved CSCFserver be exactly aware of the value of 119861119899minus1119899 at time 119905119899minus1With this piece of information it will bargain with 5G corefor the amount of 119861119899minus1119899 bandwidth using COPS messagesAt the next time slot [119905119899 119905119899+1] IoT traffic may change to anew value of 119861119899119899+1 and the improved CSCF server will haveto rebargain with 5G core network again
It is nevertheless impossible for the improved CSCFserver to achieve the value of 119861119899minus1119899 before the time of 119905119899Thus at time 119905119899minus1 we can estimate the approximate value119861119899minus1119899 which is denoted by 1198611015840119899minus1119899 If 1198611015840119899minus1119899 lt 119861119899minus1119899 the localHetNet will suffer from the shortage of bandwidth resourceto accommodate all video streaming sessions As a solutionCAC component must reject some of connection requests inorder to guarantee overall quality of service
On the other hand if 1198611015840119899minus1119899 gt 119861119899minus1119899 part of bandwidthresource is simply overbooked and useless In this sense thetraffic prediction is critical to ensure a good performance ofour proposed scheme
Figure 7 presents the general framework of IoT trafficprediction module with the component of traffic load sam-pling and the component of traffic prediction algorithm
Wireless Communications and Mobile Computing 7
We will give more elaboration on these two componentsnext
42 IoT Traffic Sampling An IoT device initiates a videostream session by sending an INVITE message to theimproved CSCF server in a bid to indicate called URL andbandwidth requirement The improved CSCF server savesevery request in the record for IoT traffic sampling usageregardless the request is accepted by CAC or not
In this paper we use 119861119899minus1119899 to represent the averageIoT traffic interval Δ119905119899minus1119899 = 119905119899 minus 119905119899minus1 In practice thesystem achieves 119861119899minus1119899 from the record in the improved CSCFserver Conventional approach conducts traffic samplingwithconstant rate which rules
Δ11990501 = Δ11990512 = sdot sdot sdot = Δ119905119899minus1119899 = (119899 = 0 1 2 ) (1)
Let 119861(119905) be the traffic load function varying with time tand let 119861(119891) be the frequency counterpart of 119861(119905) with theupper bound of 119891max Nyquist theorem regulates the idea thatsampling ratemust be at least twice of119891max to prevent aliasing[19] In IoT video streaming 119891max could be high in somecomplicated scenarios which make constant sampling ratenot applicable at all
43 Compressed Sensing Compressed sensing theory is atheory that has been widely used in recent years and iswidely used in signal processing It is also called compressedsampling or sparse sampling CS wants to sample the signalat a sampling rate much lower than Shannons Law andfind a solution for underdetermined linear systems whilemaintaining the basic information of the signal [20ndash24]
In order to implement the compressed sensing theory thefollowing solution can be used to estimate the sparse vectorx isin C119873 using the observed measurement vector 119910 isin C119872A method based on measurement equations is applied Theformula is as follows
119910 = Φ119909 + 119908 (2)
The measurement matrix is represented by Φ isin C119872times119873119908 isin C119873 represents the unknown vector of measurementnoise and modeling error The reconstruction process canonly occur under ideal conditions ie only when the signalx is S sparse (119878 ≪ 119873 ) That is there can be at most S nonzeroitems During the operation the number of observations issignificantly smaller than the number of variables ie≫ 119872
In practice the signal we are dealing with is not a sparsesignalThe processingmethod in this time is to express Ψ119873119894=1on some basis with the corresponding sparse coefficient 120579119894Processing equation (1) yields
119910 = Φ119909 + 119908 = ΦΨ120579 + 119908 = Θ120579 + 119908 (3)
where 120579 isin 119862119871 is a L dimension sparse vector of coefficientsbased on the basismatrixΨ isin 119862119873times119871Θ = ΦΨ is a119872times119871matrix(119872 ≪ 119871 )
However because fewer measurements are applied thanthe entries in 120579 the resulting solution is uncertain In
order to recover the 119878 sparse signal 1198970 norm of the mostsparse sequence should be found from all feasible solutionsMinimization can be used for the reconstruction processTheformula is expressed as follows
120579 = min 1205790st 1003817100381710038171003817Θ120579 minus 11991010038171003817100381710038172 le Ξ (4)
where 120579 refers to the estimated vector for 120579 and w le Ξis noise tolerance However for practical operations the 1198970normminimization has huge computational complexity anda reconstruction algorithm with lower computational costmust be developed
Studies have shown that the reconstruction process of sig-nal 119909must satisfy a condition that matrix Θ cannot map twodifferent S sparse signals to the same set of samplesThereforethematrixΘmust satisfy the restricted equidistance property(RIP) [25 26]
44 Adaptive Sampling Rate (a)The definition of traditionalcompressive sensing sparse signals can be recovered from asmall number of nonadaptive linearmeasurements Let signal119909 be 119870 sparse in basisdictionary Ψ For example Ψ is DFTmatrix if the signal is frequency domain sparse Ψ can beexpressed as a matrix as shown in Figure 8(a) In Figure 9if signal 119909 is sparse then we can use compressive sensing by119910 = Φ times 119909
Φ is the measurement matrix and the theory of CS statesthat random Φ will work How to understand this Forexample if the signal is sparse in frequency domain then wemake a few random samplings in time domain In contrast ifthe signal is sparse in time domain we make a few randomsamplings in frequency domain
(b) Address the situation that the appearance time ofsparse signal can be predicted In this case Ψ is identitymatrix I as shown in Figure 8(b) Then measure matrix Φis part of I which involves the time points where the sparsesignal appears Φ is not random anymore In practice wepropose variable sampling rate scheme to predict these timepoints (or the measurement matrix)
If a fixed sampling interval applies as mentioned beforethe sampling rate must double the highest frequency of trafficcurve to achieve accurate prediction
Accordingly it is significant to develop an inconstantsampling rate scheme with much lower overhead
Variable sampling rate comes from the observation ofthe traffic curve in Figure 10 with combination of slow andfast changing periods The fluctuation of traffic significantlydepends on the timely feature of IoT devices For examplethe cameras monitor the parking carsrsquo moving and stoppingaccording to peoplersquos behavior which is obviously related tothe number of car arrivals and departures In Figure 8 fastchanging zone represents the period of rapid event varyingsuch as morning peak hour slow changing zone representsthe period of steady event numbers such as at night
With the goal of reducing the sampling overhead astraightforward adaptive rate approach is to utilize low ratein slow changing period and high rate in fast changing
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 3
INDUSTRY
SECURITY RETAIL
SOCIETY HEALTHCARE
HOME ENERGY
MOBILITY
SMART CITY
Figure 1 5G IoT application in smart cities
BS
5G
Server
Mobile clientManagement
cloud platform
Figure 2 Smart parking system
in real time through target recognition technology Thenthe video pile transmits the information to the base stationvia the wireless network The base station forwards theinformation from different video pile to the server and thecloud management platformThe whole system adopts intel-ligent management and the big data technology is adoptedin the cloud management platform to carry out regionallevel parking space scheduling to prevent the problem ofuneven resource allocationTheuser can query the remainingnumber of parking spaces and make a parking space reser-vation in real time through the mobile phone client andif the user is not familiar with the terrain the system canrecommend the best parking space for the user and provideprecise navigation services Throughout the service process
the system performs automatic billing and supports mobileclient online payment
3 IoT Video Communication overMultilocation HetNets
31 5G Core Networks Bridged by MPLS VPN In the IoTsHetNets networks consist of eNode BS and low-power nodes(LPNs) aka access points a unit that makes it up Differentlayers are covered by the twonodes according to the transmis-sion power [13 14] This approach densifies the topology andimproves space utilization and spectrum reuseThe structureof the heterogeneous network satisfies the requirements of 5Gtechnology and greatly improves the spectral efficiency [15]
4 Wireless Communications and Mobile Computing
Cellular BS
Operator deployed smallcell
User deployed smallcell
Cloud
Macrocell link
Small cell link
Relaying
Gateway
DataCenter
Servers
Figure 3 A typical 5G IoT heterogeneous network
In order to solve this problem 5G wireless networkshave received attention and research in order to provideconnectivity and meet the requirements of different qualityof service (QoS) The key technologies of 5G enable thedeployment of heterogeneous networks (HetNets) to supporta large number of IoT data requirements by deploying a largenumber of small cells
