h.26l pre-standard evaluation - faculty.engineering.asu.edu
TRANSCRIPT
H.26L Pre-Standard Evaluation
F. Fitzek, P. Seeling, M. Reisslein∗
acticom GmbH – mobile networksR & D Group
Germany[fitzek|seeling]@acticom.de
Arizona State UniversityDepartment of Electrical Engineering
30. November 2002
Technical Report acticom-02-002
In this report we give an overview of our first evaluations of the upcoming H.26Lvideo encoding standard of the ITU–T. Since this standard is still in the develop-ment phase, all of the results presented here are to be seen as preliminary, as is theintroduction to the standard itself. The video encodings analyzed in this report havebeen generated with a preliminary version of the H.26L encoder. The key charac-teristics of the final H.26L encoder are expected to be very close to the preliminarycoder used in our experiments. The traffic characterisations given in this report givetherefore a very close approximation of the video traffic and quality produced by thefinal encoder. In this report we first outline the current state of the standard, ourmeasurement setup, and give an introduction to the analyzed statistical measures.We then present and interpret the statistical characteristics of the H.26L encodedvideo. We conclude by stating the current problems and outline future work.
∗The work of M. Reisslein is supported in part by the National Science Foundation through Grant No. Career ANI-0133252 and Grant No. ANI-0136774. Any opinions, findings, and conclusions or recommendations expressedin this material are these of the authors and do not necessarily reflect the views of the National ScienceFoundation.
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 1
Contents
1 Introduction 4
2 Video Basics 4
3 The H.26L Standard 5
4 Measurement Setup 10
5 Statistical Results of Video Trace Files 105.1 Frame–based Statistical Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 125.2 GoP–based Statistical Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125.3 Frame Size Traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135.4 Frame Size Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135.5 Autocorrelation Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165.6 R/S Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215.7 Variance Time Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245.8 Periodogram Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6 Conclusion 28
A Carphone 32
B Claire 37
C Container 42
D Foreman 47
E Grandma 52
F Mobile 57
G Mother and Daughter 62
H News 67
I Salesman 72
J Silent 77
K Tempete 82
List of Tables
1 Overview of Evaluated Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 Single frame statistics for different quality levels using Paris . . . . . . . . . . . . 143 GoP statistics for different quality levels using Paris . . . . . . . . . . . . . . . . 15
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 2
4 Single frame statistics for Carphone . . . . . . . . . . . . . . . . . . . . . . . . . 335 GoP statistics for Carphone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 Single frame statistics for Claire . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 GoP statistics for Claire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398 Single frame statistics for Container . . . . . . . . . . . . . . . . . . . . . . . . . 439 GoP statistics for Container . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4410 Single frame statistics for Foreman . . . . . . . . . . . . . . . . . . . . . . . . . . 4811 GoP statistics for Foreman . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4912 Single frame statistics for Grandma . . . . . . . . . . . . . . . . . . . . . . . . . . 5313 GoP statistics for Grandma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5414 Single frame statistics for Mobile . . . . . . . . . . . . . . . . . . . . . . . . . . . 5815 GoP statistics for Mobile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5916 Single frame statistics for MotherDaughter . . . . . . . . . . . . . . . . . . . . . . 6317 GoP statistics for MotherDaughter . . . . . . . . . . . . . . . . . . . . . . . . . . 6418 Single frame statistics for News . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6819 GoP statistics for News . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6920 Single frame statistics for Salesman . . . . . . . . . . . . . . . . . . . . . . . . . . 7321 GoP statistics for Salesman . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7422 Single frame statistics for Silent . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7823 GoP statistics for Silent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7924 Single frame statistics for Tempete . . . . . . . . . . . . . . . . . . . . . . . . . . 8325 GoP statistics for Tempete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 3
1 Introduction
H.26L is currently in the development status and is due to become standard according to theITU–T and ISO/MPEG groups in late 2002. It will then become part of the H.2xx, H.3xx andMPEG standard families. H.26L is expected to replace its predecessors H.261 and H.263, whichwere widely used in integrated circuits for telephone and video equipment for ISDN services overtelephone networks. H.26L has been designed with packet-switched networks in mind and has inits current implementation a complete network adaptation layer. Due to the joint developmentof the ITU and ISO bodies, it is also known as H.264, to furthermore express these joint efforts.The development goal — to reach DVD quality video with data rates of about 1 MBit/s — isalso referred to as “Advanced Video Coding” (AVC). Standardization bodies in Europe, suchas the DVB-Consortium, as well as its American counterpart, the Advanced Television SystemsCommittee (ATSC), are considering to employ H.26L in their respective standards. H.26L is alsowidely viewed as a promising standard for wireless video streaming and is expected to largelyreplace MPEG–4 and H.263+.
Given the expected popularity and widespread use of the new H.26L video encoding stan-dard, the bandwidth demands of H.26L encoded video need to be taken into consideration whendesigning future wired and wireless (e.g., wireless LAN and 3G) networks. It is therefore veryimportant to understand the characteristics of the traffic produced by this new standard. In thisreport we examine the traffic (bit rate) characteristics of video encoded with the H.26L encoder.
We have generated traces which contain the sizes (in byte) of the encoded video frames. Ourvideo traces serve as the basis for our statistical analysis of the H.26L video traffic. The tracesmay also be used by other researchers as a basis for the development of models of the H.26Lvideo traffic. The traces may also be used to evaluate networking protocols and mechanismswith trace–driven simulations of the H.26L video traffic.
This report has four main parts. We first give a brief introduction to the latest proposal forthe H.26L standard. We then describe the general setup of our video trace generation and givea brief statistical evaluation of the traces, followed by a description of the currently existingproblems. We finish with an outlook of our future work.
2 Video Basics
The digitized video is generated by sampling the analog video signal as it is received by theA/D converter hardware. The rate of pictures per second (or frames per second, fps) that isgenerated is different for the two major standards, PAL (Phase Alternation by Line) has 25 fpsand NTSC (National Television Standards Committee) has 30 fps. The main picture formatscurrently used for video compression studies are CIF and QCIF. The CIF picture size is 352columns by 288 lines, the QCIF format is 176x144 (i.e., half the size of CIF in each dimension).The video signal is sampled according to the picture size and with respect to the sensivity char-acteristics of the human eye. In contrast to the RGB format — which generates any color bycombining red, green and blue components — the YUV format combines the luminance compo-nent and the two chrominance components hue and intensity (saturation). Since the eye is farmore sensitive to the luminance level than to coloring information, the YUV formats subsamplethe chrominance information. (Although YUV is often referred to as lossless (or raw) pictureinformation, when sampling into YUV, some chrominance information is lost.) The two mostcommon YUV sampling formats are 4:1:1 as illustrated in Figure 1 and 4:2:0 as illustrated in
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 4
Figure 2. Both formats store one set of hue and intensity samples for four luminance samples(pixels), i.e., 176x144 luminance samples and 88x72 samples for each hue and intensity in case ofa QCIF format frame. The 4:1:1 subsampling format stores one set of hue and intensity samplesfor four luminance samples grouped in a row, whereas the 4:2:0 subsampling format stores oneset of hue and intensity samples for four luminance samples grouped in a rectangle. The 4:2:0format is most commonly used since is has proven to give the best trade–off between samplingefficiency and accuracy.Each sample is typically stored in an 8–bit value. Thus, the size of one YUV frame with 4:2:0
Y Y Y Y U V
Figure 1: YUV 4:1:1 subsampling
YY
Y Y
UV
Figure 2: YUV 4:2:0 subsampling
(or 4:1:1) chrominance subsampling in the QCIF format is
176 · 144 ·(
8 bit +2 · 8 bit
4
)= 304128 bit = 38016 byte. (1)
Similarly, the size of of one YUV frame in the CIF format is
352 · 288 ·(
8 bit +2 · 8 bit
4
)= 1216512 bit = 152064 byte. (2)
The corresponding bit rates, with the NTSC frame rate of 30 frames per second are
Tx,QCIF = 30 Hz · 304128 bit = 9123840 bps ≈ 9.12 Mbps (3)
for QCIF andTx,CIF ≈ 35.5 Mbps (4)
for CIF if no encoding is applied. As is obvious by these rates, compression schemes have to beemployed in order to achieve data rates suitable for transmission over wireless networks.
3 The H.26L Standard
We used a development version of the future H.26L standard implementation software, the JMrev. 2 (dated April 11th, 2002) in our experiments. Since the standard is not yet final anddiscussions concerning the header and other parts of the encoded video bitstream are currentlyongoing, we will focus on the main coding algorithm. We leave out the other details (such asheader formats) for further examination when the standard has been adopted. The upcomingH.26L standard differs from its predecessors (the ITU–T H.26x video standard family and theMPEG standards MPEG–2 and MPEG–4) in providing a high compression video coding layer(VCL) for storage optimization as well as a network adaption layer (NAL) for the packetization of
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 5
Data Partitioning
Video Coding Layer
Co
ntr
ol D
ata
...H.320 H.324x H.323/IP
Network Adaption Layer
Slice / Partition
Macroblock
Figure 3: Block diagram of an H.26L coder.
the encoded bitstream according to transmission requirements [14]. An overview of these layersis given in Figure 3. The network adaption layer will be left out in the following discussion, sinceits functionality will vary according to the underlying network type (e.g. 802.3, 802.11x, UMTS,and others) and will not be a subject of our evaluations. The encoded bitstream will thereforenot be sliced (partitioned) further but remain a single sequence. (Slices represent independentcoding units that can be decoded without referencing other slices of the same frame. Theyconsist typically of several consecutive macroblocks. Slicing can be utilized to achieve a highererror–robustness.)
The standard is based on a block–oriented and motion–compensating hybrid transformationprocess. Similar to other video coding standards [14, 15, 6], the standard will only specify thedecoding process to allow for maximum customization possibilities in the encoding routines. Theencoding is done on a macroblock level. Each CIF format picture is subdivided into 18 (lines)× 22 (rows) macroblocks, for a total of 396 macroblocks. As illustrated in Figure 4, each QCIFformat picture is subdivided into 9 × 11 macroblocks for a total of 99 macroblocks.
11
9
Figure 4: Sample QCIF image layout.
As noted above, compression is needed in order to enable efficient video transmission overdata networks. Several types of redundancy can be exploited to achieve compression. The mostcommonly exploited redundancy is the temporal interdependence of consequtive video frames,
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 6
which typically leads to the highest achievable compression gains. (Additional compressionschemes are also in development, but not commonly applied up to now, such as the exploitationof object recognition techniques.) There are three methods for encoding the original pictures: I(Intra), P (Inter), and B (Bi–directional). These encoding methods are applied on the macroblocklevel. An intra–coded frame consists exclusively of intra-coded macroblocks. Thus, an intra–coded frame contains the compressed image information (without any prediction information),resulting in a large frame size (compared to the size of the inter– or bidirectional–coded frames).Intra–coding uses well-known compression schemes such as JPEG or wavelet–based approachesto compress the image information.
The inter–coded frames use a motion estimation relying on the previous inter– or intra–codedframe, whereas the bi–directional encoded frames rely on a previous as well as a following intra–oder inter–coded frame. This prediction information results in smaller frame sizes for the P–frames and even smaller frame sizes for the B–frames. The relationship between the encodingtypes and how frames rely on each other in a typical frame sequence [5] is illustrated in Figure 5.(Note that when B frames do not have any following I– or P–frames they can be referenced to,no encoding or decoding is possible, as illustrated in Figure 5.) The sequence of frames between
Figure 5: Typical frame sequence and dependencies for one GoP
the intra–coded frames is referred to as Group of Pictures (GoP). (Note that it is not necessaryto have more than one I–frame at the beginning of the video sequence, in which case the entireframe sequence is a single GoP.)
