h.26l pre-standard evaluation - faculty.engineering.asu.edu

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

[email protected]

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

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


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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,


acticom-02-002

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


acticom-02-002

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-


acticom-02-002

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.


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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.


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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.


acticom-02-002

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



0

200

400

600

800

1000

1200

1400

1600

0 200 400 600 800 1000

fram

e si

ze [b

yte]

frame index




acticom-02-002

0

10000

20000

30000

40000

50000

60000

70000

0 200 400 600 800 1000

fram

e si

ze [b

yte]

frame index


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



0

50

100

150

200

250

300

350

400

0 200 400 600 800 1000

fram

e si

ze [b

yte]

frame index




acticom-02-002

Figure 18: Quantization effect for Paris (quantization 40, PSNR for this frame: 27.4271)




acticom-02-002

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]



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]




acticom-02-002

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,


acticom-02-002

-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]



-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]




acticom-02-002

-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]



-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]




acticom-02-002

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)


acticom-02-002

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)



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)




acticom-02-002

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)


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)



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)




acticom-02-002

-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)



-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)




acticom-02-002

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


acticom-02-002

-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)



-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)




acticom-02-002

-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)


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)



-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)




acticom-02-002

[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


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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 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)


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

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


acticom-02-002

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


acticom-02-002

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)


acticom-02-002

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)


acticom-02-002

h.26l pre-standard evaluation - faculty.engineering.asu.edu

Documents