Network density is themost effectivemeans inmanywaysto increase network capacity such as spectrum expansion andspectrum efficiency enhancement [16] By densely deploy-ing small cells indoors and outdoors network capacity isincreased and network latency is reduced Small cells include
microcells picocells and femtocells which are closer to theuser Figure 3 is a schematic diagram of deployment of a smallcell which can be widely deployed indoors and outdoorssuch as in offices intersections and plazas Among themthe data traffic of indoor users can be provided by Wi-FiThe next generation of Wi-Fi 80211ac is expected to growrapidly providing multigigabit transmission ratesThis led tothe concept of HetNets a multilayer network with multipleradio access technologies
A 5G core network could be enhanced by network slicingwhich allows different clients to bridge their private sites overa common provider-owned cloud Using MPLS or another
Wireless Communications and Mobile Computing 5
5G core networks5G HetNet
CE
PEPE
5G HetNet
5G HetNet
CE
CE
PE
PE
Figure 4 Network slicing based 5G core network connecting several local HetNets
CE PE
IMS
Improved CSCF Server
Figure 5 IMS based IoT video streaming over 5G HetNets
similar technology [17] network slicing becomes a noveltrend and has a wonderful capability to offer QoS guaran-teed services with flexibility Figure 4 shows an example ofnetwork slicing based 5G core that connects several localHetNets Using MPLS technology the core network containsthe following types of equipment
(a) customer edge (CE) router which serves to connectwith client local network directly
(b) provider edge (PE) router which serves as the inter-face between 5G HetNet and optical core
(c) provider (P) router which serves to forward the usertraffic arranged by CE and PE routers
32 IMS Based IoT Video Streaming Internet of things (IoT)is regarded recently as the most promising form of thewireless communication environments
In Internet devices a large amount of data is generatedand accumulated into a large amount of data streams overtime Therefore higher requirements are imposed on theprocessing and decomposition of online data In manyapplications a large amount of data storage is generated suchas monitoring the generated video stream data
5G HetNets transport the video streaming traffic bychanging an IoT sensor into an IP Multimedia Subsystem(IMS) terminal IMS realizes robust IMS service functionalityby working closely with Call Session Control Function(CSCF)TheCSCFnode facilitates Session Initiation Protocol(SIP) to operate session setup and teardown [18]
As a popular video streaming protocol CSCF conductssignaling and session management and enables connectioninformation to go across network boundaries
Figure 5 gives an example of IMS based IoT videostreaming over 5G HetNets The improved CSCF serveracts as an intermediate entity that receives and forwardssignaling messages Improved server offers a bunch of cru-cial functions such as authentication authorization androuting
33 Model of Improved CSCF Server By definition the mainmodule of the IMS system is call session control to allowvideovoice communication over the convergence of differentaccess networks It comprises of all the functional modulesrequired to handle all the signaling from end-user to servicesand other networks However conventional CSCF cannot
6 Wireless Communications and Mobile Computing
SIPampCOPSProtocol Stacks
Bandwidth
Negotiation
Traffic
Prediction
Connection
Admission
Control
Original CSCF Server
Improved CSCF Server
SIP Messages
COPS Messages
Enhanced Modules
Figure 6 Model of improved CSCF server
Traffic Load Sampling
Traffic Prediction Algorithm
Real Traffic Load
Predicted Traffic Load
Traffic Load Prediction Module
Figure 7 General framework of IoT traffic prediction
100 meet the requirement of IoT video streaming and bemodernized
Figure 6 shows that an improved CSCF server containsthree more modules than the original Our design has themerit that the improved CSCF server bargains with theoptical core on behalf of all IoT devices of the local HetNetin advance
For this reason the paths required by video streaming arepreconstructed before the stream even starts to reduce theunnecessary signaling delay
4 Traffic Prediction in IoT UsingCompressed Sensing
Traffic prediction in the IoT has been widely studied andapplied in recent years Sampling and prediction representusually two independent parts in the past For example mostconventional approaches keep the sampling rate unchangedwhich may result in low efficiency in IoT video streamingTo overcome this problem this paper develops a novelcompressed sensing based traffic prediction where error offorecast algorithm goes back to control the sampling intervalOur proposed scheme can significantly reduce the samplingoverhead without sacrificing the accuracy
41 Problem Formulation and Architecture Design Let 119861119899minus1119899be the sum of IoT video streaming loads from a local HetNetat time slot [119905119899minus1 119905119899](119899 = 0 1 2 ) Let improved CSCFserver be exactly aware of the value of 119861119899minus1119899 at time 119905119899minus1With this piece of information it will bargain with 5G corefor the amount of 119861119899minus1119899 bandwidth using COPS messagesAt the next time slot [119905119899 119905119899+1] IoT traffic may change to anew value of 119861119899119899+1 and the improved CSCF server will haveto rebargain with 5G core network again
It is nevertheless impossible for the improved CSCFserver to achieve the value of 119861119899minus1119899 before the time of 119905119899Thus at time 119905119899minus1 we can estimate the approximate value119861119899minus1119899 which is denoted by 1198611015840119899minus1119899 If 1198611015840119899minus1119899 lt 119861119899minus1119899 the localHetNet will suffer from the shortage of bandwidth resourceto accommodate all video streaming sessions As a solutionCAC component must reject some of connection requests inorder to guarantee overall quality of service
On the other hand if 1198611015840119899minus1119899 gt 119861119899minus1119899 part of bandwidthresource is simply overbooked and useless In this sense thetraffic prediction is critical to ensure a good performance ofour proposed scheme
Figure 7 presents the general framework of IoT trafficprediction module with the component of traffic load sam-pling and the component of traffic prediction algorithm
Wireless Communications and Mobile Computing 7
We will give more elaboration on these two componentsnext
42 IoT Traffic Sampling An IoT device initiates a videostream session by sending an INVITE message to theimproved CSCF server in a bid to indicate called URL andbandwidth requirement The improved CSCF server savesevery request in the record for IoT traffic sampling usageregardless the request is accepted by CAC or not
In this paper we use 119861119899minus1119899 to represent the averageIoT traffic interval Δ119905119899minus1119899 = 119905119899 minus 119905119899minus1 In practice thesystem achieves 119861119899minus1119899 from the record in the improved CSCFserver Conventional approach conducts traffic samplingwithconstant rate which rules
Δ11990501 = Δ11990512 = sdot sdot sdot = Δ119905119899minus1119899 = (119899 = 0 1 2 ) (1)
Let 119861(119905) be the traffic load function varying with time tand let 119861(119891) be the frequency counterpart of 119861(119905) with theupper bound of 119891max Nyquist theorem regulates the idea thatsampling ratemust be at least twice of119891max to prevent aliasing[19] In IoT video streaming 119891max could be high in somecomplicated scenarios which make constant sampling ratenot applicable at all
43 Compressed Sensing Compressed sensing theory is atheory that has been widely used in recent years and iswidely used in signal processing It is also called compressedsampling or sparse sampling CS wants to sample the signalat a sampling rate much lower than Shannons Law andfind a solution for underdetermined linear systems whilemaintaining the basic information of the signal [20ndash24]
In order to implement the compressed sensing theory thefollowing solution can be used to estimate the sparse vectorx isin C119873 using the observed measurement vector 119910 isin C119872A method based on measurement equations is applied Theformula is as follows
119910 = Φ119909 + 119908 (2)
The measurement matrix is represented by Φ isin C119872times119873119908 isin C119873 represents the unknown vector of measurementnoise and modeling error The reconstruction process canonly occur under ideal conditions ie only when the signalx is S sparse (119878 ≪ 119873 ) That is there can be at most S nonzeroitems During the operation the number of observations issignificantly smaller than the number of variables ie≫ 119872
In practice the signal we are dealing with is not a sparsesignalThe processingmethod in this time is to express Ψ119873119894=1on some basis with the corresponding sparse coefficient 120579119894Processing equation (1) yields
119910 = Φ119909 + 119908 = ΦΨ120579 + 119908 = Θ120579 + 119908 (3)
where 120579 isin 119862119871 is a L dimension sparse vector of coefficientsbased on the basismatrixΨ isin 119862119873times119871Θ = ΦΨ is a119872times119871matrix(119872 ≪ 119871 )