To handle abrupt changes in the bitstream and the loss of parts of pictures or structures,the H.26L standard provides the possibility of refreshing the pictures on a macroblock level.Additionally, refresh frames (intra picture refresh) are used to stop the prediction process offrames that are referencing lost or errorneous frames. Furthermore, the standard will providethe possibility to switch between several different bitrate streams to avoid high computationaleffort (and thus high power consumption)for the encoding and decoding. In order to providequantized values, an inverse discrete cosine transformation (IDCT) is utilized. The IDCT inH.26L is performed in the same manner as in H.263 [17]. The representation of the image in afinite numberspace done by the quantization based on the IDCT coefficients is the main reason forlosses and compression. The quantization parameter defines the fidelity of the picture encoding,since the smaller the quantization parameter, the more values are available to express the valueof each coefficient resulting from the transformation.
The H.26L standard takes the characteristics of wireless environments where the availablebitrate may change often and over larger ranges into account. The value of the quantizationparameter can be changed on a frame–by–frame basis as well as on a macroblock–by–macroblock
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 7
basis. This — in addition to the stream switching functionality — allows for a fast response of thereal–time encoding process to changing bandwidths. The stream switching functionality allowsfor non–realtime encoding and real–time, bandwidth–based selection of streams encoded withdifferent quantization and/or GoP settings. In this paper we will not perform any evaluations ofthese advanced features. Instead, we focus on the non–real–time behavior of the standard andassume that the encoder has no information about the underlying channel charactersistics.
After quantizing the IDCT coefficients, the temporal redundancy of the picture informationis removed by applying motion estimation. This is especially useful for high frame rates, wheresuccessive frames are highly correlated. The motion estimation is performed for multiple referenceframes (see H.263++ standard, Annex U – long term memory prediction) and works beyond thepicture boundaries as illustrated in Figure 6.
B frame
2
I or P frame
3
I or P frame
1
� �� �� �� �� �� �� �� �
� �� �� �� �� �� �� �� �
� �� �� �� �� �� �� �� �
� �� �� �� �� �� �� �� �
Complete Movement
4
Backward motionestimation
Forward motionprediction
� � �� � �� � �� � �� � �� � �� � �
� �� �� �� �� �� �� �
� � �� � �� � �� � �� � �� � �� � �
� � �� � �� � �� � �� � �� � �� � �
� � �� � �� � �� � �� � �� � �� � �
Figure 6: Illustration of motion estimation
Motion Compensation is utilizing so–called motion vectors to exploit the temporal redundancymore efficiently in order to gain higher compression. A motion vector uses reference frames (orfields of frames) in the past and/or future. This two–dimensional vector provides an offset fromthe coordinate in the current picture to the coordinates in the referenced frame. For illustration,the fourth frame in Figure 6 represents an already identified moving object with its full path. Thefirst to third frames are illustrating the motion vector generation. For a backward prediction, anearlier reference is used to derive the coordinate change, whereas for forward prediction, a laterreference is utilized. For the different encoding types, different prediction modes are implemented.The enhancement of normal motion vectors is the revocation of picture boundaries as limits forthe validity of a vector’s target, also known as unrestricted or extended motion vector mode.Frame three in Figure 6 gives such an example. Since there is no content and thus data availablefor the outside of a picture, the pixels at the border are simply replicated to fill the nonexistentvalues needed as references. Figure 7 illustrates this scheme.
Each macroblock can be subdivided into smaller fragments in order to provide a finer granu-
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 8
Repeated edgepixels
Current frame Next reference frame
Figure 7: Illustration of unrestricted motion estimation.
larity and higher quality. The different subdivision formats are illustrated in Figure 8.
Mode 1 Mode 3 Mode 4
Mode 5 Mode 6 Mode 7
Mode 2
Figure 8: Different macroblock subdivision modes.
After the discrete cosine transformation and motion estimation are completed, some redun-dancy is typically still left in the video data. This remaining redundancy can be exploited byentropy–coding techniques such as the universal variable length coding (UVLC) or the context–adaptive binary arithmetic coder (CABAC) [17]. The latter approach uses probability distri-butions to further reduce the space needed to store the encoded frame. Shorter symbols areassigned to bit patterns with a high probability of occurence and longer symbols to bit patternswith a smaller probability of occurence. This mapping process achieves lossless compression.The UVLC uses an infinite set of code words and is applied only on the mapping of the symbolsand thus reduces the necessity of redefining codewords [4]. The coding is based on a single, statictable of codewords which results in a simple mapping process. As an alternative to the UVLC,the CABAC–technique can be used. This algorithm encodes the sequence of symbols into aninterval of real numbers between 0 and 1 and is able to do this with respect to the symbol’s prob-ability at the source. It is therefore exploiting additional correlation of symbols at the encodingside for further reduction of data to be stored for each frame.
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 9
4 Measurement Setup
We used the reference JM2–encoder version 3.6 which is publicly available (for more recentreleases refer to [13]). This reference encoder conforms to the current standard developmentand includes the currently proposed features. Nevertheless, as changes are ongoing, the softwaremay lack the most recently adopted features. Since the purpose of our study is to generate andstatistically evaluate the frame sizes of the encoded videostreams, we disabled some of the moreadvanced encoder features.
The disabled features included the slice mode that is providing error resilience features bycoding fixed macroblocks or fixed bytes per slice. A slice is an individual entity no relyingon other data inside a frame. We also used only the CABAC–technique to remove intersymbolcorrelation. The network adaption layer was also not used, as were restriction to the search range.We were therefore only using the basic features such as inter–, intra–, and bidirectional predictionand motion estimation. Additionally, we used a fixed GoP and motion–vector resolution settingfor the prediction modes. The result is a setup being very close to the most basic encodingsettings used in previous video trace file generation processes such as [5].
Overall, we believe that the differences between the encoder version used in our experimentsand the final encoder version are negligible as far as the video traffic characterization is concernedfor these basic settings.
We did not specify a target bit rate, since rate–adaptive encoding is not available at present.Instead, we used static quality levels (quantization parameters) which we set for all three frametypes to 1, 5, 10, 15, 20, 25, 30, 35, 40, 45, and 51. For ease of comparison with the al-ready existing video trace files (H.261, H.263, and MPEG–4, see [5]) we used the GoP structureIBBPBBPBBPBB. Note that the encoder has to encode the referenced frames first, thus theresulting frame sequence is IPBBPBBPBBIBBP....
We initially used the freely available and widely used YUV testing sequences [16] in ourexperiments. These sequences are in the NTSC format and have a frame rate of 30 frames persecond.
For each of the studied quality levels we encoded the YUV–files into the H.26L bitstreamoff–line (thus there was no frame drop during the encoding).
The encoder status output was parsed to generate the traces.For each quantization level and test sequence we generated a terse and a verbose trace file, as
illustrated in Figure 9. The traces were then used for the statistical analysis of the video traffic.(We found that working with the encoded file itself is yet not very advisable, see discussion onproblems encountered) The verbose trace gives for each frame the type (I, P, or B), the playouttime (= frame number/30) in msec, and the frame size in byte. The terse trace gives only thesequence of frame sizes in byte as generated by the encoder IPBBPBBPBBIBBP.....
We used bytes instead of bits, since every frame start code has to be byte–aligned in the H.26Lbitstream. Note that in the encodings the last GoP is incomplete, since the last two B–framesare referencing a frame that is not available. The traces and the statistics are publicly availableon our website [1] for viewing and downloading.
5 Statistical Results of Video Trace Files
The following results are only intended to give a first impression of the capabilities of the futureH.26L standard. The longest testing sequence we were able to utilize has 1000 pictures. T he
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 10
configuration file(quantization)
encoder
input YUV files
trace file
evaluation
encoded bitstream
Figure 9: Setup of the evaluation process
following detailed discussion focuses on this video, called Paris, since the statistical evaluationof the shorter sequences is far more prone to residual errors and inconsistencies. A screenshotgiving an impression of the content (discussion with some movements of bodies and ball) of theParis seqeunce is given in Figure 10 We provide the results for the other video traces with aquantization parameter setting of 25 in abbreviated form in the corresponding appendeces. Anoverview of the evaluated sequences is given in Table 1. For the statistical evaluation of thetraces we introduce the following notation.
Tables 2 and 3Let N denote the number of considered frames, in case of Paris this would be N = 1000. The
individual frame sizes are denoted by X1, . . . , XN . The mean frame size is estimated as
X =1N·
N∑i=1
Xi. (5)
(6)
The variance is estimated as
S2X =
1N − 1
·N∑
i=1
(Xi −X
)2(7)
=1
N − 1
N∑i=1
X2i −
1N·(
N∑i=1
Xi
)2 . (8)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 11
Figure 10: Screenshot of Paris in CIF format
The Coefficient of Variation is given by
CoV =SX
X. (9)
5.1 Frame–based Statistical Overview
Table 2 provides an overview of the basic statistics of the Paris traces for the different quanti-zation parameter settings.
5.2 GoP–based Statistical Overview
We also evaluated the traces at an aggregation level of 12 frames, i.e., at the GoP level, seeTable 3. This fixed–length moving average analysis gives a more stationary impression of thevideo trace since the frame type differences are smoothed out.
In the folowing sections, we give some graphical representations of the frame size traces, thedistribution, the autocorrelation function, and the R/S plots. The R/S plots are used to findthe Hurst parameters for the three quantization settings ql = 1, 25, 51 that are stated in eachtitle as QPql. The main usage and conclusion of these plots will be described in the followingpart of the evaluation. These figures should give a graphical overview of some characteristicalbehavior (e.g. the long time dependency of the frame trace) which is regarded as a time seriesin statistical means.
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 12
Table 1: Overview of Evaluated SequencesName of Video Sequence Number of Frames FormatCarphone 382 QCIFClaire 494 QCIFContainer 300 QCIFForeman 400 QCIFGrandma 870 QCIFMobile 300 CIFMother and Daughter 961 QCIFNews 300 QCIFParis 1000 CIFSalesman 449 QCIFSilent 300 QCIFTempete 260 CIF
5.3 Frame Size Traces
The main purpose for evaluating the behavior of the frame sizes is to achieve a statistically soundbase for the modeling and simulation of video traffic. The frame sizes reflect the video contentand its dynamic behavior. With any block– and motionvector–based encoding process, the framesizes are larger if the movie content is more dynamic and richer in texture. As can be seen in theframe traces of Carphone, the frame size is rising around frame 150. This is due to a shift of thelandscape in the back, viewable through the car window. Before, the view is a clear sky, onlyoccasionally interrupted by moving objects (e.g. lanterns, street signs) — after frame 150, theview is a forest, with a rich texture. Figure 11 gives an impression on the changing backgroundsand the resulting frame sizes for a GoP–aggregation.
Furthermore, the frame sizes are larger when smaller quantization parameters are used (whichin turn give higher video quality). These factors are interdependent, i.e., high dynamics pairedwith finer quantization results in larger frame sizes, and vice versa. We observe from Figures 12,13, and 14, that the range of frame sizes is extremely different within a given GoP.
