KR20080093923A - A sub-block based cyclic updating method for low-delay fountain coding - Google Patents

Abstract

A coding method and a decoding method are provided to suppress delay of data effectively, when transmitting the data through a wireless Internet, which causes the loss in a real-time multimedia service sensitive to delay. An input symbol is divided into the predetermined number of same sub-blocks. An encoding time section is divided into same sub-time sections corresponding to the number of the divided sub-blocks. Initial input symbols are generated by using symbols having sizes corresponding to the sub-blocks. Initial output symbols are generated by using the initial input symbols. Symbols corresponding to a new sub-block, which is to be encoded every sub-time section, are sequentially received and then added to the input symbols(710). A symbol corresponding to the first received sub-block is removed to update the input symbol(720). An output symbol related to the first received sub-block is updated. A packet is generated by using the updated output symbol(725).

Description

Low delay fountain coding method using sequential update method for each subblock {A SUB-BLOCK BASED CYCLIC UPDATING METHOD FOR LOW-DELAY FOUNTAIN CODING}

The present invention relates to video coding, and more particularly, to a method of minimizing delay when using a fountain code for packet loss recovery in a delay sensitive real-time multimedia service.

When the code rate of channel coding is expressed as '(amount of existing information) / (amount of encoded data)', the fountain code is also called a 'rateless' code because the amount of encoded data is not predetermined. The fountain code has the advantage of perfect reception without error even when bidirectional information transmission is difficult, such as when there is insufficient information on the receiver or when the number of receivers is very large. Therefore, it is necessary for multicast in computer network. This was raised. Fountain code is based on the following: First, the transmitter makes a packet continuously encoded using a file to be transmitted and transmits it. Then, each receiving end receives and decodes only packets that can be decoded without requiring feedback. This allows the fountain code to provide asynchronous reception to the receiver, eliminating retransmission requests that cause network overload.

The principles and advantages of fountain codes are introduced in the following documents.

Byers, John W. et al .; "A Digital Fountain Approach to Reliable Distribution of Bulk Data"; 1998, Computer Communication Review, Association for Computing Machinery, vol. 28, No. 4, pp. 56-67.

As such, fountain codes are meaningful in that they can be processed by receiving as much data as you want, regardless of where the data comes from. The fountain code can generate an infinite number of output symbols using k input symbols. Each output symbol is randomly generated using input symbols. The output symbols are not correlated with other output symbols and it is possible to recover the original input symbols using m output symbols that are slightly larger than k. In a peer-to-peer network, the receiving side recovers the original data by connecting to multiple transmitting parties that transmit the same data and receiving slightly more m symbols than k, regardless of which sender sent the data. Can be. This means that when the receiving side is connected to multiple transmitting stations with different network conditions, it can receive more symbols from the transmitting side with better network performance than the transmitting side with poor network performance, and each symbol is not related and therefore the total If the number of received symbols exceeds m, the original data can be recovered.

The first invention that corresponds to the fountain code is the LT code (Luby Transform codes) published in 1998, and in 2004, Raptor code was developed by 'Amin Shokrollahi'.

Raptor code has been adopted as the standard in the application layer of DVB-H (Digital Video Broadcasting-Handheld). This code has better performance in terms of the amount of computation in the decoding process than the existing code, LT code. The biggest drawback of LT codes is that the amount of computation required to recover the source symbols is not linear, and the Raptor code compensates for these drawbacks and allows to obtain the amount of decoding operations in linear time while keeping the amount of coding within linear time.

Thus, raptor code is an advanced form that takes advantage of LT code and tornado code. By adding a low density parity check (LDPC) preprocessing step to the LT code, the decoding rate in the small amount of information is improved. Also, LT code is not systematic code, but raptor code is characterized by systematic code by adding a preprocessing step. In addition, unlike tornadoes, raptor and LT codes do not have a fixed rate. This means that an infinite number of additional packets can be generated as needed. Raptor code also produces graphs with log density, unlike tornado code. In addition, the raptor coding method performs very fast probabilistic decoding by using a linear bitpartite graph.

In systematic code in which n-k parity symbols are calculated for k data symbols, the first k symbols become data symbols and the next n-k symbols become parity symbols. In Reed Solomon code, all n-k parity symbols are generated depending on all k data symbols. However, in fountain code, one parity symbol is made dependent on some data symbols that are different sets. Information about which set of data symbols is generated should be transmitted with the parity symbols.