However because fewer measurements are applied thanthe entries in 120579 the resulting solution is uncertain In
order to recover the 119878 sparse signal 1198970 norm of the mostsparse sequence should be found from all feasible solutionsMinimization can be used for the reconstruction processTheformula is expressed as follows
120579 = min 1205790st 1003817100381710038171003817Θ120579 minus 11991010038171003817100381710038172 le Ξ (4)
where 120579 refers to the estimated vector for 120579 and w le Ξis noise tolerance However for practical operations the 1198970normminimization has huge computational complexity anda reconstruction algorithm with lower computational costmust be developed
Studies have shown that the reconstruction process of sig-nal 119909must satisfy a condition that matrix Θ cannot map twodifferent S sparse signals to the same set of samplesThereforethematrixΘmust satisfy the restricted equidistance property(RIP) [25 26]
44 Adaptive Sampling Rate (a)The definition of traditionalcompressive sensing sparse signals can be recovered from asmall number of nonadaptive linearmeasurements Let signal119909 be 119870 sparse in basisdictionary Ψ For example Ψ is DFTmatrix if the signal is frequency domain sparse Ψ can beexpressed as a matrix as shown in Figure 8(a) In Figure 9if signal 119909 is sparse then we can use compressive sensing by119910 = Φ times 119909
Φ is the measurement matrix and the theory of CS statesthat random Φ will work How to understand this Forexample if the signal is sparse in frequency domain then wemake a few random samplings in time domain In contrast ifthe signal is sparse in time domain we make a few randomsamplings in frequency domain
(b) Address the situation that the appearance time ofsparse signal can be predicted In this case Ψ is identitymatrix I as shown in Figure 8(b) Then measure matrix Φis part of I which involves the time points where the sparsesignal appears Φ is not random anymore In practice wepropose variable sampling rate scheme to predict these timepoints (or the measurement matrix)
If a fixed sampling interval applies as mentioned beforethe sampling rate must double the highest frequency of trafficcurve to achieve accurate prediction
Accordingly it is significant to develop an inconstantsampling rate scheme with much lower overhead
Variable sampling rate comes from the observation ofthe traffic curve in Figure 10 with combination of slow andfast changing periods The fluctuation of traffic significantlydepends on the timely feature of IoT devices For examplethe cameras monitor the parking carsrsquo moving and stoppingaccording to peoplersquos behavior which is obviously related tothe number of car arrivals and departures In Figure 8 fastchanging zone represents the period of rapid event varyingsuch as morning peak hour slow changing zone representsthe period of steady event numbers such as at night
With the goal of reducing the sampling overhead astraightforward adaptive rate approach is to utilize low ratein slow changing period and high rate in fast changing
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
4 Wireless Communications and Mobile Computing
Cellular BS
Operator deployed smallcell
User deployed smallcell
Cloud
Macrocell link
Small cell link
Relaying
Gateway
DataCenter
Servers
Figure 3 A typical 5G IoT heterogeneous network
In order to solve this problem 5G wireless networkshave received attention and research in order to provideconnectivity and meet the requirements of different qualityof service (QoS) The key technologies of 5G enable thedeployment of heterogeneous networks (HetNets) to supporta large number of IoT data requirements by deploying a largenumber of small cells
Network density is themost effectivemeans inmanywaysto increase network capacity such as spectrum expansion andspectrum efficiency enhancement [16] By densely deploy-ing small cells indoors and outdoors network capacity isincreased and network latency is reduced Small cells include
microcells picocells and femtocells which are closer to theuser Figure 3 is a schematic diagram of deployment of a smallcell which can be widely deployed indoors and outdoorssuch as in offices intersections and plazas Among themthe data traffic of indoor users can be provided by Wi-FiThe next generation of Wi-Fi 80211ac is expected to growrapidly providing multigigabit transmission ratesThis led tothe concept of HetNets a multilayer network with multipleradio access technologies
A 5G core network could be enhanced by network slicingwhich allows different clients to bridge their private sites overa common provider-owned cloud Using MPLS or another
Wireless Communications and Mobile Computing 5
5G core networks5G HetNet
CE
PEPE
5G HetNet
5G HetNet
CE
CE
PE
PE
Figure 4 Network slicing based 5G core network connecting several local HetNets
CE PE
IMS
Improved CSCF Server
Figure 5 IMS based IoT video streaming over 5G HetNets
similar technology [17] network slicing becomes a noveltrend and has a wonderful capability to offer QoS guaran-teed services with flexibility Figure 4 shows an example ofnetwork slicing based 5G core that connects several localHetNets Using MPLS technology the core network containsthe following types of equipment
(a) customer edge (CE) router which serves to connectwith client local network directly
(b) provider edge (PE) router which serves as the inter-face between 5G HetNet and optical core
(c) provider (P) router which serves to forward the usertraffic arranged by CE and PE routers
32 IMS Based IoT Video Streaming Internet of things (IoT)is regarded recently as the most promising form of thewireless communication environments
In Internet devices a large amount of data is generatedand accumulated into a large amount of data streams overtime Therefore higher requirements are imposed on theprocessing and decomposition of online data In manyapplications a large amount of data storage is generated suchas monitoring the generated video stream data
5G HetNets transport the video streaming traffic bychanging an IoT sensor into an IP Multimedia Subsystem(IMS) terminal IMS realizes robust IMS service functionalityby working closely with Call Session Control Function(CSCF)TheCSCFnode facilitates Session Initiation Protocol(SIP) to operate session setup and teardown [18]
As a popular video streaming protocol CSCF conductssignaling and session management and enables connectioninformation to go across network boundaries
Figure 5 gives an example of IMS based IoT videostreaming over 5G HetNets The improved CSCF serveracts as an intermediate entity that receives and forwardssignaling messages Improved server offers a bunch of cru-cial functions such as authentication authorization androuting
33 Model of Improved CSCF Server By definition the mainmodule of the IMS system is call session control to allowvideovoice communication over the convergence of differentaccess networks It comprises of all the functional modulesrequired to handle all the signaling from end-user to servicesand other networks However conventional CSCF cannot
6 Wireless Communications and Mobile Computing
SIPampCOPSProtocol Stacks
Bandwidth
Negotiation
Traffic
Prediction
Connection
Admission
Control
Original CSCF Server
Improved CSCF Server
SIP Messages
COPS Messages
Enhanced Modules
Figure 6 Model of improved CSCF server
Traffic Load Sampling
Traffic Prediction Algorithm
Real Traffic Load
Predicted Traffic Load
Traffic Load Prediction Module
Figure 7 General framework of IoT traffic prediction
100 meet the requirement of IoT video streaming and bemodernized
Figure 6 shows that an improved CSCF server containsthree more modules than the original Our design has themerit that the improved CSCF server bargains with theoptical core on behalf of all IoT devices of the local HetNetin advance
For this reason the paths required by video streaming arepreconstructed before the stream even starts to reduce theunnecessary signaling delay
4 Traffic Prediction in IoT UsingCompressed Sensing
Traffic prediction in the IoT has been widely studied andapplied in recent years Sampling and prediction representusually two independent parts in the past For example mostconventional approaches keep the sampling rate unchangedwhich may result in low efficiency in IoT video streamingTo overcome this problem this paper develops a novelcompressed sensing based traffic prediction where error offorecast algorithm goes back to control the sampling intervalOur proposed scheme can significantly reduce the samplingoverhead without sacrificing the accuracy
41 Problem Formulation and Architecture Design Let 119861119899minus1119899be the sum of IoT video streaming loads from a local HetNetat time slot [119905119899minus1 119905119899](119899 = 0 1 2 ) Let improved CSCFserver be exactly aware of the value of 119861119899minus1119899 at time 119905119899minus1With this piece of information it will bargain with 5G corefor the amount of 119861119899minus1119899 bandwidth using COPS messagesAt the next time slot [119905119899 119905119899+1] IoT traffic may change to anew value of 119861119899119899+1 and the improved CSCF server will haveto rebargain with 5G core network again