The GoP–smoothed traces in Figures 15, 16, and 17 give a clearer impression of the trafficdynamics. We observe that the plots do not indicate any large dynamic change. This is becausethe used tesing sequences typically have only little dynamic change in their content. In fact,these testing sequences are typically employed to study video encoding at the time scale of avideo frame or smaller. The study of the impact of dynamic changes of the video content on thevideo traffic requires longer test videos, which we will study in future work. A clear observationfrom the figures is that frame sizes are larger for smaller quantization parameters
5.4 Frame Size Distribution
The distribution of the frame sizes is needed in order to make any statistical modeling of thetraffic possible. Frame size histograms or probability distributions allow us to make observationsconcerning the variability of the encoded data and the necessary requirements for the purposeof real–time transport of the data over a combination of wired and wireless networks. In thefollowing we present the probability density function p as a function of the frame size. For the
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 13
Tab
le2:
Sing
lefr
ame
stat
isti
csfo
rdi
ffere
ntqu
ality
leve
lsus
ing
Par
is
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
43390
29037
12061
3930
1288
418
119
67
29
25
16
[byte
]
Xm
ax
95525
81139
62578
46474
33824
23919
15746
9840
5763
3214
1448
[byte
]
X54345.2
839572.0
522062.5
111395.9
46331.7
23827.6
92145.8
61182.1
0647.4
0348.3
4161.3
0[b
yte
]
XI−
fr
am
e94414.2
480066.8
361447.8
545408.6
132699.1
822964.6
214945.3
29248.7
05445.1
03035.9
01377.3
5[b
yte
]
XP−
fr
am
e58793.2
743068.2
924494.7
011899.6
05928.8
83337.2
21748.5
1854.7
6421.3
8208.5
864.7
5[b
yte
]
XB−
fr
am
e47621.8
733152.2
016182.0
16916.9
93157.3
11598.1
4680.6
7287.5
7127.1
361.8
344.1
6[b
yte
]
S2 X
174399635.3
4172302917.0
0158433906.5
9112674047.0
365962417.2
734444608.4
115328149.0
66054760.3
72134816.6
5668818.2
7136146.8
3
CoV
0.2
40.3
30.5
70.9
31.2
81.5
31.8
22.0
82.2
62.3
52.2
9
Mean
bit
rate
13042866.9
69497292.4
85295003.3
62735025.3
61519611.8
4918646.3
2515007.1
2283704.0
0155376.2
483601.8
438711.0
4[b
it/s]
Peak
bit
rate
25473333.3
321637066.6
716687466.6
712393066.6
79019733.3
36378400.0
04198933.3
32624000.0
01536800.0
0857066.6
7386133.3
3[b
it/s]
Peak
to
mean
1.9
52.2
83.1
54.5
35.9
46.9
48.1
59.2
59.8
910.2
59.9
7
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 14
Tab
le3:
GoP
stat
isti
csfo
rdi
ffere
ntqu
ality
leve
lsus
ing
Par
is
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P93578
79234
60610
44535
31944
22312
14461
8896
5273
2955
1348
[byte
]
Xm
ax
,Go
P721737
539976
324552
179212
101868
61343
33712
17741
9496
5035
2362
[byte
]
XG
oP
650782.2
3473844.2
3264149.3
0136385.8
275734.0
545763.5
825640.1
014112.8
27725.0
24154.1
81924.1
1[b
yte
]
S2 X
,Go
P512264567.3
5432694471.1
8366798439.6
5199274339.9
374414914.8
026030350.1
26610073.9
41420858.6
9319171.8
860188.6
410792.1
5
CoV
Go
P0.0
30.0
40.0
70.1
00.1
10.1
10.1
00.0
80.0
70.0
60.0
5
Mean
GoP
rate
14461.8
310529.8
75869.9
83030.8
01682.9
81016.9
7569.7
8313.6
2171.6
792.3
242.7
6[b
it/s]
Peak
GoP
rate
16038.6
011999.4
77212.2
73982.4
92263.7
31363.1
8749.1
6394.2
4211.0
2111.8
952.4
9[b
it/s]
Peak
to
mean
1.1
11.1
41.2
31.3
11.3
51.3
41.3
11.2
61.2
31.2
11.2
3
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 15
0
200
400
600
800
1000
1200
1400
1600
1800
2000
0 50 100 150 200 250 300 350
fram
e si
ze [
byte
]
frame index
Frame Size Trace of Carphone_QP25 − Aggregation 12
Figure 11: Impact of changing background dynamics on frame sizes.
probability distribution function as well as the inverse probalbility distribution function, we referto our web page [1]. We observe for all the different quality levels a large spread of the frame sizes.We observe that the distribution is spreading out more for smaller quantization parameters.
This is expectedly derived by comparing the differences in the frame sizes for the differentframe types (which normally tend to be high for I–frames, intermediate for P–frames, and lowfor B–frames). With lower fidelity (i.e. higher quantization), the differentiation between thesetypes regarding the frame size is decreasing due to the more forcefully applied quantization. Theviewable result is characterized by a total loss of clear differences between objects, colors andso forth. Figures 18, 19, and 20 give an overview of these quantization effects (please note thatthese images were scaled down to fit on a single page).
The overall distribution may very roughly be seen as normal or Gaussian, what should beeasing future modeling efforts. A short warning here, again, with respect to the length of thetraces evaluated up to now.
5.5 Autocorrelation Coefficient
The autocorrelation [3] function can be used for the detection of non–randomness in data oridentification of an appropriate time series model if the data are not random. One basic assump-tion is that the observations are equi-spaced. The autocorrelation is a correlation coefficient andthus referred to as autocorrelation coefficient (acc). However, instead of the correlation betweentwo different variables, the correlation is between two values of the same variable at times Xt
and Xt+k. When the autocorrelation is used to detect non-randomness, it is usually only thefirst (lag k = 1) autocorrelation that is of interest. When the autocorrelation is used to identifyan appropriate time series model, the autocorrelations are usually plotted for a range of lags k.
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 16
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
0 200 400 600 800 1000
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Paris_QP01 - Aggregation 1
Figure 12: Frame size trace of Paris (quantization 01)
0
5000
10000
15000
20000
25000
0 200 400 600 800 1000
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Paris_QP25 - Aggregation 1
Figure 13: Frame size trace of Paris (quantization 25)
0
200
400
600
800
1000
1200
1400
1600
0 200 400 600 800 1000
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Paris_QP51 - Aggregation 1
Figure 14: Frame size trace of Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 17
0
10000
20000
30000
40000
50000
60000
70000
0 200 400 600 800 1000
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Paris_QP01 - Aggregation 12
Figure 15: Frame size trace of Paris averaged over one GoP (quantization 01)
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
0 200 400 600 800 1000
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Paris_QP25 - Aggregation 12
Figure 16: Frame size trace of Paris averaged over one GoP (quantization 25)
0
50
100
150
200
250
300
350
400
0 200 400 600 800 1000
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Paris_QP51 - Aggregation 12
Figure 17: Frame size trace of Paris averaged over one GoP (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 18
Figure 18: Quantization effect for Paris (quantization 40, PSNR for this frame: 27.4271)
Figure 19: Quantization effect for Paris (quantization 45, PSNR for this frame: 24.2853)
Figure 20: Quantization effect for Paris (quantization 51, PSNR for this frame: 20.3898)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 19
0.001
0.0012
0.0014
0.0016
0.0018
0.002
0.0022
40000 50000 60000 70000 80000 90000 100000
p
frame size [byte]
Probability Density Function (p) for Paris_QP01
Figure 21: Frame size distribution for Paris (quantization 01)
0.001
0.0015
0.002
0.0025
0.003
0.0035
0.004
0.0045
0 5000 10000 15000 20000 25000
p
frame size [byte]
Probability Density Function (p) for Paris_QP25
Figure 22: Frame size distribution for Paris (quantization 25)
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0 200 400 600 800 1000 1200 1400 1600
p
frame size [byte]
Probability Density Function (p) for Paris_QP51
Figure 23: Frame size distribution for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 20
With our notation the acc can be estimated by
ρX(k) =1
N − k·
N−k∑i=1
(Xi −X
)·(Xi+k −X
)S2
X
(10)
In 10 the lag (i.e. for either single frames or aggregated for one or multiple GoPs) is denotedas k, with k = 0, 1, . . . , N . The autocorrelation function for the single frame aggregation levelshows the similarity within a GoP, whereas higher aggregation levels give an indication of thelong–term self–similarity. We observe from Figures 24, 25, and 26, that there large spikes spaced12 frames apart. These are due to repetive GoPs, which contain 12 frames each. Thus for a lag of12 frames, I frames correlate with I frames, P frames with P frames, and B frames with B frames.The intermediate spikes that are spaced three frames apart are due to the correlations betweenI and P frames. We observe that the intermediate spikes are decreasing with the fidelity of theencoded bitstream. This appears to be due to the wider spread of the frame size distribution forlarger quantization parameters.
We observe from Figures 27, 28, and 29 that the GoP–based autocorrelation tends to fall offslower than an exponential, suggesting the presence of long-range dependencies.
5.6 R/S Plots
The Hurst parameter, or self–similarity parameter, H, is a key measure of self-similarity [7, 8].H is a measure of the persistence of a statistical phenomenon and is a measure of the lengthof the long range dependence of a stochastic process. A Hurst parameter of H = 0.5 indicatesabsence of self-similarity whereas H = 1 indicates the degree of persistence or a present long–range dependence. The H parameter can be estimated from a graphical interpolation of theso–called R/S plot. The R/S plot gives the graphical interpretation of the rescaled adjustedrange statistic by utilizing the following method [2, 9].The length of the complete series N has to be subdivided into blocks with a length of k, forwhich the partial sums Y (k) have to be calculated as in Equation 11. Following the varianceof all these aggregations has to be calculated. The resulting R/S value is derived as shown inEquation 13 for a single block.
Y (k) =k∑
i=1
Xi (11)
S2X(k) =
1k·
k∑i=1
[X2
i −(
1k
)2
· Y (k)2]
(12)
R
S(N) =
1SX(k)
[max0≤t≤k
(Y (t)− t
k· Y (k)
)−min0≤t≤k
(Y (t)− t
k· Y (k)
)](13)
If plotted on an log/log scale for R/S versus differently sized blocks, the result will be severaldifferent points. This plot is also called the pox plot for the R/S statistic. The Hurst parameterH can then be estimated by fitting a line to the points of the plot, normally by using a leastsquare fit, neglecting the residual values at the lower and upper borders ( since those are typicallytransient zones that represent the short rage dependencies, which exist on a GoP–level as studiedearlier). The larger the resulting Hurst parameter, the higher the degree of long range dependencyof the time series. As can be seen from the following Figures 30, 31, 31 on a single–frame basis,
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 21
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12 14 16
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Paris_QP01
Figure 24: Autocorrelation coefficients for Paris (quantization 01)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12 14 16 18
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Paris_QP25
Figure 25: Autocorrelation coefficients for Paris (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Paris_QP51
Figure 26: Autocorrelation coefficients for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 22
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12 14 16
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Paris_QP01
Figure 27: GoP autocorrelation coefficients for Paris (quantization 01)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12 14 16 18
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Paris_QP25
Figure 28: GoP autocorrelation coefficients for Paris (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6 7 8 9
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Paris_QP51
Figure 29: GoP autocorrelation coefficients for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 23
and 33, 34, 35 on a GoP–basis, the hurst parameters stay well above 0.5, reflecting the presenceof long–term dependence.
We applied the 4σ–test [11] to eliminate all outlying residuals for a better estimation of thehurst parameter.
5.7 Variance Time Plot
The variance time plot is applied to a time series to show the development of the varianceas in Equation 8 over different aggregation levels. This provides another test for long–rangedependency [10, 5, 8, 2]. It is furthermore used to derive an estimation of the Hurst parameter.In order to obtain the plot, the normalized variance as given in Equation 14 of the trace isplotted as a function of different aggregation levels k of the single frame sizes in a log–log plot.For each agggregation level k the total amount of frames N is divided into blocks and the variancecalculated as shown before in Equations 11 and 12.