1 is an exemplary diagram showing each symbol in an operation stage of conventional raptor coding. Referring to FIG. 1, first, input symbols 110 are converted into intermediate symbols 120 through pre-coding (inverse LT encoding, LDPC & Half symbol encoding) for input symbols 110. The intermediate symbol 120 may be LT coded to generate an output symbol 130 having an infinite size. Also, in order to have a systematic characteristic, the first K output symbols 130 have the same value as the input symbol 110, and each output symbol 130 is an XOR result between arbitrarily selected intermediate symbols 120. appear.

Meanwhile, a method of restoring an erasure using the Reed Solomon method is as follows. With RS [n, k], when k data packets are transmitted, n-k parity packets are generated, and when n-k or less packets are lost during transmission, all can be recovered at the receiving side. This method is used in a real-time service because it can save time than the method requiring a retransmission. The number of packets lost can be determined by looking at the sequence number in the RTP packet header. This loss of knowing the lost location is called an erasure. If more loss occurs than n-k, none of the lost packets can be recovered.

Unlike convolutional code, block code such as Reed Solomon code divides the input symbols into k units and calculates n output symbols. That is, n output symbols may be calculated by multiplying a generator matrix of size nⅹk by a vector having a size k. In this case, if the front of the generation matrix is a kⅹk indentity matrix, it is a systematic code, and the first k of n output the input symbols as they are. The remaining output is calculated by multiplying and adding one or more input symbols with coefficients. Since k inputs are cut off and processed and n outputs are not restored to the receiver, there is always a delay as long as k inputs are generated (or the time spent at the receiver). . To minimize this delay, reducing the input block size k degrades the loss resilience even at the same code rate (k / n). Therefore, there is a need for a method where k is large enough and the coding delay can be reduced.

The present invention is to provide a loss recovery method that can reduce the delay when data is transmitted through a wireless Internet network that causes loss in a delay sensitive real-time multimedia service such as a video phone.

According to one embodiment of the present invention for achieving this, in a low delay fountain coding method using a sequential update method for each subblock, dividing an input symbol into a predetermined number of identical subblocks and a predetermined number of input symbols Dividing the encoding time of the input symbol into the same sub-time intervals so as to correspond to the division into the same sub-block of?, Receiving a symbol having a size corresponding to the sub-block, generating initial input symbols and generating the initial input A process of generating and outputting an initial output symbol by using a symbol, sequentially receiving symbols corresponding to a new subblock to be encoded for each sub-time interval, and adding the symbols to the input symbol and correspondingly to a previously inputted subblock. The input symbol by removing a symbol from the input symbol A characterized in that it comprises the step and the step of the process of updating the output symbols associated with the most previously received sub-blocks and to generate a packet using the updated output symbols for updating.

According to another aspect of the present invention, in the low-delay fountain decoding method using a sequential update method for each subblock, receiving and decoding a packet generated using an initial output symbol to generate a reconstruction symbol, and an encoded subblock Restoring the original subblock by using the reconstruction symbol when receiving a packet including the information, and adding the symbols corresponding to the newly received subblock in the reconstruction symbol, and corresponding to the previously received subblock. And removing the symbols to update the reconstructed symbols.

According to the present invention, when a multimedia stream is transmitted in a lossy network environment, a delay caused by a loss recovery code such as a fountain code used for an interactive service such as a video phone can be minimized.

In particular, when using a fountain code such as a raptor code or a low density parity check (LDPC), the block size (number of input symbols) k must be at least 1000 to achieve performance. It can effectively reduce the delay while maintaining.

1 is an exemplary diagram showing symbols of each step in a coding operation in a conventional Raptor coding scheme.

2 is a block diagram showing a hierarchical structure of a transceiver in 3GPP (The 3rd Generation Partnership Project) Multimedia Broadcast / Multicast Service (MBMS).

3 is an exemplary diagram illustrating an update operation of a symbol during a coding operation according to an embodiment of the present invention.

4 is an exemplary diagram illustrating a frame encoding operation according to time in a low delay coding method according to an embodiment of the present invention.

5 is a diagram illustrating a frame encoding operation according to time in a low delay coding method according to an embodiment of the present invention.