It is nevertheless impossible for the improved CSCFserver to achieve the value of 119861119899minus1119899 before the time of 119905119899Thus at time 119905119899minus1 we can estimate the approximate value119861119899minus1119899 which is denoted by 1198611015840119899minus1119899 If 1198611015840119899minus1119899 lt 119861119899minus1119899 the localHetNet will suffer from the shortage of bandwidth resourceto accommodate all video streaming sessions As a solutionCAC component must reject some of connection requests inorder to guarantee overall quality of service
On the other hand if 1198611015840119899minus1119899 gt 119861119899minus1119899 part of bandwidthresource is simply overbooked and useless In this sense thetraffic prediction is critical to ensure a good performance ofour proposed scheme
Figure 7 presents the general framework of IoT trafficprediction module with the component of traffic load sam-pling and the component of traffic prediction algorithm
Wireless Communications and Mobile Computing 7
We will give more elaboration on these two componentsnext
42 IoT Traffic Sampling An IoT device initiates a videostream session by sending an INVITE message to theimproved CSCF server in a bid to indicate called URL andbandwidth requirement The improved CSCF server savesevery request in the record for IoT traffic sampling usageregardless the request is accepted by CAC or not
In this paper we use 119861119899minus1119899 to represent the averageIoT traffic interval Δ119905119899minus1119899 = 119905119899 minus 119905119899minus1 In practice thesystem achieves 119861119899minus1119899 from the record in the improved CSCFserver Conventional approach conducts traffic samplingwithconstant rate which rules
Δ11990501 = Δ11990512 = sdot sdot sdot = Δ119905119899minus1119899 = (119899 = 0 1 2 ) (1)
Let 119861(119905) be the traffic load function varying with time tand let 119861(119891) be the frequency counterpart of 119861(119905) with theupper bound of 119891max Nyquist theorem regulates the idea thatsampling ratemust be at least twice of119891max to prevent aliasing[19] In IoT video streaming 119891max could be high in somecomplicated scenarios which make constant sampling ratenot applicable at all
43 Compressed Sensing Compressed sensing theory is atheory that has been widely used in recent years and iswidely used in signal processing It is also called compressedsampling or sparse sampling CS wants to sample the signalat a sampling rate much lower than Shannons Law andfind a solution for underdetermined linear systems whilemaintaining the basic information of the signal [20ndash24]
In order to implement the compressed sensing theory thefollowing solution can be used to estimate the sparse vectorx isin C119873 using the observed measurement vector 119910 isin C119872A method based on measurement equations is applied Theformula is as follows
119910 = Φ119909 + 119908 (2)
The measurement matrix is represented by Φ isin C119872times119873119908 isin C119873 represents the unknown vector of measurementnoise and modeling error The reconstruction process canonly occur under ideal conditions ie only when the signalx is S sparse (119878 ≪ 119873 ) That is there can be at most S nonzeroitems During the operation the number of observations issignificantly smaller than the number of variables ie≫ 119872
In practice the signal we are dealing with is not a sparsesignalThe processingmethod in this time is to express Ψ119873119894=1on some basis with the corresponding sparse coefficient 120579119894Processing equation (1) yields
119910 = Φ119909 + 119908 = ΦΨ120579 + 119908 = Θ120579 + 119908 (3)
where 120579 isin 119862119871 is a L dimension sparse vector of coefficientsbased on the basismatrixΨ isin 119862119873times119871Θ = ΦΨ is a119872times119871matrix(119872 ≪ 119871 )
However because fewer measurements are applied thanthe entries in 120579 the resulting solution is uncertain In
order to recover the 119878 sparse signal 1198970 norm of the mostsparse sequence should be found from all feasible solutionsMinimization can be used for the reconstruction processTheformula is expressed as follows
120579 = min 1205790st 1003817100381710038171003817Θ120579 minus 11991010038171003817100381710038172 le Ξ (4)
where 120579 refers to the estimated vector for 120579 and w le Ξis noise tolerance However for practical operations the 1198970normminimization has huge computational complexity anda reconstruction algorithm with lower computational costmust be developed
Studies have shown that the reconstruction process of sig-nal 119909must satisfy a condition that matrix Θ cannot map twodifferent S sparse signals to the same set of samplesThereforethematrixΘmust satisfy the restricted equidistance property(RIP) [25 26]
44 Adaptive Sampling Rate (a)The definition of traditionalcompressive sensing sparse signals can be recovered from asmall number of nonadaptive linearmeasurements Let signal119909 be 119870 sparse in basisdictionary Ψ For example Ψ is DFTmatrix if the signal is frequency domain sparse Ψ can beexpressed as a matrix as shown in Figure 8(a) In Figure 9if signal 119909 is sparse then we can use compressive sensing by119910 = Φ times 119909
Φ is the measurement matrix and the theory of CS statesthat random Φ will work How to understand this Forexample if the signal is sparse in frequency domain then wemake a few random samplings in time domain In contrast ifthe signal is sparse in time domain we make a few randomsamplings in frequency domain
(b) Address the situation that the appearance time ofsparse signal can be predicted In this case Ψ is identitymatrix I as shown in Figure 8(b) Then measure matrix Φis part of I which involves the time points where the sparsesignal appears Φ is not random anymore In practice wepropose variable sampling rate scheme to predict these timepoints (or the measurement matrix)
If a fixed sampling interval applies as mentioned beforethe sampling rate must double the highest frequency of trafficcurve to achieve accurate prediction
Accordingly it is significant to develop an inconstantsampling rate scheme with much lower overhead
Variable sampling rate comes from the observation ofthe traffic curve in Figure 10 with combination of slow andfast changing periods The fluctuation of traffic significantlydepends on the timely feature of IoT devices For examplethe cameras monitor the parking carsrsquo moving and stoppingaccording to peoplersquos behavior which is obviously related tothe number of car arrivals and departures In Figure 8 fastchanging zone represents the period of rapid event varyingsuch as morning peak hour slow changing zone representsthe period of steady event numbers such as at night
With the goal of reducing the sampling overhead astraightforward adaptive rate approach is to utilize low ratein slow changing period and high rate in fast changing
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 5
5G core networks5G HetNet
CE
PEPE
5G HetNet
5G HetNet
CE
CE
PE
PE
Figure 4 Network slicing based 5G core network connecting several local HetNets
CE PE
IMS
Improved CSCF Server
Figure 5 IMS based IoT video streaming over 5G HetNets
similar technology [17] network slicing becomes a noveltrend and has a wonderful capability to offer QoS guaran-teed services with flexibility Figure 4 shows an example ofnetwork slicing based 5G core that connects several localHetNets Using MPLS technology the core network containsthe following types of equipment
(a) customer edge (CE) router which serves to connectwith client local network directly
(b) provider edge (PE) router which serves as the inter-face between 5G HetNet and optical core
(c) provider (P) router which serves to forward the usertraffic arranged by CE and PE routers
32 IMS Based IoT Video Streaming Internet of things (IoT)is regarded recently as the most promising form of thewireless communication environments
In Internet devices a large amount of data is generatedand accumulated into a large amount of data streams overtime Therefore higher requirements are imposed on theprocessing and decomposition of online data In manyapplications a large amount of data storage is generated suchas monitoring the generated video stream data
5G HetNets transport the video streaming traffic bychanging an IoT sensor into an IP Multimedia Subsystem(IMS) terminal IMS realizes robust IMS service functionalityby working closely with Call Session Control Function(CSCF)TheCSCFnode facilitates Session Initiation Protocol(SIP) to operate session setup and teardown [18]
As a popular video streaming protocol CSCF conductssignaling and session management and enables connectioninformation to go across network boundaries
Figure 5 gives an example of IMS based IoT videostreaming over 5G HetNets The improved CSCF serveracts as an intermediate entity that receives and forwardssignaling messages Improved server offers a bunch of cru-cial functions such as authentication authorization androuting
33 Model of Improved CSCF Server By definition the mainmodule of the IMS system is call session control to allowvideovoice communication over the convergence of differentaccess networks It comprises of all the functional modulesrequired to handle all the signaling from end-user to servicesand other networks However conventional CSCF cannot
6 Wireless Communications and Mobile Computing
SIPampCOPSProtocol Stacks
Bandwidth
Negotiation
Traffic
Prediction
Connection
Admission
Control
Original CSCF Server
Improved CSCF Server
SIP Messages
COPS Messages
Enhanced Modules
Figure 6 Model of improved CSCF server
Traffic Load Sampling
Traffic Prediction Algorithm
