increasing lpwan scalability by means of concurrent

21
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000. Digital Object Identifier 10.1109/ACCESS.2017.DOI Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case TONI ADAME 1 , ALBERT BEL 2 , AND BORIS BELLALTA 3 1 Network Technologies and Strategies research group, Universitat Pompeu Fabra, 08018 Barcelona, Spain (e-mail: [email protected]) 2 Network Technologies and Strategies research group, Universitat Pompeu Fabra, 08018 Barcelona, Spain (e-mail: [email protected]) 3 Wireless Networking research group, Universitat Pompeu Fabra, 08018 Barcelona, Spain (e-mail: [email protected]) Corresponding author: Toni Adame (e-mail: [email protected]). This work has been partially supported by the Spanish Ministry of Economy and Competitiveness under the Maria de Maeztu Units of Excellence Programme (MDM-2015-0502) and the Open Call of INTER-IoT, Research and Innovation action - Horizon 2020 European Project, Grant Agreement #687283, financed by the European Union. It also received funding from the Catalan government through projects SGR-2017-1188 and SGR-2017-1739, and from the Spanish government under the project TEC2016-79510-P. The authors also acknowledge the technical support received by Josep Pujol from Universitat Pompeu Fabra, Universitat Politècnica de Valencia, Fundacion Valenciaport, and Friopuerto Valencia S.L. ABSTRACT One of the most important challenges of the Internet of Things (IoT) in the next years will be the smooth incorporation of millions of smart devices into its communications paradigm. The greater coverage area of sub-1GHz low-power wide area networks (LPWANs) makes them a suitable technology to easily encompass hundreds of end devices under a single base station. However, LPWAN inherent simplicity affects negatively on scalability, as these networks are not flexible enough to deal with a high number of nodes unless their traffic load is really low, which limits many potential use-cases. This paper analyzes the scalability issue in LPWAN and proposes the INTER-HARE protocol: a solution based on the use of concurrent multiband IoT technologies, where an 868 MHz LPWAN acts as transparent backhaul for a set of subnetworks working at 2.4 GHz. The implementation of the INTER-HARE protocol on a real IoT platform was assessed both in several laboratory testbeds and in a pilot developed in the premises of an industrial company, proving its suitability in non-delay sensitive monitoring applications with end devices scattered throughout the targeted area. INDEX TERMS LPWAN, scalability, Internet of Things (IoT), multiband, wireless sensor networks (WSN), power consumption, HARE, INTER-HARE, Industry 4.0. I. INTRODUCTION T HE number of connected Internet of Things (IoT) de- vices worldwide is exponentially growing in the last years, with up to 25 billion devices expected to be part of the IoT ecosystem by the year 2020 [1]. The integration of this huge amount of devices into existing and upcoming technological infrastructure was already considered in the past as a vital challenge for the consolidation of the digital industrial economy, also known as Industry 4.0 [2]. After being identified for a long time as one of the most far-reaching trends in IoT [3], low-power wide area networks (LPWANs) have truly come of age. Originally intended for long-range applications, LPWAN technologies are expected to be used by hundreds or even thousands of stations (STAs) deployed on a limited geographical area. They are also char- acterized by transmitting small data packets at rates of up to tens of kilobits per second (kbps), and delivering several years of device operation on a single battery. LPWAN topology is characteristically a single-hop star, where end devices are directly connected to the base station by means of ALOHA-based medium access control (MAC) protocols [4], greatly simplifying the network and endowing it with robustness and centralized control. However, this uncontrolled medium access leads to interference or packet collisions among uncoordinated devices, acutely affecting reliability and scalability in dense networks [5]–[7]. The INTER-HARE protocol presented in this article sets out to increase typical LPWAN scalability by creating a cluster-tree network [8] based on interoperable multiband IoT technologies, where the LPWAN acts not only as data VOLUME 4, 2016 1

Upload: others

Post on 11-May-2022

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Increasing LPWAN scalability by means of concurrent

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.

Digital Object Identifier 10.1109/ACCESS.2017.DOI

Increasing LPWAN scalability by meansof concurrent multiband IoTtechnologies: An industry 4.0 use caseTONI ADAME1, ALBERT BEL2, AND BORIS BELLALTA31Network Technologies and Strategies research group, Universitat Pompeu Fabra, 08018 Barcelona, Spain (e-mail: [email protected])2Network Technologies and Strategies research group, Universitat Pompeu Fabra, 08018 Barcelona, Spain (e-mail: [email protected])3Wireless Networking research group, Universitat Pompeu Fabra, 08018 Barcelona, Spain (e-mail: [email protected])

Corresponding author: Toni Adame (e-mail: [email protected]).

This work has been partially supported by the Spanish Ministry of Economy and Competitiveness under the Maria de Maeztu Units ofExcellence Programme (MDM-2015-0502) and the Open Call of INTER-IoT, Research and Innovation action - Horizon 2020 EuropeanProject, Grant Agreement #687283, financed by the European Union. It also received funding from the Catalan government throughprojects SGR-2017-1188 and SGR-2017-1739, and from the Spanish government under the project TEC2016-79510-P.The authors also acknowledge the technical support received by Josep Pujol from Universitat Pompeu Fabra, Universitat Politècnica deValencia, Fundacion Valenciaport, and Friopuerto Valencia S.L.

ABSTRACT One of the most important challenges of the Internet of Things (IoT) in the next years willbe the smooth incorporation of millions of smart devices into its communications paradigm. The greatercoverage area of sub-1GHz low-power wide area networks (LPWANs) makes them a suitable technology toeasily encompass hundreds of end devices under a single base station. However, LPWAN inherent simplicityaffects negatively on scalability, as these networks are not flexible enough to deal with a high number ofnodes unless their traffic load is really low, which limits many potential use-cases. This paper analyzesthe scalability issue in LPWAN and proposes the INTER-HARE protocol: a solution based on the use ofconcurrent multiband IoT technologies, where an 868 MHz LPWAN acts as transparent backhaul for a set ofsubnetworks working at 2.4 GHz. The implementation of the INTER-HARE protocol on a real IoT platformwas assessed both in several laboratory testbeds and in a pilot developed in the premises of an industrialcompany, proving its suitability in non-delay sensitive monitoring applications with end devices scatteredthroughout the targeted area.

INDEX TERMS LPWAN, scalability, Internet of Things (IoT), multiband, wireless sensor networks(WSN), power consumption, HARE, INTER-HARE, Industry 4.0.

I. INTRODUCTION

THE number of connected Internet of Things (IoT) de-vices worldwide is exponentially growing in the last

years, with up to 25 billion devices expected to be part ofthe IoT ecosystem by the year 2020 [1]. The integrationof this huge amount of devices into existing and upcomingtechnological infrastructure was already considered in thepast as a vital challenge for the consolidation of the digitalindustrial economy, also known as Industry 4.0 [2].

After being identified for a long time as one of the mostfar-reaching trends in IoT [3], low-power wide area networks(LPWANs) have truly come of age. Originally intended forlong-range applications, LPWAN technologies are expectedto be used by hundreds or even thousands of stations (STAs)deployed on a limited geographical area. They are also char-

acterized by transmitting small data packets at rates of upto tens of kilobits per second (kbps), and delivering severalyears of device operation on a single battery.

LPWAN topology is characteristically a single-hop star,where end devices are directly connected to the base stationby means of ALOHA-based medium access control (MAC)protocols [4], greatly simplifying the network and endowingit with robustness and centralized control. However, thisuncontrolled medium access leads to interference or packetcollisions among uncoordinated devices, acutely affectingreliability and scalability in dense networks [5]–[7].

The INTER-HARE protocol presented in this article setsout to increase typical LPWAN scalability by creating acluster-tree network [8] based on interoperable multibandIoT technologies, where the LPWAN acts not only as data

VOLUME 4, 2016 1

Page 2: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

collector at 868 MHz, but also as backhaul network forseveral so-called low-power local area networks (LPLANs)working at 2.4 GHz. Main contributions of the current workcan be summarized into:

1) Evaluation of the scalability problem in LPWAN andreview of state-of-the-art solutions.

2) Design, development and implementation of INTER-HARE on a real IoT platform as an upgrade of theHARE protocol stack [9], by adding the followingfeatures:

• Adaptation of the MAC sublayer to cluster-treemultiband requirements.

• Development of a more flexible beaconing system.• Integration of three different data delivery models:

continuous, query-driven for downlink (DL) spe-cific requests, and event-driven for alarms.

• Introduction of a traffic prioritization mechanismamong data delivery models.

3) Experimental validation of INTER-HARE in a labora-tory testbed and in a real industrial pilot connected toan external cloud-based platform.

Overall, this paper is a proof of concept in which multi-band LPWANs may solve the scalability issue, providing aspecific but complete reference technology named INTER-HARE. Beyond showing how this technology offers lowpower consumption and high reliability, the authors aim toinspire future work on these multiband LPWANs that cantake this paper as a baseline.

The remainder of this paper is organized as follows: Sec-tion II goes in depth on the scalability issue in LPWAN.Section III comprehensively describes the INTER-HAREprotocol and Section IV details its technological implemen-tation. Section V refers to the laboratory testbed employedto validate INTER-HARE operation. Similarly, Section VIdescribes the industrial pilot and compiles the obtained re-sults from different experiments. Lastly, Section VII elabo-rates on the lessons learned from the entire development ofINTER-HARE and Section VIII presents the conclusions anddiscusses open challenges.

II. ON THE LPWAN SCALABILITY LIMITSThe full spectrum of LPWAN technologies can be classi-fied according to their operating frequency into 1) licensedspectrum LPWANs, using bands from cellular operators, and2) unlicensed spectrum LPWANs, using any of the publiclyavailable industrial, scientific, and medical (ISM) bands [10].

This article focuses on unlicensed spectrum LPWANs,whose main representative technologies are SIGFOX™ [11],LoRa™ [12], Weightless™ [13], DASH7™ [14], HARE[9], IEEE 802.15.4-based [15], and IEEE 802.11ah (WiFiHaLow) [16], among others. They all rely on sub-1GHz radiofrequencies to cover up to several tens of km, thus supportinghigh numbers of connected devices per base station.

Despite the variety of the aforementioned technologies,their MAC protocol is mostly derived from ALOHA or its

variation with carrier sensing, i.e., carrier sense multipleaccess (CSMA) [17], where any STA from a shared channeltransmits a packet with a certain probability whenever ithas data in the buffer. Whereas these protocols operate verywell with a low number of simultaneous contending userstransmitting small data packets, they suffer from scalabilityissues as the traffic load and/or the node density increases ona geographical area [18].

In the following lines, several initiatives to address theissue of efficiently handling a massive number of wireless de-vices without compromising overall network’s performanceare reviewed.

