WO2021073338A1

WO2021073338A1 - Decoding method and decoder

Info

Publication number: WO2021073338A1
Application number: PCT/CN2020/115383
Authority: WO
Inventors: 马亮; 魏岳军; 梁璟
Original assignee: 华为技术有限公司
Priority date: 2019-10-14
Filing date: 2020-09-15
Publication date: 2021-04-22
Also published as: CN112737600A; CN112737600B

Abstract

A cascade decoder (600) and a decoding method. The cascade decoder (600) comprises a parallel decoder (620) and a serial decoder (630), and iterative decoding is performed for a maximum of T iterations; the parallel decoder (620) is configured to decode one or more LLR sequences with a length of N; the serial decoder (630) is configured to decode one or more LLR sequences with a length of N_S, wherein N_S<N, and t-th iteration is performed. The method comprises: the parallel decoder (620) decodes L(t) second input LLR sequences to obtain L(t) second output LLR sequences (710); the serial decoder (630) decodes L(t) third input LLR sequences to obtain M(t) decoding paths (720), each of the third input LLR sequences comprising Ns LLRs in the corresponding second output LLR sequence, respectively; the serial decoder (630) determines whether to terminate an iteration (730). By adoption of the cascade decoder (600), the throughput rate of the decoding is improved by means of the parallel decoder (620), and the performance of parallel decoding is improved by means of the serial decoder (630), so that the decoding performance and the throughput rate are improved as a whole.

Description

Decoding method and decoder

Technical field

This application relates to the field of communications, and more specifically, to a decoding method and decoder of polarization codes in the field of communications.

Background technique

Polar code was first proposed by Arikan in 2008, and it was proven to reach Shannon's limit capacity. The encoding of Polar codes is mainly based on the theory of channel polarization. The process of channel polarization is mainly divided into channel combination and channel decomposition. When the number of channel combinations tends to be infinite, channel polarization will occur. The phenomenon of channel polarization is that the channel capacity obviously tends to two levels, one part tends to a pole with a channel capacity of 1, that is, a noiseless channel; and the other part tends to a pole with a channel capacity of 0, that is, a full-noise channel. Since the channel is polarized, we can reasonably use the polarization feature to encode, set the information bit on the noise-free channel, that is, transmit the information bit; and set the freeze bit on the all-noise channel, that is, transmit the fixed bit without information. , Such as 0 or 1, these bits can also be called frozen bits. The decoding calculation complexity of the polarization code is O(N log ₂ N), where N is the code length.

Generally, P(N,K) can be used to represent a polarization code with a ^{code length of N=2 n and K information bits.} Encoder pair information sequence

Perform Polar encoding to get the encoded sequence

The recursive generation process of polarization codes can be expressed in the form of matrix multiplication as follows:

Where matrix

Is the polarization matrix

The matrix obtained by the n-order Kronecker product.

When constructing the polarization code, the relatively reliable K bits are selected for the transmission of information bits, that is, the bits corresponding to the above-mentioned noise-free channel, and the set is represented by A; and the value of the NK relatively unreliable bits is set to one A fixed value, for example, set to 0 or 1. These bits are called frozen bits, and the positions of these bits can also be called fixed bits, that is, the bits corresponding to the above-mentioned full-noise channel.

At this stage, Polar’s main decoding methods include serial decoding methods and parallel decoding methods. Serial decoding methods include: serial cancellation (successive cancellation, SC) decoding method, list serial cancellation (successive cancellation list, SCL) decoding method and cyclic redundancy check (CRC) assisted SCL (CRC aided SCL, CA-SCL) decoding method, etc. Parallel decoding methods include: Belief Propagation (BP)/ The minimum sum (Min-Sum, MS) decoding method and the deep neutral network (DNN), etc. These decoding methods have their own strengths. From the perspective of decoding performance, the serial decoding method has better decoding performance, but the decoding delay is large, and the decoding throughput rate is limited; while the parallel decoding method has high parallelism , But the decoding performance often has a big gap compared with the serial decoding method.

3GPP selected polarization codes for 5G enhanced mobile broadband (enhance mobile broadband, eMBB) control channel coding, and low density parity check (LDPC) was used for 5G eMBB service data channel coding.

The Polar code decoder needs to be designed to better meet the actual application requirements of the polar code in the communication system.

Summary of the invention

In view of this, the various embodiments of the present application provide parallel decoders for cascaded decoding, cascaded decoders, and methods for cascaded decoding, which are used to improve the decoding performance of parallel decoding while maintaining relatively high performance. High throughput rate. The embodiment of the present application also provides a multi-cascade decoder, which can further reduce the hardware overhead of the multi-mode decoder in the communication device.

In the first aspect, the embodiments of the present application provide a parallel decoder for cascaded decoding, which is used to decode one or more _{LLR sequences of length N in} with a maximum of S times. , And provide one or more input LLR sequences of _{length N out to the next decoder in each iteration.} The parallel decoder includes: a first-stage update unit and a second-stage update unit. For the sth iteration,

s=1, the first-stage update unit is used to perform the first-stage update on the input LLR sequence of the parallel decoder to obtain the output LLR sequence, and the output LLR sequence is used to provide the input LLR sequence of the next-stage decoder , Wherein the input LLR sequence of the parallel decoder includes N _in LLRs, the output LLR sequence includes N _in LLRs, and the input LLR sequence of the next-level decoder includes N in the output LLR sequence. _out LLRs;

The second-stage update unit is used to obtain the output LLR sequence of the next-stage decoder, and the output LLR sequence of the next-stage decoder includes N _out LLRs.

s>1,

The second-stage update unit is configured to perform a second-stage update on a second update input LLR sequence to obtain a second update output LLR sequence, and the second update input LLR sequence is obtained according to the obtained s-1th iteration The output LLR sequence of the next-level decoder and the sequence obtained from the output LLR sequence in the s-1th iteration, including N _in LLRs;

The first-stage update unit is configured to perform a first-stage update on the second updated output LLR sequence to obtain an output LLR sequence, and the output LLR sequence is used to provide an input LLR sequence of a next-stage decoder.

The above-mentioned parallel decoder can realize large-block-length decoding calculations in cascaded decoding, and can be adapted to the next-stage decoder with better decoding performance for small-block lengths, which helps to improve the overall decoding throughput Rate, and use the decoding performance of the next-level decoder.

With reference to the first aspect, in a first possible implementation manner, the second update input LLR sequence is the (s-2)·N _out +1th in the output LLR sequence in the s-1th iteration LLR to the second (s-1) · N _out a replacement for the first LLR s-1 iterations LLR output sequence of the next stage of the decoder LLR obtained number N _out.

Combining the first aspect and any one of the foregoing possible implementation manners, in a second possible implementation manner, the parallel decoder outputs the output LLR sequence to the next-stage decoder, so that the The next-stage decoder obtains its input LLR sequence according to the (s-1)·N _out +1 th LLR to the s·N _out LLR of the output LLR sequence.