Snorm =S2
X(k)S2
X
(14)
If no long range dependency is present, the slope of the function would be −1. For slopes largerthan −1, a dependency is present. For simple reference we plot a reference line with a slope of−1 in the figures. We did not apply any regression–fits up to now but plan to do so in the future.
Our plots in Figures 36, 37, and 38 indicate a certain degree of long term dependency since theestimated slope is less than −1. We estimate that this is due to the occurence of the I–framesevery 12 frames.
5.8 Periodogram Plot
A periodogram is a graphical data analysis technique for examining frequency–domain modelsof an equi–spaced time series. The periodogram is the Fourier transform of the autocovariancefunction. This calculation is currently employed in measuring the spectral density, followingthe idea that this spectrum is actually the variance at a given frequency. Therefore additionalinformation about the magnitude of the variance of a given time series can be obtained byidentifying the frequency component. This is done by correlating the series against the sine/cosinefunctions, leading to the Fourier frequencies [12]. For the calculation of the periodogram plot,the frame sizes xi of N frames are aggregated into equidistant blocks k. For each block, themoving averages and their according logarithms are calculated as in Equations 15 and 16 withn = 1, . . . , N/k.
Y (k)n =
1k·
Nk∑
i=1
xi (15)
Z(k)n = log10Y
(k)n (16)
In order to determine the frequency part of the periodogram, we calculate λk as in Equation 17.The periodogram itself is then derived as given in Equation 18. For each different aggregationlevel, we plot the resulting I(λk) and λk in a log/log–plot.
λk =2πiNk
, i = 1, . . . ,M − 1
2(17)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 24
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
log(
R/S
)
log(d)
R/S Plot for Paris_QP01 (H=0.474323708362334)
Figure 30: Single Frame R/S plot and H parameter for Paris (quantization 01)
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
log(
R/S
)
log(d)
R/S Plot for Paris_QP25 (H=0.335127777077995)
Figure 31: Single Frame R/S plot and H parameter for Paris (quantization 25)
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
log(
R/S
)
log(d)
R/S Plot for Paris_QP51 (H=0.168249895272951)
Figure 32: Single Frame R/S plot and H parameter for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 25
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
log(
R/S
)
log(d)
R/S Plot for Paris_QP01 (H=0.842531505526994)
Figure 33: GoP R/S plot and H parameter for Paris (quantization 01)
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
log(
R/S
)
log(d)
R/S Plot for Paris_QP25 (H=0.795946215846184)
Figure 34: GoP R/S plot and H parameter for Paris (quantization 25)
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
log(
R/S
)
log(d)
R/S Plot for Paris_QP51 (H=0.685029343419222)
Figure 35: GoP R/S plot and H parameter for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 26
-2
-1
0
1
2
3
4
1 1.2 1.4 1.6 1.8 2 2.2 2.4
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Paris_QP01
Figure 36: Variance time plot for Paris (quantization 01)
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
1 1.2 1.4 1.6 1.8 2 2.2 2.4
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Paris_QP25
Figure 37: Variance time plot for Paris (quantization 25)
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
1 1.2 1.4 1.6 1.8 2 2.2 2.4
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Paris_QP51
Figure 38: Variance time plot for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 27
I(λk) =1
2π Nk
·
∣∣∣∣∣∣∣Nk−1∑
l=0
Z(a)l · e−jlλk
∣∣∣∣∣∣∣2
(18)
The resulting plots are shown in Figures 39, 40, and 41 for a single frame aggregation andFigures 42, 43, and 44 for an aggregation level of a single GoP.
The Hurst paramter is estimated as H = (1 − β1)/2, using a least squares regression onthe samples. The Equations 20 and 21 are applied to determine the slope of the fitted liney = β0 + β1x and H.
K = 0.7 ·Nk − 2
2(19)
β1 =K ·
∑Ki=1 xiyi −
(∑Ki=1 xi
)·(∑K
i=1 yi
)K ·
(∑Ki=1 x2
i
)−(∑K
i=1 yi
)2 , (20)
β0 =∑K
i=1 yi − β1 ·∑K
i=1 xi
K(21)
6 Conclusion
In this report we have reported on our pre-standard evaluation of the H.26L video compressionstandard. Although the standard is not finalized as of the writing of this report and somechanges in the algorithms employed as well as the output format generated are possible, it isexpected that the basic routines utilized in our study will not change. As a consequence, ourtraffic characterizations give very close approximations of the final H.26L standard.
In the ongoing research of H.26L video compression we want to make comparisons with thecurrently utilized standard video encoding formats according to compression and statistical be-havior. Once the final H.26L standard has been approved by the ITU–T, we will encode full–length movies to make more appropriate and consistent evaluations. The currently evaluatedsequences do lack length and differ in content from what is expected to be a typical movie thatwould be transmitted over wireless networks in the future. Therefore the presented results willlikely differ from what could be expected by a more lengthy evaluation. Since dynamic behaviorand content are correlated, we will examine different categories such as action, comedy, animated,and news. Additionally, we want to expand the statistical evaluation further with additional longterm dependency evaluations and outlier elimination.
References
[1] acticom GmbH. Publicly available research results. http://www.acticom.info. 10, 16
[2] J. Beran. Statistics for Long–Memory Processes. Chapman and Hall, London, 1994. 21, 24
[3] Box, G. E. P., and Jenkins, G. Time Series Analysis: Forecasting and Control. Holden-Day,1976. 16
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 28
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
-2.5 -2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Paris_QP01 Agg. 1 (H=0.605923440415645)
Figure 39: Single frame periodogram plot for Paris (quantization 01)
-6
-5
-4
-3
-2
-1
0
1
-2.5 -2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Paris_QP25 Agg. 1 (H=0.702824720266338)
Figure 40: Single frame periodogram plot for Paris (quantization 25)
-6
-5
-4
-3
-2
-1
0
1
-2.5 -2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Paris_QP51 Agg. 1 (H=0.585729792652431)
Figure 41: Single frame periodogram plot for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 29
-3.8
-3.7
-3.6
-3.5
-3.4
-3.3
-3.2
-3.1
-3
-2.9
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Paris_QP01 Agg. 12 (H=0.468869712079431)
Figure 42: Single GoP periodogram plot for Paris (quantization 01)
-3.4
-3.2
-3
-2.8
-2.6
-2.4
-2.2
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Paris_QP25 Agg. 12 (H=0.478428776597514)
Figure 43: Single GoP periodogram plot for Paris (quantization 25)
-3.1
-3
-2.9
-2.8
-2.7
-2.6
-2.5
-2.4
-2.3
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Paris_QP51 Agg. 12 (H=0.41808930209482)
Figure 44: Single GoP periodogram plot for Paris (quantization 51)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 30
[4] Kristofer Dovstam. Video Coding in H.26L Video Coding in H.26L. PhD thesis, 2000. 9
[5] Frank H.P. Fitzek and Martin Reisslein. MPEG–4 and H.263 Video Traces for NetworkPerformance Evaluation. Technical report, Technical University of Berlin, 2000. TKN–00–06. 7, 10, 24
[6] Bernd Girod and Niko Farber. Compressed Video Over Networks, chapter Wireless Video.November 1999. 6
[7] H.E. Hurst. Long–Term Storage Capacity of Reservoirs. Proc. American Society of CivilEngineering, 76(11), 1950. 21
[8] Will E. Leland, Murad S. Taqq, Walter Willinger, and Daniel V. Wilson. On the self-similarnature of Ethernet traffic. In Deepinder P. Sidhu, editor, ACM SIGCOMM, pages 183–193,San Francisco, California, 1993. 21, 24
[9] A.W. Lo. Long–term Memory in Stock Market Prices. Economatria, (59):1276–1313, 1991.21
[10] P. Morin. The impact of self-similarity on network performance analysis, 1995. 24
[11] Sachs, Lothar. Angewandte Statistik. Springer–Verlag, 2002. 24
[12] Stoffer David S. Shumway, Robert H. Time Series Analysis and Its Applications. Springer,New York, 2000. 24
[13] Suehring, Carsten. H.26l software coordination. http://bs.hhi.de/ suehring/tml/. 10
[14] Wiegand Thomas Sullivan, Garry J. and Thomas Stockhammer. Using the Draft H.26LVideo Coding Standard for Mobile applications. 6
[15] D. S. Turaga and T. Chen. Fundamentals of Video Coding: H.263 as an example CompressedVideo over Networks, chapter Fundamentals of Video Coding: H.263 as an example. TheSignal Processing Series. MarcelDekker, 2000. 6
[16] Wenger, Stephan. The tml project web-page and archive. http://kbs.cs.tu-berlin.de/ stewe/vceg/. 10
[17] Th. Wiegand. H.26L Test Model Long–Term Number 9 (TML-9) draft0. ITU-T StudyGroup 16, Dec. 2001. 7, 9
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 31
Statistical Results for Other Reference Videos
A Carphone
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
0 50 100 150 200 250 300 350
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Carphone_QP25 - Aggregation 1
Figure 45: Frame size trace for Carphone (quantization 25)
0
200
400
600
800
1000
1200
1400
1600
1800
2000
0 50 100 150 200 250 300 350
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Carphone_QP25 - Aggregation 12
Figure 46: Frame size trace for one GoP Carphone (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 32
Tab
le4:
Sing
lefr
ame
stat
isti
csfo
rC
arph
one
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
11263
7497
2986
1026
318
134
43
20
18
13
13
[byte
]
Xm
ax
22367
18807
14005
10059
6986
4751
2977
1730
991
541
243
[byte
]
X15030.9
411131.9
76398.5
43525.1
71949.0
41094.1
4563.9
7288.2
5148.5
579.8
940.8
1[b