A. INCREASING SCALABILITY AT MAC LAYER LEVEL

There already exist several improvements that can be devel-oped at MAC layer level of LPWAN technologies to alleviatethe lack of medium access coordination in dense networks[6]:

• Scheduled MAC protocols: Also known as time divi-sion multiple access (TDMA)-like protocols, they allo-cate sensor nodes to specific slots in a frame to transmitand/or receive data, so nodes are only active in thoseslots and remain asleep in the rest [19]. While providingreduced or zero collision probabilities, scheduled MACprotocols suffer from high latency as the number ofSTAs grows. In addition, the necessity of a centralizedscheduler could lead to unfeasible overheads. Time slot-ted channel hopping (TSCH), one of the MAC modesdefined in the IEEE 802.15.4e standard [20], is the mostrepresentative example for LPWANs.

• Station grouping algorithms: In combination with ascheduled MAC protocol, the distribution of STAs intogroups can be used not only for organizational purposesbut also for properly allocating available channel re-sources, as in the TIM and Page Segmentation schemeof IEEE 802.11ah [21] or in the hybrid MAC sublayer ofHARE [9]. Thus, time domain is successively restrictedto a different group of STAs contending for the samechannel.

• Adaptive transmission mode: The optimization of dif-ferent transmission parameters (bandwidth, coding rate,maximum payload, or output power, among others)can alleviate the scalability issue in densely populatednetworks. The adaptive data rate (ADR) mechanism ofLoRa [22] determines several transmission parametersaccording to the environmental conditions between enddevices and the gateway (GW). The scalability issuein LoRa is specifically addressed in [23], by efficientlyselecting the transmission mode of end devices.

• Adaptive power control: Power regulation mechanism(PRM) of HARE [9] dynamically adapts the trans-mission power level based on the channel conditionsbetween sender and receiver, being specially useful indensely deployed areas where a short transmission rangeis sufficient to reach the sink or the next hop. Con-

2 VOLUME 4, 2016

Page 3: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

sequently, reduced range of transmissions results in afewer number of potentially interfering STAs.

B. EXTENSION OF RANGE COVERAGE BY MEANS OFMULTI-HOP TOPOLOGIESUnlike most LPWAN technologies relying on single-hoplinks, several novel initiatives have been proposed in recenttimes to extend the operating area by employing multi-hopapproaches:

• To improve data transmission efficiency and extendrange coverage, IEEE 802.11ah allows for using a dual-hop relay system in between end devices and the AP[24]. By carefully choosing the position of the relay andthe AP, the expected range in DL is up to 800 m, whereasin uplink (UL) it is about 550 m [25].

• LoRaBlink [26] and CT-LoRa [27] are protocols on topof LoRa’s physical layer designed to support reliableand energy-efficient multi-hop communications, wheretime synchronization is used to define slotted chan-nel access. In both cases, DL messages are distributedthrough flooding.

• One of the main singularities of DASH7 is enabling bothstar and tree topologies to facilitate the management oflarge networks [28]. In the latter case, STAs not directlyreaching the GW may transmit data to sub-controllers orother STAs, which in turn are responsible for forwardingthe messages.

• Sensor Network Over White Spaces (SNOW) [29] isproposed as an LPWAN architecture to support wide-area wireless sensor networks (WSNs) by exploiting TVwhite spaces. Range coverage is extended by means ofmultiple SNOWs organized into a cluster-tree topologymanaged by a single entity.

C. INCREASING SCALABILITY WITH MULTIBANDTECHNOLOGIESThe scalability limit in ALOHA-based LPWANs, such asSIGFOX™ or LoRa™ , where hundreds or even thousandsof devices share the same wireless medium, motivates thearticle’s hypothesis of using concurrent multiband technolo-gies and paves the way for the development of the INTER-HARE protocol. To illustrate it, a set of metrics (throughput,efficiency, and delay) are evaluated in function of the numberof devices in two different networks: a single-hop ALOHA-based network and a two-tier cluster-tree multiband network.

It is worth noting that regional regulations about duty cyclelimits in ISM bands (e.g., overall duty cycle limit for anEU device is 1%) are excluded from the current analysis.If such constraints were considered, they would severelyreduce the airtime occupied by transceivers, thus impactingthe aforementioned set of evaluation metrics.

1) ALOHA-based networkGiven a single-hop network with n STAs and a GW, wherethe ALOHA protocol is used as communication system, the

total normalized load (GALOHA) is expressed as

GALOHA = n ·λ · τ = n ·λ · Lr, (1)

where λ is the mean STA packet generation rate from a Pois-son process, τ is the packet transmission time, L is the packetlength, and r is the data rate. Assuming no backoff and noretransmissions, metrics of normalized throughput (SALOHA),efficiency (ηALOHA) and delay (DALOHA) are obtained from

SALOHA = GALOHA · e−2GALOHA , (2)ηALOHA = e−2GALOHA , (3)

and

DALOHA = τ =Lr. (4)

2) Multiband networkFigure 1 shows a two-tier cluster-tree multiband network,where the first tier is made up of a set of c non-overlappingclusters. Each cluster contains a multiband device, namedfrom now on cluster-head (CH), which simultaneously be-longs to the first and the second tier, and is responsible forgathering information from its corresponding set of STAs andthen retransmitting it to the GW located in the second tier.

GW

CH

STA

CH

CH

CH

Tier 1

(f1, r1)

Tier 2

(f2, r2)

Multiband

device

(f1, r1)

(f2, r2)

FIGURE 1: Two-tier cluster-tree multiband network repre-sentation.

Under this setting, a single GW controls n devices uni-formly scattered throughout the different clusters of the net-work. As there are c CHs (one per each cluster), the resultingnumber of STAs is n− c. Assuming that all clusters containthe same number of STAs, this value is equal to n

c −1.Whereas communication within each cluster is based on

the ALOHA protocol, packet transmissions from CHs tothe GW are scheduled by means of the TDMA channelaccess method. Both communication protocols are conducted

VOLUME 4, 2016 3

Page 4: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

simultaneously, as frequency of the first tier ( f1) is differentfrom that of the second tier ( f2). Assuming f1 > f2 andgreater bandwidth (B) of communication channels at f1, ahigher data rate in the first tier can then be achieved (i.e.,r1 > r2).

CHs are the actual single multiband devices of the wholenetwork, as their radio modules can simultaneously commu-nicate with STAs of the first tier at f1 and with the GW ofthe second tier at f2. Unlike them, the STAs and the GW canonly use a single frequency: f1 in the first case and f2 in thesecond. Finally, data rate employed by devices in the first andthe second tier is r1 and r2, respectively.

All STAs and CHs create data packets of length L at rateλ following a Poisson process, and having τ1 and τ2 aspacket transmission times, respectively. Normalized trafficload of each kind of device is represented by GSTA and GCH,1

respectively, and can be expressed as

GSTA = λ · τ1 = λ · Lr1

and GCH = λ · τ2 = λ · Lr2. (5)

Then, normalized traffic load generated by all STAs belong-ing to a cluster (Gc) is defined as

Gc =(n

c−1)·GSTA =

(nc−1)·λ · L

r1. (6)

Resulting normalized throughput in each cluster (Sc) whenSTAs use the ALOHA protocol can be obtained from

Sc = Gc · e−2Gc . (7)

Consequently, normalized throughput in the first tier (ST1) iscalculated from

ST1 = c ·Sc = c ·Gc · e−2Gc . (8)

In the second tier, the TDMA communication protocolallocates a time slot to each CH. This time period is usedby a CH to transmit an aggregated packet2 containing up toMagg packets from its corresponding STAs plus one packetgenerated by the CH itself. The TDMA time slot duration Tsis adjusted to the maximum length of the aggregated packet,being

Ts = (1+Magg) · τ2 = (1+Magg) ·Lr2. (9)

As each TDMA time frame contains c time slots (one pereach CH), its duration Tf is

Tf = c ·Ts = c · (1+Magg) ·Lr2. (10)

Each CH continuously receives packets from its corre-sponding STAs and generates its own ones, creating a new

1GCH load refers only to packets generated by the CH itself, withoutconsidering those received from STAs.

2Aggregated packet term is used here to distinguish it from packets createdevery λ by STAs and CHs. Benefits of packet aggregation in terms ofoverhead reduction are not considered in the proposed theoretical approach.

aggregated packet every Tf seconds. The number of packetsincluded in an aggregated packet, Nagg(Sc), accounts for

Nagg(Sc) = NSTA(Sc)+NCH, (11)

where NSTA(Sc) and NCH are the number of packets comingfrom the STAs and the CH itself, respectively, being

NSTA(Sc) =

Scτ1·Tf i f Sc

τ1·Tf < Magg

Magg otherwise(12)

and

NCH =

λ ·Tf i f λ ·Tf < 1

1 otherwise(13)

Due to the limited duration of Ts, if NSTA(Sc) or NCHexceeded their corresponding maximum value, the CH wouldonly include a limited number of packets from each datasource in the aggregated packet. As no buffer in the CH isconsidered, the rest of packets would be lost.

Once the aggregated packet is created, CHs employ theirallocated TDMA time slot to transmit the gathered informa-tion to the GW. Resulting throughput (in bps) in the GW,S∗MB(Sc), can then be computed as

S∗MB(Sc) =c ·L ·Nagg(Sc)

Tf=

Nagg(Sc) · r2

(1+Magg). (14)

Efficiency of the proposed multiband network, ηMB(Sc), iscomputed as the quotient between S∗MB(Sc) and the total load(in bps) generated by the network (G∗

MB), with

G∗MB = c · (Gc · r1 +GCH · r2) = λ ·L ·n, (15)

and therefore

ηMB(Sc) =S∗MB(Sc)

G∗MB

=Nagg(Sc) · r2

(1+Magg) ·λ ·L ·n. (16)

DMB(Sc) represents the delay of a packet since it is gener-ated by an STA until it is received by the GW. It is computedas

DMB(Sc) =Tf

2+

L ·Nagg(Sc)

r2=

Lr2

·[

c · (1+Magg)

2+Nagg(Sc)

]. (17)

3) Performance comparison: ALOHA-based network vs. firsttier of multiband networkAssuming λ = 1 packet/s and L = 800 bits, Figure 2 showsnormalized throughput in the ALOHA network (SALOHA) andin the first tier of the proposed multiband network (ST1).Whereas the ALOHA network uses r = 50 kbps, two differentdata rates are considered for the clusters located in the firsttier of the multiband network: r1 = 50 kbps and r1 = 250kbps.