With reference to the first aspect and any one of the foregoing possible implementation manners, in a third possible implementation manner, the parallel decoder determines the (s-1)·N _out +1th from the output LLR sequence The LLR to the s·N _out- th LLR are output to the next-stage decoder as the input LLR sequence of the next-stage decoder.

With reference to the first aspect or any one of the foregoing possible implementation manners, in a fourth possible implementation manner, the parallel decoder includes at least the n _out +1 th layer to the n _in +1 th layer n _{in of the factor graph.} -n _out decoding layers.

With reference to the fourth possible implementation manner, in the fifth possible implementation manner, the first-stage update unit is specifically configured to update _{the N LLR nodes of the n in} +1th layer to the input LLR Sequence, and perform soft value update layer by layer _{from the n in} +1 layer to the n _out +1 layer to obtain the N _in _{LLR nodes of the n out} +1 layer, and the output LLR sequence includes the n _out + _{N in} LLR nodes of layer 1; the second-stage update unit is specifically configured to update _{the N LLR nodes of the n out} + 1th layer to the second update input LLR sequence, and start from the n _out + 1-layer to the first layer n _in +1 direction to perform a soft layer to obtain a first value updating n _{_in} N _in a layer LLR + 1 nodes, the second output LLR update sequence comprising n _{_in} N _in the first layer +1 LLR nodes.

With reference to the first aspect or any of the foregoing possible implementation manners, in a sixth possible implementation manner, the first-stage update unit is specifically further configured to determine the LDPC check matrix corresponding to the input LLR sequence, and the The first-stage update unit and the second-stage update unit update the input LLR sequence based on the LDPC check matrix. In this way, the Polar code is decoded by the calculation unit of the LDPC code, which makes it possible to decode the Polar code in common mode with the LDPC code and save overhead.

With reference to the first aspect or any of the foregoing possible implementation manners, in a seventh possible implementation manner, the second-stage update unit is further configured to return the decoded LLR sequence to the upper-level decoder, and the decode The LLR sequence includes the second update output sequence.

In another possible implementation manner, the parallel decoder includes one or more of the following: a BP decoder, an MS decoder, or a DNN decoder.

In another possible implementation, the _{value of N in} can be any of the following: 8192,4096,2048,1024,512,256,128,64,32; the _{value of N out} can be any of the following: 1024,512,256,128,64, 32,16,8,4,2,1128,64,32,16,8,4,2,1.

In the second aspect, an embodiment of the present application provides a cascaded decoding method, which performs a maximum T iterations of cascaded decoding on an input LLR sequence, where the t-th iteration includes:

Use a parallel decoding algorithm to decode L(t) second input LLR sequences to obtain L(t) second output LLR sequences;

Use a serial decoding algorithm to decode L(t) third input LLR sequences to obtain M(t) decoding paths;

Determine to continue the iteration, or determine to terminate the iteration;

The length of each second input LLR sequence is N, the length of each second output LLR sequence is N, and each third input LLR sequence includes Ns LLRs in the corresponding second output LLR sequence.

The above method uses parallel decoding algorithms to perform parallel decoding calculations for large block lengths, which improves the throughput of decoding, and provides small block lengths to the next level of serial decoding algorithms, and uses small block length serial decoding algorithms to improve Decoding performance, combining the advantages of the two, enables decoding to take into account both throughput and decoding performance at the same time.

With reference to the second aspect, in the first possible implementation manner of the second aspect, each of the third input LLR sequences includes the (t-1)×N _S +1 th LLR in the corresponding second output LLR sequence To the t×N _S LLR.

With reference to the second aspect or the first possible implementation manner of the second aspect, in the second possible implementation manner of the second aspect, the parallel decoding algorithm is used to perform L(t) second input LLR sequences Decode and obtain L(t) second output LLR sequences, including:

A parallel decoding algorithm is used to perform a first stage update on the L(t) second input LLR sequences respectively to obtain the L(t) second output LLR sequences.

In combination with the second aspect or any of the foregoing possible implementation manners, in the third possible implementation manner of the second aspect,

t=1, L(t)=1, and the second input LLR sequence is the input LLR sequence of the cascaded decoder.

In combination with the second aspect or any of the foregoing possible implementation manners, in the fourth possible implementation manner of the second aspect,

t>1, the t-1th iterative decoding also includes:

According to the M(t-1) decoding paths, M(t-1) third output LLR sequences are obtained, and each of the third output LLR sequences includes Ns LLRs.

Combined with the fourth possible implementation, in the fifth possible implementation of the second aspect,

The t-th iteration further includes:

Acquiring M(t-1) second output LLR sequences corresponding to the M(t-1) third output LLR sequences of the t-1th iteration;

According to the M(t-1) third output LLR sequence and the M(t-1) second output LLR sequence of the t-1th iteration to obtain the second L(t)th iteration Update the input LLR sequence, L(t)=M(t-1);

The parallel decoding algorithm is used to perform a second stage update on the L(t) second updated input LLR sequences to obtain the L(t) second input LLR sequences.

With reference to the second aspect or any of the foregoing possible implementation manners, in a sixth possible implementation manner of the second aspect, the parallel decoding algorithm includes n _s +1 th to n +1 th layer nn _s translations. Code layer,

Each layer includes N LLR nodes; the first stage update includes:

Updating the N LLR nodes of the n+1th layer to N LLRs in the second input LLR sequence,

Perform soft value update from the n+1th layer to the n _s +1th layer to obtain _{N LLR nodes of the n s} +1th layer, and the corresponding second output LLR sequence includes the first N LLR nodes of n _{s +1 layer;}

The second phase update includes:

Updating the N LLR nodes of the n _s +1 th layer to the N LLRs in the second updated input LLR sequence,

Perform a soft value update from the n _s +1 th layer to the n+1 th layer to obtain N LLR nodes of the n+1 th layer, and the corresponding second input LLR sequence includes the n th layer N LLR nodes at level +1.

With reference to the second aspect or any of the foregoing possible implementation manners, in the seventh possible implementation manner of the second aspect, a serial decoding algorithm is used to decode the L(t) third input LLR sequences Obtain M(t) decoding paths, including:

Decoding the L(t) third input LLR sequences to obtain L(t)*2 ^k decoding paths, where k is a positive integer;

The M(t) decoding paths are M(t) decoding paths with the ^{largest path metric among the L(t)*2 k} decoding paths, or the M(t) decoding paths Among the L(t)*2 ^k decoding paths, M(t) decoding paths have the largest path metric and pass the CRC check.