Real Traffic Load
Predicted Traffic Load
Traffic Load Prediction Module
Figure 7 General framework of IoT traffic prediction
100 meet the requirement of IoT video streaming and bemodernized
Figure 6 shows that an improved CSCF server containsthree more modules than the original Our design has themerit that the improved CSCF server bargains with theoptical core on behalf of all IoT devices of the local HetNetin advance
For this reason the paths required by video streaming arepreconstructed before the stream even starts to reduce theunnecessary signaling delay
4 Traffic Prediction in IoT UsingCompressed Sensing
Traffic prediction in the IoT has been widely studied andapplied in recent years Sampling and prediction representusually two independent parts in the past For example mostconventional approaches keep the sampling rate unchangedwhich may result in low efficiency in IoT video streamingTo overcome this problem this paper develops a novelcompressed sensing based traffic prediction where error offorecast algorithm goes back to control the sampling intervalOur proposed scheme can significantly reduce the samplingoverhead without sacrificing the accuracy
41 Problem Formulation and Architecture Design Let 119861119899minus1119899be the sum of IoT video streaming loads from a local HetNetat time slot [119905119899minus1 119905119899](119899 = 0 1 2 ) Let improved CSCFserver be exactly aware of the value of 119861119899minus1119899 at time 119905119899minus1With this piece of information it will bargain with 5G corefor the amount of 119861119899minus1119899 bandwidth using COPS messagesAt the next time slot [119905119899 119905119899+1] IoT traffic may change to anew value of 119861119899119899+1 and the improved CSCF server will haveto rebargain with 5G core network again
It is nevertheless impossible for the improved CSCFserver to achieve the value of 119861119899minus1119899 before the time of 119905119899Thus at time 119905119899minus1 we can estimate the approximate value119861119899minus1119899 which is denoted by 1198611015840119899minus1119899 If 1198611015840119899minus1119899 lt 119861119899minus1119899 the localHetNet will suffer from the shortage of bandwidth resourceto accommodate all video streaming sessions As a solutionCAC component must reject some of connection requests inorder to guarantee overall quality of service
On the other hand if 1198611015840119899minus1119899 gt 119861119899minus1119899 part of bandwidthresource is simply overbooked and useless In this sense thetraffic prediction is critical to ensure a good performance ofour proposed scheme
Figure 7 presents the general framework of IoT trafficprediction module with the component of traffic load sam-pling and the component of traffic prediction algorithm
Wireless Communications and Mobile Computing 7
We will give more elaboration on these two componentsnext
42 IoT Traffic Sampling An IoT device initiates a videostream session by sending an INVITE message to theimproved CSCF server in a bid to indicate called URL andbandwidth requirement The improved CSCF server savesevery request in the record for IoT traffic sampling usageregardless the request is accepted by CAC or not
In this paper we use 119861119899minus1119899 to represent the averageIoT traffic interval Δ119905119899minus1119899 = 119905119899 minus 119905119899minus1 In practice thesystem achieves 119861119899minus1119899 from the record in the improved CSCFserver Conventional approach conducts traffic samplingwithconstant rate which rules
Δ11990501 = Δ11990512 = sdot sdot sdot = Δ119905119899minus1119899 = (119899 = 0 1 2 ) (1)
Let 119861(119905) be the traffic load function varying with time tand let 119861(119891) be the frequency counterpart of 119861(119905) with theupper bound of 119891max Nyquist theorem regulates the idea thatsampling ratemust be at least twice of119891max to prevent aliasing[19] In IoT video streaming 119891max could be high in somecomplicated scenarios which make constant sampling ratenot applicable at all
43 Compressed Sensing Compressed sensing theory is atheory that has been widely used in recent years and iswidely used in signal processing It is also called compressedsampling or sparse sampling CS wants to sample the signalat a sampling rate much lower than Shannons Law andfind a solution for underdetermined linear systems whilemaintaining the basic information of the signal [20ndash24]
In order to implement the compressed sensing theory thefollowing solution can be used to estimate the sparse vectorx isin C119873 using the observed measurement vector 119910 isin C119872A method based on measurement equations is applied Theformula is as follows
119910 = Φ119909 + 119908 (2)
The measurement matrix is represented by Φ isin C119872times119873119908 isin C119873 represents the unknown vector of measurementnoise and modeling error The reconstruction process canonly occur under ideal conditions ie only when the signalx is S sparse (119878 ≪ 119873 ) That is there can be at most S nonzeroitems During the operation the number of observations issignificantly smaller than the number of variables ie≫ 119872
In practice the signal we are dealing with is not a sparsesignalThe processingmethod in this time is to express Ψ119873119894=1on some basis with the corresponding sparse coefficient 120579119894Processing equation (1) yields
119910 = Φ119909 + 119908 = ΦΨ120579 + 119908 = Θ120579 + 119908 (3)
where 120579 isin 119862119871 is a L dimension sparse vector of coefficientsbased on the basismatrixΨ isin 119862119873times119871Θ = ΦΨ is a119872times119871matrix(119872 ≪ 119871 )
However because fewer measurements are applied thanthe entries in 120579 the resulting solution is uncertain In
order to recover the 119878 sparse signal 1198970 norm of the mostsparse sequence should be found from all feasible solutionsMinimization can be used for the reconstruction processTheformula is expressed as follows
120579 = min 1205790st 1003817100381710038171003817Θ120579 minus 11991010038171003817100381710038172 le Ξ (4)
where 120579 refers to the estimated vector for 120579 and w le Ξis noise tolerance However for practical operations the 1198970normminimization has huge computational complexity anda reconstruction algorithm with lower computational costmust be developed
Studies have shown that the reconstruction process of sig-nal 119909must satisfy a condition that matrix Θ cannot map twodifferent S sparse signals to the same set of samplesThereforethematrixΘmust satisfy the restricted equidistance property(RIP) [25 26]
44 Adaptive Sampling Rate (a)The definition of traditionalcompressive sensing sparse signals can be recovered from asmall number of nonadaptive linearmeasurements Let signal119909 be 119870 sparse in basisdictionary Ψ For example Ψ is DFTmatrix if the signal is frequency domain sparse Ψ can beexpressed as a matrix as shown in Figure 8(a) In Figure 9if signal 119909 is sparse then we can use compressive sensing by119910 = Φ times 119909
Φ is the measurement matrix and the theory of CS statesthat random Φ will work How to understand this Forexample if the signal is sparse in frequency domain then wemake a few random samplings in time domain In contrast ifthe signal is sparse in time domain we make a few randomsamplings in frequency domain
(b) Address the situation that the appearance time ofsparse signal can be predicted In this case Ψ is identitymatrix I as shown in Figure 8(b) Then measure matrix Φis part of I which involves the time points where the sparsesignal appears Φ is not random anymore In practice wepropose variable sampling rate scheme to predict these timepoints (or the measurement matrix)
If a fixed sampling interval applies as mentioned beforethe sampling rate must double the highest frequency of trafficcurve to achieve accurate prediction
Accordingly it is significant to develop an inconstantsampling rate scheme with much lower overhead
Variable sampling rate comes from the observation ofthe traffic curve in Figure 10 with combination of slow andfast changing periods The fluctuation of traffic significantlydepends on the timely feature of IoT devices For examplethe cameras monitor the parking carsrsquo moving and stoppingaccording to peoplersquos behavior which is obviously related tothe number of car arrivals and departures In Figure 8 fastchanging zone represents the period of rapid event varyingsuch as morning peak hour slow changing zone representsthe period of steady event numbers such as at night
With the goal of reducing the sampling overhead astraightforward adaptive rate approach is to utilize low ratein slow changing period and high rate in fast changing
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
6 Wireless Communications and Mobile Computing
SIPampCOPSProtocol Stacks
Bandwidth
Negotiation
Traffic
Prediction
Connection
Admission
Control
Original CSCF Server
Improved CSCF Server
SIP Messages
COPS Messages
Enhanced Modules
Figure 6 Model of improved CSCF server
Traffic Load Sampling
Traffic Prediction Algorithm
Real Traffic Load
Predicted Traffic Load
Traffic Load Prediction Module
Figure 7 General framework of IoT traffic prediction
100 meet the requirement of IoT video streaming and bemodernized
Figure 6 shows that an improved CSCF server containsthree more modules than the original Our design has themerit that the improved CSCF server bargains with theoptical core on behalf of all IoT devices of the local HetNetin advance