As it can be observed, maximum normalized throughputin each configuration is achieved for a different maximumnumber of network devices (nmax). Thus, maximum achieved

4 VOLUME 4, 2016

Page 5: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

n (devices)

0

0.25

0.5

0.75

1

1.25

1.5

1.75

2

2.25

2.5

S

SALOHA

(r=50 kbps)

ST1

(r1=250 kbps, c=2)

ST1

(r1=250 kbps, c=6)

ST1

(r1=250 kbps, c=12)

ST1

(r1=50 kbps, c=2)

ST1

(r1=50 kbps, c=6)

ST1

(r1=50 kbps, c=12)

SALOHA

(max) = 0.184

FIGURE 2: Normalized throughput (S) in function of the number of devices (n) with λ = 1.

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

n (devices)

0

0.2

0.4

0.6

0.8

1

ALOHA

MB (c=2)

MB (c=4)

MB (c=6)

MB (c=8)

MB (c=10)

(a) Comparison of network’s efficiency (η).

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

n (devices)

0

0.2

0.4

0.6

0.8

1

1.2

D (

s)

DALOHA

DMB

(c=2)

DMB

(c=4)

DMB

(c=6)

DMB

(c=8)

DMB

(c=10)

DALOHA

= L/r2 = 0.016 s

(b) Comparison of delay (D).

FIGURE 3: Network’s efficiency (η) and delay (D) in function of the number of devices (n) with λ = 130 and Magg = 10.

throughput in the ALOHA network (SALOHA = 0.184) cor-responds to a very low number of devices (nmax ∼ 32).In contrast, nmax value is multiplied by c in the multibandnetwork when data rate within clusters is r1 = 50 kbps.Furthermore, if a higher data rate such as r1 = 250 kbps isused, the original nmax value of ALOHA is multiplied by ac · r1

r factor.

These results exemplify how clusters help to increase nmaxin a given area without impacting network’s reliability aslong as they are placed properly enough to avoid overlap-ping (by means of the combination of an appropriate deviceplacement and a selection of frequencies where f1 > f2).Additionally, the use of higher data rates within clusters (dueto the closer distance between STAs and CH) allows for aneven greater node density in the targeted area.

4) Performance comparison: ALOHA-based network vs.entire multiband network

When the complete two-tier cluster-tree multiband networkis considered, the comparison with the ALOHA network ishere made by means of two evaluation metrics: efficiency anddelay. Figure 3a shows obtained efficiency in both networksby using parameters from Table 1, Magg = 10 and severaldifferent c values.

TABLE 1: Network parameters.

Parameter Valuer 50 kbpsr1 250 kbpsr2 50 kbpsf1 2.4 GHzf2 868 MHzλ

130 packets/s

L 800 bits

VOLUME 4, 2016 5

Page 6: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

For any n value, ηMB is much higher than ηALOHA. Ifsetting a network’s efficiency goal, for instance η = 90%,the corresponding nmax differs by one order of magnitude(100 stations in the ALOHA network vs. 1000 or more inthe multiband one). As for the visible change of trend in themultiband network at n ∼ 2000, it starts when the number ofpackets received by the CH is higher than Magg, so that theyall cannot be retransmitted to the GW and an increasing shareof them are lost as n grows.

Figure 3b represents average packet delay from STAs tothe GW. Whereas delay always keeps the same reduced valuein the ALOHA network, it increases together with n untiln ∼ 2000 in the multiband network. Then, it remains stableas Nagg(Sc) has achieved its maximum value. The effect ofthe selected number of clusters is also observable, as highc values enlarge Tf and, consequently, the computed delay.This fact exemplifies the existing trade-off between reliabilityand delay when increasing c.

D. MULTIBAND TECHNOLOGIESLiterature and current IoT ecosystem offer multiple combina-tions of multiband technologies. In the following lines, theyare classified into the number of employed radio transceivers:

• Single radio transceiver: academical work from [30]presents a 433/868 MHz multiband wireless sensor plat-form that allows switching between the two consideredISM bands during normal operation, under the controlof a host microcontroller.3

• Multiradio development platforms: most scientific ini-tiatives can be classified among those operating at 433MHz and 2.4 GHz [33], [34], those operating at 868/920MHz and 2.4 GHz [35]–[37], and those combining twodifferent technologies at 2.4 GHz [38], [39].4

III. INTER-HARE PROTOCOLConceived as an innovative evolution of the HARE protocolstack [9], the INTER-HARE protocol takes advantage of thebenefits of multiband IoT technologies. Hence, the originalLPWAN does not only act as data collector at 868 MHz,but also as backhaul network for several concurrent LPLANsworking at 2.4 GHz, thus ensuring interoperability betweendevices working at one or both frequency bands.

A two-tier cluster-tree network topology is then created,with a GW as the main element of a LPWAN working at868 MHz, and dual-band CHs (868 MHz/2.4 GHz) entitledto manage the 2.4 GHz data acquisition devices (DADs) oftheir corresponding LPLAN in a hierarchic way, as in theexample shown in Figure 4.

The aforementioned CHs shall be built on a dual-band de-vice working at both 868 MHz and 2.4 GHz frequency bands.

3The most remarkable commercial options with a single dual-band sub-1GHz/2.4 GHz RF transceiver are the Texas Instruments™ CC1352R MCU[31] and the Silicon Labs™ EFR32FG SoC [32].

4Commercial plug-and-play products tend to combine 868/920 MHz and2.4 GHz (as in Zolertia™ RE-Mote [40], Zolertia™ Orion Router [41], orOpenMote B [42]), even integrating other technologies such as cellular orRFID, as in Libellium™ Waspmote [43].

FIGURE 4: Example of INTER-HARE network topologyand addressing system.

It can be achieved either by means of a single dual-band radiotransceiver controlled by a microprocessor or by means ofa multiradio development platform (either controlled by asingle microprocessor or in a master-slave approach usingtwo microprocessors).

The use of the HARE protocol stack ensures transmis-sion reliability, low energy consumption, self-organization,and resilience. Additionally, separated frequency bands inoverlapping networks result in an overall reduction of in-terference. Lastly, thanks to the hierarchic system proposed,scalability is enforced by a management scheme based onsubnetworking techniques.

A. PHY LAYERAs in HARE, there is not a predefined PHY layer to runthe INTER-HARE protocol, as it is only required to fulfilla minimum set of functions; namely, availability of differentoperational states both in the microprocessor (processing andlow power mode) and in the radio module (receiving, trans-mitting, and sleeping), selection of different transmissionlevels in the radio transceiver, and ability to execute low leveltasks required by typical shared medium access techniques.

B. LINK LAYERLink layer communication within the INTER-HARE proto-col is based on the ring-based transmission scheme of HARE(see Table 2). More specifically, LPLANs form an additionalring in the HARE network topology, so that transmissionsfrom DADs are allocated into a single 2.4 GHz TDMA slot

6 VOLUME 4, 2016

Page 7: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

TABLE 2: INTER-HARE link layer.

MAC SublayerFrequency

band Scheduled-basedprotocol

Contention-basedprotocol

Radio dutycycling(RDC)

sublayerLPWAN at868 MHz

TDMA(1 slot per ring)

CSMA/CA,or other

X-MAC,or other

LPLAN at2.4 GHz

TDMA(1 concurrent slot

for all DADs)

CSMA/CA,or other

X-MAC,or other

(a) Continuous data delivery model

(b) Query-driven and event-driven data delivery model

FIGURE 5: Transmission scheduling schemes.

placed immediately before the 868 MHz TDMA slots (seeFigure 5). CSMA/CA technique is internally performed inall slots by their corresponding contenders.

DADs transmit data to their corresponding CH during theaforementioned 2.4 GHz TDMA slot, regardless the position(ring) of the CH in the network. If concurrent LPLANs areproperly deployed (i.e., distant enough), there will be nointerference between them, making the system able to ex-tend LPWAN boundaries beyond typical 868 MHz coveragerange and easily integrating devices coming from these non-overlapping 2.4 GHz clusters.

Adaptation of the different link layer mechanisms derivedfrom HARE into the INTER-HARE protocol is described inthe following lines:

1) Beaconing SystemThe beaconing system holds a double function: synchroniz-ing network devices and scheduling the different actions tobe performed in a time-division multiplexing scheme (seeFigure 6). Two types of beacons are used for this purpose,primary and secondary beacons:

• Primary beacons, emitted every Tp seconds, include atimestamp, the action to be taken by the network, andthe time until the next primary beacon. There are threedifferent primary beacons according to their function:

– Network association primary beacons announcethe beginning of a new network association phase.

– Data primary beacons announce the start of a newUL data transmission phase to send application

data or statistics packets. An STA association phaseis always included to accept devices not yet asso-ciated. Lastly, query-driven requests/responses andevent-driven alarms may also be included in theDL request and UL response/alarm time slots. Adetailed view of the elements contained in a dataprimary beacon is offered in Figure 7.

– Void primary beacons only allow the transmis-sion of query-driven requests/responses and event-driven alarms in the corresponding DL request andUL response/alarm time slots.

• Secondary beacons are emitted every Ts seconds, whereTp = (ks+1) ·Ts, with ks being the number of secondarybeacons transmitted after every primary beacon. Theyinclude the same information as primary ones, andare used to guarantee information and synchronizationredundancy for already associated devices as well as toaccelerate network discovery for non-associated ones.However, no action is performed by devices after asecondary beacon.

The use of two different frequency bands makes beaconsbe first transmitted by the GW in the LPWAN at 868 MHzand immediately repeated (except for the network associationprimary beacons, as CHs must be first associated to thenetwork in order to retransmit them) by each CH in its ownLPLAN at 2.4 GHz.

The developed beaconing system increases network’s flex-ibility with respect to the previous HARE version by incor-porating three major improvements:

• Introduction of void primary beacons to modify datacollection frequency during network operation whilekeeping devices synchronized.

• Allocation of specific time periods to notify changesregarding network topology (e.g., new associations) andtransmit collected data to external platforms.

• Integration of query-driven and event-driven data deliv-ery models together with the already existing continuousdata delivery model into a single system, as detailed inthe next subsection.

2) Data delivery models and traffic prioritizationLink layer inherited from HARE has been adapted tosmoothly include three different data delivery models:

• Continuous: As in HARE, end devices transmit to theGW, at a predefined rate, sensing data into applicationdata packets or performance metrics into statistics pack-ets. Data flow is UL through a single-hop link in theLPLAN and a multi-hop route in the LPWAN.

• Query-driven: The GW initiates the communicationimmediately after the transmission of a data or a voidprimary beacon, during the DL request time slot, bycodifying in a broadcast message one or several datarequests of the following types:

– Global requests are addressed to all network de-vices (both CHs and DADs).

VOLUME 4, 2016 7

Page 8: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

FIGURE 6: INTER-HARE Beaconing system by type of device and operating frequency band.

FIGURE 7: Detail of a data primary beacon without including time periods corresponding to communication with externalplatforms. Description of time variables is included in Table 4.