With reference to the second aspect, in an eighth possible implementation manner of the second aspect, the parallel decoding algorithm is used to decode the L(t) second input LLR sequences to obtain L(t) second output LLRs Sequence, including:

Respectively determining a corresponding LDPC check matrix for the L(t) second input LLR sequences;

Decoding the L(t) second input LLR sequences based on the LDPC check matrix to obtain L(t) second output sequences.

In each of the above possible implementations, the parallel decoding algorithm includes: BP decoding algorithm, MS decoding algorithm, or DNN decoding algorithm; the serial decoding algorithm includes: SCL decoding algorithm or CA-SCL Decoding algorithm.

In each of the above possible implementations, the value of N is any of the following: 1024, 512, 256, 128, 64, 32; the value of Ns is any of the following: 128, 64, 32, 16, 8, 4, 2, 1 .

With reference to the second aspect or any of the foregoing possible implementation manners, in a ninth possible implementation manner of the second aspect, t<T and the tth iteration does not meet the early termination condition, the tth iteration also includes :

According to the M(t) decoding paths, M(t) third output LLR sequences are obtained.

With reference to the second aspect or any of the foregoing possible implementation manners, in the tenth possible implementation manner of the second aspect, t=T, or the t-th iteration satisfies the early termination condition; the t-th iteration also include:

The serial decoder obtains the decoding result according to the M(t) decoding paths, and terminates iteration.

In each of the above possible implementations, the input LLR sequence is the information sequence

Corresponding LLR sequence, the information sequence includes a plurality of information bits, or one or more information bits and one or more frozen bits; said obtaining a decoding result according to the M(t) decoding paths includes: Perform a hard decision on one decoding path with the largest path metric value or the largest path metric value and a successful CRC check among the M(t) decoding paths to obtain each information bit in the information sequence.

In the third aspect, an embodiment of the present application provides a cascaded decoder, the cascaded decoder includes a serial decoder, and the second aspect of the first aspect or any possible implementation of the first aspect. Parallel decoder, where N _in =N, N _out =N _S , S=T, iteratively decodes the input LLR sequence for a maximum of T times, and the second parallel decoder is used to decode one or more lengths of The log-likelihood ratio LLR sequence of N is decoded, and the serial decoder is used to decode one or more LLR sequences of _{length N S} _{, N S} <N, where the t-th iterative decoding includes:

The second parallel decoder decodes L(t) second input LLR sequences to obtain L(t) second output LLR sequences, each second input LLR sequence has a length of N, and each second input LLR sequence has a length of N. The length of the output LLR sequence is N;

The serial decoder decodes the L(t) third input LLR sequences to obtain M(t) decoding paths, and each of the third input LLR sequences includes the corresponding second output LLR sequence. Ns LLRs;

The serial decoder determines to continue the iteration, or the serial decoder determines to terminate the iteration.

The above-mentioned cascaded decoder uses a parallel decoder to perform parallel decoding calculations for large block lengths to improve decoding throughput, and provides small block length decoding to the serial decoder, and uses small block length serial translation The encoder improves the decoding performance and combines the advantages of the two, so that the decoding can take into account both the throughput and the decoding performance at the same time.

With reference to the third aspect, in the first possible implementation manner of the third aspect, each of the third input LLR sequences includes the (t-1)×N _S +1 th LLR in the corresponding second output LLR sequence To the t×N _S LLR.

Combining the third aspect and the first possible implementation manner, in the second possible implementation manner of the third aspect, the second parallel decoder decodes L(t) second input LLR sequences to obtain L(t) second output LLR sequence, including:

The second parallel decoder performs a first stage update on the L(t) second input LLR sequences respectively to obtain the L(t) second output LLR sequences.

With reference to the third aspect or any of the foregoing possible implementation manners, in the third possible implementation manner of the third aspect, t=1, L(t)=1, and the second input LLR sequence is the stage The input LLR sequence of the connected decoder.

In combination with the third aspect or any of the foregoing possible implementation manners, in the fourth possible implementation manner of the third aspect,

t>1, the t-1th iterative decoding also includes:

The serial decoder obtains M(t-1) third output LLR sequences according to the M(t-1) decoding paths, and each third output LLR sequence includes Ns LLRs.

With reference to the fourth possible implementation manner of the third aspect, in the fifth possible implementation manner of the third aspect, the t-th iteration further includes:

Acquiring, by the second parallel decoder, M(t-1) second output LLR sequences corresponding to the M(t-1) third output LLR sequences of the t-1th iteration;

The second parallel decoder obtains L(t) according to the M(t-1) third output LLR sequences and the M(t-1) second output LLR sequences of the t-1th iteration The second updated input LLR sequence of the t-th iteration, L(t)=M(t-1);

The second parallel decoder performs a second stage update on the L(t) second updated input LLR sequences to obtain the L(t) second input LLR sequences.

With reference to the fifth possible implementation manner of the third aspect, in the sixth possible implementation manner of the third aspect, the second parallel decoder includes at least the n _s +1 th to n +1 th layer nn _s Decoding layers,

Each layer includes N LLR nodes;

The second parallel decoder performs a first stage update on each second input LLR sequence to obtain a corresponding second output LLR sequence, including:

The second parallel decoder updates the N LLR nodes of the n+1th layer to N LLRs in the second input LLR sequence,

Said second decoder performs parallel from the first layer n + 1 to the first direction of the soft layer n _s +1 to obtain the updated value of n _s N + 1 th layer LLR nodes, the corresponding first The second output LLR sequence includes the N LLR nodes of _{the n s +1 th layer;}

The second parallel decoder performs a second stage update on each second update input LLR sequence to obtain the corresponding second input LLR sequence, including:

The second parallel decoder _{updates the N LLR nodes of the n s} +1 th layer to the N LLRs in the second updated input LLR sequence,

The second parallel decoder _{performs a soft value update from the n s} +1 th layer to the n+1 th layer to obtain the N LLR nodes of the n+1 th layer, and the corresponding second The input LLR sequence includes N LLR nodes of the n+1th layer.

With reference to the third aspect or any of the foregoing possible implementation manners, in a seventh possible implementation manner of the third aspect, the serial decoder interprets the L(t) third input LLR sequences The code gets M(t) decoding paths, including:

The serial decoder decodes the L(t) third input LLR sequences to obtain L(t)*2 ^k decoding paths, where k is a positive integer;

With reference to the third aspect or any one of the foregoing possible implementation manners, in an eighth possible implementation manner of the third aspect, the second parallel decoder is configured to perform processing on L(t) of the second input LLR sequences Determine the corresponding LDPC check matrix respectively;

The second parallel decoder decodes the L(t) second input LLR sequences based on the LDPC check matrix to obtain L(t) second output sequences.

With reference to the third aspect or any of the foregoing possible implementation manners, in a ninth possible implementation manner of the third aspect, the second parallel decoder includes one or more of the following: BP decoder or MS A decoder or a DNN decoder, and the serial decoder includes an SCL decoder or a CA-SCL decoder.