For this reason the paths required by video streaming arepreconstructed before the stream even starts to reduce theunnecessary signaling delay
4 Traffic Prediction in IoT UsingCompressed Sensing
Traffic prediction in the IoT has been widely studied andapplied in recent years Sampling and prediction representusually two independent parts in the past For example mostconventional approaches keep the sampling rate unchangedwhich may result in low efficiency in IoT video streamingTo overcome this problem this paper develops a novelcompressed sensing based traffic prediction where error offorecast algorithm goes back to control the sampling intervalOur proposed scheme can significantly reduce the samplingoverhead without sacrificing the accuracy
41 Problem Formulation and Architecture Design Let 119861119899minus1119899be the sum of IoT video streaming loads from a local HetNetat time slot [119905119899minus1 119905119899](119899 = 0 1 2 ) Let improved CSCFserver be exactly aware of the value of 119861119899minus1119899 at time 119905119899minus1With this piece of information it will bargain with 5G corefor the amount of 119861119899minus1119899 bandwidth using COPS messagesAt the next time slot [119905119899 119905119899+1] IoT traffic may change to anew value of 119861119899119899+1 and the improved CSCF server will haveto rebargain with 5G core network again
It is nevertheless impossible for the improved CSCFserver to achieve the value of 119861119899minus1119899 before the time of 119905119899Thus at time 119905119899minus1 we can estimate the approximate value119861119899minus1119899 which is denoted by 1198611015840119899minus1119899 If 1198611015840119899minus1119899 lt 119861119899minus1119899 the localHetNet will suffer from the shortage of bandwidth resourceto accommodate all video streaming sessions As a solutionCAC component must reject some of connection requests inorder to guarantee overall quality of service
On the other hand if 1198611015840119899minus1119899 gt 119861119899minus1119899 part of bandwidthresource is simply overbooked and useless In this sense thetraffic prediction is critical to ensure a good performance ofour proposed scheme
Figure 7 presents the general framework of IoT trafficprediction module with the component of traffic load sam-pling and the component of traffic prediction algorithm
Wireless Communications and Mobile Computing 7
We will give more elaboration on these two componentsnext
42 IoT Traffic Sampling An IoT device initiates a videostream session by sending an INVITE message to theimproved CSCF server in a bid to indicate called URL andbandwidth requirement The improved CSCF server savesevery request in the record for IoT traffic sampling usageregardless the request is accepted by CAC or not
In this paper we use 119861119899minus1119899 to represent the averageIoT traffic interval Δ119905119899minus1119899 = 119905119899 minus 119905119899minus1 In practice thesystem achieves 119861119899minus1119899 from the record in the improved CSCFserver Conventional approach conducts traffic samplingwithconstant rate which rules
Δ11990501 = Δ11990512 = sdot sdot sdot = Δ119905119899minus1119899 = (119899 = 0 1 2 ) (1)
Let 119861(119905) be the traffic load function varying with time tand let 119861(119891) be the frequency counterpart of 119861(119905) with theupper bound of 119891max Nyquist theorem regulates the idea thatsampling ratemust be at least twice of119891max to prevent aliasing[19] In IoT video streaming 119891max could be high in somecomplicated scenarios which make constant sampling ratenot applicable at all
43 Compressed Sensing Compressed sensing theory is atheory that has been widely used in recent years and iswidely used in signal processing It is also called compressedsampling or sparse sampling CS wants to sample the signalat a sampling rate much lower than Shannons Law andfind a solution for underdetermined linear systems whilemaintaining the basic information of the signal [20ndash24]
In order to implement the compressed sensing theory thefollowing solution can be used to estimate the sparse vectorx isin C119873 using the observed measurement vector 119910 isin C119872A method based on measurement equations is applied Theformula is as follows
119910 = Φ119909 + 119908 (2)
The measurement matrix is represented by Φ isin C119872times119873119908 isin C119873 represents the unknown vector of measurementnoise and modeling error The reconstruction process canonly occur under ideal conditions ie only when the signalx is S sparse (119878 ≪ 119873 ) That is there can be at most S nonzeroitems During the operation the number of observations issignificantly smaller than the number of variables ie≫ 119872
In practice the signal we are dealing with is not a sparsesignalThe processingmethod in this time is to express Ψ119873119894=1on some basis with the corresponding sparse coefficient 120579119894Processing equation (1) yields
119910 = Φ119909 + 119908 = ΦΨ120579 + 119908 = Θ120579 + 119908 (3)
where 120579 isin 119862119871 is a L dimension sparse vector of coefficientsbased on the basismatrixΨ isin 119862119873times119871Θ = ΦΨ is a119872times119871matrix(119872 ≪ 119871 )
However because fewer measurements are applied thanthe entries in 120579 the resulting solution is uncertain In
order to recover the 119878 sparse signal 1198970 norm of the mostsparse sequence should be found from all feasible solutionsMinimization can be used for the reconstruction processTheformula is expressed as follows
120579 = min 1205790st 1003817100381710038171003817Θ120579 minus 11991010038171003817100381710038172 le Ξ (4)
where 120579 refers to the estimated vector for 120579 and w le Ξis noise tolerance However for practical operations the 1198970normminimization has huge computational complexity anda reconstruction algorithm with lower computational costmust be developed
Studies have shown that the reconstruction process of sig-nal 119909must satisfy a condition that matrix Θ cannot map twodifferent S sparse signals to the same set of samplesThereforethematrixΘmust satisfy the restricted equidistance property(RIP) [25 26]
44 Adaptive Sampling Rate (a)The definition of traditionalcompressive sensing sparse signals can be recovered from asmall number of nonadaptive linearmeasurements Let signal119909 be 119870 sparse in basisdictionary Ψ For example Ψ is DFTmatrix if the signal is frequency domain sparse Ψ can beexpressed as a matrix as shown in Figure 8(a) In Figure 9if signal 119909 is sparse then we can use compressive sensing by119910 = Φ times 119909
Φ is the measurement matrix and the theory of CS statesthat random Φ will work How to understand this Forexample if the signal is sparse in frequency domain then wemake a few random samplings in time domain In contrast ifthe signal is sparse in time domain we make a few randomsamplings in frequency domain
(b) Address the situation that the appearance time ofsparse signal can be predicted In this case Ψ is identitymatrix I as shown in Figure 8(b) Then measure matrix Φis part of I which involves the time points where the sparsesignal appears Φ is not random anymore In practice wepropose variable sampling rate scheme to predict these timepoints (or the measurement matrix)
If a fixed sampling interval applies as mentioned beforethe sampling rate must double the highest frequency of trafficcurve to achieve accurate prediction
Accordingly it is significant to develop an inconstantsampling rate scheme with much lower overhead
Variable sampling rate comes from the observation ofthe traffic curve in Figure 10 with combination of slow andfast changing periods The fluctuation of traffic significantlydepends on the timely feature of IoT devices For examplethe cameras monitor the parking carsrsquo moving and stoppingaccording to peoplersquos behavior which is obviously related tothe number of car arrivals and departures In Figure 8 fastchanging zone represents the period of rapid event varyingsuch as morning peak hour slow changing zone representsthe period of steady event numbers such as at night
With the goal of reducing the sampling overhead astraightforward adaptive rate approach is to utilize low ratein slow changing period and high rate in fast changing
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 7
We will give more elaboration on these two componentsnext
42 IoT Traffic Sampling An IoT device initiates a videostream session by sending an INVITE message to theimproved CSCF server in a bid to indicate called URL andbandwidth requirement The improved CSCF server savesevery request in the record for IoT traffic sampling usageregardless the request is accepted by CAC or not
In this paper we use 119861119899minus1119899 to represent the averageIoT traffic interval Δ119905119899minus1119899 = 119905119899 minus 119905119899minus1 In practice thesystem achieves 119861119899minus1119899 from the record in the improved CSCFserver Conventional approach conducts traffic samplingwithconstant rate which rules
Δ11990501 = Δ11990512 = sdot sdot sdot = Δ119905119899minus1119899 = (119899 = 0 1 2 ) (1)
Let 119861(119905) be the traffic load function varying with time tand let 119861(119891) be the frequency counterpart of 119861(119905) with theupper bound of 119891max Nyquist theorem regulates the idea thatsampling ratemust be at least twice of119891max to prevent aliasing[19] In IoT video streaming 119891max could be high in somecomplicated scenarios which make constant sampling ratenot applicable at all