yte
]
XI−
fr
am
e20617.1
217027.6
912166.5
98438.1
65760.3
43831.9
12350.2
51392.0
9807.2
5460.1
2215.4
4[b
yte
]
XP−
fr
am
e15991.0
711868.8
66993.2
23934.0
22211.0
41243.7
5630.7
3315.6
8158.8
081.0
530.7
6[b
yte
]
XB−
fr
am
e13964.2
910110.6
95447.0
92751.6
91369.8
6692.6
8313.6
9138.8
161.6
931.5
522.6
1[b
yte
]
S2 X
5764401.4
45835062.7
35447232.4
83609468.0
11967678.9
9940211.7
1370694.3
5131964.3
344803.5
414335.5
42890.4
5
CoV
0.1
60.2
20.3
60.5
40.7
20.8
91.0
81.2
61.4
21.5
01.3
2
Mean
bit
rate
3607426.1
82671672.4
61535648.8
0846040.8
4467770.6
8262593.9
3135352.4
669179.6
935651.3
119174.2
49794.7
6[b
it/s]
Peak
bit
rate
5964533.3
35015200.0
03734666.6
72682400.0
01862933.3
31266933.3
3793866.6
7461333.3
3264266.6
7144266.6
764800.0
0[b
it/s]
Peak
to
mean
1.6
51.8
82.4
33.1
73.9
84.8
25.8
76.6
77.4
17.5
26.6
2
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 33
Tab
le5:
GoP
stat
isti
csfo
rC
arph
one
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P20435
16906
12000
8264
5363
3464
2054
1243
724
428
209
[byte
]
Xm
ax
,Go
P210971
163857
106085
64375
38104
22389
12165
6356
3206
1535
674
[byte
]
XG
oP
179504.0
0132977.0
676476.2
942142.6
823306.5
213083.1
66741.7
73447.5
51778.0
0955.3
9488.5
5[b
yte
]
S2 X
,Go
P322008210.1
3303239404.1
3276673349.4
8159149249.4
970175931.9
926848586.2
17878839.7
81791887.7
9362037.5
356501.6
54601.5
2
CoV
Go
P0.1
00.1
30.2
20.3
00.3
60.4
00.4
20.3
90.3
40.2
50.1
4
Mean
GoP
rate
3988.9
82955.0
51699.4
7936.5
0517.9
2290.7
4149.8
276.6
139.5
121.2
310.8
6[b
it/s]
Peak
GoP
rate
4688.2
43641.2
72357.4
41430.5
6846.7
6497.5
3270.3
3141.2
471.2
434.1
114.9
8[b
it/s]
Peak
to
mean
1.1
81.2
31.3
91.5
31.6
31.7
11.8
01.8
41.8
01.6
11.3
8
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 34
0.002
0.003
0.004
0.005
0.006
0.007
0.008
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
p
frame size [byte]
Probability Density Function (p) for Carphone_QP25
Figure 47: Frame size distribution for Carphone (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Carphone_QP25
Figure 48: Autocorrelation coefficients for Carphone (quantization 25)
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Carphone_QP25
Figure 49: GoP autocorrelation coefficients for Carphone (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 35
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6
log(
R/S
)
log(d)
R/S Plot for Carphone_QP25 (H=0.758091918469651)
Figure 50: Single Frame R/S plot and for Carphone (quantization 25)
-1
-0.5
0
0.5
1
1.5
2
2.5
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Carphone_QP25
Figure 51: Variance time plot for Carphone (quantization 25)
-7
-6
-5
-4
-3
-2
-1
0
1
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Carphone_QP25 Agg. 1 (H=0.746120744195317)
Figure 52: Single frame periodogram plot for Carphone (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 36
B Claire
0
500
1000
1500
2000
2500
0 50 100 150 200 250 300 350 400 450
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Claire_QP25 - Aggregation 1
Figure 53: Frame size trace for Claire (quantization 25)
0
100
200
300
400
500
600
700
800
0 50 100 150 200 250 300 350 400 450
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Claire_QP25 - Aggregation 12
Figure 54: Frame size trace for one GoP Claire (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 37
Tab
le6:
Sing
lefr
ame
stat
isti
csfo
rC
laire
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
7374
4175
910
237
62
28
19
18
16
13
12
[byte
]
Xm
ax
14857
11505
7172
4885
3376
2282
1465
873
522
356
218
[byte
]
X9185.8
85809.7
02204.4
71063.3
7558.7
6319.8
0184.5
5107.1
367.8
949.0
135.1
6[b
yte
]
XI−
fr
am
e14670.4
811334.1
77010.1
44729.3
63250.1
92187.3
61404.9
3803.6
9486.1
2321.5
0199.4
0[b
yte
]
XP−
fr
am
e10026.2
86424.7
52469.3
31199.4
6601.6
0312.3
9157.2
185.5
950.0
330.3
516.8
9[b
yte
]
XB−
fr
am
e8168.4
44871.6
51489.7
9542.9
1198.0
683.4
338.5
426.0
221.0
421.1
220.9
8[b
yte
]
S2 X
3534886.2
73368796.3
72393434.5
41363059.7
2717295.0
4339236.0
0142775.8
246314.3
616607.0
96986.9
62533.5
8
CoV
0.2
00.3
20.7
01.1
01.5
21.8
22.0
52.0
11.9
01.7
11.4
3
Mean
bit
rate
2204612.2
51394327.4
6529072.4
5255208.1
1134102.0
776750.8
344292.9
025712.1
316294.6
911762.9
28437.9
7[b
it/s]
Peak
bit
rate
3961866.6
73068000.0
01912533.3
31302666.6
7900266.6
7608533.3
3390666.6
7232800.0
0139200.0
094933.3
358133.3
3[b
it/s]
Peak
to
mean
1.8
02.2
03.6
15.1
06.7
17.9
38.8
29.0
58.5
48.0
76.8
9
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 38
Tab
le7:
GoP
stat
isti
csfo
rC
laire
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P14838
11442
7113
4811
3277
2137
1339
763
452
321
202
[byte
]
Xm
ax
,Go
P116473
74897
30900
15862
8468
4868
2704
1506
926
648
450
[byte
]
XG
oP
109731.9
569374.2
226288.0
012656.0
56633.2
43788.3
72182.6
11266.8
5802.8
5580.2
9416.8
5[b
yte
]
S2 X
,Go
P12825367.0
07667306.0
84689651.3
02051947.5
5707354.1
4214328.0
944381.4
411194.3
82814.4
8952.8
6229.1
8
CoV
Go
P0.0
30.0
40.0
80.1
10.1
30.1
20.1
00.0
80.0
70.0
50.0
4
Mean
GoP
rate
2438.4
91541.6
5584.1
8281.2
5147.4
184.1
948.5
028.1
517.8
412.9
09.2
6[b
it/s]
Peak
GoP
rate
2588.2
91664.3
8686.6
7352.4
9188.1
8108.1
860.0
933.4
720.5
814.4
010.0
0[b
it/s]
Peak
to
mean
1.0
61.0
81.1
81.2
51.2
81.2
81.2
41.1
91.1
51.1
21.0
8
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 39
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0 500 1000 1500 2000 2500
p
frame size [byte]
Probability Density Function (p) for Claire_QP25
Figure 55: Frame size distribution for Claire (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Claire_QP25
Figure 56: Autocorrelation coefficients for Claire (quantization 25)
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Claire_QP25
Figure 57: GoP autocorrelation coefficients for Claire (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 40
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6
log(
R/S
)
log(d)
R/S Plot for Claire_QP25 (H=0.24608005215917)
Figure 58: Single Frame R/S plot and for Claire (quantization 25)
-1.5
-1
-0.5
0
0.5
1
1.5
2
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Claire_QP25
Figure 59: Variance time plot for Claire (quantization 25)
-6
-5
-4
-3
-2
-1
0
1
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Claire_QP25 Agg. 1 (H=0.727728889354086)
Figure 60: Single frame periodogram plot for Claire (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 41
C Container
0
500
1000
1500
2000
2500
3000
3500
4000
4500
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Container_QP25 - Aggregation 1
Figure 61: Frame size trace for Container (quantization 25)
0
200
400
600
800
1000
1200
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Container_QP25 - Aggregation 12
Figure 62: Frame size trace for one GoP Container (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 42
Tab
le8:
Sing
lefr
ame
stat
isti
csfo
rC
onta
iner
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
66
53
49
47
46
29
19
18
17
18
12
[byte
]
Xm
ax
20643
17174
12824
9287
6444
4287
2653
1590
951
512
252
[byte
]
X10713.6
67197.3
63562.5
71718.5
5887.3
0484.3
6274.4
7164.7
0100.7
861.9
938.2
5[b
yte
]
XI−
fr
am
e19746.0
816383.5
012187.4
68819.2
76101.5
84054.4
22509.4
21500.0
4880.5
0473.9
2233.9
6[b
yte
]
XP−
fr
am
e12150.6
98341.6
74467.2
72241.4
3897.5
9336.9
5133.7
171.9
643.0
728.5
717.7
7[b
yte
]
XB−
fr
am
e9000.5
65574.0
52102.0
8599.3
8205.5
875.5
336.7
025.8
821.0
520.9
720.4
8[b
yte
]
S2 X
10925540.0
210368803.6
18629068.1
35565628.4
02806810.0
21280270.8
9497151.7
5177050.8
760277.8
316780.4
13776.4
5
CoV
0.3
10.4
50.8
21.3
71.8
92.3
42.5
72.5
52.4
42.0
91.6
1
Mean
bit
rate
2571278.6
71727366.1
1855017.9
4412451.5
6212950.9
6116245.3
265871.6
339527.4
424186.5
814877.6
19179.0
0[b
it/s]
Peak
bit
rate
5504800.0
04579733.3
33419733.3
32476533.3
31718400.0
01143200.0
0707466.6
7424000.0
0253600.0
0136533.3
367200.0
0[b
it/s]
Peak
to
mean
2.1
42.6
54.0
06.0
08.0
79.8
310.7
410.7
310.4
99.1
87.3
2
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 43
Tab
le9:
GoP
stat
isti
csfo
rC
onta
iner
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P9871
6609
3098
1359
524
281
125
91
80
82
71
[byte
]
Xm
ax
,Go
P131676
89129
45717
22736
11806
6611
3825
2290
1369
843
479
[byte
]
XG
oP
128166.1
286094.2
842618.0
420562.9
610632.6
45808.1
63294.9
61976.5
61208.4
4741.6
0456.4
0[b
yte
]
S2 X
,Go
P13760651.7
85555771.9
61843470.6
2590539.2
9225383.9
1104331.8
929999.5
49025.6
72394.0
1844.9
2254.0
0
CoV
Go
P0.0
30.0
30.0
30.0
40.0
40.0
60.0
50.0
50.0
40.0
40.0
3
Mean
GoP
rate
2848.1
41913.2
1947.0
7456.9
5236.2
8129.0
773.2
243.9
226.8
516.4
810.1
4[b
it/s]
Peak
GoP
rate
2926.1
31980.6
41015.9
3505.2
4262.3
6146.9
185.0
050.8
930.4
218.7
310.6
4[b
it/s]
Peak
to
mean
1.0
31.0
41.0
71.1
11.1
11.1
41.1
61.1
61.1
31.1
41.0
5
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 44
0
0.005
0.01
0.015
0.02
0.025
0.03
0 500 1000 1500 2000 2500 3000 3500 4000 4500
p
frame size [byte]
Probability Density Function (p) for Container_QP25
Figure 63: Frame size distribution for Container (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Container_QP25
Figure 64: Autocorrelation coefficients for Container (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Container_QP25
Figure 65: GoP autocorrelation coefficients for Container (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 45
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
log(
R/S
)
log(d)
R/S Plot for Container_QP25 (H=0.191241178623913)
Figure 66: Single Frame R/S plot and for Container (quantization 25)
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Container_QP25
Figure 67: Variance time plot for Container (quantization 25)
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Container_QP25 Agg. 1 (H=0.549621887472763)