In combination with the third aspect or any of the foregoing possible implementation manners, in the tenth possible implementation manner of the third aspect, the value of N is any one of the following: 1024, 512, 256, 128, 64, 32; the value of Ns is the following Any item: 128,64,32,16,8,4,2,1.

With reference to the third aspect or any of the foregoing possible implementation manners, in another possible implementation manner, the serial decoder determines to continue iteration, including:

The serial decoder determines that t<T and the t-th iteration does not meet the early termination condition;

The t-th iteration further includes:

The serial decoder obtains M(t) third output LLR sequences according to the M(t) decoding paths.

With reference to the third aspect or any one of the foregoing possible implementation manners, in another possible implementation manner, the serial decoder determines to terminate the iteration, including:

The serial decoder determines that t=T, or the t-th iteration satisfies an early termination condition;

The t-th iteration further includes:

Wherein, the input LLR sequence of the cascaded decoder is an information sequence

Corresponding to the LLR sequence, the serial decoder performs a hard decision on one of the M(t) decoding paths with the largest path metric value or one decoding path with the largest path metric value and a successful CRC check, and obtains the result Each information bit in the information sequence.

In a fourth aspect, the embodiments of the present application provide a multi-cascade decoding method, including two-stage iterative decoding, in which the first-stage iterative decoding performs a maximum of I iterations on the initial LLR sequence, I=N _p /N, The second-level iteration is a cascaded decoding iteration like the second aspect or any possible implementation of the second aspect, the maximum number of iterations is T, the i-th iteration, i<I, including:

Using the first parallel decoding algorithm to perform the first-level iterative decoding of K(i) first input LLR sequences to obtain K(i) first output LLR sequences, the length of each first input LLR sequence is N _p , The length of each first output LLR sequence is N _p ;

Perform the second-level iterative decoding on K(i) second input LLR sequences using any possible implementation manner as in the second aspect or the second aspect, and each of the second input LLR sequences respectively includes a corresponding first output LLR N LLRs in the sequence.

Since LDPC codes are usually decoded by parallel decoding algorithms, the multi-cascade decoding method can further share part of the parallel decoding units of the LDPC codes, saving system overhead.

With reference to the fourth aspect, in the first possible implementation manner of the fourth aspect, each of the second input LLR sequences includes the (i-1)×N+1th in the corresponding first output LLR sequence LLR to the i×Nth LLR.

With reference to the fourth aspect or the first possible implementation manner of the fourth aspect, in the second possible implementation manner of the fourth and fifth aspects, the first parallel decoding algorithm is used to perform the processing of the K(i) The first input LLR sequence is decoded to obtain K(i) first output LLR sequences, including:

The first parallel decoding algorithm performs a first stage update on the K(i) first input LLR sequences respectively to obtain the K(i) first output LLR sequences.

Wherein, i=1, K(i)=1, and the first input LLR sequence is the initial LLR sequence.

i>1, the i-1th iterative decoding also includes:

Obtain M(i-1,t) second decoded LLR sequences, where the second decoded LLR sequence includes N LLRs.

The i-th iteration further includes:

Acquiring M(i-1,t) first output LLR sequences of the i-1th iteration corresponding to the M(i-1,t) second decoded LLR sequences of the i-1th iteration;

According to the M(i-1,t) second decoded LLR sequences and the first output LLR sequence of the M(i-1,t) i-1th iteration, K(i) i-th The second update input LLR sequence of the second iteration, K(i)=M(i-1,t);

Using the first parallel decoding algorithm to perform a second stage update on the K(i) second updated input LLR sequences to obtain the L(i, 1) first input LLR sequences.

In yet another possible implementation manner, the first parallel decoding algorithm includes at least n+1th to _np +1th layers n _p -n decoding layers,

Each layer includes N _p LLR nodes;

The first stage update of each first input LLR sequence by using the first parallel decoding algorithm to obtain the corresponding first output LLR sequence includes:

The number of n _{_p} N _p + 1 layer node LLR updated LLR input to the first sequence of N _p LLR,

Perform soft value update from the n _p +1th layer to the n+1th layer to obtain N _p LLR nodes of the n+1th layer, and the corresponding first output LLR sequence includes the first output LLR sequence. _{N p} LLR nodes of n+1 layer;

The second stage update of each second updated input LLR sequence by using the first parallel decoding algorithm to obtain the corresponding first input LLR sequence includes:

The LLR of N _p n + 1 node in the first layer to said second update input LLR update sequence of N _p LLR,

Perform soft value update from the n+1th layer to the _np +1th layer to obtain _Np _{LLR nodes of the np} +1th layer, and the corresponding first input LLR sequence includes the _{N p} LLR nodes of the n _p +1th layer.

In another possible implementation manner, t<T and the tth iteration does not meet the early termination condition, the tth iteration of the cascade decoding method includes: decoding according to the M(i,t) The path gets M(i,t) third output LLR sequences.

In another possible implementation manner, t=T or the second-level iteration meets the early termination condition, and i<I and the i-th first-level iteration does not meet the early termination condition. The t iterations include:

According to the M(i,t) decoding paths, M(i,t) third output LLR sequences are obtained.

Acquiring M(i,t) second output LLR sequences corresponding to the M(i,t) third output LLR sequences of the t-th iteration;

According to the M(i,t) third output LLR sequence and the M(i,t) second output LLR sequence of the tth iteration, the second output of the M(i,t) tth iteration is obtained. Update the input LLR sequence;

Performing a second stage update on the M(i,t) second updated input LLR sequences to obtain M(i,t) second decoded LLR sequences.

In another possible implementation manner, i=I and t=T, or the t-th iteration meets the early termination condition of the second-level iteration and the i-th meets the early termination condition of the first-level iteration. The t-th iteration of the associative decoding method includes:

Obtain the decoding result according to the M(i,t) decoding paths, and terminate the iteration.

In another possible implementation manner, the first parallel decoder algorithm includes one or more of the following: BP decoding algorithm, or MS decoding algorithm or DNN decoding algorithm.

In another possible implementation manner, the _{value of N p} includes any one of the following: 8192, 4096, 2048, 1024, 512, 256, 128.

In a fifth aspect, an embodiment of the present application provides a multi-cascade decoder, including a first parallel decoder as in the first aspect or any possible implementation of the first aspect, where N _in =N _p , N _out =N, S=I, and the cascade decoder of the third aspect or any one of the first to tenth possible implementation manners of the third aspect. The cascaded decoder is a lower-level decoder of the first parallel decoder, and the first parallel decoder is used to decode one or more LLR sequences with _{a length of N p.} A parallel decoder performs a maximum of I iterations on the initial LLR sequence, I=N _p /N, the i-th iteration, i<I, including:

The first parallel decoder performs the first stage iterative decoding on K(i) first input LLR sequences to obtain K(i) first output LLR sequences, and the length of each first input LLR sequence is N _p , The length of each first output LLR sequence is N _p ;

The cascaded decoder performs a second stage iterative decoding on K(i) second input LLR sequences, and each of the second input LLR sequences includes N LLRs in the corresponding first output LLR sequence.