43 Compressed Sensing Compressed sensing theory is atheory that has been widely used in recent years and iswidely used in signal processing It is also called compressedsampling or sparse sampling CS wants to sample the signalat a sampling rate much lower than Shannons Law andfind a solution for underdetermined linear systems whilemaintaining the basic information of the signal [20ndash24]
In order to implement the compressed sensing theory thefollowing solution can be used to estimate the sparse vectorx isin C119873 using the observed measurement vector 119910 isin C119872A method based on measurement equations is applied Theformula is as follows
119910 = Φ119909 + 119908 (2)
The measurement matrix is represented by Φ isin C119872times119873119908 isin C119873 represents the unknown vector of measurementnoise and modeling error The reconstruction process canonly occur under ideal conditions ie only when the signalx is S sparse (119878 ≪ 119873 ) That is there can be at most S nonzeroitems During the operation the number of observations issignificantly smaller than the number of variables ie≫ 119872
In practice the signal we are dealing with is not a sparsesignalThe processingmethod in this time is to express Ψ119873119894=1on some basis with the corresponding sparse coefficient 120579119894Processing equation (1) yields
119910 = Φ119909 + 119908 = ΦΨ120579 + 119908 = Θ120579 + 119908 (3)
where 120579 isin 119862119871 is a L dimension sparse vector of coefficientsbased on the basismatrixΨ isin 119862119873times119871Θ = ΦΨ is a119872times119871matrix(119872 ≪ 119871 )
However because fewer measurements are applied thanthe entries in 120579 the resulting solution is uncertain In
order to recover the 119878 sparse signal 1198970 norm of the mostsparse sequence should be found from all feasible solutionsMinimization can be used for the reconstruction processTheformula is expressed as follows
120579 = min 1205790st 1003817100381710038171003817Θ120579 minus 11991010038171003817100381710038172 le Ξ (4)
where 120579 refers to the estimated vector for 120579 and w le Ξis noise tolerance However for practical operations the 1198970normminimization has huge computational complexity anda reconstruction algorithm with lower computational costmust be developed
Studies have shown that the reconstruction process of sig-nal 119909must satisfy a condition that matrix Θ cannot map twodifferent S sparse signals to the same set of samplesThereforethematrixΘmust satisfy the restricted equidistance property(RIP) [25 26]
44 Adaptive Sampling Rate (a)The definition of traditionalcompressive sensing sparse signals can be recovered from asmall number of nonadaptive linearmeasurements Let signal119909 be 119870 sparse in basisdictionary Ψ For example Ψ is DFTmatrix if the signal is frequency domain sparse Ψ can beexpressed as a matrix as shown in Figure 8(a) In Figure 9if signal 119909 is sparse then we can use compressive sensing by119910 = Φ times 119909
Φ is the measurement matrix and the theory of CS statesthat random Φ will work How to understand this Forexample if the signal is sparse in frequency domain then wemake a few random samplings in time domain In contrast ifthe signal is sparse in time domain we make a few randomsamplings in frequency domain
(b) Address the situation that the appearance time ofsparse signal can be predicted In this case Ψ is identitymatrix I as shown in Figure 8(b) Then measure matrix Φis part of I which involves the time points where the sparsesignal appears Φ is not random anymore In practice wepropose variable sampling rate scheme to predict these timepoints (or the measurement matrix)
If a fixed sampling interval applies as mentioned beforethe sampling rate must double the highest frequency of trafficcurve to achieve accurate prediction
Accordingly it is significant to develop an inconstantsampling rate scheme with much lower overhead
Variable sampling rate comes from the observation ofthe traffic curve in Figure 10 with combination of slow andfast changing periods The fluctuation of traffic significantlydepends on the timely feature of IoT devices For examplethe cameras monitor the parking carsrsquo moving and stoppingaccording to peoplersquos behavior which is obviously related tothe number of car arrivals and departures In Figure 8 fastchanging zone represents the period of rapid event varyingsuch as morning peak hour slow changing zone representsthe period of steady event numbers such as at night
With the goal of reducing the sampling overhead astraightforward adaptive rate approach is to utilize low ratein slow changing period and high rate in fast changing
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
8 Wireless Communications and Mobile Computing
(a) DFT matrix (b) Identity matrix I
Figure 8 The matrix involved in the compressive sensing
Mtimes 1
measurements
y
=
Mtimes N
K lt M ≪ N
xΦ
Ntimes 1
sparsesignal
K
nonzeroentries
Figure 9 The process of compressive sensing
Fast Changing ZoneFast Changing Zone
Traffic Load
Slow ChangingZone
0
500
1000
1500
2000
2500
Traffi
c Loa
d (M
bps)
2 4 6 8 10 12 14 16 18 20 22 240Time (Hours)
Figure 10 IoT traffic load curve of a parking monitor camera
period The adaptive sampling interval aims to sampleand reconstruct the traffic curve efficiently and effectivelywhile keeping the same prediction accuracy as constant rateapproach does
Next we study a typical scenario as follows Letrsquos dividean IoT traffic curve into a number of slow changing and fastchanging periods by using 119891119904max and 119891119891max to represent themaximum frequencies respectively Nyquist sampling theorystates that the sampling rate must be greater than 2119891119904max incase of slow changing zone and 2119891119891max in case of fast changingAfter that we can achieve the average rate of sampling 119877avg
119881119878119877by
119877119886V119892119881119878119877 =2119891119904max119879119904 + 2119891119891max119879119891
119879119904 + 119879119891 (5)
where 119879s and 119879119891 are time of slow and fast changing Since119891119904max lt 119891119891max lt 119891max we have 119877avg
119881119878119877 lt 2119891max implyingthat ASR approach has much less sampling overhead than itsconstant counterpart
45 Traditional Linear Predictor The traditional predictorused most often in practice is the least mean square (LMS)linear filter A k-step LPmakes a guess of the value of 119909(119899+119896)by a linear sum of the current and past x(n) Mathematicallythe pth-order k-step is given by
119909 (119899 + 119896) =119901minus1
sum119894=0
119908119894119909 (119899 minus 119894) = 119908119879119909 (119899) (6)
where 119909(119899 + 119896) is the estimated value of 119909(119899 + 119896) 119908 =[1199080 1199081 119908119901minus1]119879 is the weights and 119909(119899) = [119909(119899) 119909(119899 minus1) 119909(119899 minus 119901 + 1)]119879 is priori knowledge We further definethe prediction error as
119890 (119899) = 119909 (119899 + 119896) minus = 119909 (119899 + 119896) minus 119908119879119909 (119899) (7)Then LP updates 119908(119899) using the following recursive
equation
w (119899 + 1) = w (119899) + 120583119890 (119899) x (119899)x (119899)2 (8)
where 120583 is the step size that may keep constant (constant stepsize (CSS)) or adaptive (adaptive step size (ASS))
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 9
ASR Sampler w(n)x(n)
Linear Predictor
e(n)
Traffic Curve
+
-
x(n+k)
x(n+k)
Figure 11 ASR-LP working with variable sampling rate
46 Adaptive Sampling Rate Linear Predictor Realisticallythe traffic curve is unknown at the time of prediction andthus it is a mission impossible to chop it into slow andfast changing periods We therefore develop an adaptiveprediction scheme ASR-LP as shown in Figure 11
ASR-LP rules the sampling rate at time 119905119899 as 119877119904(119899) =1Δ119905119899 minus 1 and then update it in integration by
1198771199041015840 (119899 + 1) = 119877119904 (119899) [119902 + (1 minus 119902) |119890 (119899)|119864119887 ] (9)
with 0 lt 119902 lt 1 119864119887 gt 0 and
119877119904 (119899 + 1) =
119877max119904 119894119891 1198771199041015840 (119899 + 1) gt 119877max
119904
119877min119904 119894119891 1198771199041015840 (119899 + 1) lt 119877min
119904
1198771199041015840 (119899 + 1) 119900119905ℎ119890119903119908119894119904119890(10)
where 0 lt 119902 lt 1 is the remembering factor Eb is the targetedprediction error bound and 119877smax and 119877smax are the upperand lower bounds of sampling rate
The scheme of ASR-LP optimizes sampling interval bythe following principal The larger the prediction error isthe sharper the traffic curve changes As a result we mustincrease the sampling rate for satisfying prediction accuracyIn contrast the smaller the prediction error is the lowerthe sampling rate goes Particularly network operator candetermine the value of targeted prediction error bound Eb If|119890(119899)| gt 119864119887 the sampling rate should be increased if |119890(119899)| lt119864119887 the sampling rate should be decreased if |119890(119899)| = 119864119887 thesampling rate should be kept unchanged