– Cluster requests target a specific CH and all itscorresponding DADs.

– Single requests might be sent to individual devices(whether CHs or DADs).

When necessary, CHs retransmit requests to DADs oftheir own cluster. CHs then wait for response packetsfrom their DADs, aggregate them (even with their ownresponse packet, if requested) and send them to the GW

8 VOLUME 4, 2016

Page 9: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

TABLE 3: Main features of the different data delivery models in INTER-HARE. Note that HARE original configuration iswritten in italics.

Data delivery model(packet type)

Trafficdirection

Trafficpriority

Type of message Transmissionpower level Routing Acknowledgement

LPWAN868 MHz

LPLAN2.4 GHz

LPWAN868 MHz

LPLAN2.4 GHz

LPWAN868 MHz

LPLAN2.4 GHz

LPWAN868 MHz

LPLAN2.4 GHz

Continuous(application data andstatistics packets)

UL Low Unicast Unicast PRM PRM Multi-hop Single-hop Link layer &Transport layer Link layer

Query-driven(request packets) DL Medium Broadcast Broadcast Max. Max. Single-hop Single-hop Link layer Link layer

Query-driven(response packets) UL Medium Unicast Unicast Max. Max. Single-hop Single-hop Link layer Link layer

Event-driven(alarm packets) UL High Unicast Unicast Max. Max. Single-hop Single-hop Link layer Link layer

during the UL response/Alarm time slot.• Event-driven: The activation of an alarm in a CH or a

DAD generates a UL alarm packet in the next availableUL response/Alarm time slot of a data or void primarybeacon.

Traffic from different data delivery models is prioritizedby means of the time slot allocation order. Hence, alarmpackets are always sent before response packets in the ULresponse/Alarm time slot. Continuous traffic is relegated tothe subsequent UL data transmission time slot. A summaryof the different data delivery models together with their maincharacteristics is provided in Table 3.

3) Wakeup PatternsBoth CHs and DADs remain asleep out of their correspond-ing TDMA slots in order to save energy. In addition, theradio duty cycling (RDC) sublayer keeps the radio moduleturned off as much time as possible during active slots whileproviding enough rendezvous points for two nodes to be ableto communicate with each other.

4) Data Acquisition, Transmission, Aggregation, andSegmentationData acquisition routine (for example, from connected envi-ronmental sensors) is run by CHs and DADs only once justafter receiving a data or a void primary beacon. Collectedinformation is then stored into an application data packet,whose content keeps immutable until the next data acquisi-tion routine is started.

Resulting application data packet is analyzed in case itfulfills any of the predefined alarm conditions. If so, a newalarm packet is generated. Next, CHs and DADs keep awakeduring the DL request waiting for new query-driven requests.Again, a response packet is generated only if a new requestis detected. On the other hand, statistics packets follow abackground routine compiling not only internal (i.e., fromthe device itself) but also external (i.e., from the network)information and generating a packet whenever a data primarybeacon asks specifically for it.

Application data and statistics packets are sent duringthe UL data transmission time slot and subsequently ac-knowledged at link layer level by the receiver both in theLPWAN and in the LPLAN. As in HARE, the LPWANUL data transmission time slot is split into w consecutive

transmission windows (i.e., transmission opportunities), eachone consisting of R+1 equal ring time slots, being R the totalnumber of LPWAN network rings, as in Figure 5a.5

Data aggregation is performed twofold: 1) in the LPLAN,CHs are responsible for aggregating application data orstatistics packets coming from their respective DADs; and2) in the LPWAN, CHs in their way to the GW attachapplication data or statistics packets from CHs located athigher rings (i.e., their children).6 If the total amount ofdata aggregated by a CH exceeded the maximum payloadsupported by the hardware, it would be split into segmentssent consecutively through the LPWAN.

Figure 8 shows the network operation in the continuousdata delivery model with blue envelopes. In the example,DAD 45.51 sends an application data packet with its owncollected data to CH 45.5 located in ring 2, which aggregatesinformation coming from other DADs and from itself. A newapplication data packet is then generated and sent to CH 45.3located in ring 1, which in turn has also previously receivedother application data packets from its own DADs. In the laststep, all data are packed together and sent to the GW.

On the contrary, transmissions of query- and event-driventraffic are carried out in the UL response/Alarm time slot(represented in Figure 5b) and only consist of two hops:one inside the LPLAN and the other inside the LPWAN. Asshown in Figure 8 with red envelopes, a response or an alarmpacket generated in DAD 45.80 is first transmitted to CH45.8, which directly forwards it in just one hop to the GW.In this case, aggregation is performed only once by the CHwhen attaching alarm or response packets coming from itscorresponding DADs and from itself.

5) Power Regulation Mechanism (PRM)In INTER-HARE, PRM is available for both CHs and DADsin their transmissions under the continuous data deliverymodel. Therefore, they can dynamically adapt the transmis-sion power level of their outgoing packets according to amechanism that maintains received signal strength indicator(RSSI) between two thresholds: RSSImax and RSSImin.

5Although R is computed by the GW before every primary beacon, eachtransmission window consists of R + 1 ring time slots as a contingencymeasure in case the association of a new CH in the previous STA associationphase has added one ring to the LPWAN size.

6To allow aggregation capabilities, INTER-HARE requires a higher max-imum data packet length than other LPWAN technologies.

VOLUME 4, 2016 9

Page 10: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

FIGURE 8: Packet transmission examples in INTER-HARE under different data delivery models.

C. NETWORK LAYERNetwork communications follow a centralized scheme,where the GW adopts the main role and assumes the respon-sibility of managing network associations, allocating networkaddresses, and periodically notifying the start of new routingprocesses.

1) Addressing SystemThe addressing system is managed by the GW, which allo-cates a unique network address to each node. Nodes willmaintain the same network address as long as they do notleave the network. A dynamic record matching the physicaland the network address of all CHs and DADs is stored inthe GW. Similarly, each CH keeps a record of the DADsassociated in its own cluster.

The size of the network address is configurable and itsvalue determines the addressing range. Due to the inherenttwo-tier nature of the network, LPLANs can be consideredas subnetworks, so that an addressing system flexible enoughto allow subnetting is highly advisable.

In our particular implementation, the addressing systemis based on the Rime format [44], consisting of two 8-bitnumbers. Whereas the first number identifies the networkprefix shared by all devices (i.e., the network part), thesecond one is the host part, whose value for the GW is always0, for CHs is a number from 1 to 9, and for DADs is a numberfrom 10 to 99.

For the sake of simplicity, the host part of any DADconsists of two digits, being the first one the same as the hostpart of its corresponding CH, as shown in Figure 4. Lastly, itis worth noting that neither specific multicast nor broadcastaddresses are reserved in the Rime format.7

7The lowest level operation in Rime is the anonymous best-effort broad-cast layer, providing a 16-bit channel abstraction but no node addressing.

2) AssociationThe scheme of two different association mechanisms fromHARE is still used, with an active, global, scheduled one,called network association mechanism; and a passive, singu-lar one, called STA association mechanism.

The network association mechanism starts immediately af-ter the GW broadcasts a network association primary beaconwith the network association phase and follows the structurefrom Figure 9. Firstly, CHs determine their association turn(from 1 to at_net ) according to the RSSI computed from theGW’s beacon and send a discovery message via broadcastduring a randomly chosen association slot (from 1 to as),which is responded by the GW and all already associatedCHs within the coverage range.

Then, among the received responses, CHs send an associ-ation request to the device (the GW or other CH) with theminimum computed score (S) from

S = a1 · (PTXmax −RSSITX)+a2 · (PTXmax −RSSIRX)+

+ a3 · r+a4 · c, (18)

where PTXmax is the maximum transmission power of theCH’s transceiver (in dBm), RSSITX is the RSSI receivedat the candidate (in dBm), RSSIRX is the RSSI received atthe CH itself (in dBm), r is the ring to which the candidatebelongs, and c is the current number of candidate’s children[9]. The a weights are selected empirically according tochannel conditions and distributed by the GW into eachprimary beacon.8

Lastly, the GW distributes a list of the newly associatedCHs in an association summary sent via broadcast. Thewhole process is repeated as many times as association turnsat_net are defined.

Once finished all LPWAN association turns, the associ-ated CHs retransmit the network association primary beacon

8A study on selection of optimal routing configuration parameters inmulti-hop LPWANs like HARE can be found in [45].

10 VOLUME 4, 2016

Page 11: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

FIGURE 9: Example of the network association mechanism. The STA association mechanism follows the same structure, butwith a different number of LPWAN association turns (at_STA).

at 2.4 GHz and then starts the single association turn forLPLANs. DADs evaluate the computed RSSI from thosebeacons and send an association request to the CH withthe highest received RSSI. If the CH is able to admit newDADs, it retransmits all received requests in a single messageto the GW. Then, the GW evaluates association requestsfrom DADs of the whole network and emits an associationsummary via broadcast, which is in turn retransmitted byCHs in their respective clusters.

The STA association mechanism is executed in every dataprimary beacon, immediately after the time period devotedto query-driven and alarm-driven traffic, during the STA as-sociation phase. It follows the same structure as the networkassociation mechanism, except for the number of LPWANassociation turns (at_STA in this case).

3) Routing

The distance vector routing protocol of HARE here onlyapplies to the continuous data delivery model through theLPWAN. To avoid latency caused by packet processingand transmission in intermediate nodes, response and alarmpackets are directly sent by CHs to the GW in a single-hopapproach (see Figure 8). Within each LPLAN, connectionsalso follow a single-hop approach between DADs and theircorresponding CH.

D. TRANSPORT LAYER

Reliable end-to-end UL communications between CHs andthe GW under the continuous data delivery model are guar-anteed by means of the transport layer defined in HARE.Particularly, additional transmission windows and end-to-endACKs are maintained as a way to ensure correct packet recep-tion by the GW after packets go from CH to CH through thedifferent LPWAN rings. Other HARE mechanisms to reducethe time CHs are awake in error-prone UL communications

such as the poisoning mechanism and the distributed cachingare also applied.

There is no explicit end-to-end acknowledgement forDADs in the LPLAN, as their packets are aggregated by CHsand acknowledged at link layer level. Consequently, packetsfrom DADs not properly received by CHs during the single2.4 GHz TDMA slot are lost, as they cannot be retransmittedin subsequent transmission windows.

IV. IMPLEMENTATIONThe implementation of the INTER-HARE protocol in realhardware encompassed the adaptation of commercial devicesto all considered network device roles (GW, CH, and DAD, asshown in Figure 10) and the programming of the aforemen-tioned routines and network mechanisms.9 Additionally, inthe context of the collaboration with the INTER-IoT H2020European Project,10 the whole network was integrated into acloud-based framework that allowed interoperability amongdifferent IoT platforms.

