Download Error Resilient Encoding & Reversible Codes: Joint Source-Channel for Multimedia Comms - P and more Study notes Electrical and Electronics Engineering in PDF only on Docsity!
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Joint Source- Joint Source
-Channel Approach forChannel Approach for
Multimedia CommunicationsMultimedia Communications
Min Wu
Electrical & Computer Engineering
Univ. of Maryland, College Park
http://www.ece.umd.edu/class/enee739m/
[email protected]
ENEE739M Spring 2002^ ENEE739M Spring 2002LectureLecture-
-15^15
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [2]
Review of Last ClassReview of Last Class "^
MDC via pairwise correlating transform^ ñ
Performance
"
Error resilient communications^ ñ
Causes and consequences of errors in MM comm. ñ^
Transport-level error control tools ñ^
Error resilient source coding
"
Today:^ ñ
Discussions on RVLC ñ^
Joint source-channel approaches for MM comm.
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [3]
Recall: Error Resilient Encoding of MM Source Recall: Error Resilient Encoding of MM Source "^
Purpose^ ñ
May add controlled redundancy hence less efficient than codec optimizedfor efficient coding rateñ Under targeted channel conditions
#^
Achieve best decoded quality for a given amount of redundancy, orminimize incurred redundancy while maintaining a prescribed quality #^
prevent excess error propagation&facilitate concealment)
"
Types of approaches^ $
Error isolation $
Robust entropy encoding
#^
prevent excess error propagation & facilitate concealment
Error resilient prediction %
Layered coding with unequal error protection %
Multiple description coding $
Other joint-source-and-channel approaches
==> next time
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [5]
Recall: Reversible Variable Length Codes (RVLC)Recall: Reversible Variable Length Codes (RVLC)^ "
Built-in error detection capabilities of VLCs^ ñ
A damaged bitstream is not necessarily decodable ñ^
But error could affect decoding of future codewords
"
Reversible VLC^ ñ
Prefix property (same as regular VLC) & suffix property=> two-way decodability ñ^
Help to narrow down the undecodable data
From Sun-Reibman Fig.8.
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [7]
Discussion on Last QFT Discussion on Last QFT "^
Is RVLC necessarily lead to reduction in coding efficiency?^ ñ
Give specific examples to demonstrate your answer
Ö
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [8]
More on RVLC: ReversibleMore on RVLC: Reversible Golomb
Golomb-
-Rice CodeRice Code
"
Golomb-Rice codes^ ñ
Near-optimal class of VLC for exponentially distr. non-negative integers^ #
e.g., magnitude of residue in predictive image/video coding
ñ^
Nicely structured^ #
adjustable parameters choose ìkî ~ determine by exp. distr. param. # represent an integer n with (quotient, remainder): n = q 2
k^ + r
#^
GR codeís prefix: unary representation of q (i.e., 0, 10, 110, 1110, Ö ) #^
GR codeís suffix: k-bit fixed length binary representation of r
ñ^
E.g. for k=2,^ #
6 = 1 x 2
2 + 2 => 10
10
#^
9 = 2 x 2
2 + 1 => 110
01
ñ^
of codewords at each length is constant
"
Reversible GR codes^ ñ
Need only make prefix (variable-length) to be reversible ñ^
Change unary representation to symmetric: 0, 11, 101, 1001, Ö ñ^
Code length unchanged!
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [9]
ReversibleReversible-
-GR Codes (contíd)GR Codes (contíd)
"
Extension: Exponential-Golomb codes & Reversible version^ ñ
Allow # of codewords at each length L to grow exponentially w.r.t. Lñ Well match run-length coded data from quantized image transf. coeff.
#^
More stable even under mismatch actual source distr. & model
From Wen et al. ICIPí97 Table
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [10]
Error Resilient Entropy Codes (EREC) Error Resilient Entropy Codes (EREC) "
Distribute several VLC codewords of groups of blocks intoa set of equal-size slots^ ñ
Little overhead in bitrate ñ^
One bit error only affect codewords in one slot
From Sun-Reibman Fig.8.
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [16]
Multi- Multi
-resolution Channel Coding/Modulationresolution Channel Coding/Modulation
"
Desired multi-resolution characteristics in source and ch.^ ñ
For source coding ~ to adapt to bit rate changes (scalable coding, etc.)ñ For ch. coding/modulation ~ to adapt to varying ch. distortions via ahierarchy of ìresolutionsî in noise immunity
"
Multi-resolution ch. coding/mod as useful tools for JSCC^ ñ
Allows efficient adaptation to different ch. impairment and to differentimportance levels of sourceñ Techniques reflecting digital and/or analog philosophy
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [17]
Unequal Error Protection CodesUnequal Error Protection Codes^ "
UEP channel codes^ ñ
Families of ch. codes that can change their coding strengths flexibly andefficiently ñ^
Often prefer codes with different protection have same encoding/decoding structures ñ^
Useful for dealing with time-varying channel conditions
"
E.g. RCPC codes^ ñ
ìRate-Compatible Punctured Convolutional Codesî ñ^
Start with a rate 1/N conv. code as a mother code (having M mem. units) ñ^
Puncture the N ch. output periodically (with period p) and align survivedoutput serially
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [18]
An Example of RCPC CodeAn Example of RCPC Code
From Cox et al. RCPCpaper 8/91 Fig.
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [19]
More on RCPC CodeMore on RCPC Code^ "
Rate of RCPC code^ ñ
of 1ís in an N x p puncturing table determines overall code rateR = p/(p+i) where i = 1, Ö , (N-1)p
ñ^
ìRate Compatibilityî ~ higher rate codes are nested in lower rate codes^ #
Bits unpunctured in higher-rate wonít be punctured in lower rate
"
Decode RCPC code^ ñ
Optimal (maximum likelihood) sequence decoding via Viterbi algorithm^ #
VA commonly used in conv. code decoding (dynamic programming) # Break distance or correlation metric into additive components # Prune intermediate suboptimal paths in trellis diagram
&
Ref.(1) Hagenauerís RCPC paper in Trans. on Comm. 4/88(2) Cox et al.ís RCPC for subband speech coding in Trans. Sig. Proc.8/
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [21]
Embedded ModulationEmbedded Modulation
"
Motivated by T.Coverís info.-theoretic study^ ñ
On broadcast ch. with receivers of different ch. condition or capabilitiesñ Optimal strategy is to superimpose/ìembedî detail info. intended forstronger reciever in the coarse info. intended for noisier receiver
"
Achieve ìembeddingî in modulation domain^ ñ
Provide analog-like gracefuldegradation & UEPñ Intracloud-to-Intercloud distanceu = d1/d
#^
to determine bit error rate ofdifferent levels
ñ^
ìSatellites-over-cloudî strategy:some similarity with data hidingvia Costaís dirty paper codes
From Pradhan et al. ICIPí97 Fig.
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [25]
SummarySummary^ "
Finish error resilient entropy coding
"
Joint Source-Channel approaches for MM comm.^ ñ
Digital vs. Analog philosophy for joint source-channel coding ñ^
Multiresolution channel coding and modulation techniques^ #
UEP codes # Embedded modulation
ñ^
(A simple example of multiresolution joint source-channel coding)
"
Next time:^ ñ
(Yefengís survey) Decoder-side error concealment & Encoder-decoderinteractive error recovery ñ^
(Charlesí survey) MM over DSL
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [26]
Suggested ReadingSuggested Reading "^
Error resilient entropy coding^ ñ
Chapt. 8 of Sun-Reibmanís bookñ Wen et al. ICIPí97 paper on RVLCñ Takishima et al. Trans. on Comm. 2/95 paper on RVLC
"
Joint Source-Channel approaches^ ñ
Chapt. 14.4.6 of Wangís video textbookñ Chapt. 7 (by Ramchandran-Vitterli) of Poor-Wornellís ìWireless Commî
M. Wu: ENEE739M Multimedia Comm. & Info. Security (S'02)
Lec15 ñ Joint Source-Ch. Approach 4/2/02 [27]
Questions for Today^ Questions for Today "^
Alice has L parallel channels to send messages to Bob^ ñ
Each channel
i^
is independently affected by slow time-varying distortion
ñ^
Both Alice and Bob know a set of data referred as ìtraining blockî^ #
Bob can use training block to determine current channel conditions # Data blocks sent soon after training can be decoded with less errors
ñ^
For each channel and at each given time slot, Alice can choose to send atraining block or a real data block
Suggest to Alice nice strategies to send training block atwhat time slot and which channel^ [case-1] for delivering email messages; [case-2] for delivering image/video
Time
Ö...
Ö...
Training Block
Data Block 1
2
3
BER
1 < BER
2 < BER
Ö... 3 <
Ö...
Ö...
Fig. is from Y.Sun DSPtalk 3/