Figure 68: Single frame periodogram plot for Container (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 46
D Foreman
0
1000
2000
3000
4000
5000
6000
7000
0 50 100 150 200 250 300 350 400
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Foreman_QP25 - Aggregation 1
Figure 69: Frame size trace for Foreman (quantization 25)
0
200
400
600
800
1000
1200
1400
1600
0 50 100 150 200 250 300 350 400
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Foreman_QP25 - Aggregation 12
Figure 70: Frame size trace for one GoP Foreman (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 47
Tab
le10
:Si
ngle
fram
est
atis
tics
for
Fore
man
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
11934
8191
3896
1437
366
98
25
18
18
18
13
[byte
]
Xm
ax
26002
22598
18056
13621
9624
6330
3711
1942
998
563
229
[byte
]
X15859.5
712027.0
97430.0
74143.1
62083.8
51038.2
8508.0
8268.8
2149.7
688.4
846.3
0[b
yte
]
XI−
fr
am
e22705.4
119203.4
714466.4
110417.5
37064.9
74517.9
12676.3
81518.4
7855.1
8479.1
5202.2
4[b
yte
]
XP−
fr
am
e17019.5
112977.5
38255.1
34791.1
22489.0
31263.0
0597.0
6305.1
6173.1
1103.9
847.0
7[b
yte
]
XB−
fr
am
e14548.4
710752.5
16220.5
23097.5
91294.8
5509.0
3197.4
795.4
250.8
232.7
226.0
9[b
yte
]
S2 X
6843517.9
77032503.6
96655068.2
44976908.8
52950084.0
41384889.1
9517421.7
8166606.6
652443.3
416148.9
12621.9
1
CoV
0.1
60.2
20.3
50.5
40.8
21.1
31.4
21.5
21.5
31.4
41.1
1
Mean
bit
rate
3806296.8
02886502.8
01783216.8
0994359.6
0500124.6
0249187.2
0121938.6
064516.2
035942.4
021235.2
011113.2
0[b
it/s]
Peak
bit
rate
6933866.6
76026133.3
34814933.3
33632266.6
72566400.0
01688000.0
0989600.0
0517866.6
7266133.3
3150133.3
361066.6
7[b
it/s]
Peak
to
mean
1.8
22.0
92.7
03.6
55.1
36.7
78.1
28.0
37.4
07.0
75.4
9
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 48
Tab
le11
:G
oPst
atis
tics
for
Fore
man
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P22042
18466
13665
9672
6462
4014
2240
1238
665
356
161
[byte
]
Xm
ax
,Go
P215273
169960
115495
67853
34703
18636
9719
5142
2746
1592
925
[byte
]
XG
oP
189043.2
1143259.3
988366.7
949212.3
024725.6
712321.8
26030.4
53193.3
01781.0
91053.4
5552.0
6[b
yte
]
S2 X
,Go
P137575944.5
5124844010.0
0115464961.1
759348651.0
922080629.9
86719176.2
21854297.0
7530968.0
9143635.7
745380.4
412853.4
3
CoV
Go
P0.0
60.0
80.1
20.1
60.1
90.2
10.2
30.2
30.2
10.2
00.2
1
Mean
GoP
rate
4200.9
63183.5
41963.7
11093.6
1549.4
6273.8
2134.0
170.9
639.5
823.4
112.2
7[b
it/s]
Peak
GoP
rate
4783.8
43776.8
92566.5
61507.8
4771.1
8414.1
3215.9
8114.2
761.0
235.3
820.5
6[b
it/s]
Peak
to
mean
1.1
41.1
91.3
11.3
81.4
01.5
11.6
11.6
11.5
41.5
11.6
8
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 49
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0 1000 2000 3000 4000 5000 6000 7000
p
frame size [byte]
Probability Density Function (p) for Foreman_QP25
Figure 71: Frame size distribution for Foreman (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Foreman_QP25
Figure 72: Autocorrelation coefficients for Foreman (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Foreman_QP25
Figure 73: GoP autocorrelation coefficients for Foreman (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 50
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6
log(
R/S
)
log(d)
R/S Plot for Foreman_QP25 (H=0.539676937394127)
Figure 74: Single Frame R/S plot and for Foreman (quantization 25)
-1
-0.5
0
0.5
1
1.5
2
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Foreman_QP25
Figure 75: Variance time plot for Foreman (quantization 25)
-6
-5
-4
-3
-2
-1
0
1
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Foreman_QP25 Agg. 1 (H=0.68443840910836)
Figure 76: Single frame periodogram plot for Foreman (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 51
E Grandma
0
500
1000
1500
2000
2500
3000
3500
4000
0 100 200 300 400 500 600 700 800
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Grandma_QP25 - Aggregation 1
Figure 77: Frame size trace for Grandma (quantization 25)
0
100
200
300
400
500
600
700
800
900
1000
0 100 200 300 400 500 600 700 800
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Grandma_QP25 - Aggregation 12
Figure 78: Frame size trace for one GoP Grandma (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 52
Tab
le12
:Si
ngle
fram
est
atis
tics
for
Gra
ndm
a
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
79
66
63
60
25
12
13
14
12
12
12
[byte
]
Xm
ax
21225
17726
12842
8941
6072
3835
2196
1164
620
351
193
[byte
]
X12517.7
28698.7
93934.2
71641.4
2819.8
5443.5
8235.3
9127.0
471.6
247.4
632.7
4[b
yte
]
XI−
fr
am
e21046.2
717563.9
312663.3
78813.0
05959.0
73752.6
02134.2
71135.7
1594.5
5336.6
4180.3
4[b
yte
]
XP−
fr
am
e13039.8
49114.0
54245.2
41668.1
2735.7
0329.5
4138.9
564.6
633.1
019.7
114.1
4[b
yte
]
XB−
fr
am
e11248.0
57426.9
32718.7
3728.7
5204.6
469.9
632.6
423.5
320.2
921.5
021.1
6[b
yte
]
S2 X
7749794.9
38077830.0
87661600.8
85000822.7
22506370.3
01024230.6
9333989.4
193752.4
925132.6
27677.7
32008.4
0
CoV
0.2
20.3
30.7
01.3
61.9
32.2
82.4
62.4
12.2
11.8
51.3
7
Mean
bit
rate
3004252.4
92087710.1
3944225.5
8393940.5
7196763.3
5106458.6
056493.9
630489.3
717189.6
211391.0
47858.5
5[b
it/s]
Peak
bit
rate
5660000.0
04726933.3
33424533.3
32384266.6
71619200.0
01022666.6
7585600.0
0310400.0
0165333.3
393600.0
051466.6
7[b
it/s]
Peak
to
mean
1.8
82.2
63.6
36.0
58.2
39.6
110.3
710.1
89.6
28.2
26.5
5
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 53
Tab
le13
:G
oPst
atis
tics
for
Gra
ndm
aQ
P01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P20793
17392
12431
8657
5841
3661
2078
1112
584
337
185
[byte
]
Xm
ax
,Go
P159866
113306
55601
25427
12850
6719
3382
1712
947
620
412
[byte
]
XG
oP
150107.8
3104326.1
047213.8
319703.7
29826.3
85308.5
02814.1
21517.7
1855.2
5566.5
0390.7
1[b
yte
]
S2 X
,Go
P44043766.9
633422801.6
924505395.4
111775843.9
82540153.9
3419184.9
056350.7
37809.5
61474.0
8435.3
896.7
7
CoV
Go
P0.0
40.0
60.1
00.1
70.1
60.1
20.0
80.0
60.0
40.0
40.0
3
Mean
GoP
rate
3335.7
32318.3
61049.2
0437.8
6218.3
6117.9
762.5
433.7
319.0
112.5
98.6
8[b
it/s]
Peak
GoP
rate
3552.5
82517.9
11235.5
8565.0
4285.5
6149.3
175.1
638.0
421.0
413.7
89.1
6[b
it/s]
Peak
to
mean
1.0
71.0
91.1
81.2
91.3
11.2
71.2
01.1
31.1
11.0
91.0
5
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 54
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
0 500 1000 1500 2000 2500 3000 3500 4000
p
frame size [byte]
Probability Density Function (p) for Grandma_QP25
Figure 79: Frame size distribution for Grandma (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Grandma_QP25
Figure 80: Autocorrelation coefficients for Grandma (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Grandma_QP25
Figure 81: GoP autocorrelation coefficients for Grandma (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 55
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
log(
R/S
)
log(d)
R/S Plot for Grandma_QP25 (H=0.299358787559301)
Figure 82: Single Frame R/S plot and for Grandma (quantization 25)
-1.5
-1
-0.5
0
0.5
1
1.5
2
1 1.2 1.4 1.6 1.8 2 2.2 2.4
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Grandma_QP25
Figure 83: Variance time plot for Grandma (quantization 25)
-5
-4
-3
-2
-1
0
1
2
-2.5 -2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Grandma_QP25 Agg. 1 (H=0.641022082408271)
Figure 84: Single frame periodogram plot for Grandma (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 56
F Mobile
0
5000
10000
15000
20000
25000
30000
35000
40000
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Mobile_QP25 - Aggregation 1
Figure 85: Frame size trace for Mobile (quantization 25)
0
2000
4000
6000
8000
10000
12000
14000
16000
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Mobile_QP25 - Aggregation 12
Figure 86: Frame size trace for one GoP Mobile (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 57
Tab
le14
:Si
ngle
fram
est
atis
tics
for
Mob
ile
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
79
75
74
74
73
72
74
67
65
64
46
[byte
]
Xm
ax
118449
104711
86616
69247
53186
39074
26588
16708
9866
5577
2736
[byte
]
X78447.4
563461.2
645667.8
530220.4
618019.5
910082.9
44901.1
12375.4
01233.0
3664.5
5332.9
8[b
yte
]
XI−
fr
am
e110923.5
897590.0
080210.5
063816.3
848859.6
535838.6
524329.0
815357.5
49079.1
25041.8
52286.1
5[b
yte
]
XP−
fr
am
e81525.0
765983.0
047931.8
032101.5
919313.7
310757.0
84997.1
52123.8
1992.9
7508.3
1236.5
9[b
yte
]
XB−
fr
am
e73071.4
558078.8
740328.3
225147.5
613525.0
86481.9
02339.4
5782.0
7303.0
6154.1
0115.2
2[b
yte
]
S2 X
173892464.3
2172401258.9
5162907058.3
6143381795.8
2112798949.3
074132068.2
639752235.7
217204631.1
86228006.6
11934733.6
6386327.7
1
CoV
0.1
70.2
10.2
80.4
00.5
90.8
51.2
91.7
52.0
22.0
91.8
7
Mean
bit
rate
18827388.4
415230702.1
910960283.3
27252910.0
34324701.9
32419905.6
51176265.5
1570096.4
8295926.3
8159493.1
679916.0
1[b
it/s]
Peak
bit
rate
31586400.0
027922933.3
323097600.0
018465866.6
714182933.3
310419733.3
37090133.3
34455466.6
72630933.3
31487200.0
0729600.0
0[b
it/s]
Peak
to
mean
1.6
81.8
32.1
12.5
53.2
84.3
16.0
37.8
28.8
99.3
29.1
3
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 58
Tab
le15
:G
oPst
atis
tics
for
Mob
ileQ
P01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P71069
56372
38929
24530
13338
6845
2715
1175
637
425
343
[byte
]
Xm
ax
,Go
P1003069
822632
606466
413229
254439
145046
68723
30744
15726
8452
4410
[byte
]
XG
oP
938793.0
0759538.2
4546694.6
4361875.1
2215866.1
2120834.6
058777.8
828496.2
014792.1
67968.0
03983.4
8[b
yte
]
S2 X
,Go
P2784281504.0
82413707187.7
72067802530.9
11533374464.7
8904409033.1
9357694740.4
249164767.6
12893049.3
3539169.3
1131375.1
747563.0
1
CoV
Go
P0.0
60.0
60.0
80.1
10.1
40.1
60.1
20.0
60.0
50.0
50.0
5
Mean
GoP
rate
20862.0
716878.6
312148.7
78041.6
74797.0
22685.2
11306.1
8633.2
5328.7
1177.0
788.5
2[b
it/s]
Peak
GoP
rate
22290.4
218280.7
113477.0
29182.8
75654.2
03223.2
41527.1
8683.2
0349.4
7187.8
298.0
0[b
it/s]
Peak