A. PHYSICAL DEVICES

1) Gateway (GW)

The GW is built on a Zolertia™ RE-Mote board [40] workingat 868 MHz (r = 50 kbps and B = 0.25 MHz) and connected tothe electrical supply. It is also equipped with a 8-GB microSDcard to store generated logs on network’s performance. Tocommunicate the network with the INTER-IoT cloud plat-form, the GW is connected via USB to a Raspberry Pi 3B+[46] responsible for managing the data exchange.

9Except for the PRM mechanism, which was only implemented in LP-WAN transmissions.

10INTER-IoT H2020 project main website:http://www.inter-iot-project.eu/.

VOLUME 4, 2016 11

Page 12: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

FIGURE 10: Hardware employed in the implementation ofthe INTER-HARE protocol.

2) Cluster-head (CH)The CH consists of two Zolertia™ RE-Mote boards con-nected by means of a serial cable under a master-slavecommunication scheme. The master board works at 868 MHz(r = 50 kbps and B = 0.25 MHz) and controls the activationcycles of the slave board, which works at 2.4 GHz (r = 250kbps and B = 5 MHz). Energy can be obtained either fromtwo 800 mAh LiPo batteries (one per each board) or, whenpossible, from the electrical supply. The CH is also equippedwith two different analogical sensors for monitoring pur-poses: a DHT22 temperature + humidity sensor [47] and aGrove luminance sensor [48].

3) Data acquisition device (DAD)DAD is based on a Zolertia™ RE-Mote board working at 2.4GHz (r = 250 kbps and B = 5 MHz) and equipped with a800 mAh LiPo battery. It is also equipped with the sametemperature + humidity and luminance sensors as the CH.

B. SOFTWARE DEVELOPMENTThe INTER-HARE protocol was developed on Contiki 3.0OS [49] as a new hardware independent module workingin coordination with the already existing IEEE 802.15.4gunderlying communication standard. Specific interactions ofINTER-HARE with hardware PHY layer were programmedseparately.

A purpose-oriented application layer was programmed, inwhich CHs and DADs encapsulated into application datapackets of LD = 10 bytes11 the measurements from theirsensors together with other complementary information, suchas the Rime address, the packet sequence number, and thebattery level. Response and alarm packets shared the same

11Implementation of IEEE 802.15.4g in Contiki OS increases the mini-mum length of any transmitted packet up to 43 bytes after including headersand, if necessary, applying padding.

structure and length (LR and LA, respectively), with the singledifference that the latter codified the alarm type in bytenumber 10, which otherwise kept void.

Statistics packets were designed to provide informationabout device performance.12 They followed the same trans-mission rules as application data packets, but consisted ofLS = 20 bytes and compiled, since the last network asso-ciation primary beacon, the following metrics: number ofsent packets, acknowledged packets, sent ACKs, averageround trip time (RTT) at link level, average RTT at transportlevel, % of time in each operational state, and maximum andminimum employed transmission power level.

As part of the collaboration with the INTER-IoT project,a device controller embedded as a Java-based OSGi module[50] was programmed and installed in the Raspberry Pi,responsible for managing the serial communication protocolwith the GW and establishing a reliable connection with theINTER-IoT cloud platform via Ethernet or WiFi.

The implementation of the INTER-HARE protocol in-volved the setting of a wide series of configuration param-eters, both new and derived from the HARE stack. Table 4compiles the most important ones into different subsystems,remaining unchanged through the whole experimentation.

V. LABORATORY TESTBEDInitial performance evaluation was performed in several adhoc testbeds located on the 2nd floor of the Tànger building,at Universitat Pompeu Fabra (UPF) facilities.13 The spaceis characterized by a transversal corridor consisting on twosections of 50 m and 36 m long with office rooms at bothsides. Office rooms have sizes between 20 m2 and 32 m2.

A. RANGE COVERAGEThis test assessed the suitability of the selected hardware inindoor environments when working at the 868 MHz band.Two Zolertia RE-Mote units were used: one acting as a GWplaced in a static position (on the table of an office), and theother as a CH trying to communicate with the GW from a setof 133 locations from 3 different areas (see Figure 11):

• 16 Locations inside the same office as the GW.• 14 Locations inside another office.• 103 Locations along the floor central corridor (one loca-

tion every 80 cm).From each of the 133 positions, the CH sent 10 packets

(one every 3 seconds) to the GW at maximum transmissionpower level (i.e., PTX = 14 dBm) and data rate r = 50 kbps.14

The GW always received enough signal (i.e., there was fullcoverage in the whole floor when using the 868 MHz band)and stored the RSSI values from received packets.

RSSI values were later used together with the computeddistance between the GW and the CH in each position to

12The current implementation of INTER-HARE only supports statisticsgathering in the CHs.

13Tànger building at UPF campus:https://www.upf.edu/web/campus/tanger.

14Note that the sensitivity of the receiver at 868 MHz is Smin = -109 dBm.

12 VOLUME 4, 2016

Page 13: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

FIGURE 11: Laboratory testbed for range coverage at 868 MHz, where the G bullet corresponds to the GW location and redbullets correspond to the selected CH locations.

TABLE 4: INTER-HARE configuration parameters.

General parametersParameter Description Value

ksNumber of secondary beacons pereach primary beacon 2

w Number of transmission windows 5

at_netNumber of association turns(Network association mechanism) 2

at_STANumber of association turns(STA association mechanism) 1

as Number of association slots per turn 4Scheduling

Parameter Description ValueTreq DL request time slot 8 sTres UL response / Alarm time slot 8 sTds LPWAN Discovery time slot 2 s

Tahw LPWAN Association handshake time slot 10 sTahl LPLAN Association handshake time slot 10 s

Tassoc_STASTA Association phase time slotTassoc_STA = at_STA · (as ·Tds +Tahw)+Tahl

28 s

Tassoc_netNetwork association phase time slotTassoc_net = at_net · (as ·Tds +Tahw)+Tahl

46 s

Tdl LPLAN UL data transmission time slot 5 sTr LPWAN Ring time slot 5 s

TwWindow time slotTw = (R+1) ·Tr

R dependant

TdwLPWAN UL data transmission time slotTdw = w ·Tw = w · (R+1) ·Tr

R dependant

Node discovery configuration parametersParameter Description Value

PTXmax Maximum transmission power 14 dBma1 Weight of received RSSI 10a2 Weight of received RSSI at destination 10a3 Weight of ring number 1a4 Weight of children number 5

Power regulation mechanism (PRM)Parameter Description Value

RSSImax PRM Upper threshold -80 dBmRSSImin PRM Lower threshold -110 dBm

Packet lengthParameter Description Value

LD Application data packet length 10 bytesLS Statistics packet length 20 bytesLR Response packet length 10 bytesLA Alarm packet length 10 bytes

model the propagation channel, as shown in Figure 12. Thelog-distance path loss model, PL(d), obtained by using theleast square approximation method, is expressed as

PL(d) =−23.7268−34.9844 · log10(d), (19)

where the obtained path loss exponent (γ ∼ 3.5) fits into thetypical ranges of an office building [51].

As for the 2.4 GHz band, range coverage tests were limitedto prove that a CH using that frequency band had full visioninside two selected offices.

0 2 4 6 8 10 12 14 16 18 20

10·log10

(d)

-120

-100

-80

-60

-40

-20

0

RS

SI

(dB

m)

Least Squares

Robust

FIGURE 12: Least square approximation of the simplifiedpath loss propagation model at 868 MHz.

B. ASSOCIATIONA network consisting of a GW, 2 CHs and 8 DADs wasdeployed in two different offices, as shown in Figure 13. TheGW was configured to run the INTER-HARE protocol, send-ing a primary beacon every Tp = 2 min and interspersing onenetwork association primary beacon with one data primarybeacon asking for statistics packets.

By using this beacon scheduling, all devices were discon-nected from the network every 4 minutes (when receiving anetwork association primary beacon) and immediately after

VOLUME 4, 2016 13

Page 14: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

FIGURE 13: Laboratory testbed for association and datatransmission experimentation. Warning signs are added tothose DADs with the ability of generating alarms.

that, a new network association phase was initiated. If adevice did not complete a successful association, it would usethe STA association phase of the next data primary beaconto restart the process.

The experiment consisted of 50 cycles (each cycle witha network association primary beacon and a data primarybeacon) and lasted 200 minutes. The different groups ofdevices (CHs and DADs) got an association success rate ofmore than 95%, as detailed in Table 5. While both CHs werealways connected (100%), DADs suffered from sporadicinterference, with a success rate of 95.75%.

It is also observed from results that devices got connectedmainly during the network association phase (i.e., in their 1sttry), but they also often took advantage from the additionalchance offered by the STA association phase (2nd try).

As for the association delay (computed as the time dif-ference between the emission of the network associationprimary beacon of a cycle and the actual association of adevice to the network), it was 12.35 s on average for CHsand 52.11 s for DADs. The latter value is higher than thenetwork association phase time slot due to those cases whereDADs must wait for the STA association phase correspond-ing to the next primary beacon. Though rare, collisions ofassociation requests, sudden channel alterations and/or clockdesynchronization may prevent DADs from satisfactorilyconcluding their association process before the end of theLPLAN association handshake time slot, delaying it until thenext opportunity.

As expected, DADs tended to connect to the networkthrough the CH of their own office (more than 90% in allstudied cases, as shown in Figure 14). However, there werealso some cases (due to interference or the unsuccessfulassociation of a CH during the network association phase)in which DADs considered the CH placed in the other officeas their own parent (i.e., their next hop in the path to the GW).

DAD #9 DAD #11 DAD #13 DAD #15 DAD #6 DAD #8 DAD #10 DAD #120

10

20

30

40

50

60

70

80

90

100

Fre

qu

en

cy o

f C

H s

ele

ctio

n (

%)

CH #1

CH #2

FIGURE 14: Distribution of selected CH among DADs.

C. DATA TRANSMISSIONThe behaviour of a real IoT application was emulated in thistest, where requests and responses, alarms, application dataand statistics were transmitted over the same network. Devicedeployment was maintained, with the single particularity thattemperature + humidity sensors from DADs #6 and #9 weredeliberately switched off to provoke alarms. In addition, aglobal request (i.e., an application data request addressed toall associated devices) was generated by the GW in the firstemitted void beacon after a data primary beacon.

The GW was configured to send a primary beacon everyTp = 2 min, according to the beacon scheduling providedin Table 6. This beacon scheduling constituted a 20-minutecycle, which was continuously repeated by the GW. Theexperiment consisted of 12 operation cycles (i.e., a total timeof T = 4 hours). The number of expected packets per type(Nx) can be derived from the expression

Nx = NSTAx ·bx ·Nc, (20)

where x represents the packet type (D: application data, R:response, and A: alarm), NSTAx is the number of devices (CHsand DADs) sending packets of type x, bx is the number ofpackets per cycle and device of type x, and Nc is the numberof test cycles.