6 is an exemplary view schematically showing an operation of updating and transmitting in a low delay coding method according to an embodiment of the present invention;

7 is a flowchart of a low-delay coding operation of a transceiver according to an embodiment of the present invention.

8 is a diagram schematically illustrating a low delay coding operation according to an embodiment of the present invention.

9 is a diagram schematically illustrating a low delay decoding operation according to an embodiment of the present invention.

10 is a linear diagram showing an update probability according to M size in a low delay coding method according to an embodiment of the present invention

11 is a linear diagram showing update probability according to degree in a low delay coding method according to an embodiment of the present invention

Hereinafter, an apparatus and an operation method of constructing the present invention will be described in detail with reference to the accompanying drawings. In the following description, specific details such as specific components are shown, which are provided to help a more general understanding of the present invention, and it is understood that these specific details may be changed or changed within the scope of the present invention. It is self-evident to those of ordinary knowledge in Esau. In addition, in describing the present invention, when it is determined that the detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted.

In conventional Reed Solomon code, n output symbols are generated for k input symbols. Thus, a delay occurs by the time T at which k inputs are made. For example, for a video that passes 30 frames per second, applying a code every second will cause a delay of at least T = 1 second. This delay is not a problem for streaming services, such as video on demand. However, this delay is not allowed for interactive services such as video phones. Most lossy recovery codes have the same code rate (k / n), so that the larger k, the better the recovery capability. Therefore, trade-off is established between the restoring capability and the coding delay. The present invention provides a method of minimizing delay while maintaining resilience.

The present invention updates k output symbols every T / M time by dividing k input symbols generated every T time by M in time. It is preferable that M becomes a divisor of k. This reduces the delay time from the original delay time T to T / M. Every T / M time, the oldest k / M data is deleted and new k / M data is input to generate n outputs for the total k symbols. In this case the invention only updates the output symbols associated with the new k / M inputs. This ensures that only new incoming data and updated output symbols are transmitted during every T / M time. These outputs are made sequentially. First, n outputs for k inputs are exchanged at the transmitting and receiving sides, and then the outputs for k / M inputs can be sequentially transmitted in k / M units.

The probability of updating can be estimated simply as follows. If the average number of edges per output symbol is d and k input blocks are divided by M, the probability that one edge does not belong to the sub-block being updated is (M-1) / M. The probability that the d edges do not belong to the updated sub block is Premain = ((M-1) / M) e. The probability of being updated is Pupdate = 1-Premain and the overhead of the present invention is determined by how large this ratio is than 1 / M.

Hereinafter, with reference to the drawings will be described in more detail the present invention.

FIG. 2 is a block diagram illustrating a hierarchical structure of a transceiver in 3GPP (The 3rd Generation Partnership Project) Multimedia Broadcast / Multicast Service (MBMS) (see 3GPP TS 26.346, 2006.12).

MBMS is a multimedia broadcast / multicast service standard under development in 3GPP, which is a service for simultaneously transmitting data packets to multiple users, and aims to transmit multimedia data. Referring to FIG. 2, first of all, audio, video, and text data of a sender is used for real-time transport control protocol (RTP & RTCP) for real-time transmission, and security of media. Secure RTP (SRTP) is used, and key handling uses MIKEY (Multimedia Internet KEYing), a key management protocol. The data are FEC encoded at the FEC layer for error correction and transmitted to a plurality of users for MBMS using User Datagram Protocol (UDP). A receiver receives data transmitted from the transmitter, corrects an error lost in the FEC layer through UDP, and recovers data. In the FEC layer, a loss recovery code is used for error correction, and the present invention can be applied to the loss recovery code.

3 is an exemplary view illustrating an update operation of a symbol during a coding operation according to an embodiment of the present invention. In the loss recovery code, even if the same code rate (k / n), the larger k, the better the loss recovery ability. However, if k is large, the delay is more delayed. The present invention divides a block of size k into M sub-blocks in order to minimize delay while maintaining the value of k in accordance with the characteristics of the present invention. Update. The subblock is not necessarily divided by M, but it does not matter if the receiver knows the total length of the code. In addition, the receiver should be able to know which part of the code has already been restored. As shown in Fig. 3, in the example of Fig. 3, k is 16 and M is 4. First, n outputs are delivered to the receiver for the first k inputs, and the output symbols with edges (if the coefficient is non-zero) are selected in the oldest k / M subblocks among the next n output symbols. Update. There can be only one edge in this output symbol. A new output symbol is created by subtracting the number multiplied by the corresponding input symbol and the coefficient of the corresponding edge from the output symbol and multiplying the new input symbol at the same location by the coefficient of the corresponding edge to the modified corresponding output symbol.