Since the LDPC decoder usually uses a parallel decoding algorithm for decoding, the multi-cascade decoder can further share part of the parallel decoding unit of the LDPC decoder to save system overhead.

With reference to the fifth aspect, in the first possible implementation manner of the fifth aspect, each of the second input LLR sequences includes the (i-1)×N+1th in the corresponding first output LLR sequence LLR to the i×Nth LLR.

With reference to the fifth aspect or the first possible implementation manner of the fifth aspect, in the second possible implementation manner of the fifth aspect, the first parallel decoder receives the K(i) first inputs The LLR sequence is decoded to obtain K(i) first output LLR sequences, including:

The first parallel decoder performs a first stage update on the K(i) first input LLR sequences respectively to obtain the K(i) first output LLR sequences.

i>1, the i-1th iterative decoding also includes:

The cascaded decoder outputs M(i-1,t) second decoded LLR sequences to the first parallel decoder, and the second decoded LLR sequence includes N LLRs.

The i-th iteration further includes:

The first parallel decoder obtains M(i-1,t) corresponding to the i-1th iteration of M(i-1,t) second decoding LLR sequences of the i-1th iteration ) First output LLR sequence;

The first parallel decoder according to the M(i-1,t) second decoding LLR sequences and the M(i-1,t) first output LLR sequences of the i-1th iteration Obtain K(i) the second updated input LLR sequence of the i-th iteration, K(i)=M(i-1,t);

The first parallel decoder performs a second stage update on the K(i) second updated input LLR sequences to obtain the L(i, 1) first input LLR sequences.

In yet another possible implementation manner, the first parallel decoder includes at least n+1th to _np +1th layers n _p -n decoding layers,

Each layer includes N _p LLR nodes;

The first parallel decoder performs a first stage update on each first input LLR sequence to obtain the corresponding first output LLR sequence, including:

Parallel to said first coder _p +1 said first layer of N _p n of nodes updated LLR LLR input to the first sequence of N _p LLR,

The first parallel decoder _{performs a soft value update from the np} +1th layer to the n+1th layer to obtain the _Np LLR nodes of the n+1th layer, and the corresponding An output LLR sequence includes N _p LLR nodes of the n+1th layer;

The first parallel decoder performs a second stage update on each second update input LLR sequence to obtain the corresponding first input LLR sequence, including:

The first parallel decoder LLR the nodes of N _p n + 1, the second layer updates the LLR updated input sequence of N _p LLR,

The first parallel decoder performs a soft value update from the n+1th layer to the _np +1th layer to obtain the _Np LLR nodes of the _np +1th layer, and the corresponding The first input LLR sequence includes N _p LLR nodes of _{the n p +1 th layer.}

In another possible implementation manner, the serial decoder determines that t<T and the tth iteration does not meet the early termination condition, and the tth iteration of the cascaded decoder includes: the serial The decoder obtains M(i,t) third output LLR sequences according to the M(i,t) decoding paths.

In another possible implementation manner, the serial decoder determines that t=T or the second-level iteration meets the early termination condition, and i<I and the i-th first-level iteration does not meet the multi-cascade The decoder terminates conditions early, and the t-th iteration of the cascaded decoder includes:

The serial decoder obtains M(i,t) third output LLR sequences according to the M(i,t) decoding paths.

Acquiring, by the second parallel decoder, M(i,t) second output LLR sequences corresponding to the M(i,t) third output LLR sequences of the t-th iteration;

The second parallel decoder obtains M(i,t) according to the M(i,t) third output LLR sequences and the M(i,t) second output LLR sequences of the tth iteration The second updated input LLR sequence of the t-th iteration;

The second parallel decoder performs a second stage update on the M(i,t) second update input LLR sequences to obtain M(i,t) second decoded LLR sequences;

The second parallel decoder outputs the M(i,t) second decoded LLR sequences to the first parallel decoder.

In another possible implementation manner, the serial decoder determines that i=I and t=T, or the t-th iteration satisfies the early termination condition of the cascaded decoder and the i-th is the first The stage iteration satisfies the early termination condition of the multi-cascade decoder, and the t-th iteration of the cascade decoder includes:

The serial decoder obtains the decoding result according to the M(i,t) decoding paths, and terminates the iteration.

In another possible implementation manner, the first parallel decoder includes one or more of the following: a BP decoder, an MS decoder, or a DNN decoder.

In a sixth aspect, an embodiment of the present application provides a decoding device, which has the function of implementing the method described in any one of the possible designs of the second aspect and the fourth aspect. The function can be realized by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules or units corresponding to the above-mentioned functions.

In a possible design, when part or all of the functions are realized by hardware, the decoding device includes: an input interface circuit for obtaining the LLR sequence corresponding to the bit sequence to be decoded; and a logic circuit for executing The method described in the second aspect or the fourth aspect or any one of the possible designs of the foregoing two aspects; an output interface circuit for outputting information bits.

Optionally, the decoding device may be a chip or an integrated circuit.

In a possible design, when part or all of the function is realized by software, the decoding device includes: a memory for storing a program; a processor for executing the program stored in the memory, when When the program is executed, the decoding device can implement the method described in the second aspect or the fourth aspect or any one of the possible designs of the foregoing two aspects.

Optionally, the foregoing memory may be a physically independent unit, or may be integrated with the processor.

In a possible design, when part or all of the functions are implemented by software, the decoding device includes a processor. The memory for storing the program is located outside the decoding device, and the processor is connected to the memory through a circuit/wire for reading and executing the program stored in the memory.

In a possible design, the communication device provided in the sixth aspect includes a processor and a transceiver component, and the processor and the transceiver component can be used to implement the functions of each part of the foregoing encoding or decoding method. In this design, if the communication device is a terminal, a base station or other network equipment, its transceiver component can be a transceiver. If the communication device is a baseband chip or a baseband single board, its transceiver component can be a baseband chip or a baseband single board. The input/output circuit is used to realize the receiving/sending of input/output signals. Optionally, the communication device may further include a memory for storing data and/or instructions.

In a seventh aspect, an embodiment of the present application provides a network device, including any possible decoder as in the first aspect, the third aspect, or the fifth aspect, or the decoding device of the sixth aspect.

In an eighth aspect, an embodiment of the present application provides a terminal device, including any possible decoder as in the first aspect, the third aspect, or the fifth aspect, or the decoding device in the sixth aspect.

In a ninth aspect, an embodiment of the present application provides a communication system, which includes the network device of the seventh aspect and the terminal device of the eighth aspect.

In a tenth aspect, an embodiment of the present application provides a computer storage medium that stores a computer program, and the computer program includes instructions for executing the method described in any one of the above-mentioned second or fourth aspects.

In an eleventh aspect, a computer program product containing instructions is provided, which when running on a computer, causes the computer to execute the method described in any one of the possible designs of the second or fourth aspects.

Description of the drawings

Figure 1 is an architecture diagram of the communication system provided by this application;

Figure 2a is a schematic diagram of the decoding path of an SCL decoding algorithm provided by this application;

2b is a schematic diagram of the decoding path of an SCL decoding algorithm provided by this application;

Fig. 3a is a schematic diagram of a basic processing unit of a parallel decoding algorithm provided by this application;

FIG. 3b is a schematic diagram of iterative calculation of a parallel decoding algorithm butterfly network provided by this application;

3c is a schematic diagram of an iterative operation unit of a DNN decoding algorithm provided by this application;

FIG. 4 is an example of a Tanner graph of an LDPC code provided by an embodiment of the application;

FIG. 5 is a schematic structural diagram of a parallel decoder provided by an embodiment of this application;

FIG. 6 is a schematic structural diagram of a cascaded decoder provided by an embodiment of this application;

FIG. 7 is a flowchart of a cascaded decoding method provided by an embodiment of this application;

FIG. 8 is a schematic structural diagram of a multi-cascade decoder provided by an embodiment of this application;

FIG. 9 is a flowchart of a cascaded decoding method provided by an embodiment of this application;

Fig. 10 is a decoding performance diagram of the cascaded decoding method and other decoding algorithms provided by the implementation of this application.

Detailed ways

The embodiments of the present application can be applied to various fields that adopt Polar coding, such as: data storage field, optical network communication field, wireless communication field, and so on. Among them, the wireless communication systems involved in the embodiments of the present application include but are not limited to: global system for mobile communications (GSM) system, code division multiple access (CDMA) system, and broadband code division multiple access (GSM) system. wideband code division multiple access, WCDMA) system, general packet radio service (GPRS), long term evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division Duplex (time division duplex, TDD), universal mobile telecommunication system (UMTS), worldwide interoperability for microwave access (WiMAX) communication system, the future 5th generation (5G) ) System or new radio (NR), vehicle-to-X V2X, where V2X can include vehicle-to-network (V2N), vehicle-to-vehicle (V2V) ), Vehicle to Infrastructure (V2I), Vehicle to Pedestrian (V2P), etc., Long Term Evolution-Vehicle (LTE-V) of Workshop Communication, Internet of Vehicles, Machine Communication ( Machine type communication (MTC), Internet of Things (IoT), Long Term Evolution-Machine (LTE-M), Machine to Machine (M2M), etc. Of course, the areas where Polar encoding is used can also be other, which is not specifically limited in this application.

The communication device involved in this application may be a chip (such as a baseband chip, or a data signal processing chip, or a general-purpose chip, etc.), a terminal, a base station, or other network equipment. Among them, a terminal is a device with a communication function, which can communicate with one or more core networks via a radio access network (Radio Access Network, RAN). The terminal may include a handheld device with a wireless communication function, a vehicle-mounted device, a wearable device, a computing device, or other processing device connected to a wireless modem. Terminals can be called different names in different networks, such as: user equipment (UE), mobile station (Mobile Station, MS), subscriber unit, station, cellular phone, personal digital assistant, wireless modem, wireless communication equipment , Handheld devices, laptop computers, cordless phones, wireless local loop stations, etc. For the convenience of description, it is simply referred to as a terminal in this application. A base station (BS), also called a base station device, is a device deployed on a wireless access network to provide wireless communication functions. The name of the base station may be different in different wireless access systems. For example, in the Universal Mobile Telecommunications System (UMTS) network, the base station is called NodeB (NodeB), while in the LTE network, the base station is called NodeB. The base station is called an evolved NodeB (evolved NodeB, eNB or eNodeB), and the base station in the new radio (NR) network is called the transmission reception point (TRP) or the next generation node B (gNB) ), or the base station can also be a relay station, access point, in-vehicle device, wearable device, network equipment in the future evolution of the Public Land Mobile Network (PLMN), or base station in various other evolved networks Other names may also be used. The present invention is not limited to this.

Figure 1 is an architecture diagram of the communication system provided by this application. It should be noted that FIG. 1 merely illustrates an architecture diagram of a communication system in the form of an example, and is not a limitation on the architecture diagram of the communication system.

Please refer to FIG. 1, which includes a communication device 101 and a communication device 102. For the convenience of description, this article uses the communication device 101 as a transmitting end device to send a signal, and the communication device 102 as a receiving end device to receive a signal as an example for description. Of course, the communication device 102 can also send information to the communication device 101, and the communication device 101 receives the signal accordingly, then the communication device 102 is the transmitting end device, and the communication device 101 is the receiving end device. The embodiment of the present invention is not limited to this. The transmitting end device includes an encoder, and the receiving end device includes a decoder. Since the communication device may be a transmitting end device or a receiving end device, it may include an encoder and a decoder.

The communication device 101 can respond to the sequence of information to be sent

Such as the signaling transmitted on the control channel, perform polar encoding and output the encoded sequence

Encoded sequence

After rate matching, interleaving, and modulation, it is transmitted to the communication device 102 on the control channel. The communication device 102 performs processing such as demodulation on the received signal to obtain a Likelihood Rate (LLR) sequence

LLR sequence

The number of LLR soft values included in is the same as the number of bits included in the information sequence, and both are N. It can also be said that its length is N, and N is a positive integer greater than 0. The communication device 102 performs Polar decoding according to the received LLR sequence. Wherein, regardless of whether the communication device 101 sends a bit 1 or a bit 0, the communication device 102 may make a misjudgment. For the signal r, the ratio of the probability p(r|b=0) that is correctly judged as 0 by the receiving end device and the probability p(r|b=1)] that is correctly judged as 1 is the likelihood ratio. In order to facilitate the calculation and processing, the natural logarithm of the likelihood ratio can be used to obtain the log-likelihood ratio, that is, LLR=ln[p(r|b=0)/p(r|b=1)]. LLR can be a floating point number.

Several decoding algorithms are described below.

1. Serial decoding algorithm

Serial decoding algorithms mainly include SC decoding algorithm and SCL decoding algorithm. Among them, there are many improved decoding algorithms based on SCL decoding algorithm, such as CA-SCL algorithm with CRC check.

Information sequence

Including information bits and frozen bits, Polar encoded and _{sent through the channel W N} , output

Its transition probability is

The SC decoder makes a decision on the LLR value related _{to each bit u i} in the information sequence from the first (i=1) to the last (i=N) in order to obtain its estimated value

Wherein the information bits for the index set A, set to freeze bit index A ^c. If _{i ∈ A} ^c , u i is a frozen bit, and its value is known, for example, it is fixed to 0 or 1, so it can be directly judged

And the decision result of this bit is used for the decision of the next bit u _i+1 ; if _{i ∈ A} , u i is the information bit, and the decision result of each bit before the bit is obtained

After that, the decoding LLR is

And make a hard judgment on the LLR to get the judgment result

And the decision result of this bit is used for the decision of the next bit u _i+1 . The decision function of the above polarization code is as follows:

Hard decision function

It can be expressed as follows:

The decoding LLR corresponding to u _{i is defined as follows:}

Is the polarized sub-channel with sequence number i of channel W _N,

Yes

The transition probability function of, expressing through the polarized sub-channel

Send information bits u _{i to} get output

with

The probability of can be calculated recursively:

Among them, sign(L ₁ ,L ₂ ) represents _{the symbolic product of L 1} and L ₂ , |x| represents the absolute value of x, and min(x,y,...) represents the minimum value of the values in parentheses. In formula (5)

among them,

Is the sequence

A subsequence of odd-numbered elements in the middle,

Is the sequence

A subsequence composed of elements with an even index in the middle.

It can be seen that the LLR sequence of length N can be reduced to two LLR sequences of length N/2 for calculation, and according to the above recursive process recursion, it can be calculated by multiple reductions to LLR sequences of length 1, that is, to 1 A soft value of LLR is calculated. For example: N=8, the first recursive process can be reduced to two LLR sequence calculations of length 4, and for each LLR sequence of length 4, it can be reduced to two LLR sequences of length 2 according to the above recursive process. The LLR sequence of, a total of 4 LLR sequences of length 2 are calculated, and further recursive, reduced to 8 LLR sequences of length 1. The calculation of the soft value of 1 LLR can be obtained according to the following formula:

The SC decoding process can be described as a depth-first search process on a code tree. Figure 2a shows an example of a code tree. The code length is N=4, corresponding to a full binary tree with a depth of N. Each layer corresponds to an information bit or a frozen bit, and the two edges between each parent node and two child nodes are marked as 0 path and 1 path respectively, and a total of 2 ^N paths can be expanded. The SC decoder starts decoding from the root node u ₁ , and selects 0 path or 1 path each time according to the decision result of the current bit. After reaching the leaf node, the N bit decision ends, and the path of the SC decoder in the code tree is Is the decoding result, as shown in Figure 2a, the SC decoding result is

The SC decoding algorithm selects 0 path or 1 path at each node according to the current decision result, and each step is a local optimal choice. If a certain bit is judged incorrectly, it will continue to expand along the path. The current error cannot be corrected, and the current error will affect the subsequent decoding process. The SCL decoding algorithm changes the hard decision in the SC decoding algorithm to a soft decision, that is, the L paths with a decision of 0 or 1 are retained, where L is the search width. For each path expansion, the path metric (PM) must be calculated. PM is the probability of the decoding sequence corresponding to a path, which is usually expressed in logarithmic form as follows:

Each time the path is expanded, the SCL decoding algorithm sorts the PM value and outputs the L decoding paths with the largest PM value. At the last bit, the path with the largest PM value is selected as the decoding output and input.

The calculation process of PM is as follows:

If u _i are information bits or correct frozen bits, and

Then the path metric is calculated as follows:

If u _i are information bits or correct frozen bits, and

Then the path metric is calculated as follows:

among them

The calculation of is the same as the SC decoding algorithm. If u _i is a frozen bit and the value is wrong, the path metric value is calculated as follows:

The CA-SCL decoding algorithm is an optimization of the SCL decoding algorithm. The CRC is introduced into the information sequence, and the CRC is used to assist the decision, and the path that passes the CRC check and the PM value is the largest is selected. In the CA-SCL decoding algorithm, the path formed from the root node to any node in the code tree corresponds to a path metric value; each time the path is expanded, the L paths with the largest path metric value in the current layer are selected . After reaching the leaf node, the decoding sequences corresponding to the L paths are output in the order of the metric value from small to large, forming a set of candidate decoding sequences. Perform a CRC check on the candidate decoding sequence, and select the path with the largest path metric value that can pass the CRC check as the final decoding result.

Figure 2b shows the path search process of an SCL decoding algorithm with code length N=4 and L=2. Finally, the path with the largest path metric value is output among the two decoding paths {0011} and {1000} as the final decoding result. If the CA-SCL decoding algorithm is used, the one that passes the CRC check and the path metric value is the largest is output among the two decoding paths as the final decoding result.

Although the decoding performance of the serial decoding algorithm is better, its decoding process is serial, which cannot meet the throughput requirements of the communication system.

2. Parallel decoding algorithm

Parallel decoding algorithms include BP decoding algorithm, MS decoding algorithm and DNN decoding algorithm. It can be used for Polar code decoding and LDPC code decoding.

1. The parallel decoding algorithm for Polar codes is decoded based on the factor graph of the generator matrix G. The following takes the BP decoding algorithm as an example to introduce.

For Polar codes with P=(N,K), the factor graph generally includes n+1 decoding layers, each with N nodes, a total of N×(n+1) nodes, n=log ₂ N, where Each layer has N/2 basic processing elements PE (process element, PE), as shown in Figure 3a, a schematic diagram of a basic processing unit, including two input variables and two output variables, each node (i, j ) Contains two kinds of information, the information transmitted from left to right, namely the right information

Information passed from right to left, left information

Among them, j represents the layer number, j=1, 2,...,n+1, i represents the sequence number of the node in each layer, i=1, 2,...,N. Usually the leftmost layer, that is, j = 1, each node in the first layer represents the information sequence

Each bit u ₁ , u ₂ ,..., u _{N in the} , the rightmost layer, that is, j=n+1, each node in the n+1th layer represents

Each LLR soft value in y ₁ , y ₂ ,..., y _N.

Generally, the information will be transferred layer by layer from left to right, that is, from the first layer to the n+1 layer. It is called the right operation R operation. From right to left, that is, from the n+1 layer to the first layer. The layer transfer information is called the left operation L operation. It should be noted that sometimes the transfer of information layer by layer is also referred to as the information update process. among them,

with

All are LLRs, where t is the number of iterations, 0≤t≤T, and T is the maximum number of iterations of the BP decoding algorithm. The calculation formula of the output node in Figure 3a is as follows:

As shown in Fig. 3b, it is a schematic flow diagram of one iteration in the decoding process of the Polar code with a code length of N=8 using the BP decoding algorithm. j represents the number of decoding layers updated by the operation, there are log ₂ 8+1=4 layers in total, and i represents the sequence number of the node in each layer. Among them, the basic processing unit is represented as a butterfly operation unit in the figure, for example: a butterfly unit connecting nodes (1,1), (1,2), (2,1) and (2,2), connecting nodes ( 2,2), (2,3), (4,2) and (4,3) butterfly elements, connecting nodes (4,3), (4,4), (8,3) and (8, 4) Butterfly unit and so on. In each iteration of BP decoding, the LLR L operation is updated layer by layer from right to left. After reaching the left end, the LLR R operation update is performed layer by layer. When all nodes are visited once, an iteration operation is completed. After each iteration is completed, the LLR value of the corresponding information bit is hard-decided and the CRC is checked. If the CRC passes or reaches the maximum number of iterations, the iteration is stopped, otherwise the iteration is continued.

