CN101056414A - A joint coding method for video data stream source and channel - Google Patents

Abstract

The invention provides a video data stream and source channel combination decoding method, performing channel decoding on video communication data stream by using an LDPC decoder, then obtaining video images by decoding the channel decoded result via a video decoder, a particular process of a combination iterative decoding of the LDPC decorder and the video decorder comprises: (1) at first decoding the channel decoded result by the video decoder, detecting wrong codes during the decoding process by using video protocol phraseology, semantic analysis phraseology units, if a wrong code is detected, an error range of the video stream is given and is converted into a probability likelihood ratio of a corresponding wrong inforamtion node; (2) inputting the probability likelihood ratio of the corresponding wrong inforamtion node into the LDPC decoder to perform internal iterative decoding in the LDPC decoder; (3) for the latest obtained channel decoded result, repeating proceses from the step (1) to a next combination iterative decoding till the phraseology units of the channel decoded result is accorded with a video protocol.

Description

A kind of joint coding method for video data stream source and channel

Technical field

The invention belongs to the wireless video communication technical field, particularly joint coding method for video data stream source and channel.

Background technology

Video traffic is one of most important business on the third generation the 4th third-generation mobile communication platform.In video communication, the challenging real-time Transmission that in wireless network, realizes video beyond doubt of tool, this must overcome, and the video data volume is big, delay variation, transmission bandwidth fluctuation, error burst packet loss, multipath fading and multiple access reception or the like problem.The error code that video code flow after the high compression produced when transmission is very responsive, in case the recovery that error code not only influences this error code data has taken place, also can influence the recovery of other associated data, cause " error code diffusion " (Error Propagation), make the signal that recovers to come out changed beyond recognition, have a strong impact on the quality of video communication, the bit error correcting capability that improves communication system is the key of dealing with problems.

At present communication system all is basically under the guidance of Shannon's theorems, communication system is divided into source encoding and chnnel coding designs respectively, and main purpose is: global optimization is decomposed into some local optimums.The optimization principles of source encoding is compression efficiency and goes back proper mass, and the optimization principles of chnnel coding is error correcting capability and coding cost.And in the design of wireless video communication system, limitation has appearred in the local optimum principle of shannons coding theorem, it is supposed the carrying out of information source and channel is desirable: information source satisfies steady, the asymptotic property cut apart, each attitude experience property, the characteristics such as channel has steadily, independent random distribution of being equal to.But for video communication, the data importance of information source is different, and the probability statistics of appearance are also inequality, for the degrees of tolerance difference of channel error code.And Shannon's theorems is an existence theorem, realizes that reliable communication needs very long chnnel coding length, and is not suitable for the optimization wireless video communication system.As seen prior art is difficult to provide the bit error correcting capability that the effect technique means improve video communication system, and this ability is also referred to as the robustness of video communication system to channel error code in this area.

Summary of the invention

The object of the invention is to solve the prior art deficiency, and a kind of video data stream decoding method that can improve video communication system bit error correcting capability is provided.For achieving the above object, the present invention adopts ldpc decoder and Video Decoder associating iterative decoding, and the detailed process of associating iterative decoding is,

(1) at first by the channel decoding result being carried out in the decode procedure at Video Decoder, utilize video protocols grammer, semantic analysis syntactic units, detect whether error code is arranged, when detecting error code, provide the video flowing fault domain, and be converted to the probability likelihood ratio of corresponding information node of makeing mistakes;

(2), carry out the inner iterative decoding of ldpc decoder with the probability likelihood ratio substitution ldpc decoder of corresponding information node of makeing mistakes;

(3) the channel decoding result that newly obtains is repeated to begin to carry out next round associating iterative decoding by step (1), meet video protocols up to channel decoding result's syntactic units.

And, after video decoder decodes obtains video image, judge by analyzing macroblock image whether picture decoding is correct, when the picture decoding mistake, return and carry out ldpc decoder and Video Decoder associating iterative decoding, when picture decoding is correct, stop iteration.

And step detects when not having error code in (1), and further the acknowledgement channel decode results is correct, heightens the probability likelihood ratio of corresponding information bit.

And Video Decoder adopts the MPEG4 video protocols.

Technical scheme of the present invention is utilized the iterative decoding process of Video Decoder to the analytical information participation LDPC sign indicating number of bit stream, forms message source and channel dual decoding device associating iterative decoding.This scheme improves the error correcting capability of channel by the high level syntax information of Video Decoder, improve the decode procedure of Video Decoder to video code flow, can prevent the error code diffusion, improve the robustness of video communication, guarantee the wireless video communication quality channel error code.

Description of drawings

Fig. 1 is video data information bit of the present invention and LDPC chnnel coding hierarchical relationship figure.

Fig. 2 is a video flowing associating iterative decoding process schematic diagram of the present invention.

Fig. 3 is that Video Decoder of the present invention is checked wrong schematic diagram

Embodiment

Referring to accompanying drawing 2, video data stream decoding method and technology scheme provided by the invention is: adopt ldpc decoder that video data stream is carried out channel decoding, adopt video decoder decodes channel decoding result to obtain video image then, the process of associating iterative decoding is, ldpc decoder and Video Decoder associating iterative decoding, the detailed process of associating iterative decoding is

(1) at first by the channel decoding result being carried out in the decode procedure at Video Decoder, utilize video protocols grammer, semantic analysis syntactic units, detect whether error code is arranged, when detecting error code, provide the video flowing fault domain, and be converted to the probability likelihood ratio of corresponding information node of makeing mistakes;

(2), carry out the inner iterative decoding of ldpc decoder with the probability likelihood ratio substitution ldpc decoder of corresponding information node of makeing mistakes;

(3) the channel decoding result that newly obtains is repeated to begin to carry out next round associating iterative decoding by step (1), meet video protocols up to channel decoding result's syntactic units.

According to technical scheme provided by the invention, the basic process of video data stream encoding and decoding is: elder generation was through video coding, LDPC encoding process before video data stream sent, launch by modulator then, channel transmission back is accepted by demodulator, correct error code through the associating iterative decoding then, afterwards again by the video decode output video image.

Key improvements of the present invention is row ldpc decoder and Video Decoder associating iterative decoding, adopts this scheme based on following consideration:

Practical variable length code generally all is a prefix code, and promptly any one code word is not the prefix of another one code word.Traditional variable length code decoder utilizes this rule to decipher just, when reading in input bit successively, judge the some bits read in whether just with certain code word coupling, just corresponding symbol is exported if mate.Prefix decoding computing is very simple, and the memory space demand is also very little, has therefore obtained using widely.Yet, having under the situation of error code, if certain bit that the prefix decoder reads in is made mistakes, just may cause and lose between the codec synchronously, at this moment not only current sign can be translated mistake, and several symbols of back also can be translated mistake, and the result that the error code diffusion causes is catastrophic sometimes.Because the various piece of video communication data is different to the noise effect sensitivity, partial data is very responsive to noise simultaneously.Therefore the present invention considers to be used to carry out soft source decoding from the prior information of information source and the soft output of channel decoder.Employing is handled variable length code decoding based on the algorithm of sequencal estimation, the benefit of this method is that not only the simple relatively prefix decoding of decoding performance can make moderate progress, and can accept to realize the associating iterative decoding between the bit sequence of soft input and the channel decoder, thereby improve the performance of decoder greatly.Therefore chnnel coding of the present invention has been selected practical LDPC sign indicating number.

Consider the mechanism of chnnel coding and source encoding, can think like this: video encoder is bit sequence according to video standard with the image sequence compressing and converting, and Video Decoder obtains the continuous videos image by analyzing the bit sequence decoding.Therefore Video Decoder is to obtain by the relation between video information bit and the information bit to the understanding of video code flow as can be seen, and channel decoder is obtaining by the relation between information bit and the correction bits to the bit correct understanding.Both are the difference understanding to same code stream, and both understanding has difference, can eliminate this difference by iteration and reach same recognition results, the reliability of deepening the understanding.Therefore utilize the high level syntax information of Video Decoder to carry out the error correcting capability that iterative decoding will be hopeful further to improve channel, break through the performance boundary of present irregular LDPC codes.

Therefore, the present invention proposes to utilize the information exchange of ldpc decoder and Video Decoder, realizes the associating iterative decoding, improves decoded result with this.Each iteration, the probability likelihood ratio substitution ldpc decoder of adjusting through Video Decoder, and each iteration result of all having utilized last iteration to obtain.After the several times iteration, ldpc decoder is just consistent in the judgement of all bits with Video Decoder, has finally improved the error correcting capability of communication system.

Whether above-mentioned base case provided by the invention is based on unites iterative decoding to the differentiation decision of bit, further scheme is based on: after video decoder decodes obtains video image, judge by analyzing macroblock image whether picture decoding is correct, when the picture decoding mistake, return and carry out ldpc decoder and Video Decoder associating iterative decoding, when picture decoding is correct, stop iteration.Whether increase unites the judgement of iterative decoding by the evaluation of image decision, is in order to differentiate based on macroblock image, the approach of another kind of discovery error code to be provided, helping further improving the error correcting capability of communication system.

Because it often is not the accurate position that mistake takes place that decoder detects the position of mistake, so decoder is after detecting mistake, need take heavy synchronization policy regain with encoder synchronously.H.26x, current video encoding standard all takes one of heavy synchronization policy exactly each two field picture to be divided into several piece groups (GOB or Slice) with MPEG-x, and decoder can utilize the initial code of GOB as heavy synchronous code.But this heavy synchronization policy, the distance between heavy synchronous code is unequal, for the violent part of image motion, code word is bigger at interval, in case make a mistake, the time that decoder recovers synchronous is longer, affected data are more, are unfavorable for the further mistake recovery of decoder processing.MPEG (being Moving Picture Experts Group moving image panel of expert) is an international standard, promptly so-called ISO11172.In the MPEG-4 agreement, defined a kind of new heavy synchronization policy, made the heavy synchronous code word space in the MPEG-4 video flowing keep equal substantially based on video packets (Video Packet).This strategy is that encoder is divided into some video packets with every two field picture, and video packets is made up of complete macro block, and its length is determined that by pre-set threshold encoder is realized synchronously heavy by inserting heavy synchronization code word at each place, video packet header.Therefore, Video Decoder of the present invention adopts the MPEG4 video protocols.Also can adopt MPEG1, MPEG2 agreement during concrete enforcement.

Below video data stream decoding method provided by the invention is specified so that implement:

Referring to Fig. 1, the LDPC sign indicating number is a kind of special linear block codes, and its check matrix H is a sparse matrix, and when representing with Tanner figure, there is M node C the top ₁, C ₂C _M, each check-node is represented a checksum set of code word, is called check-node; There is N node V the centre of figure ₁, V ₂V _N, each node is represented an information bit position of code word, is called variable node; There is W node I in the bottom of figure ₁, I ₂I _W(W gets 4 among the figure), each node are represented a grammatical symbol of compressed video stream.And have fillet between corresponding check-node of " 1 " element and the variable node in the check matrix, the present invention is called adjacent node with the node at two ends, a limit.The limit number that each node links to each other is called the number of degrees of this node.With the corresponding node I of MPEG4 video protocols ₁, I ₂I _WWith variable node V ₁, V ₂V _NBetween have fillet, reflect many-to-one mapping relations, the representative information bit is formed the syntactic element of video by certain sequence.For regular LDPC sign indicating number, the number of " 1 " was identical during each row and each were listed as in its check matrix; To non-regular code, the number of degrees of variable node and check-node are unfixed, and they are to select according to certain distribution.Can distribute by the number of degrees of following defined variable node and check-node:

$\mathrm{\μ}\left(\mathrm{\χ}\right)=\underset{i=2}{\overset{d_v}{\mathrm{\Σ}}}{\mathrm{\μ}}_i{\mathrm{\χ}}^{i-1},v\left(x\right)=\underset{j=2}{\overset{d_c}{\mathrm{\Σ}}}{v}_j{\mathrm{\χ}}^{j-1}$

μ (χ) the expression number of degrees are variable node shared ratio in total node of i, and υ (χ) the expression number of degrees are check-node shared ratio, the wherein d in total node of j _vBe the number of times of maximum variable node, d _cIt is the number of times of the check-node of maximum.

Ldpc decoder carries out inside decoding.Belief propagation algorithm is based on the probability inference algorithm that graph model carries out local message transmission and renewal.

Anticipated and the situation that there is content false in certain video image occurs in Fig. 1 illustration of passing the imperial examinations at the provincial level, the corresponding bit mistake of video image content mistake is present in node I ₂, I ₃, judge video flowing fault domain (dotting among the figure) then in view of the above, corresponding information node of makeing mistakes is node V ₅, V ₆V _N-3, V _N-2And adjust the probability likelihood ratio of corresponding information node of makeing mistakes.More than belong to the process of adjusting according to the video decode result, the probability likelihood ratio that adjustment obtains is brought ldpc decoder into and is carried out once more inner iterative decoding.

The process that ldpc decoder carries out channel decoding belongs to prior art, may further comprise the steps,

Step a: initialization:

$\mathrm{LLR}\left(p_i\right)=\frac{2}{{\mathrm{\σ}}^2}{R}_i,1\≤i\≤N$

Suppose that channel is additive white Gaussian noise channel, wherein σ ²Be the standard deviation of interchannel noise, q _IjFor send to the external information of check-node j, r from variable node i _IjFor send to the external message of variable node i from check-node j.R _iBe the information after the demodulation of channel reception, p _iBe the initialized logarithm probability likelihood ratio of variables corresponding node i, initialization, LLR (r _Ij)=0.

Step b:LDPC decoding comprises two steps:

Send information to check-node from variable node.

$\left(q_{\mathrm{ij}}\right)=\underset{j^{\′}\≠1}{\underset{j^{\′}\∈\mathrm{Col}\left[i\right]}{\mathrm{\Σ}}}\mathrm{LLR}\left(r_{\mathrm{ij}}\right)+\mathrm{LLR}\left(p_i\right)$

Col[i wherein] refer in check matrix the element of non-zero in the i row.

From the check-node to the variable node

$\mathrm{LLR}\left(r_{\mathrm{ij}}\right)={\mathrm{\Φ}}^{-1}\left(\underset{j^{\′}\≠i}{\underset{i^{\′}\∈\mathrm{Row}\left[j\right]}{\mathrm{\Π}}}\mathrm{\Φ}(\mathrm{LLR}\left(q_{i^{\′}j}\right)\right)$

Row[j wherein] refer in check matrix i capable in the element of non-zero.Definition

$\mathrm{\Φ}\left(\mathrm{\χ}\right)=\tanh (-\frac{1}{2}x)=\frac{e^x-1}{e^x+1}$

Step c: channel decoding output is defined as

$\mathrm{LLR}\left(q_i\right)=\underset{j^{\′}\∈\mathrm{Col}\left[i\right]}{\mathrm{\Σ}}\mathrm{LLR}\left(r_{\mathrm{ij}}\right)+\mathrm{LLR}\left(p_i\right)$

Carry out hard criterion output then:

${\hat{u}}_i=\left\{\begin{array}{ccc}1& \mathrm{if}& \mathrm{LLR}\left(q_i\right)>0\\ 0& \mathrm{if}& \mathrm{LLR}\left(q_i\right)<0\end{array}\right.$

Hard herein criterion output is the channel decoding result.Q wherein _iBe defined as the final probability likelihood ratio of the iteration output of information node i.

Video Decoder participates in the process of iteration and is initiated by the present invention, and the wrong detailed process of check is referring to Fig. 3 among the embodiment:

At first check all syntactic units whether to meet MEPG4 video protocols standard.Comprise the encoded video header of video parameter information for some: sequence head, image sets head, image head, can be according to video grammar and semantic directly analysis.The header of MPEG4 agreement for example: object video head, video object plane head, User Data Header.Analyze corresponding video parameter content, monitor whether find mistake promptly to have error code.Be decoded into by the channel decoding result in the process of image,, need further decode for DCT coefficient that has comprised video content information and motion vector part.In decode procedure, whether the syntactical unit of at first analyzing these variable-length encodings meets the variable length coding table in the video protocols, if be not inconsistent situation, then has error code, need correct.In above-mentioned checkout procedure, if find error code then write down the information bit of corresponding mistake, and be converted to the probability likelihood ratio of corresponding information bit, and return ldpc decoder and carry out corresponding modify generation.If there is no error code, can confirm further that LDPC decoding is correct, heighten the probability likelihood ratio of corresponding information bit, the coefficient of change is made as w, the probability likelihood ratio of whole information bit sequence generation time ldpc decoder, the convergence of ldpc decoder iteration can be better so then.

Then, in order to differentiate based on macroblock image, need recover corresponding macroblock image, continuity of the average gray by analyzing continuity between the macroblock image, macroblock edges continuity, macro block or the like standard judges whether picture decoding is correct.This analytic process can be carried out by software, and the present invention is called the decoded picture evaluator.If picture decoding is incorrect, need determine corresponding information bit of makeing mistakes by analyzing, corresponding probability likelihood ratio need be adjusted, and carries out corresponding modify for returning ldpc decoder, promptly carries out ldpc decoder and Video Decoder associating iterative decoding once more.Judge that the reliability of judgement is less than the result who judges by video grammar owing to carry out error correction by the parameter of image, so it is little to adjust the adjustment of numeric ratio grammer error correction accordingly, the coefficient of change is made as v, v＜w.

MPEG4 video protocols standard comprises multiple reference model, present embodiment provides the picture appraisal process of I frame and P frame: at the I frame, after detecting the video parameter content, check sync mark and decoded video packet header, the MCBPC (macro block (mb) type and chroma coder pattern) that decodes then obtains the parameter information of macro block, analyze MCBPC, monitor whether find mistake; 6 block messages of then decoding are analyzed variable-length encoding, monitor whether find mistake; Be based on macroblock image at last and differentiate, rebuild decoding back picture material, the authentication image content monitors whether find mistake.All write down the information bit of corresponding mistake after the discovery mistake, be converted to the probability likelihood ratio of corresponding information bit, and carry out corresponding modify for returning ldpc decoder.Comprised kinetic characteristic in the P frame, therefore needed decoding motion vectors simultaneously, analyzed motion vector value, monitored whether find mistake at 6 block messages of decoding.Generally can not use the B frame in real time communication, the evaluation method of B frame should be consistent with the P frame, will not give unnecessary details.

The associating iterative decoding needs the design iteration condition as a kind of iterative algorithm:

To the judgement information of channel-decoding and the judged result of decoding picture appraisal device, whether decision-making needs to carry out iterative decoding according to Video Decoder; Just the adjusted probability likelihood of process Video Decoder ratio, the substitution ldpc decoder carries out new inside iterative decoding to iteration, realizes once uniting iterative decoding if desired; If do not need iterative decoding, the associating iterative decoder of channel (being the combination of ldpc decoder and Video Decoder) just carries out hard criterion, and decode results is sent to the upper-layer protocol stack, finishes the decode procedure of channel.Specific algorithm in the associating iterative decoding process is realized adopting DSP (Digital Signal Processing) technology.

Claims (5)

1. joint coding method for video data stream source and channel, at first adopt ldpc decoder that the video communication data flow is carried out channel decoding, obtain video image by video decoder decodes channel decoding result then, it is characterized in that: ldpc decoder and Video Decoder associating iterative decoding, the detailed process of associating iterative decoding is

(1) at first by the channel decoding result being carried out in the decode procedure at Video Decoder, utilize video protocols grammer, semantic analysis syntactic units, detect whether error code is arranged, when detecting error code, provide the video flowing fault domain, and be converted to the probability likelihood ratio of corresponding information node of makeing mistakes;

(2), carry out the inner iterative decoding of ldpc decoder with the probability likelihood ratio substitution ldpc decoder of corresponding information node of makeing mistakes;

(3) the channel decoding result that newly obtains is repeated to begin to carry out next round associating iterative decoding by step (1), meet video protocols up to channel decoding result's syntactic units.

2. video data stream decoding method according to claim 1, it is characterized in that: after video decoder decodes obtains video image, judge by analyzing macroblock image whether picture decoding is correct, when the picture decoding mistake, return and carry out ldpc decoder and Video Decoder associating iterative decoding, when picture decoding is correct, stop iteration.

3. video data stream decoding method as claimed in claim 1 or 2, it is characterized in that: step detects when not having error code in (1), and further the acknowledgement channel decode results is correct, heightens the probability likelihood ratio of corresponding information bit.

4. video data stream decoding method as claimed in claim 1 or 2, it is characterized in that: Video Decoder adopts MPEG4 video protocols standard.

5. state the video data stream decoding method as claim 3, it is characterized in that: Video Decoder adopts MPEG4 video protocols standard.

Images