Table 7 compiles observed results from Nx, the actual num-ber of packets received by the GW (N∗

x ), and the associatedpacket delivery ratio (PDRx). Latency (lx) is computed as thetime difference between the emission of a primary beaconand the reception of the corresponding packet by the GW.Lastly, network throughput (Sx) is defined by

Sx =N∗

x ·Lx

T, (21)

where Lx is the packet length in bits.Results show PDRx values roughly or above 90% for

all considered packet types, being alarm packets the mostreliable due to their reduced number of simultaneous con-tenders. As expected, the largest latency corresponds to dataapplication packets, in accordance with the lowest prioritygiven to the UL data transmission time slot. Throughputvalues confirm the low use of wireless resources in a typicalIoT application.

14 VOLUME 4, 2016

Page 15: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

TABLE 5: Association results in the laboratory testbed.

Associated devices (%) Associationdelay (s)

CHs DADsStage Cycles During the networkassociation phase

During the STAassociation phase Total During the network

association phaseDuring the STA

association phase Total CHs DADs

Total 1-50 98.00% 2.00% 100.00% 89.50% 6.25% 95.75% 12.35 52.11

TABLE 6: Beacon scheduling of a cycle during the datatransmission test and expected packets.N: Network association primary beaconD: Data primary beacon (application data packets)S: Data primary beacon (statistics packets)V: Void primary beacon

Beaconindex #1 #2 #3 #4 #5 #6 #7 #8 #9 #10

Beacontype N D V V V D V V V S

Datarequest? - - Yes - - - Yes - - -

VI. PILOTThe suitability of the INTER-HARE protocol in a warehousestorage monitoring use case was assessed in a pilot developedduring three working days of September 2018 in several se-lected cold warehouses from Friopuerto15 company, locatedat the port of Valencia. As shown in Figure 15, the entirebuilding consists of an office area (in white in the figure)and a main docking area (in green), connected to other coldchambers (in different blue tonalities according to the innertarget temperature) by means of a transversal corridor. Viewsof the main corridor and one of the cold chambers are offeredin Figure 16a and Figure 16b, respectively.

FIGURE 15: Friopuerto facilities plan.

A. RANGE COVERAGEThe coverage of the 868 MHz band in an industrial envi-ronment was assessed in this test. Again, two Zolertia RE-Mote units were used: one acting as a GW placed in a static

15Friopuerto main website: http://www.friopuerto.com/.

position (specifically, inside a cabin in the middle of the maindocking area), and the other as a CH trying to communicatewith the GW from a set of 17 locations from 4 different areas(see Figure 17 for the deployment map):

• 6 Locations on the main docking area.• 3 Locations inside cold chamber #1.• 3 Locations on the transversal corridor.• 5 Locations inside cold chamber #6.

From each position, the CH sent 10 packets (at a rate of1 measurement every 5 seconds) to the GW at maximumtransmission power level (i.e., PTX = 14 dBm) and data rater = 50 kbps. The main outcome from analyzing the computedRSSI values is that full coverage was achieved in the maindocking area and in any selected location from chambers #1and #6 when working at 868 MHz frequency band.

Tests at 2.4 GHz, aimed to validate the connectivity ofdevices in the main docking area and in chamber #1, wereused to plan the network deployment of subsequent tests.

B. ASSOCIATION AND RESILIENCEAs in the laboratory testbed, a network consisting of a GW,2 CHs and 8 DADs was deployed to run this test. The GWwas placed inside the cabin as in the previous experiment,CH #1 was placed on a platform above the same cabin, andCH #2 was situated on a table inside the office area. Locationof DADs was chosen to cover the main docking area and coldchamber #1 (see Figure 18 for a more detailed view).

Again, the GW set Tp = 2 min, interspersing one networkassociation primary beacon with one data primary beaconasking for statistics packets, so that all devices were discon-nected from the network every 4 minutes. The experimentlasted approximately 16 hours, containing 239 cycles (eachcycle with a network association primary beacon and a dataprimary beacon).

To evaluate the resilience in the association process, thewhole experiment was split into 2 different stages, that is,before and after the deliberate switching off of CH #2 in cycle#130. Table 8 compiles the association success rate and itscorresponding delay, both total and for each of the two stages.While CH #2 was still on (stage 1), both CHs and DADsgot an association success rate of more than 95%. In stage2 the network proved its resilience, as DADs did not sufferfrom performance degradation despite the unavailability ofCH #2. It it also worth noting the low impact of the STAassociation phase (even less than in the previous laboratorytestbed), as most devices got associated during the earliernetwork association phase.

VOLUME 4, 2016 15

Page 16: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

TABLE 7: Data transmission results in the laboratory testbed.

Type of packet (x) NSTAxbx

(packets/cycle)c

(cycles)Nx

(packets)N∗

x(packets)

PDRx(%)

lx(s)

Sx(bps)

D: Application data 10 2 12 240 214 89.17 54.81 1.19R: Response 10 2 12 240 221 92.08 19.94 1.23A: Alarm 2 9 12 216 202 93.52 12.12 1.12

TABLE 8: Association results in the pilot.

State Associated devices (%) Associationdelay (s)

CHs DADs

Stage Cycles CH #1 CH #2During the

networkassociation

phase

During theSTA

associationphase

Total

During thenetwork

associationphase

During theSTA

associationphase

Total CHs DADs

Stage 1 1-130 ON ON 96.92% 1.54% 98.46% 94.04% 2.02% 96.06% 11.61 47.23Stage 2 131-239 ON OFF 100.00% 0% 100.00% 94.61% 3.44% 98.05% 9.12 49.37Total 1-239 ON ON/OFF 97.83% 1.08% 98.92% 94.30% 2.67% 96.97% 10.76 48.20

(a) Main docking area. (b) Cold chamber #1.

FIGURE 16: Images of the pilot application areas in Friopuerto company.

FIGURE 17: Pilot testbed for range coverage at 868 MHz, where the G bullet corresponds to the GW location and red bulletscorrespond to the selected CH locations.

16 VOLUME 4, 2016

Page 17: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

FIGURE 18: Devices’ distribution at Friopuerto premises.

DAD #9 DAD #11 DAD #13 DAD #15 DAD #6 DAD #8 DAD #10 DAD #120

10

20

30

40

50

60

70

80

90

100

Fre

qu

en

cy o

f C

H s

ele

ctio

n (

%)

CH #1

CH #2

FIGURE 19: Distribution of selected CH among DADs instage 1.

Figure 19 shows the distribution of selected CH amongDADs in stage 1. Only DADs #8 and #10 preferentially choseCH #2 (located inside the office area), as they received thehighest RSSI values from its beacons. During stage 2, allDADs always chose CH #1, as it was the single active CHfrom which they received beacons.

The association delay of CHs decreased from 11.61 s instage 1 to 9.12 s in stage 2, as only CH #1 remained active.Conversely, DADs spent some more time to get associated(from 47.23 s to 49.37 s), as more devices competed forcommunicating with the single active CH.

C. DATA TRANSMISSIONThe suitability of the INTER-HARE protocol for an in-dustrial monitoring application was assessed in the samedeployment from the previous test. The GW was configuredto send a primary beacon every Tp = 2 min, according to thebeacon scheduling of a cycle provided in Table 6. In this case,however, no specific DL requests from the GW (as defined inSection III) were programmed.

The experiment consisted of 33 operation cycles of 20minutes each one (i.e., a total time of T = 11 hours). Alarmswere automatically generated by CHs or DADs in case any

TABLE 9: Data transmission results (detail) in the pilot.

PDRD (%)Device ND

(packets)N∗

D(packets) Per

devicePer typeof device

CH #1 66 65 98.48%CH #2 66 64 96.97% 97.73%

DAD #6 66 63 95.45%DAD #8 66 59 89.39%

DAD #10 66 55 83.33%DAD #12 66 63 95.45%DAD #9 50 39 78.00%

DAD #11 66 64 96.97%DAD #13 66 57 86.36%DAD #15 66 65 98.48%

90.82%

Total 644 594 92.24%

of the following circumstances was detected:• A malfunctioning or an out-of-bounds value from the

battery sensor.• A malfunctioning or an out-of-bounds value from the

temperature + humidity sensor.• A malfunctioning or an out-of-bounds value from the

luminance sensor.If more than one of the aforementioned circumstances wasdetected by a device, a single cumulative alarm packet withthe corresponding codification would be built to notify allevents.

According to (20), ND = 10 ·2 ·33 = 660 application datapackets were expected to be received in the GW. However,and due to the unexpected switch off of DAD #9, the ac-tual value was ND = 644. As for stored logs, 594 appli-cation data packets were actually received, accounting forPDRD = 92.24%. Table 9 details the achieved PDR in eachdevice, where all DADs were well above 80% of reliabilityexcept for DAD #9, with CHs achieving even better overallperformance. With respect to effectively received applicationdata packets, obtained values of latency (lD = 54.99 s) andthroughput (SD = 1.20 bps) kept high similarity with thosefrom the laboratory testbed.

Further analysis of logs revealed that DAD #9 ran out ofbattery 2.5 hours before the end of the test, which explainsits poor performance. The location of DAD #9 on the floorof chamber #1, affected by the cold stratification,16 couldcertainly have altered the discharging behaviour of its battery,as it can be seen in Figure 20.

With regard to alarms, they all were set off by CH #1,probably caused by the infiltration of condensation waterdrops into the plastic junction box hosting the CH electronics,as it could not be perfectly sealed. Particularly, out of 176alarm packets received, 125 were set off by lecture errors inthe battery sensor, 50 by lecture errors in the battery sensor incombination with an out-of-bounds value from the luminancesensor, and 1 by a lecture error in the temperature + humiditysensor. Values of averaged alarm latency and throughputwere lA = 12.78 s and SA = 0.36 bps, respectively.

16Cold stratification is a phenomenon that accumulates the coldest air ofa warehouse in the lower levels [52].

VOLUME 4, 2016 17

Page 18: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

0 5 10 15 20 25 30 35 40 45 50 55 60 65

Index of expected application packet

90

91

92

93

94

95

96

97

98

99

100

Battery

level (%

)

CH #1

CH #2

DAD #9

DAD #11

DAD #13

DAD #15

DAD #6

DAD #8

DAD #10

DAD #12

FIGURE 20: Battery discharging process in function of time.

From the analysis of statistics packets sent by CHs, furtherinformation regarding PRM operation and energy consump-tion of devices can be obtained. As PRM only acted inthe transmission of application data and statistics packets,transmission power was reduced in up to 3 levels in eachcycle (from 14 dBm to 11 dBm) to be again reset to 14 dBmwith the reception of the network association primary beaconcorresponding to the next cycle.