For example, suppose the four symbols on the left of the past are removed and four new inputs are input from the right to the left, as shown in FIG. 3, and the output symbol y6 has an edge connected to x2. The x2 and the new x2 should be xored at y6 once to update to the new y6 and send to the receiver. However, y7 is independent of the four newly updated inputs and does not need to be retransmitted.

In Reed Solomon, the coefficient is one of several values in the Galois Field, but only 0 and 1 are possible in the Low Density Parity Check (LDPC) or Raptor code, so if the edge is not zero, If you XOR the input symbol and the new input symbol once, a new output symbol is created. In the case of Reed Solomon, all output symbols are related to all input symbols except for the transmission of input symbols as they are. That is, most of the coefficients are mostly nonzero. In this case, almost all of the output symbols must be updated, which increases the overhead of the present invention. However, in the case of LDPC or Raptor code, since the density is low, the overhead of applying the present invention is not large.

According to the present invention, the initial k input symbols during encoding are generated using the following method. Initially, k (M-1) / M virtual input symbols are set in advance by the transmitting side and the receiving side, and a total of k initial input symbols are generated by adding k / M symbols corresponding to the last subblock in the actual input. In this case, the virtual input symbols are all set to 1 or 0.

In addition, the present invention may accumulate a situation in which the loss is not restored when a lot of losses occur to prevent the loss from propagating, and thus the resilience may be reduced, and a refresh block may be inserted to start a new one at a predetermined time. This new start is also necessary for random access. Again, it is preferable that all virtual input symbols be either 1 or 0. As the refresh block is inserted more frequently, the coding efficiency decreases. Therefore, the refresh block period is appropriately determined according to the loss rate and the random access conditions.

In addition, the present invention can operate adaptively according to the current loss rate using back signaling. If the current loss recovery rate can be received from the receiver, the rate of parity symbols can be adjusted accordingly to maintain the coding rate as necessary. In other words, if the loss recovery rate is worse, more parity symbols are transmitted, and if it is better, less parity is sent. This adaptability is important because current restoration rates affect the future. Therefore, if the current loss recovery rate is bad, the future loss recovery rate can be increased in the above manner. If it is lost beyond restoration at all, the refresh block is transmitted. To use this adaptive method, the round trip delay time must be much shorter than the refresh period. The round trip delay time is preferably about 1/4 of the refresh period.

In addition, the present invention can reduce the number of output symbols to be updated by adjusting the edge distribution. Creating code so that the distribution of the edges for each output symbol is distributed over consecutive (M-1) / M can minimize the number of symbols in the output being updated. If all output symbols satisfy this condition, only n / M output variables need to be updated for one subblock.

Meanwhile, the subblocks do not have to be all the same size. The sum of the sizes of all subblocks should be the size k of the block.

The method for real-time generation of the fountain code regardless of the edge is as follows. Like the raptor code, the sender and receiver have a lot of static edge generation methods, and when the input symbol is a systematic code existing in the output symbol, the received code is used again for real-time recovery of the code to be received in the future. Can be used as a parity. In this method, when a repair code is generated using k input symbols (k does not need to be a static value each time) using a fountain code, information corresponding to M frames is inserted into k symbols. The symbols for the M frames are sequential and the last frame is the most recently encoded frame or the information to be transmitted next not currently transmitted. After this, if a new frame is encoded or the information to be transmitted next is prepared, information about the previous frame is removed from k and the next information is sequentially added to k to generate a new k '. This k 'is used for fountain coding to generate a repair code. The generated fountain code is systemic, so if the original symbols sent in the past exist and the receiver is successfully restored, the receiver can know these symbols. Therefore, only the fountain code corresponding to the newly input information of k 'and the generated repair codes need to be transmitted to the receiver. When the receiver recovers k symbols, the oldest information is removed and the rest can be used to recover the next information to be newly received. The past, future, or recent frame unit can be set as a bundle of several frames instead of a single frame.

