video transmission over high-speed wpan's based on low-power … · 2014-02-05 · video...

Video transmission over high-speed WPAN’s based onlow-power data compression

Eugeniy Belyaev1, Andrey Turlikov1 and Anna Ukhanova2

1State University of Aerospace Instrumentation, Saint-Petersburg2DTU Fotonik, Technical University of Denmark

November 4, 2009

Eugeniy Belyaev (SUAI) Video transmission over WPAN’s November 4, 2009 1 / 21

Video transmission over high-speed WPAN’s 1 2 3

HDTV1920x1080, 60 frames/s

IEEE 802.15.3c

System properties:I Transmission rate up to 6Gbps;I Low-power data transmitter.

1http://www.ieee802.org/15/pub/TG3c.html2http://www.wirelesshd.org/3http://wirelessgigabitalliance.org/Eugeniy Belyaev (SUAI) Video transmission over WPAN’s November 4, 2009 2 / 21

System restrictions and requirements

System restrictions:I Low-memory and low-complexity video processing;I One-pass processing only is possible;

System requirements:I Very low transmission latency (1-3 ms);I Continuous video playback at the receiver;I Acceptable visual quality for wide types of video sources:

F sequences of computer graphics, snapshots;F natural and mixed images.


Possible solutions

Uncompressed video transmission1;Intra single-layer video coding2;Intra scalable video coding3;Distributive video coding4.

1H. Singh, Jisung Oh, Changyeul Kweon, Xiangping Qin, Huai-Rong Shao and ChiuNgo, “A 60 GHz wireless network for enabling uncompressed videocommunication“, IEEE Communications Magazine, 2008

2E. Belyaev, A. Dogadaev and A. Ukhanova. “MINMAX rate control in near-losslessvideo encoders for real-time data transmission“, XII International Symposium onProblems of Redundancy in Information and Control Systems, 2009.

3M.Gallant and F. Kossentini, “Rate-distortion optimized layered coding with unequalerror protection for robust Internet video“, IEEE Transactions on Circuits and Systemsfor Video Technology, 2001.

4T. Kuganeswaran, X. Fernando, L. Guan, “Distributed video coding andtransmission over wireless fading channel“, Canadian Conference on Electrical andComputer Engineering, 2008.


Possible solutions comparison

Solution Encodercomplexity

Decodercomplexity

Compressionefficiency

Link/PHYlayers mod-ification

Uncompressedvideo transmission

very low very low very low yes

Intra single-layervideo coding

low low medium no

Intra scalable videocoding

medium medium low partly

Distributive videocoding

low high high yes


AgendaIntra single-layer video compression and transmission over high-speed WPAN’s

H.264/AVC low-complexity single-layer video compression;Video transmission system description

I End-to-end distortion in video transmission system;I End-to-end latency in video transmission system;I MINMAX optimization task description.

Main idea of the video source rate control;Video transmission system model;Practical results;Future work.


H.264/AVC low-complexity single-layer video compression

Inputframe

Video sourcerate control

DCT 4x4 Quantization EncoderBuffer Channel

Number of bits

Channel info

InverseQuantization

Inverse DCT4x4

IntraDC-prediction

QP/Skip

CAVLC NAL

IPCM / Intramode decision

Hash valuecalculation

Low-memory INTRA solution (32 pixel lines);DC-Prediction only, 4x4 DCT only, CAVLC;Hash function calculation for temporal redundancy removal (by usingSKIP macroblock type);


Proposed video transmission system scheme

Encoder buffer

Video sourcerate control Channel

Decoder buffer

Videoencoder

Videodecoder

Input macroblocksequence

Reconstructedmacroblock sequence

eT∆

dT∆

ebT∆

dbT∆

cT∆

End-to-endlatency

dT∆

)(tbtq

),( tqtmr

tm

Channelrate control

Transmitter

Receiver

Channelfeedback

Coderate

Modulationtype

1−tc

Packetsize

tbT∆

rbT∆

End-to-enddistortion

ARQ

Quantizationdistortion

Channeldistortion

Channel rate controller chooses the transmission scheme thatmaximizes the channel throughput;Video source rate controller chooses the quantization step qt andmacroblocks type mt ∈ {intra, skip} to provide acceptable visualquality and continuous video playback for given channel throughput.


End-to-end distortion in video transmission system

The end-to-end distortion dt for unit t:dt = d(qt) + dc ,

d(qt) is quantization distortion, dc is distortion caused by channel errors.

MINMAX quality criteria1:minimize max

tdt .

Packet loss probability for ARQ with n retransmissions:pnt = (1− (1− BER)l )n,

BER is bit error rate, l is packet length.

If BER < 10−4 then l and n can be chosen to guarantee that pnt < 10−10.

Therefore, for further optimization we can disregard packet losses andminimize d(qt) only.

1N. Cherniavsky, G. Shavit, M.F. Ringenburg, R.E. Ladner, E.A. Riskin,“Multistage:A MINMAX bit allocation algorithm for video coders“, IEEE Transactions on Circuitsand Systems for Video Technology, 2007.