to
mean
1.0
71.0
81.1
11.1
41.1
81.2
01.1
71.0
81.0
61.0
61.1
1
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 59
0.003
0.0035
0.004
0.0045
0.005
0.0055
0.006
0.0065
0.007
0 5000 10000 15000 20000 25000 30000 35000 40000
p
frame size [byte]
Probability Density Function (p) for Mobile_QP25
Figure 87: Frame size distribution for Mobile (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Mobile_QP25
Figure 88: Autocorrelation coefficients for Mobile (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Mobile_QP25
Figure 89: GoP autocorrelation coefficients for Mobile (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 60
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
log(
R/S
)
log(d)
R/S Plot for Mobile_QP25 (H=0.471436506988997)
Figure 90: Single Frame R/S plot and for Mobile (quantization 25)
-1
-0.5
0
0.5
1
1.5
2
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Mobile_QP25
Figure 91: Variance time plot for Mobile (quantization 25)
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Mobile_QP25 Agg. 1 (H=0.55999620632099)
Figure 92: Single frame periodogram plot for Mobile (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 61
G Mother and Daughter
0
500
1000
1500
2000
2500
3000
3500
4000
0 100 200 300 400 500 600 700 800 900
fram
e si
ze [b
yte]
frame index
Frame Size Trace of MotherDaughter_QP25 - Aggregation 1
Figure 93: Frame size trace for MotherDaughter (quantization 25)
0
200
400
600
800
1000
1200
0 100 200 300 400 500 600 700 800 900
fram
e si
ze [b
yte]
frame index
Frame Size Trace of MotherDaughter_QP25 - Aggregation 12
Figure 94: Frame size trace for one GoP MotherDaughter (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 62
Tab
le16
:Si
ngle
fram
est
atis
tics
for
Mot
herD
augh
ter
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
10214
6465
1825
155
31
20
18
17
13
12
12
[byte
]
Xm
ax
21320
17861
12948
9061
6086
3850
2203
1242
678
388
186
[byte
]
X13283.9
29445.4
04715.6
42155.9
61033.4
2540.1
4276.7
7148.0
783.3
752.3
832.8
1[b
yte
]
XI−
fr
am
e20627.1
917094.1
012171.1
68452.2
65714.8
43623.9
42086.5
91153.4
9639.5
4359.0
4172.8
3[b
yte
]
XP−
fr
am
e14342.7
110355.7
35460.3
72632.1
71222.8
8567.8
0252.7
2121.1
860.6
531.6
517.1
2[b
yte
]
XB−
fr
am
e11957.5
08135.9
83492.7
81180.5
1369.8
8139.4
856.7
330.9
121.5
021.3
420.9
8[b
yte
]
S2 X
6762779.2
76964248.1
76448300.7
24378181.7
92260550.7
8942619.1
5316267.5
296161.9
529104.7
28773.7
91823.3
5
CoV
0.2
00.2
80.5
40.9
71.4
51.8
02.0
32.0
92.0
51.7
91.3
0
Mean
bit
rate
3188141.2
72266894.9
01131754.3
4517431.2
6248021.3
9129634.4
666424.8
135537.7
320009.6
612571.1
67874.8
0[b
it/s]
Peak
bit
rate
5685333.3
34762933.3
33452800.0
02416266.6
71622933.3
31026666.6
7587466.6
7331200.0
0180800.0
0103466.6
749600.0
0[b
it/s]
Peak
to
mean
1.7
82.1
03.0
54.6
76.5
47.9
28.8
49.3
29.0
48.2
36.3
0
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 63
Tab
le17
:G
oPst
atis
tics
for
Mot
herD
augh
ter
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P20705
17150
12261
8521
5702
3609
2069
1136
620
354
166
[byte
]
Xm
ax
,Go
P179334
132017
74349
39366
20465
10895
5412
2627
1368
759
441
[byte
]
XG
oP
159059.7
7113088.1
656449.5
625789.8
812342.0
46442.3
43297.8
21763.8
8993.1
6624.2
4391.4
4[b
yte
]
S2 X
,Go
P92664051.7
780182193.0
771449789.4
937333825.1
010344312.7
52623134.2
0509903.8
988790.9
513823.7
82401.8
3295.4
6
CoV
Go
P0.0
60.0
80.1
50.2
40.2
60.2
50.2
20.1
70.1
20.0
80.0
4
Mean
GoP
rate
3534.6
62513.0
71254.4
3573.1
1274.2
7143.1
673.2
839.2
022.0
713.8
78.7
0[b
it/s]
Peak
GoP
rate
3985.2
02933.7
11652.2
0874.8
0454.7
8242.1
1120.2
758.3
830.4
016.8
79.8
0[b
it/s]
Peak
to
mean
1.1
31.1
71.3
21.5
31.6
61.6
91.6
41.4
91.3
81.2
21.1
3
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 64
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
0.01
0.011
0 500 1000 1500 2000 2500 3000 3500 4000
p
frame size [byte]
Probability Density Function (p) for MotherDaughter_QP25
Figure 95: Frame size distribution for MotherDaughter (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for MotherDaughter_QP25
Figure 96: Autocorrelation coefficients for MotherDaughter (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for MotherDaughter_QP25
Figure 97: GoP autocorrelation coefficients for MotherDaughter (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 65
0.4
0.6
0.8
1
1.2
1.4
1.6
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
log(
R/S
)
log(d)
R/S Plot for MotherDaughter_QP25 (H=0.452602492677487)
Figure 98: Single Frame R/S plot and for MotherDaughter (quantization 25)
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
1 1.2 1.4 1.6 1.8 2 2.2 2.4
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for MotherDaughter_QP25
Figure 99: Variance time plot for MotherDaughter (quantization 25)
-5
-4
-3
-2
-1
0
1
2
-2.5 -2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for MotherDaughter_QP25 Agg. 1 (H=0.767168339987136)
Figure 100: Single frame periodogram plot for MotherDaughter (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 66
H News
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of News_QP25 - Aggregation 1
Figure 101: Frame size trace for News (quantization 25)
0
200
400
600
800
1000
1200
1400
1600
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of News_QP25 - Aggregation 12
Figure 102: Frame size trace for one GoP News (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 67
Tab
le18
:Si
ngle
fram
est
atis
tics
for
New
s
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
66
53
49
47
35
29
24
25
18
19
12
[byte
]
Xm
ax
20607
16954
12400
9038
6532
4578
2986
1884
1136
659
322
[byte
]
X8178.2
45858.8
43241.4
01878.4
01143.0
0708.1
4420.7
0249.2
2145.5
585.6
547.1
8[b
yte
]
XI−
fr
am
e19658.6
216123.7
711767.0
48538.0
86129.6
24253.1
52774.9
21748.5
01062.0
4613.5
8300.8
8[b
yte
]
XP−
fr
am
e8678.8
16299.2
83612.4
12106.3
61223.2
1698.4
5375.3
2203.4
7113.1
159.6
725.5
7[b
yte
]
XB−
fr
am
e6498.0
84359.2
31993.9
3927.1
6464.6
6250.9
1131.6
671.4
738.5
726.7
722.3
0[b
yte
]
S2 X
15257550.6
211957828.8
78091292.9
74830720.2
22639248.1
91307971.6
8568005.9
2228165.5
884816.0
827956.9
66394.7
0
CoV
0.4
80.5
90.8
81.1
71.4
21.6
21.7
91.9
22.0
01.9
51.6
9
Mean
bit
rate
1962778.2
11406121.7
3777935.6
8450816.4
8274320.0
0169952.6
9100967.4
459813.4
234931.5
620556.2
811323.0
6[b
it/s]
Peak
bit
rate
5495200.0
04521066.6
73306666.6
72410133.3
31741866.6
71220800.0
0796266.6
7502400.0
0302933.3
3175733.3
385866.6
7[b
it/s]
Peak
to
mean
2.8
03.2
24.2
55.3
56.3
57.1
87.8
98.4
08.6
78.5
57.5
8
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 68
Tab
le19
:G
oPst
atis
tics
for
New
sQ
P01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P5516
3673
1596
679
291
203
108
97
91
92
82
[byte
]
Xm
ax
,Go
P110009
81418
48524
29314
17599
10398
5936
3502
2022
1185
699
[byte
]
XG
oP
97934.2
070170.2
038838.0
422518.5
213710.6
08494.0
05049.4
82989.9
61744.5
61024.6
0562.9
2[b
yte
]
S2 X
,Go
P18214575.1
713682735.2
59666299.6
24703273.5
11590853.8
3503967.3
3164397.0
172089.5
423891.0
98519.0
01724.7
4
CoV
Go
P0.0
40.0
50.0
80.1
00.0
90.0
80.0
80.0
90.0
90.0
90.0
7
Mean
GoP
rate
2176.3
21559.3
4863.0
7500.4
1304.6
8188.7
6112.2
166.4
438.7
722.7
712.5
1[b
it/s]
Peak
GoP
rate
2444.6
41809.2
91078.3
1651.4
2391.0
9231.0
7131.9
177.8
244.9
326.3
315.5
3[b
it/s]
Peak
to
mean
1.1
21.1
61.2
51.3
01.2
81.2
21.1
81.1
71.1
61.1
61.2
4
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 69
0.003
0.004
0.005
0.006
0.007
0.008
0.009
0.01
0.011
0.012
0.013
0.014
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
p
frame size [byte]
Probability Density Function (p) for News_QP25
Figure 103: Frame size distribution for News (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for News_QP25
Figure 104: Autocorrelation coefficients for News (quantization 25)
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for News_QP25
Figure 105: GoP autocorrelation coefficients for News (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 70
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
log(
R/S
)
log(d)
R/S Plot for News_QP25 (H=0.253659412037892)
Figure 106: Single Frame R/S plot and for News (quantization 25)
-1
-0.5
0
0.5
1
1.5
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for News_QP25
Figure 107: Variance time plot for News (quantization 25)
-5
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for News_QP25 Agg. 1 (H=0.482451099660243)
Figure 108: Single frame periodogram plot for News (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 71
I Salesman
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
0 50 100 150 200 250 300 350 400
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Salesman_QP25 - Aggregation 1
Figure 109: Frame size trace for Salesman (quantization 25)
0
200
400
600
800
1000
1200
1400
1600
0 50 100 150 200 250 300 350 400
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Salesman_QP25 - Aggregation 12
Figure 110: Frame size trace for one GoP Salesman (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 72
Tab
le20
:Si
ngle
fram
est
atis
tics
for
Sale
sman
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
8785
5424
1409
279
63
21
19
17
17
13
12
[byte
]
Xm
ax
23117
19514
14851
10909
7551
4923
2886
1602
827
414
180
[byte
]
X11398.2
77892.7
53651.0
21883.5
01063.3
7620.9
8339.8
1186.3
796.9
553.2
131.8
4[b
yte
]
XI−
fr
am
e23019.9
519409.4
214755.3
910818.4
57489.5
34858.8
92829.5
01556.7
1804.1
3390.1
8167.7
9[b
yte
]
XP−
fr
am
e11584.3
58008.3
83634.4
61730.0
8880.1
1464.2
9232.7
9120.0
356.6
625.8
915.2
0[b
yte
]
XB−
fr
am
e9846.3
76380.7
12241.2
5801.8
1312.8
1139.4
762.5
633.8
521.9
220.5
020.7
6[b
yte
]
S2 X
13435514.3
813087284.3
012064243.9
97697474.6
13940524.2
61703347.4
6585238.8
0179803.1
446879.7
810598.0
71726.9
2
CoV
0.3
20.4
60.9
51.4
71.8
72.1
02.2
52.2
82.2
31.9
31.3
1
Mean
bit
rate
2735583.7
51894258.9
3876244.8
2452040.0
0255209.4
6149035.7
181555.0
044729.8
023268.7
512769.2
97641.4
3[b
it/s]
Peak
bit
rate
6164533.3
35203733.3
33960266.6
72909066.6
72013600.0
01312800.0
0769600.0
0427200.0
0220533.3
3110400.0
048000.0
0[b
it/s]
Peak
to
mean
2.2
52.7
54.5
26.4
47.8
98.8
19.4
49.5
59.4
88.6
56.2
8
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 73
Tab
le21
:G
oPst
atis
tics
for
Sale
sman
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P23069
19511
14835
10909
7536
4851
2846
1548
797
395
165
[byte
]
Xm
ax
,Go
P152586
109126
57158
32090
18538
10714
5586
2809
1387
733
421
[byte
]
XG
oP
135974.4
694106.3
543445.5
722362.1
912606.9
57356.5
74023.4
32171.3
01148.0
5630.5
7378.1
9[b
yte
]
S2 X
,Go
P37528558.8
128171359.4
624440230.1
411699782.6
03926095.7
71190301.3
6266349.8
657525.9
98130.9
41162.1
4175.1
0
CoV
Go
P0.0
50.0
60.1
10.1
50.1
60.1
50.1
30.1
10.0
80.0
50.0
3
Mean
GoP
rate
3021.6
52091.2
5965.4
6496.9
4280.1
5163.4
889.4
148.2
525.5
114.0
18.4
0[b
it/s]
Peak
GoP
rate
3390.8
02425.0
21270.1
8713.1
1411.9
6238.0
9124.1
362.4
230.8
216.2
99.3
6[b
it/s]
Peak
to
mean
1.1
21.1
61.3
21.4
41.4
71.4
61.3
91.2
91.2
11.1
61.1
1
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 74
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
p
frame size [byte]
Probability Density Function (p) for Salesman_QP25
Figure 111: Frame size distribution for Salesman (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Salesman_QP25
Figure 112: Autocorrelation coefficients for Salesman (quantization 25)
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Salesman_QP25
Figure 113: GoP autocorrelation coefficients for Salesman (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 75
0.4
0.5
0.6
0.7
0.8
0.9
1
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6
log(
R/S
)
log(d)
R/S Plot for Salesman_QP25 (H=0.294441155331762)
Figure 114: Single Frame R/S plot and for Salesman (quantization 25)
-1.5
-1
-0.5
0
0.5
1
1.5
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Salesman_QP25
Figure 115: Variance time plot for Salesman (quantization 25)
-5
-4
-3
-2
-1
0
1
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Salesman_QP25 Agg. 1 (H=0.726112561906172)
Figure 116: Single frame periodogram plot for Salesman (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 76
J Silent
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Silent_QP25 - Aggregation 1
Figure 117: Frame size trace for Silent (quantization 25)
0
200
400
600
800
1000
1200
1400
0 50 100 150 200 250 300
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Silent_QP25 - Aggregation 12
Figure 118: Frame size trace for one GoP Silent (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 77
Tab
le22
:Si
ngle
fram
est
atis
tics
for
Sile
nt
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
66
53
49
47
39
32
23
24
17
12
12
[byte
]
Xm
ax
23229
19695
15219
11179
7697
4889
2811
1575
828
412
183
[byte
]
X10871.7
67506.2
43928.3
42266.2
91317.9
0779.2
4423.6
8230.8
4120.5
362.8
134.8
1[b
yte
]
XI−
fr
am
e21828.4
618427.0
414120.0
010307.1
97018.8
54476.1
52557.1
91437.9
2772.5
8369.0
4165.8
5[b
yte
]
XP−
fr
am
e11272.4
37819.6
83899.2
72240.1
11269.6
0728.5
7389.3
2208.7
2107.1
352.7
724.3
3[b
yte
]
XB−
fr
am
e9297.1
45968.9
92614.3
31230.7
9594.8
9317.6
4159.2
182.2
240.8
026.7
721.7
1[b
yte
]
S2 X
14380827.5
213635582.9
711236120.9
76880304.3
63410205.9
81422954.6
6471319.3
0150086.9
943514.7
69530.5
41708.8
5
CoV
0.3
50.4
90.8
51.1
61.4
01.5
31.6
21.6
81.7
31.5
51.1
9
Mean
bit
rate
2609221.7
91801497.4
1942802.1
3543908.5
7316296.0
8187017.4
1101683.4
655401.7
328928.3
715075.3
58355.3
5[b
it/s]
Peak
bit
rate
6194400.0
05252000.0
04058400.0
02981066.6
72052533.3
31303733.3
3749600.0
0420000.0
0220800.0
0109866.6
748800.0
0[b
it/s]
Peak
to
mean
2.3
72.9
24.3
05.4
86.4
96.9
77.3
77.5
87.6
37.2
95.8
4
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 78
Tab
le23
:G
oPst
atis
tics
for
Sile
ntQ
P01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P9208
5976
2638
1196
600
347
192
118
89
97
73
[byte
]
Xm
ax
,Go
P145961
104077
59174
36098
21447
12712
6874
3670
1869
944
500
[byte
]
XG
oP
130135.4
889875.9
247072.8
427177.0
015812.8
09350.9
25085.3
22770.1
21445.1
2750.2
8414.8
8[b
yte
]
S2 X
,Go
P98910894.0
169524304.8
343012443.3
920664140.6
77362071.0
82498743.4
9678530.6
4174606.6
937332.6
98437.6
31200.7
8
CoV
Go
P0.0
80.0
90.1
40.1
70.1
70.1
70.1
60.1
50.1
30.1
20.0
8
Mean
GoP
rate
2891.9
01997.2
41046.0
6603.9
3351.4
0207.8
0113.0
161.5
632.1
116.6
79.2
2[b
it/s]
Peak
GoP
rate
3243.5
82312.8
21314.9
8802.1
8476.6
0282.4
9152.7
681.5
641.5
320.9
811.1
1[b
it/s]
Peak
to
mean
1.1
21.1
61.2
61.3
31.3
61.3
61.3
51.3
21.2
91.2
61.2
1
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 79
0.003
0.004
0.005
0.006
0.007
0.008
0.009
0.01
0.011
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
p
frame size [byte]
Probability Density Function (p) for Silent_QP25
Figure 119: Frame size distribution for Silent (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Silent_QP25
Figure 120: Autocorrelation coefficients for Silent (quantization 25)
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Silent_QP25
Figure 121: GoP autocorrelation coefficients for Silent (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 80
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
log(
R/S
)
log(d)
R/S Plot for Silent_QP25 (H=0.363386415577168)
Figure 122: Single Frame R/S plot and for Silent (quantization 25)
-1
-0.5
0
0.5
1
1.5
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Silent_QP25
Figure 123: Variance time plot for Silent (quantization 25)
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Silent_QP25 Agg. 1 (H=0.676267115592979)
Figure 124: Single frame periodogram plot for Silent (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 81
K Tempete
0
5000
10000
15000
20000
25000
30000
0 50 100 150 200 250
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Tempete_QP25 - Aggregation 1
Figure 125: Frame size trace for Tempete (quantization 25)
0
2000
4000
6000
8000
10000
12000
14000
0 50 100 150 200 250
fram
e si
ze [b
yte]
frame index
Frame Size Trace of Tempete_QP25 - Aggregation 12
Figure 126: Frame size trace for one GoP Tempete (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 82
Tab
le24
:Si
ngle
fram
est
atis
tics
for
Tem
pete
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
66025
50772
32641
18576
9292
4014
983
213
70
37
34
[byte
]
Xm
ax
101682
87719
69159
52312
37934
26473
16875
10046
5494
2703
1128
[byte
]
X74489.6
359063.4
840669.1
025675.7
214780.3
68029.8
43656.7
61661.1
1806.5
2394.4
9177.9
4[b
yte
]
XI−
fr
am
e98993.1
885019.5
066432.6
849902.2
336070.1
824939.0
515815.2
79381.4
15175.0
92569.6
81069.0
0[b
yte
]
XP−
fr
am
e77589.5
261453.4
242704.3
527183.7
415794.3
58616.8
83961.3
81692.0
6778.8
0394.2
5129.1
5[b
yte
]
XB−
fr
am
e70183.9
854840.3
536604.6
222007.0
911674.0
55645.1
91986.4
8661.9
3258.2
2116.3
782.4
1[b
yte
]
S2 X
70930887.7
375458278.3
173218070.6
262876688.8
547089339.5
729075289.0
614794354.9
85830631.9
71845986.6
6458469.0
075280.6
3
CoV
0.1
10.1
50.2
10.3
10.4
60.6
71.0
51.4
51.6
81.7
21.5
4
Mean
bit
rate
17877511.0
414175234.9
09760584.0
96162172.3
63547287.1
01927162.0
1877621.6
2398665.9
5193564.1
794678.6
142706.1
0[b
it/s]
Peak
bit
rate
27115200.0
023391733.3
318442400.0
013949866.6
710115733.3
37059466.6
74500000.0
02678933.3
31465066.6
7720800.0
0300800.0
0[b
it/s]
Peak
to
mean
1.5
21.6
51.8
92.2
62.8
53.6
65.1
36.7
27.5
77.6
17.0
4
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 83
Tab
le25
:G
oPst
atis
tics
for
Tem
pete
QP
01
05
10
15
20
25
30
35
40
45
51
Xm
in
,Go
P100827
86869
68278
51257
36889
25535
16202
9555
5191
2539
1010
[byte
]
Xm
ax
,Go
P938493
752182
530890
342361
200628
110792
51909
23811
11396
5549
2474
[byte
]
XG
oP
884489.8
6700847.7
1481893.2
4303686.3
3174449.8
194552.3
842876.4
819428.1
09447.1
94641.0
02109.7
1[b
yte
]
S2 X
,Go
P1155746930.6
3816403714.0
1552139035.6
9341044172.6
3161522634.2
666571933.3
522432126.4
65279694.1
91096240.7
6214245.8
045613.9
1
CoV
Go
P0.0
40.0
40.0
50.0
60.0
70.0
90.1
10.1
20.1
10.1
00.1
0
Mean
GoP
rate
19655.3
315574.3
910708.7
46748.5
93876.6
62101.1
6952.8
1431.7
4209.9
4103.1
346.8
8[b
it/s]
Peak
GoP
rate
20855.4
016715.1
611797.5
67608.0
24458.4
02462.0
41153.5
3529.1
3253.2
4123.3
154.9
8[b
it/s]
Peak
to
mean
1.0
61.0
71.1
01.1
31.1
51.1
71.2
11.2
31.2
11.2
01.1
7
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 84
0.003
0.004
0.005
0.006
0.007
0.008
0.009
0.01
0.011
0.012
0 5000 10000 15000 20000 25000 30000
p
frame size [byte]
Probability Density Function (p) for Tempete_QP25
Figure 127: Frame size distribution for Tempete (quantization 25)
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12
acc
lag [frame]
Frame Autocorrelation Coefficent (acc) for Tempete_QP25
Figure 128: Autocorrelation coefficients for Tempete (quantization 25)
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12
acc
lag [GoPs]
GoP Autocorrelation Coefficent (acc) for Tempete_QP25
Figure 129: GoP autocorrelation coefficients for Tempete (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 85
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
log(
R/S
)
log(d)
R/S Plot for Tempete_QP25 (H=0.436644472770824)
Figure 130: Single Frame R/S plot and for Tempete (quantization 25)
-1
-0.5
0
0.5
1
1.5
2
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
log
varia
nces
(agg
)
log(agg)
Variance Time Plot for Tempete_QP25
Figure 131: Variance time plot for Tempete (quantization 25)
-6
-5
-4
-3
-2
-1
0
-2 -1.5 -1 -0.5 0 0.5 1
log1
0(I(
lam
bda)
)
log10(Lambda)
Periodogram Plot for Tempete_QP25 Agg. 1 (H=0.628082797246763)
Figure 132: Single frame periodogram plot for Tempete (quantization 25)
Copyright at acticom. All Rightsreserved.
acticom-02-002 Page 86