In addition to Eb there are still three other parametersin (9) and (10) ie 119877smax is the upper bound of samplingrate which can lead to lower prediction error but largercomputational complexity as the value rises 119877smin is thelower bound of sampling rate which represents the bottomline responding time of a linear predictor Last but not leastq decides how much the current sampling rate is relying onits previous value rather than the prediction error We haveto leverage the value of remembering factor 119902 for the optimalperformance of predictor with 119902 = 70 as a typical case
In our research we have conducted simulation studieson a variety of traffic curves to evaluate the performance
Traffi
c Loa
d (M
bps)
0
500
1000
1500
2000
2500
Traffic Load
10 1505 20 25 30 35 4000
Time (Hours)
Figure 12 IoT Traffic load of four hours for parking monitorcamera
of our proposed ASR-LP scheme Due to the limit of spacewe demonstrate only one typical curve in Figure 10 as anexample However the simulation results from this exampleare also largely true to the general cases
Figure 10 is collected from Melbourne on street carparking sensor data [26] which reflects the typical timingfeature of IoT users It is worth noting that the curve of IoTtraffic is dramatically different from previously investigatedmobility models in existing literature such as Brownianmotion model
We use Figure 12 as prototype traffic curve to comparethe performance of three schemes as shown in Figure 12 Inaddition to its advantage in the running time VSR-NLMSdoes have another important virtue ie a lower averagesampling rate compared to FSS-NLMS and VSS-NLMS Itshows that with the same simulation configuration as shownin Figure 12 VSR-NLMS achieves an average sampling rateof 1371 Hz which is lower than 1420 Hz for FSS-NLMS andVSS-NLMS Clearly VSR-NLMS has the lowest sampling rateand thus the least sampling overhead
Figure 13 reveals thatASR-LP can basically tieASS-LP butfar outperform CSS-LP regarding prediction accuracy
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
10 Wireless Communications and Mobile Computing
40
35
30
25
20
15
10
5
000 05 10 15 20 25 30 35 40
Time (Hours)
Erro
r (M
bps)
CSSASRASS
Figure 13 Errors of different predicting methods in four hours
119860119878119877 minus 119871119875 119864119887 = 100119872119887119901119904 119877smax = 1 = 300119867z 119877smin =1800119867z 119877s(0) = 1420119867z and 119902 = 70
119862119878119878 minus 119871119875 119901 = 4 120583 = 01 119860119878119878 minus 119871119875 120583max = 025 120583min =005
Low overhead of average sampling is a significant advan-tage of ASR-LP compared to its counterpart Intensivesimulation study reveals that ASR-LP can save at least halfsampling rate over constant sampling rate linear predictorswith the same accuracy
5 Conclusions
This paper studies the deployment of IoT video streams on5G HetNets and proposes an improved CSCF server solutionto promote IoT traffic prediction and resource reservationWe further designed a linear predictor based on compressedsensing to capture traffic patterns greatly reducing samplingoverhead and computational complexity Intensive analysisand simulation results show that the proposed scheme caneffectively improve performance and save time and cost
Data Availability
The data used to support the findings of this study areincluded within the article
Conflicts of Interest
The authors declare that they have no conflicts of interest
References
[1] M Li Y Du X Ma and S Huang ldquoA 3-param networkselection algorithm in heterogeneous wireless networksrdquo inProceedings of the IEEE 9th International Conference onCommu-nication Software and Networks (ICCSN rsquo17) pp 392ndash396 IEEEGuangzhou China May 2017
[2] G Ma Y Yang X Qiu Z Gao and H Li ldquoFault-toleranttopology control for heterogeneous wireless sensor networks
using Multi-Routing Treerdquo in Proceedings of the 15th IFIPIEEEInternational Symposium on Integrated Network and ServiceManagement (IM rsquo17) pp 620ndash623 Lisbon Portugal May 2017
[3] S Han Y Huang W Meng C Li N Xu and D ChenldquoOptimal power allocation for SCMA downlink systems basedon maximum capacityrdquo IEEE Transactions on Communicationsvol 67 no 2 pp 1480ndash1489 2019
[4] A Montazerolghaem M H Moghaddam and A Leon-GarcialdquoOpenSIP toward software-defined SIP networkingrdquo IEEETransactions on Network and Service Management vol 15 no1 pp 184ndash199 2018
[5] A K Vimal S Pandit A K Godiyal S Anand S Luthraand D Joshi ldquoAn instrumented flexible insole for wireless COPmonitoringrdquo in Proceedings of the 8th International Conferenceon Computing Communications and Networking Technologies(ICCCNT rsquo17) pp 1ndash5 Delhi India July 2017
[6] A Mirrashid and A A Beheshti ldquoCompressed remote sensingby using deep learningrdquo in Proceedings of the 9th InternationalSymposium on Telecommunications (IST rsquo18) pp 549ndash552 IEEETehran Iran December 2018
[7] S RouabahM Ouarzeddine and B Souissi ldquoSAR images com-pressed sensing based on recovery algorithmsrdquo in Proceedingsof the 2018 IEEE International Geoscience and Remote SensingSymposium (IGARSS rsquo18) pp 8897ndash8900 IEEE Valencia SpainJuly 2018
[8] S Han Y Zhang W Meng and H Chen ldquoSelf-interference-cancelation-based SLNRprecoding design for full-duplex relay-assisted systemrdquo IEEE Transactions on Vehicular Technologyvol 67 no 9 pp 8249ndash8262 2018
[9] S Han S Xu W Meng and C Li ldquoDense-device-enabledcooperative networks for efficient and secure transmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018
[10] D Ghadiyaram J Pan and A C Bovik ldquoA subjective andobjective study of stalling events in mobile streaming videosrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 29 no 1 pp 183ndash197 2019
[11] Z Zheng L Pan and K Pholsena ldquoMode decomposition basedhybrid model for traffic flow predictionrdquo in Proceedings of the3rd IEEE International Conference onData Science inCyberspace(DSC rsquo18) pp 521ndash526 IEEE Guangzhou China June 2018
[12] S Fan and H Zhao ldquoDelay-based cross-layer QoS scheme forvideo streaming in wireless ad hoc networksrdquo China Communi-cations vol 15 no 9 pp 215ndash234 2018
[13] T Zhang L Feng P Yu S Guo W Li and X Qiu ldquoAhandover statistics based approach for cell outage detection inself-organized heterogeneous networksrdquo in Proceedings of the15th IFIPIEEE International Symposium on Integrated Networkand Service Management (IM rsquo17) pp 628ndash631 IEEE LisbonPortugal May 2017
[14] S Sun M Kadoch L Gong and B Rong ldquoIntegrating net-work function virtualization with SDR and SDN for 4G5Gnetworksrdquo IEEE Network vol 29 no 3 pp 54ndash59 2015
[15] M Lu Z Qu Z Wang and Z Zhang ldquoHete MESE multi-dimensional community detection algorithm based on multi-plex network extraction and seed expansion for heterogeneousinformation networksrdquo IEEE Access vol 6 pp 73965ndash739832018
[16] Z Zhang L Song ZHan andW Saad ldquoCoalitional gameswithoverlapping coalitions for interference management in smallcell networksrdquo IEEE Transactions on Wireless Communicationsvol 13 no 5 pp 2659ndash2669 2014
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 11
[17] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoCoop-erative content caching in 5G networks with mobile edgecomputingrdquo IEEE Wireless Communications Magazine vol 25no 3 pp 80ndash87 2018
[18] J Rosenberg H Schulzrinne G Camarillo et al ldquoSIP sessioninitiation protocolrdquo RFC 3261 IETF June 2002
[19] K Zhang S Leng Y He S Maharjan and Y Zhang ldquoMobileedge computing and networking for green and low-latencyinternet of thingsrdquo IEEE CommunicationsMagazine vol 56 no5 pp 39ndash45 2018
[20] G Shrividya and S H Bharathi ldquoA study of optimum samplingpattern for reconstruction of MR images using compressivesensingrdquo in Proceedings of the Second International Conferenceon Advances in Electronics Computers and Communications(ICAECC rsquo18) pp 1ndash6 Bangalore India Feburary 2018
[21] Y Wu B Rong K Salehian and G Gagnon ldquoCloud transmis-sion a new spectrum-reuse friendly digital terrestrial broad-casting transmission systemrdquo IEEE Transactions on Broadcast-ing vol 58 no 3 pp 329ndash337 2012
[22] A Sammoud A Kumar M Bayoumi and T Elarabi ldquoReal-time streaming challenges in Internet of VideoThings (IoVT)rdquoin Proceedings of the 50th IEEE International Symposium onCircuits and Systems (ISCAS rsquo17) pp 1ndash4 Baltimore Md USAMay 2017
[23] N Eswara KManasa A Kommineni et al ldquoA continuous QoEevaluation framework for video streaming over HTTPrdquo IEEETransactions on Circuits and Systems for Video Technology vol28 no 11 pp 3236ndash3250 2018
[24] B Rong Y Qian K Lu H-H Chen and M Guizani ldquoCalladmission control optimization in WiMAX networksrdquo IEEETransactions on Vehicular Technology vol 57 no 4 pp 2509ndash2522 2008
[25] I D Irawati A B Suksmono and I Y M Edward ldquoLocalinterpolated compressive sampling for internet traffic recon-structionrdquo in Proceedings of the International Conference onAdvanced Computing and Applications (ACOMP rsquo17) pp 93ndash98Ho Chi Minh City Vietnam December 2017
[26] N Anselmi L Poli G Oliveri and A Massa ldquoThree dimen-sional imaging with the contrast source compressive samplingrdquoin Proceedings of the International Applied Computational Elec-tromagnetics Society Symposium - China (ACES rsquo18) pp 1-2IEEE Beijing China July 2018
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom