Unequal error protection for packets with variable length

[0001] [Technical field to which the invention belongs] The present invention relates to coding a data stream. [0002] The present invention further relates to the transmission and reception of data streams. [0003] [Conventional technology] Encoding of such data streams refers to the papers of M. Budagavi, W. Rabiner Heinzelman, J. Webb, R. Talluri, Wireless MPEG-4 Video Communication on DSP Chips, IEEE Signal Processing Magazine, January 2000. Has been done. In this paper, in order to make the compressed bitstream more tolerant, the MPEG-4 video compression standard incorporates several error recovery measures in its simple profile to detect, suppress, and conceal errors ( Tolerance) is disclosed. These have 10 bit errors^-3It is a powerful source coding method for combating bit errors when it occurs at the following rates: However, in today's wireless communication paths, the bit error rate (BER) can be higher. Severe conditions in mobile wireless communication paths are caused by spatial movement between transmitters and receivers, and multipath fading due to changes in surrounding terrain. Multipath fading manifests itself as a long burst error. Therefore, some interleaving and channel coding techniques are needed to improve channel conditions. By using a combination of source coding and channel coding, MPEG-4 simple profile video compression can be used to achieve acceptable visual quality over error-prone wireless channels. Become. The structure of an MPEG-4 compressed bitstream is also a method of source-communication coupled coding to ensure that critical parts of the bitstream are error-free (UEP). ) Is also suitable for use. [0004] [Means for solving problems] One object of the present invention is to provide improved data transmission. To this end, the present invention provides coding, decoding, transmission, reception, data streams, and storage media as defined in the independent claims. Suitable examples are defined in the dependent terms. [0005] The present invention is particularly suitable in the field of wireless transmission of MPEG-4 video. The inventors of the present invention need that the encoding used is variable length and that each packet has an integer number of macroblocks, so that the MPEG-4 packet is not exactly the same length, and I realized that different packets have different partition lengths. This means that the fixed UEP structure cannot be used, and in order to perform decoding with the correct code rate, the bitstream structure must be known by the receiver at the channel decoding stage. It means that. Like partitions, packets are not the same length. Therefore, the UEP structure must change dynamically for each packet, and it is required to know the length of the partition. The present invention provides UEP for variable length packets and partitions. [0006] According to the first aspect of the present invention, each partition of at least one packet of the data stream is encoded with different error protection rates, and the length of each of the respective partitions is at least said. It is determined by multiplying the length of one packet or a portion of the at least one packet by a predetermined percentage of each. By providing partitioning with a fixed percentage, UEP for variable length packets is possible. [0007] In the most practical embodiment, the length of all partitions in an entire packet is determined by the percentage of that packet length. However, the length of some partitions may be determined by a fixed, predetermined length. In this case, the length of the remaining partition is preferably determined by a percentage of its packet length. This "part" is usually equal to the sum of the lengths of the partitions where each length should be determined by a percentage (of this "part"). In the most practical embodiment, this is equal to the packet length minus the sum of the fixed length portions. Therefore, a combination of a fixed partition length and a proportional partition length is possible. [0008] It is preferred that the predetermined packet length is determined as the distance between the two markers in the data stream, and at least one of the two markers indicates the start of the packet. [0009] It is preferred that each of the predetermined percentages be determined such that the first partition of the packet has at least the first original packet partition. The first original packet partition may be the header of the packet. Under standard conditions, protecting the entire header with the same protection rate, preferably higher than for subsequent partitions, by choosing the first percentage so that the header is always contained in the first partition. Is possible. Further percentages are preferably determined so that the sum of the given partition and the previous partition contains the same number of original partitions. [0010] In the case of the decoder according to the embodiment of the present invention, it is a data stream having at least one packet having a predetermined packet length, and each partition of the at least one packet has a different error protection rate. A data stream that is encoded and whose respective length of each of the packets is determined by multiplying the packet length by a predetermined percentage of each is received, and each of the packet partitions is said to be said. Decrypted with different error protection rates. [0011] BEST MODE FOR CARRYING OUT THE INVENTION The above and other advantageous aspects of the present invention will be clarified and elucidated with reference to the examples described below. [0012] The figure merely shows the components necessary to understand the present invention. [0013] MPEG-4 bitstreams are very error-prone because they use compression, and especially predictive coding and variable length coding (VLC). R Talluri, Error-resilient video coding in the ISO MPEG-4 standard, IEEE Communication Magazine, vol. 36, no.6, The June 1988 paper describes an error recovery perspective for video coding techniques standardized in the ISO MPEG-4 standard. The specific means adopted by the ISO MPEG-4 standard to enable the communication of compressed video data over noisy wireless communication paths are disclosed in detail. These techniques include resynchronization strategies, data partitioning, reversible variable length coding, and HEC (Header Extension codes). [0014] These measures help add resistance to the MPEG-4 bitstream. By using the Resync marker, an MPEG-4 bitstream will consist of packets of approximately the same length. Despite these measures, the reception quality achieved when MPEG-4 is transmitted over a wireless communication path is still low. However, the error recovery means can further improve the received video quality when used in the channel coding step. In particular, data partitioning means perform non-uniform error protection (UEP) by separating the information bits contained in each packet into three partitions, each with different sensitivity to channel errors. It can be usefully used for the execution of. As shown in FIG. 1, for an I frame, the partition is composed of a header HI, a DC DCT coefficient, and an AC DCT coefficient separated by a DC marker DCm. For the P frame, the partition is composed of the header HP, the motion partition m, and the texture partition tp separated by the motion marker mm. [0015] Appropriate techniques that take into account the characteristics of both the wireless communication path and the application are described below. In particular, information about the different sensitivities of the source bits to channel errors must be utilized by the UEP. This technique performs error protection depending on the perceived sensitivity of the source bit to error. That is, the more sensitive bits are protected by higher protection (corresponding to lower rate coding), and the less important bits are protected by lower protection (corresponding to higher rate coding). Compared to traditional forward error correction (FEC), UEP leverages the characteristics of the source to achieve higher perceived video quality at the same bit rate. [0016] In the proposed configuration, the three partitions are protected by different code rates, depending on the intrinsic importance of the information associated with each partition. The information contained in the header is crucial for the subsequent decryption of the packet, so they must be strongly protected. In the intraframe, the DC coefficient is of greater intrinsic importance than the AC coefficient. Therefore, the DC coefficient must be more protected than the AC coefficient. With respect to the prediction frame, the motion data must be more strongly protected than the texture data, since the texture information can be partially reproduced if the motion information is correctly received. [0017] The UEP execution of the present proposal also considers different importance for different types of frames in the MPEG-4 standard. Intraframes, prediction frames, and reverse prediction frames are considered. Here, the intra frame is encoded independently of the other frames, and the predicted frame utilizes the information from the adjacent frame. [0018] Correct reception of intraframes is crucial for performing subsequent motion compensation for predicted frames. Thus, predicted frames can be encoded with a higher average coding rate (ie, lower protection), while intraframes can be encoded with a lower average communication path coding rate (ie, higher protection). Should be related. FIG. 2 graphically illustrates the above protection configuration. [0019] UEP may be performed by a rate compatible punctured convolution (RCPC) code with a code rate chosen according to the perceived importance of the bit. In this case, the code considered is obtained by puncturing the same "mother" code. Therefore, only one encoder and one decoder are required to perform coding and decoding of the entire bitstream. Such rate-compatible puncttured convolutional codes (RCPC codes) are described by J. Hagenauer, Rate-Compatible Punctured Convolutional Codes (RCPC Codes) and their Applications, IEEE Trans. Commun., vol.36, no.4, pp. 389-400, Known by April 1988. [0020] Different average coding rates are considered for the protection of different frames (I frames are encoded with higher protection / lower rate, lower protection / higher average rate for P frames. Will be considered). In addition, for each frame, the data partitioning means attached to the MPEG-4 standard are utilized to provide stronger protection for the top-level partition. If the frame is not received correctly, it may be retransmitted. [0021] [0021] MPEG-4 encoded bitstreams are Video Objects (VO), Video Object Layers (VOL), Group of Video Object Planes (GOV), Video Object Planes. (VOP) and consists of packets. For synchronization, the start of each part of the bitstream is displayed with the appropriate start sign. The start sign is a unique word that can be identified as any set of variable-length encoded words defined in MPEG4. H1 represents VO, H2 represents VOL, H3 represents GOV, H4 represents the start sign of VOP, and H5 represents the packet start sign (resynchronization). [0022] Here, one major problem is that variable length coding is used and each packet is required to have an integer number of macroblocks, so MPEG-4 packets are exactly the same length. This means that the partitions will have different lengths if the packets are different. This means that the fixed UEP configuration cannot be used, and in order to perform decoding with the correct code rate, the bitstream structure must be known on the receiver side at the communication path decoding stage. It means that. Like partitions, packets are not the same length. Therefore, it is necessary that the UEP configuration must be dynamically changed for each packet and that the partition length is known. One solution to implement UEP, namely Proportional UEP (Proportional UEP), is proposed for this problem. [0023]<u style="single">Proportional non-uniform error protection</u><u style="single">(P-UEP)</u>FIG. 6 shows the structure of proportional non-uniform error protection. Since the length of each field is unknown to the receiver, a proportional structure is used in consideration of the (variable) length of the packet. The packet length is preferably determined by receiving two appropriate start codes (at least one of which is the packet start code). Such a structure introduces a delay of one packet to fill the packet buffer. The percentage length for each partition is selected, taking into account the characteristics of the bitstream. Percentage length P₁, P₂, P₃Each of the three partitions has a rate of R₁, R₂, R₃If protected by, the average rate for I packets is [Formula 1]<img file="JP4659331B2_D0001.tif" />Given in, likewise for P packets, [Equation 2]<img file="JP4659331B2_D0002.tif" />Given in. Therefore, the length of the encoded packet is relative to the I frame. [Equation 3]<img file="JP4659331B2_D0003.tif" />And for the P frame, [Equation 4]<img file="JP4659331B2_D0004.tif" />Will be. Here, when considering the convolutional code, M is the number of memories of the code. Regarding the number of memories M of the code, the convolutional code differs from the block code in the following points. That is, the encoder has memory, and the output of the encoder in any time unit depends not only on the input in that time unit, but also in the input block M (M is the number of memory of the code) before. Is also dependent. A convolutional coder with M memories consists of M-stage shift registers, and the output of the selected stage is modulo-2 added to generate an encoded symbol. Since the convolutional coder is a sequential circuit, its operation can be described by a state diagram. The state of the observer is defined as this shift register state. Therefore, the encoder is 2^MYou may have a state. To protect the last bit of the bitstream that has the same strength as the other bits, M tail bits are added to this bit so that the encoder converges to a known state (usually the 0 state). Must be added to the stream. In fact, when the convolutional code is considered, the packet is terminated by shifting M "0" bits to the shift register so that proper trellis termination can be done. The tail bits are encoded at a higher rate. To calculate the overall average rate, we must calculate the average of the I-frame and the P-frame, and also consider the overhead introduced by the start sign replacement. [0024] This aspect of the invention captures each predetermined percentage of variable packet length as each packet partition. This percentage is the sum of the first and second partitions, with the first partition of the packet having at least the first original packet partition (eg, header), taking into account the characteristics of the data stream. It is preferred to have at least the first original packet partition and the second original packet partition, and so on. [0025] The second major problem is that even a single error in the start code can cause its detection to fail and cause loss of synchronization. The MPEG-4 start code is for errors. The point is that it is not strong. To address these issues, the present invention proposes several suitable solutions. When an error occurs, it is possible to detect the failure and emulate the start code. To solve this problem, start sign replacement is proposed. [0026]<u style="single">Start sign replacement</u>In a further embodiment, the start sign is replaced by a pseudo-noise word, which is a highly correlated sequence (eg, Gold sequence) after MPEG-4 coding (see Figure 3). These new start codes are called "Wireless Start Codes". In particular, the replacement is done for the start sign of VO, VOL, VOP, GOV and the "resynchronization" marker. FIG. 3 shows a coded data stream S with markers H1, ..., H5. These markers are replaced by markers WH1, ..., WH5, which are more resistant to communication path errors, resulting in a data stream WS suitable for wireless transmission. The data stream WS is received by the receiver as a data stream RS that is similar to the WS but may contain a communication path error. Markers WH1, ..., WH5 are WH1_R, ..., WH5_RIs received as. Marker (word) WH1_R, ..., WH5_RIs similar to WH1, ..., WH5 but may contain communication path errors. Since these markers are highly correlated, they are recognized as WH1, ..., WH5 and then replaced by markers similar to H1, ..., H5, respectively. Given that it is an MPEG-4 bit stream, the data stream in Figure 3 does not include the GOV start sign (H3). For MPEG-4 bitstreams, the VOL start sign (H2) also indicates the start of the GOV, so there is no GOV start sign (H3) after the VOL start sign (H2). [0027] On the receiver side, the positions of these radio start codes WH1, ..., WH5 are estimated by correlation prior to the communication path decoding process. There must be a trade-off between the probability of losing the start sign and the probability of emulating the start sign. Therefore, the selection of appropriate thresholds for the "radio start code" length and correlation is performed accordingly. When the detection is performed, the radio start signs WH1, ..., WH5 are replaced by the corresponding start signs H1, ..., H5 from the original set of start signs. The above replacement is thus transparent to the MPEG-4 decoder (see Figure 5). [0028] In the channel coding step, a preferred embodiment of the present invention, start code replacement in combination with proportional non-uniform error protection (P-UEP), is proposed. [0029] A simplified case where the VOP and the frame match will be described in this preferred embodiment. In FIGS. 4 and 5, the dashed line indicates the control line. [0030] FIG. 4 shows a transmitter according to the present invention. This transmitter has a start code detector 12 for detecting start codes H1, ..., H5. The detected start code is replaced by the pseudo noise word generator 13 with the corresponding pseudo noise words WH1, ..., WH5. The pseudo-noise words WH1, ..., WH5 are supplied to the multiplexer 14, and the pseudo-noise words are added to the data stream WS transmitted there. [0031] Packet buffer 10 receives the data stream S. The packet of the data stream S existing between the markers H1, ..., H5 is channel-coded by the channel encoder 11 to obtain a channel-encoded packet. These channel-encoded packets are fed to the multiplexer and added to the transmitted data stream WS. The transmitted data stream is supplied, for example, to an antenna or storage medium 15 for wireless transmission. [0032] It is preferable that the communication path coding of FIG. 4 is performed by using P-UEP as described above, but instead, another communication path coding mechanism may be used. [0033] FIG. 5 shows receiver 3 for receiving the data stream WS transmitted by the transmitter according to FIG. In order to detect the pseudo-noise word indicating the start sign in the start sign detector 32 (for example, in practice, the pseudo-noise word detector), (from a predetermined set of pseudo-noise words corresponding to the markers). ) Correlation evaluation is performed between each permissible pseudo-noise word and the associated bitstream section. The correlation is compared to the corresponding threshold th. When a pseudo-noise word is detected, the bit identifier in the bitstream shifts the appropriate number of bits, and the corresponding MPEG-4 start code H1, ..., H5 is supplied by the start code generator 33. Will be done. This start sign is inserted into the multiplexer 34, whose function is to arrange the bitstream S'as provided to the MPEG-4 decoder. When a GOV or VOP start sign is detected, the VOP identifier changes its status. [0034] When the "resynchronization" marker is detected, packet buffer 30 is initialized and the buffer is filled with subsequent bits until the next start sign is detected. Correlation evaluation is not performed until the buffer contains N (N is the minimum length of one packet) bits. When the next start sign is detected, buffer 30 contains one packet. At this time, the channel decoding is performed on the bits in the buffer of the channel decoder 31 according to the VOP identifier information and the percentage. The rate used for this configuration is preferably fixed and the same as that used for the channel coding device 11. If a variable rate is used, that rate must be received from the channel coding device 11 of transmitter 1. The communication path decrypted packet is inserted into a multiplexer 34, which arranges the bitstream for delivery to the MPEG-4 decoder. Note that de-puncturing is performed prior to decoding when RCPC codes are used. In this case, the packet is then decoded at the mother code rate. [0035] Although not shown in FIGS. 4 and 5, the data stream may be modulated by the modulator in the transmitter prior to transmission and demodulated in the receiver by the demodulator before performing decryption. Good. [0036] The above-mentioned examples are for exemplifying the present invention, do not limit the present invention, and many others without departing from the scope of the claims described by those skilled in the art. Note that it is possible to design an embodiment of. In the claims, none of the reference numbers in parentheses should be construed as limiting the claims. The term "have" does not preclude the existence of elements or processes other than those listed in the claims. The present invention can be run by hardware with several separate elements, or by a well-programmed computer. In a device claim that lists several means, some of those means can be implemented by one and the same hardware. The fact that some measures are listed in different dependent claims does not mean that it is not effective to use them in combination. [0037] In summary, in encoding a data stream, the data stream has at least one packet with a predetermined packet length, and each partition of the at least one packet has a different error protection rate. Encoding is provided in which the respective lengths of the respective partitions are determined by multiplying the packet length or a portion of the packet length by their respective predetermined percentages. .. [Simple explanation of drawings] FIG. 1 shows data partitioning of an MPEG-4 bitstream. FIG. 2 shows a schematic representation of a protective structure according to an embodiment of the present invention. FIG. 3 shows start sign replacement and (proportional) non-uniform error protection according to an embodiment of the present invention. FIG. 4 shows a transmitter according to an embodiment of the present invention, which has a start sign detecting means and a start sign replacing means. FIG. 5 shows a receiver according to an embodiment of the present invention, which has a replacement start sign detecting means and a replacement start sign restoring means. FIG. 6 shows proportional non-uniform error protection according to an embodiment of the present invention. [Explanation of symbols] 1 transmitter 3 receiver Ten Packet buffer 11 Channel coder 12 Start sign detector 13 Pseudo-noise word generator 14 Multiplexer 15 Antenna or storage medium 30 packet buffer 31 Channel Decoder 32 Start sign detector (pseudo noise word detector) 33 Start sign generator 34 Multiplexer H1, H2, H3, H4, H5, WH1, WH2, WH3, WH4, WH5, WH1_R, WH2_R, WH4_R, WH5_R Marker (start sign) S, RS, WS data stream