There are two main sources of energy consumption in theZolertia™ RE-Mote board: the microprocessor and the radiomodule (as the consumption of electronics can be consideredas negligible). It is then possible to compute the averagedcurrent (I) by means of the current values of the differentstates from Table 10 and their time share in % (shown inFigure 21) by using

I = IµP + IR, (22)

where IµP and IR correspond to the averaged current con-sumed in the microprocessor and in the radio module, respec-tively:

IµP = ICPU ·%tCPU + ILPM ·%tLPM (23)IR = ITX ·%tTX + IRX ·%tRX + ISL ·%tSL (24)

In the specific case of the CH, the master board determinesthe CH lifetime regardless the slave board, as the first oneis responsible for maintaining the normal operation of theINTER-HARE protocol established by the GW.

First, the averaged current consumed by the master boardof CHs in the conducted data transmission test of the indus-trial pilot was I = 0.584 mA (with IµP = 13.399 µA andIR = 0.571 mA). Then, CH lifetime (TCH) can be estimatedfrom the battery capacity value of the master board (Q) byusing

TCH =QI. (25)

In our particular case, with employed batteries of Q = 800mAh, estimated TCH is 57.07 days.

TABLE 10: Current values of Zolertia™ RE-Mote opera-tional states for the master board. Note that transmittingcurrent is set according to the power level range obtained bythe PRM.

Operational state CurrentMicroprocessorARM Cortex-M3 [53]

Processing (CPU) ICPU = 13 mALow power mode (LPM) ILPM = 0.4µA

Radio ModuleTI CC1200 [54]

Receiving (RX) IRX = 19 mATransmitting (TX) ITX = 50.5-61 mA

Sleeping (SL) ISL = 0.12 µA

CPU state distribution

CPU: 0.1%

LPM: 99.9%

Radio module state distribution

TX: 0.001%RX: 3%

SL: 96.999%

FIGURE 21: Energy characterization of CPU and radio mod-ule in the master board of CHs. Note that time in TX modeof the radio module in CHs is well below the EU 868 MHzduty cycle regulation limit of 1%.

VII. LESSONS LEARNEDThis section summarizes the theoretical and practical lessonslearned through the design, development, implementation,and evaluation of the INTER-HARE protocol.

18 VOLUME 4, 2016

Page 19: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

A. THEORETICAL LESSONSThe intrinsic nature of LPWANs, where the wireless mediumis shared among uncoordinated devices that trade the trans-mission time for the communication range, faces scalabilityas one of its main challenges. Whereas no major performanceproblems are detected in low-dense LPWANs, the employedMAC protocols are not able to cope with the traffic load ofan ever-growing population of devices.

The theoretical approach presented in this article of seg-menting an LPWAN into a multiband system where multipleshort-range networks (i.e., clusters) and a long-range back-haul network operate at different frequencies has proven to beuseful to increase scalability at the cost of adding complexityto the communication system.

In fact, coordination between clusters (internally runningan ALOHA-based medium access) and the backhaul networkis a responsibility of CHs, which here have been introducedin the context of LPWANs for the first time. Althoughideally CHs shall be equipped with the required technologyto be always awake at both frequencies and willing to re-ceive/transmit data from/to both environments, it is necessaryto limit their active periods in order to control the energyconsumption.

The proposed orchestration of the backhaul network relieson a TDMA system which schedules transmissions of thedifferent CHs in successive time slots, so that a CH onlyoperates concurrently at both frequencies during one timeslot of each TDMA cycle. Whereas the absence of collisionsin this transmission scheme leads to a high overall network’sefficiency, the waiting time until the targeted slot noticeablyincreases the delay.

All in all, for a given number of end devices, the number ofclusters in the proposed multiband scheme represents a trade-off between network’s reliability and packet delay. Hence,whereas reduced delay is achieved with few clusters, a highnumber of them approaches the operation of a traditionalTDMA system, thus reducing channel contention.

B. PRACTICAL LESSONSWhen exporting the theoretical model of the multiband net-work to real IoT technologies by means of the INTER-HARE protocol, some aspects must be taken into consider-ation. First, physical segmentation of the LPWAN coveragearea into equal, non-overlapping clusters is far from beinga straightforward task, due to variable channel propagationcharacteristics of the wireless medium and an end devicedeployment subject to the high-level application.

Secondly, an accurate synchronization system is requiredthroughout clusters and the backhaul network, not only toallocate the different data acquisition, transmission, and ag-gregation periods, but also to set the energy-aware wakeuppatterns that battery-powered devices will follow in order toenlarge their lifetime.

Thirdly, the necessity of two different network interfaces inthe CH adds complexity to its computing system, which mustmaintain continuous communication with the transceiver(s)

by means of a single microprocessor, or in a master-slave ap-proach using two microprocessors. As for the CH operationitself, it shall combine its subordinate role to the GW with theentitlement to manage end devices of its own cluster in termsof synchronization, network discovery and association, anddata transmission, among others.

Lastly, although the high-level application transparentlyconsiders the multiband network as a whole and both the CHand end devices as data sources, the use of subnetworkingtechniques (unlike in traditional LPWANs) is strongly re-quired to manage the addressing system in such a hierarchicalnetwork organization.

VIII. CONCLUSIONSLPWAN technology has already proven its capacity to be-come a successful IoT player in long-range, low-demandingapplications. However, scalability is still an open issue, asperformance in these kinds of wireless network dramaticallydrops when increasing the device density or the traffic loadin the coverage area.

The proposed INTER-HARE protocol opens up a new wayto alleviate the adverse effects of huge amounts of contendersin LPWANs, by means of a TDMA-like concurrent multi-band system that groups end devices into non-overlappingclusters. In addition, the designed traffic prioritization systemfacilitates the coexistence of different data delivery modelswithin a single architecture.

Due to its flexibility, easy installation and high reliabilityachieved in extensive experimentation, the INTER-HAREprotocol can perfectly fit in Industry 4.0 environments withdynamic monitoring requirements (i.e., changing sensor lo-cations, targeted metrics, type of sensors, acquisition periods,and so forth) transmitting non-delay sensitive data.

Current INTER-HARE network coverage is limited by therange of the GW. Future work will consider the use of relaysat 868 MHz to further extend it. In addition, dependency ofon-site devices (CHs and DADs) on batteries makes nec-essary a proper maintenance plan together with the usageof high capacity power sources, or even energy harvestingtechniques.

REFERENCES[1] Gartner Inc., “Leading in a digital world: The dawn of the digital industrial

economy,” October 2013.[2] J. Manyika, The Internet of Things: Mapping the value beyond the hype.

McKinsey Global Institute, 2015.[3] A. Markkanen, “LPWA: Disruptive new networks for IoT,” Metering &

Smart Energy International, no. 6, pp. 24–26, 2015.[4] C. Goursaud and J.-M. Gorce, “Dedicated networks for IoT: PHY/MAC

state of the art and challenges,” EAI endorsed transactions on Internet ofThings, 2015.

[5] M. C. Bor, U. Roedig, T. Voigt, and J. M. Alonso, “Do LoRa low-powerwide-area networks scale?,” in Proceedings of the 19th ACM InternationalConference on Modeling, Analysis and Simulation of Wireless and MobileSystems, pp. 59–67, ACM, 2016.

[6] E. De Poorter, J. Hoebeke, M. Strobbe, I. Moerman, S. Latré, M. Weyn,B. Lannoo, and J. Famaey, “Sub-GHz LPWAN network coexistence, man-agement and virtualization: an overview and open research challenges,”Wireless Personal Communications, vol. 95, no. 1, pp. 187–213, 2017.

VOLUME 4, 2016 19

Page 20: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

[7] A. Augustin, J. Yi, T. Clausen, and W. M. Townsley, “A study of LoRa:Long range & low power networks for the internet of things,” Sensors,vol. 16, no. 9, p. 1466, 2016.

[8] L.-O. Varga, Multi-hop energy harvesting wireless sensor networks: rout-ing and low duty-cycle link layer. PhD thesis, Université Grenoble Alpes,2015.

[9] T. Adame, S. Barrachina-Muñoz, B. Bellalta, and A. Bel, “HARE: Sup-porting efficient uplink multi-hop communications in self-organizing LP-WANs,” Sensors, vol. 18, no. 1, p. 115, 2018.

[10] U. Raza, P. Kulkarni, and M. Sooriyabandara, “Low power wide areanetworks: An overview,” IEEE Communications Surveys & Tutorials,vol. 19, no. 2, pp. 855–873, 2017.

[11] SIGFOX, “Sigfox website.” https://www.sigfox.com/. Accessed: 2019-03-01.

[12] LoRa Alliance, “LoRa Alliance Wide Area Networks for IoT.” https://www.lora-alliance.org/. Accessed: 2019-03-01.

[13] Weightless SIG, “Weightless - Setting the Standard for IoT.” http://www.weightless.org/. Accessed: 2019-03-01.

[14] DASH7 Alliance, “DASH7 Alliance website.” http://www.dash7-alliance.org/. Accessed: 2019-03-01.

[15] “IEEE Standard for Low-Rate Wireless Networks,” IEEE Std 802.15.4-2015 (Revision of IEEE Std 802.15.4-2011), pp. 1–709, April 2016.

[16] “IEEE Standard for Information technology–Telecommunications andinformation exchange between systems - Local and metropolitan areanetworks–Specific requirements - Part 11: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY) Specifications Amendment2: Sub 1 GHz License Exempt Operation,” IEEE Std 802.11ah-2016(Amendment to IEEE Std 802.11-2016, as amended by IEEE Std 802.11ai-2016), pp. 1–594, April 2017.

[17] A. Laya, C. Kalalas, F. Vazquez-Gallego, L. Alonso, and J. Alonso-Zarate,“Goodbye, ALOHA!,” IEEE Access, vol. 4, pp. 2029–2044, 2016.

[18] G. Chisci, H. ElSawy, A. Conti, M.-S. Alouini, and M. Z. Win, “On thescalability of uncoordinated multiple access for the Internet of Things,” inWireless Communication Systems (ISWCS), 2017 International Sympo-sium on, pp. 402–407, IEEE, 2017.

[19] C. Cano, B. Bellalta, A. Sfairopoulou, and M. Oliver, “Low energyoperation in WSNs: A survey of preamble sampling MAC protocols,”Computer Networks, vol. 55, no. 15, pp. 3351–3363, 2011.

[20] “IEEE Standard for Local and metropolitan area networks–Part 15.4: Low-Rate Wireless Personal Area Networks (LR-WPANs) Amendment 1: MACsublayer,” IEEE Std 802.15.4e-2012 (Amendment to IEEE Std 802.15.4-2011), pp. 1–225, April 2012.