In addition, the receiving side can identify the frame by the ID of the sub-block delivered as additional information of each packet. The receiving side can also create the same edge using the edge generation method in the total code received. The frame-by-frame arrangement of symbols can be changed if the receiver knows the arrangement even if they are not sequential. Meanwhile, the present invention may use a logical structure that shifts and removes a symbol for a frame in the past in an array of k symbols, and adds symbols of a frame to be updated. You can also create a fountain code by replacing the new symbol with a placeholder.

4 is an exemplary diagram illustrating a frame encoding operation over time in a low delay coding method according to an embodiment of the present invention. 4A is a diagram schematically illustrating a low delay coding scheme according to an embodiment of the present invention, in which one packet is generated in two frames, and FIG. 4B is a diagram illustrating a conventional scheme. In the conventional scheme, unlike the present invention, there is a delay as many as the number of frames in a packet.

5 is an exemplary diagram illustrating a frame encoding operation over time in a low delay coding method according to an embodiment of the present invention. FIG. 5 (a) is a diagram schematically showing a low latency coding scheme according to an embodiment of the present invention in which one packet is generated in two frames, and FIG. 5 (b) illustrates an existing scheme for low latency. The figure shown. The conventional method reduces the delay by reducing the length of the packet for encoding, but the probability of success of restoration is lowered accordingly.

6 is an exemplary view schematically illustrating an operation of updating and transmitting an output symbol n in a low delay coding method according to an embodiment of the present invention. Referring to FIG. 6, in frame Cx, each k (input symbol to be encoded) is divided into two segments (sub blocks) to generate a fountain code, and updated by one segment. In this first set, the first segment becomes dummy symbols. As the set progresses, new segments are inputted to update output symbols and the previous segment is removed. The receiving side may have a higher recovery rate even when fewer symbols are received by using the previously recovered symbols for the recovery of the future received symbols.

There are two types of loss correction codes, a convolutional code and a block code. In the block code, since the loss recovery is performed entirely with data belonging to one block, a delay of the block size occurs. While maintaining the size of a block, one block is divided into several sub-blocks to replace input symbols of one sub-block, and output symbols that need to be updated are newly transmitted.

7 is a flowchart of a low delay coding operation of a transceiver according to an embodiment of the present invention. First, in step 705, the transmitting end receives a new input symbol, sub block k / M. In step 710, the new input symbol k / M is added to the previously stored input symbol k (M-1 / M). In step 720, the last subblock previously used is removed, and a new subblock k / M is added to the input symbol k, and the output symbol is updated accordingly. In operation 725, encoding is performed using the input symbol k updated in operation 720, and a packet is generated accordingly. In step 730, the packet is transmitted to the receiving end. In this case, the present invention transmits only the newly updated output symbol to the receiver. In the next step 735, the receiving end receives the data packet from the transmitting end. In the next step 740, the code is restored using the previously restored symbol (restoration symbol). In addition, in step 740, the receiver may request the transmitter to transmit, or add or remove parity, by an initial transmission method that removes the symbol dependency. In operation 745, the newly input subblock is distinguished, and in operation 750, a restoration symbol to be used for the next restoration is stored in the memory. The reconstructed symbol can be used for reconstructing information on new symbols, in step 740. After the code is restored in step 740, the subblock k / M is restored and used in step 755. FIG.

8 is a diagram schematically illustrating a low delay coding operation according to an embodiment of the present invention. As shown in FIG. 8, low-delay coding according to an embodiment of the present invention stores an input symbol k in a memory buffer of an encoder, adds a newly input subblock k / M to a stored input symbol, and uses no previous transfer. Remove the sub block. The input symbol and the necessary information are sent to a systematic code generator to generate a systematic code. In this case, only the newly updated part and the additional recovered symbols are transmitted to the receiver except the already restored information.

9 is a diagram schematically illustrating a low delay decoding operation according to an embodiment of the present invention. As shown in FIG. 9, in the low-delay decoding according to an embodiment of the present invention, the decoder may reuse assuming that previously transmitted symbols (restoration symbols) are received at the transmitter. The actual received symbols and the parts recovered in the past (parts used for code generation at the sender) are used to recover the necessary symbols and store them in the memory buffer to use them again. The old subblock, which is no longer needed, is removed and the new subblock is added to the memory buffer and the new subblock becomes the output of the decoder.

10 is a linear diagram illustrating an update probability according to the size of M in a low delay coding method according to an embodiment of the present invention. As shown in FIG. 10, in the case of using the low-delay coding scheme according to an embodiment of the present invention, as M increases, the probability that the symbol is updated decreases. 11 is a linear diagram illustrating update probability according to degree in a low delay coding method according to an embodiment of the present invention. As shown in FIG. 11, in the case of using the low delay coding scheme according to an embodiment of the present invention, the higher the order, the greater the update probability.

If all edges are updated each time and k is constant, the existing method should generate code n * packets if the number of frames included in each k is n. On the other hand, the present invention generates and transmits a code of k in each frame, thereby generating code of m * n * packets times. However, since the fountain code is much less computational than the conventional Reed-Solomon method, the overhead is easy to process each frame at the same time in real time processing.

As described above, the operation of the low-delay fountain coding method using the sequential update method for each sub-block according to an embodiment of the present invention can be made. Meanwhile, the above-described description of the present invention has been described in detail. It may be practiced without departing from the scope of the invention.

Claims (12)

In the low delay fountain coding method using the sequential update method for each sub-block, Dividing an input symbol into a predetermined number of identical subblocks, Dividing an encoding time of the input symbol into equal sub-time intervals so as to correspond to dividing the input symbol into a predetermined number of identical sub-blocks; Receiving a symbol having a size corresponding to the sub-block, generating initial input symbols and generating and outputting an initial output symbol using the initial input symbol; Sequentially receiving symbols corresponding to a new subblock to be encoded for each sub time interval, adding the symbols to the input symbol, and removing the symbol corresponding to the previously input subblock from the input symbol to update the input symbol. Process, Updating an output symbol related to the most recently received sub-block; And generating a packet by using the updated output symbol. The method of claim 1, wherein the output symbol associated with the most recently received sub-block, A coding method, characterized in that the output symbol is connected to one of the symbols included in the previously input sub-block and the edge (the coefficient is non-zero). The method of claim 2, wherein updating the output symbol comprises: Modifying an output symbol by subtracting the number obtained by multiplying an input symbol having an edge connected to the output symbol to be updated by a coefficient of the corresponding edge, from the output symbol to be updated; The new output symbol is added to the modified output symbol by adding the input symbol used for the modification of the output symbol and the number of the input symbol included in the newly input subblock multiplied by the coefficient of the corresponding edge located at the same position in the subblock. Coding method comprising the step of updating to. The method of claim 2, wherein updating the output symbol comprises: Modifying the output symbol with a value obtained by XORing an input symbol having an edge connected to the output symbol to be updated and an output symbol to be updated; Updating the output symbol with a value obtained by XORing the input symbol included in the newly input subblock and the modified output symbol located at the same position in the subblock and the input symbol used to modify the output symbol. Coding method. The method of claim 1, wherein the initial input symbol, Coding method, characterized in that for receiving a symbol of the size corresponding to the sub-block, the remaining portion is composed of a virtual input symbol. The method of claim 5, wherein the virtual input symbol, A coding method, characterized in that all have a value of '1' or all have a value of '0'. The coding method as claimed in claim 1, further comprising receiving a refresh block that makes the input symbol the same as the initial input symbol at a predetermined time. The coding method of claim 1, further comprising: receiving a current loss recovery rate and adjusting a transmission rate of a parity symbol according to the received loss recovery rate. In the low delay fountain decoding method using a sequential update method for each sub-block, Generating a reconstructed symbol by receiving and decoding a packet generated using an initial output symbol; Restoring the original subblock by using the reconstruction symbol when receiving a packet including information of an encoded subblock; And adding symbols corresponding to the newly received sub-blocks in the reconstructed symbols, and removing symbols corresponding to the previously received sub-blocks to update the reconstructed symbols. The decoding method of claim 9, further comprising outputting a back signaling signal for receiving an initial output symbol when the loss recovery rate is lower than a preset reference. The decoding method of claim 9, further comprising outputting a back signaling signal for receiving an initial output symbol at a predetermined time. 10. The method of claim 9, further comprising outputting a back signaling signal including current loss recovery rate information to adjust a loss recovery rate in a current decoding operation.