Bit error rate in IEEE 802.15.3c1

-6

-5

-4

-3

-2

-1

00 2 4 6 8 10 12 14 16 18 20 22 24

SNR [dB]

BER

SQPSK, 1/2 2QPSK, 1/2 16QAM, 1/2

16QAM, 3/4 64QAM, 1/2 2QPSK, 3/4

10

10

10

10

10

10

10

BER<10-4

1IEEE 802.15 WPAN Millimeter Wave Alternative PHY Task Group 3c (TG3c),Contributions and documents, 2009.


End-to-end latency in video transmission systemThe number of bits in the encoder buffer after compression unit t:

be(t) = max{0, be(t − 1)− ct}+ rt(qt , mt),ct is number of bits that are transmitted during compression of the unit t,rt(qt , mt) is bit size of the compressed unit t.Assume that data on the receiver side is accumulated for some time L afterwhich the decoding and playing starts.

End-to-end latency ∆T = L, if the number of bits in the encoder buffer1 is

be(t) ≤ beff (t) =t+L·f ·N∑i=t+1

ci ,

where beff (t) is the effective buffer size2, N is a number of units in theframe and f is a frame rate.

1A.R. Reibman and B.G Haskell, “Constraints on variable bit-rate video for ATMnetworks“, IEEE Transactions on Circuits and Systems for Video Technology, 1992.

2A. Ortega and M. Khansari, “Rate control for video coding over variable bit ratechannels with applications to wireless transmission“, International Conference on ImageProcessing, 1995.


MINMAX optimization task description

For high-speed video transmission we can use high-resolution quantizationhypothesis1 that defines distortion as d(q) = q2/12, therefore MINMAXcriteria corresponds to

minimize maxt

qt . (1)

Let us formulate rate control optimization task according to the latencyrequirements and the MINMAX quality criteria. For each unit t it isnecessary to choose the quantization step qt , so that{

minimize maxt

qt

be(t) ≤ beff (t).(2)

1H. Radha, M. Dai, D. Loguinov, “Rate-distortion modeling of scalable videocoders“, International Conference on Image Processing, 2004.


Video source rate control

MINMAXoptimization task

Multi-pass optimal solutionby consecutive search

One-pass real-timevideo source rate control

qqt~= qqqt ∆+<~

for eachunit t

after adaptation,for each unit t


Solution of MINMAX task by consecutive search algorithm

t0

1

2)()(tbtb

eff

e

1q

2q3q

Theorem 1. Consider q̃ the MINMAX solution found by the consecutivesearch algorithm. There is no sequence of quantization steps y1, y2, . . . forwhich max

tyt < q̃ that does not lead to the transmitter buffer overflow.


One-pass MINMAX video source rate control algorithm

BufferAccumulation

LatencyRestoration

skipped isunit )()( tbtb eff

e >

skipped isunit restorednot islatency if

t

effe

qtbtb

withcompressed isunit )()( ≤

)0(0 qq =begin

t

tt

qqqq

withcomressed isunit

restored islatency if+∆+←

Theorem 2. Consider that consecutive search algorithm finds thequantization step value q̃. Then for the proposed algorithm with initialvalue q0 ≤ q̃, the inequality qt < q̃ + ∆q+ holds true for any timemoment t.


System model1

Encoder buffer

Video sourcerate control

Decoder buffer

Videoencoder

Videodecoder

Input macroblocksequence

Reconstructedmacroblock sequence

eT∆

dT∆

ebT∆

dbT∆

End-to-endlatency

End-to-enddistortion

dT∆

)(tebtq

),( tqtmr

tm

1−tc

SNR time variancemodel

SNR / Throughputmodel

tSNR

Channeltc

1S. Collonge, G. Zaharia, G. El Zein, “Influence of the human activity on wide-bandcharacteristics of the 60GHz indoor radio channel“, IEEE Transactions on WirelessCommunications, vol.3, pp. 2396–2406, 2004


SNR/Throughput dependence for different MCS1

0

1000

2000

3000

4000

5000

6000

0 2 4 6 8 10 12 14 16 18 20 22 24

SNR [dB]

Thro

ughp

ut [m

bps]

SQPSK, 1/22QPSK, 1/216QAM, 1/216QAM, 3/464QAM, 1/22QPSK, 3/4Best

1IEEE 802.15 WPAN Millimeter Wave Alternative PHY Task Group 3c (TG3c),Contributions and documents, 2009.


Test video sequences

Breeze Desktop


Practical results for Breeze

0500

10001500200025003000

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5time [sec]

Rate

[Mbp

s]

Channel rateVideo source rate

10203040506070

0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450frame number

RGB-

PSNR

[dB]


Practical results for Desktop

0500

10001500200025003000

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5time [sec]

Rate

[Mbp

s]

10203040506070

0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450frame number

RGB-

PSNR

[dB]

Channel rate

Video source rate


Future work

It is interesting to solve the same optimization task for others videoprocessing approaches and compare it with intra single-layer videocompression approach.


video transmission over high-speed wpan's based on low-power … · 2014-02-05 · video...

Documents