[21] T. Adame, A. Bel, B. Bellalta, J. Barcelo, and M. Oliver, “IEEE 802.11AH: the WiFi approach for M2M communications,” IEEE Wireless Com-munications, vol. 21, no. 6, pp. 144–152, 2014.

[22] N. Sornin, M. Luis, T. Eirich, T. Kramp, and O. Hersent, “LoRaWANSpecification V1.0,” LoRa alliance, 2015. Accessed: 2019-03-01.

[23] A. Tiurlikova, N. Stepanov, and K. Mikhaylov, “Method of AssigningSpreading Factor to Improve the Scalability of the LoRaWan Wide AreaNetwork,” in 2018 10th International Congress on Ultra Modern Telecom-munications and Control Systems and Workshops (ICUMT), pp. 1–4,IEEE, 2018.

[24] E. Khorov, A. Lyakhov, A. Krotov, and A. Guschin, “A survey on IEEE802.11 ah: An enabling networking technology for smart cities,” ComputerCommunications, vol. 58, pp. 53–69, 2015.

[25] E. Kocan, B. Domazetovic, and M. Pejanovic-Djurisic, “Range Extensionin IEEE 802.11 ah Systems through Relaying,” Wireless Personal Com-munications, vol. 97, no. 2, pp. 1889–1910, 2017.

[26] M. Bor, J. E. Vidler, and U. Roedig, “LoRa for the Internet of Things,”2016.

[27] C.-H. Liao, G. Zhu, D. Kuwabara, M. Suzuki, and H. Morikawa, “Multi-hop LoRa networks enabled by concurrent transmission,” IEEE Access,vol. 5, pp. 21430–21446, 2017.

[28] M. Weyn, G. Ergeerts, L. Wante, C. Vercauteren, and P. Hellinckx, “Surveyof the DASH7 alliance protocol for 433 MHz wireless sensor communica-tion,” International Journal of Distributed Sensor Networks, vol. 9, no. 12,p. 870430, 2013.

[29] M. Rahman and A. Saifullah, “Integrating Low-Power Wide-Area Net-works in White Spaces,” in Internet-of-Things Design and Implementation(IoTDI), 2018 IEEE/ACM Third International Conference on, pp. 255–260, IEEE, 2018.

[30] J. Buckley, B. O’Flynn, L. Loizou, P. Haigh, D. Boyle, P. Angove,J. Barton, C. O’Mathuna, E. Popovici, and S. O’Connell, “A novel andminiaturized 433/868 MHz multi-band wireless sensor platform for body

sensor network applications,” in 2012 Ninth International Conference onWearable and Implantable Body Sensor Networks, pp. 63–66, IEEE, 2012.

[31] T. Instruments, “CC1352R SimpleLink™ High-Performance Dual-BandWireless MCU .” http://www.ti.com/lit/ds/symlink/cc1352r.pdf, 2018. Ac-cessed: 2019-03-01.

[32] S. Labs, “EFR32 Flex Gecko Proprietary Wireless SoC.”https://www.silabs.com/products/wireless/proprietary/efr32-flex-gecko-2-4-ghz-sub-ghz, 2018. Accessed: 2019-03-01.

[33] J. Ansari, X. Zhang, and P. Mähönen, “Multi-radio medium access controlprotocol for wireless sensor networks,” in Proceedings of the 5th interna-tional conference on Embedded networked sensor systems, pp. 403–404,ACM, 2007.

[34] M. Wang, Z. Wang, E. Ding, and Y. Yang, “Dual-band sensor networkfor accurate device-free localization in indoor environment with wifi inter-ference,” IEICE TRANSACTIONS on Information and Systems, vol. 98,no. 3, pp. 596–606, 2015.

[35] R. Jurdak, K. Klues, B. Kusy, C. Richter, K. Langendoen, and M. Brunig,“Opal: A multiradio platform for high throughput wireless sensor net-works,” IEEE Embedded Systems Letters, vol. 3, no. 4, pp. 121–124, 2011.

[36] B. O’Flynn, M. De Donno, and W. Magnin, “Multiradio, multiboot capablesensing systems for home area networking,” 2016.

[37] B. Kusy, C. Richter, W. Hu, M. Afanasyev, R. Jurdak, M. Brünig, D. Ab-bott, C. Huynh, and D. Ostry, “Radio diversity for reliable communicationin WSNs,” in Information Processing in Sensor Networks (IPSN), 201110th International Conference on, pp. 270–281, IEEE, 2011.

[38] D. Lymberopoulos, N. B. Priyantha, M. Goraczko, and F. Zhao, “Towardsenergy efficient design of multi-radio platforms for wireless sensor net-works,” in Proceedings of the 7th international conference on Informationprocessing in sensor networks, pp. 257–268, IEEE Computer Society,2008.

[39] M. Ligios, M. T. Delgado, D. Conzon, R. Rossini, F. Sottile, and C. Pas-trone, “Cognitive-based multi-radio prototype for industrial environment,”Annals of Telecommunications, vol. 73, no. 9-10, pp. 665–676, 2018.

[40] Zolertia, “Zolertia RE-Mote platform datasheet.” https://github.com/Zolertia/Resources/blob/master/RE-Mote/Hardware/Revision%20A/Datasheets/ZOL-RMA001%20-%20RE-Mote%20Rev.A%20datasheet%20(draft)%20Dec%202015.pdf, 2018. Accessed: 2019-03-01.

[41] Zolertia, “Zolertia Orion Router datasheet.” https://github.com/Zolertia/Resources/raw/master/Orion%20Ethernet%20Router/Hardware/Revision%20A/Datasheets/ZOL-BO004-A%20-%20Zolertia%20Orion%20Ethernet%20Router%20revision%20A%20Datasheet%20v.1.0.0.pdf,2018. Accessed: 2019-03-01.

[42] OpenMote, “OpenMote main website.” http://www.openmote.com/, 2018.Accessed: 2019-03-01.

[43] Libellium, “Libellium Waspmote datasheet.” http://www.libelium.com/downloads/documentation/waspmote_datasheet.pdf, 2018. Accessed:2019-03-01.

[44] A. Dunkels, “Rime, a lightweight layered communication stack for sen-sor networks,” in Proceedings of the European Conference on WirelessSensor Networks (EWSN), Poster/Demo session, Delft, The Netherlands,Citeseer, 2007.

[45] S. Barrachina-Munoz, B. Bellalta, T. Adame, and A. Bel, “Multi-hopcommunication in the uplink for LPWANs,” Computer Networks, vol. 123,pp. 153–168, 2017.

[46] Raspberry Pi Foundation, “Raspberry Pi 3 Model B+ Product Brief.”https://static.raspberrypi.org/files/product-briefs/Raspberry-Pi-Model-Bplus-Product-Brief.pdf. Accessed: 2019-03-01.

[47] Guangzhou Aosong Electronics Co., Ltd., “Temperature and humid-ity module DHT22/AM2302 Product Manual.” http://akizukidenshi.com/download/ds/aosong/AM2302.pdf. Accessed: 2019-03-01.

[48] Seeed Studio, “Grove - Luminance sensor.” http://wiki.seeedstudio.com/Grove-Luminance_Sensor/. Accessed: 2019-03-01.

[49] A. Dunkels, B. Gronvall, and T. Voigt, “Contiki-a lightweight and flexibleoperating system for tiny networked sensors,” in Local Computer Net-works, 2004. 29th Annual IEEE International Conference on, pp. 455–462,IEEE, 2004.

[50] OSGi Alliance, “The Dynamic Module System for Java™.” https://www.osgi.org/. Accessed: 2019-03-01.

[51] A. Goldsmith, Wireless communications. Cambridge university press,2005.

[52] C. Porras-Amores, F. R. Mazarrón, and I. Cañas, “Study of the verticaldistribution of air temperature in warehouses,” Energies, vol. 7, no. 3,pp. 1193–1206, 2014.

20 VOLUME 4, 2016

Page 21: Increasing LPWAN scalability by means of concurrent

T. Adame et al.: Increasing LPWAN scalability by means of concurrent multiband IoT technologies: An industry 4.0 use case

[53] Texas Instruments, “CC2538 Powerful Wireless Microcontroller System-On-Chip for 2.4-GHz IEEE 802.15.4, 6LoWPAN, and ZigBee Applica-tions.” http://www.ti.com/lit/ds/symlink/cc2538.pdf. Accessed: 2019-03-01.

[54] Texas Instruments, “CC1200 Low-Power, High-Performance RFTransceiver.” http://www.ti.com/lit/ds/symlink/cc1200.pdf. Accessed:2019-03-01.

TONI ADAME received his MSc degree intelecommunications engineering from the Univer-sitat Politècnica de Catalunya (UPC) in 2009.After several years working as an IT presalesconsultant, he joined the Network Technologiesand Strategies (NeTS) group in 2013. His re-search interests are in the area of multiple accesscommunications are Wireless Sensor Networks(WSN), Low-Power Wide-Area Networks (LP-WAN), Medium Access Control (MAC) protocols,

and power-saving mechanisms. He is currently involved in several nationaland European research projects where wireless technologies (cellular, WiFi,WSN, LPWAN, RFID) act as enablers of the Internet of Things (IoT).

ALBERT BEL received his BSc degree intelecommunications engineering (2007), the MScdegree in micro and nanotechnologies (2008), anda PhD in telecommunications and systems engi-neering (2012) from the Universitat Autònomade Barcelona. In 2013 he joined the NetworkTechnologies and Strategies (NeTS) group at theUniversitat Pompeu Fabra, where he has beeninvolved in different international and nationalresearch projects. His research interests are in the

area of wireless sensor networks, mostly focused on the design of local-ization and tracking algorithms and medium access control protocols. FromJanuary 2019, he is a lecturer professor in the Department of Information andCommunication Technologies (DTIC) at Universitat Pompeu Fabra (UPF).

BORIS BELLALTA is an Associate Professor inthe Department of Information and Communica-tion Technologies (DTIC) at Universitat PompeuFabra (UPF). He obtained his degree in Telecom-munications Engineering from Universitat Politèc-nica de Catalunya (UPC) in 2002 and the PhDin Information and Communication Technologiesfrom UPF in 2007. His research interests are in thearea of wireless networks, with emphasis on thedesign and performance evaluation of new archi-

tectures and protocols. The results from his research have been published inmore than 100 international journal and conference papers. He is currentlyinvolved in several international and national research projects, includingthe coordination of the ENTOMATIC FP7 collaborative project. At UPFhe is giving several courses on networking, queueing theory and wirelessnetworks. He is co-designer and coordinator of the interuniversity (UPF andUPC) master’s degree in Wireless Communications.

VOLUME 4, 2016 21