From the butterfly network, it can be found that after the n+1th layer to the nth layer are updated, the LLR sequence of length N can be regarded as reduced to two LLR subsequences of length N/2 for the next layer update respectively. Update one layer on the left, and the length of the LLR subsequence can be reduced to half of the previous layer. For the jth layer, it can be regarded as 2 ^n+1-j LLR subsequences with a length of 2 ^j-1.

The iterative process of the MS decoding algorithm is similar to the iterative process of the BP decoding algorithm, and it will not be described separately below, and it is generally called the BP/MS decoding algorithm.

The node update process of DNN can simulate the node update process of one or more iterations in the BP/MS decoding algorithm, and it can also assign different weights to each edge during the update process to improve the decoding performance. When DNN is used to realize Polar decoding, it can imitate the structure of BP/MS decoding algorithm. As shown in Figure 3c, it is the architecture of the DNN decoding algorithm. The L operation from right to left of the butterfly unit network and the R operation from left to right are cascaded to form an iterative operation unit, which combines the iterative operation unit of T By cascading, the DNN decoding corresponding to T iterations can be completed.

2. The parallel decoding algorithm for LDPC codes is decoded based on a check matrix.

The parallel decoding algorithm for LDPC codes can also be updated in two stages, corresponding to the L operation and the R operation in the Polar code factor graph parallel decoding algorithm. Among them, according to the check matrix calculation from top to bottom row by row, the transfer information corresponds to the L operation in the Polar code factor graph parallel decoding algorithm; according to the check matrix calculation from bottom to top row by row, the transfer information corresponds to the Polar code R operation in factor graph parallel decoding algorithm. Since the LDPC matrix supports row-column exchange and does not change its decoding properties, the corresponding R operation can also be calculated row by row from bottom to top for the check matrix, and the corresponding L operation can be calculated row by row from top to bottom.

Among them, the LDPC check matrix can correspond to the Tanner graph. For example, an example of the LDPC code check matrix and its corresponding check equation is:

Among them, "+" means modulo 2 plus.

The Tanner graph corresponding to the check matrix can be represented as shown in Figure 4. Each circular node in Figure 4 is a variable node, representing a column in the check matrix H, and each square node is a check node, representing a check In a row of matrix H, each edge connecting the check node and the variable node in Figure 4 represents a non-zero element at the intersection of the row and column corresponding to the two nodes.

For specific calculations, decoding algorithms such as BP/MS can also be used. For BP, the decoding formula can be written as:

For MS, its decoding formula can be written as:

In the formula, R _ij represents the LLR that needs to be updated for the j-th variable node, and Q _ji represents the LLR passed by other variable nodes to the current check node.

Although the throughput rate of the parallel decoding algorithm is high, the decoding performance is poor, and there is a big gap between the performance of the serial decoding algorithm.

The communication system puts forward higher performance requirements and throughput requirements for decoding. At the same time, as the 5G communication system requires both Polar decoding and LDPC encoding and decoding, reducing the overhead of the decoder is also a need to solve The problem.

FIG. 5 is a schematic structural diagram of a parallel decoder 500 for cascaded decoding according to an embodiment of the application, which is used to decode one or more log-likelihood ratio LLR sequences of _{length N in.} And provide one or more input LLR sequences of _{length N out to the next stage decoder.} Wherein, N _in and N _out are both integers, and N _out <N _in . Generally, N _out and N _in are generally powers of 2. For example, the _{value of N in} can be any of the following: 8192, 4096, 2048, 1024, 512, 256, 128, 64, 32, and N _out can be any of the following One: 1024,512,256,128,64,32,16,8,4,2,1128,64,32,16,8,4,2,1.

The parallel decoder (N _in , N _out ) can be used to represent the length of the decoded input LLR sequence of the parallel decoder 500 and the input LLR sequence provided to the next stage, for example, the parallel decoder (1024, 64) represents The length of the decoded input LLR sequence is 1024, and the parallel decoder with the length of the input LLR sequence provided to the next stage is 64; the parallel decoder (8192, 512) indicates that the length of the decoded input LLR sequence is 8192, The parallel decoder with the length of the input LLR sequence provided by the next stage is 512. It should be noted that this is only an example, not limited to the above examples.

Wherein, the parallel decoder 500 can update the information of the decoded input LLR sequence according to any of the aforementioned parallel decoding algorithms, and the length of the output LLR sequence and the decoded input LLR sequence are usually equal, and both are N _in . The parallel decoder 500 provides an _{input LLR sequence with a length of N out to} _{the next-stage decoder. The parallel decoder 500 may determine N out} LLRs as the next _{LLR sequence in} the output LLR sequence with a length of N in. The input LLR sequence of the first-level decoder can also be determined by the next-level decoder according to the output LLR sequence of the _{parallel decoder 500 whose length is N in} _{. N out} LLRs are used as the input of the next-level decoder. The LLR sequence is not limited in this embodiment of the application. If multiple iterative decoding is performed, in a possible implementation, for the s-th iterative decoding, the input LLR sequence of the next-stage decoder includes the (s-1)×N _out +th in the output LLR sequence 1 LLR to the s×N _outth LLR. In this way, the maximum number of iterations is N _in /N _out .

Since the LLR sequence provided by the parallel decoder 500 to the next decoder is a part of its output LLR sequence, the decoding process is adjusted accordingly.

As shown in FIG. 5, the parallel decoder 500 may include a first-stage update unit 510 and a second-stage update unit 520. Among them, for the sth iteration, 0<s≤S, and S is the maximum number of iterations:

The first-stage update unit 510 can be used to decode the input LLR sequence of the parallel decoder 500

Perform the first stage update to obtain the output LLR sequence of the parallel decoder 500

The output LLR sequence

Used to provide the input LLR sequence of the next stage decoder

s=1, the input LLR sequence of the first stage update unit 510

The LLR sequence is input for the decoding of the parallel decoder 500.

When s>1,

The second-stage update unit 520 may be used for:

According to the output LLR sequence of the next stage decoder in the s-1th iteration

And the output LLR sequence of the parallel decoder 500

Get the second updated input LLR sequence of the sth iteration

The second updated input LLR sequence for the sth iteration

Perform the second stage update to get the second update output LLR sequence

Among them, the second update input LLR sequence

For will

Replace the (s-2) _{th N out} +1 LLR to the (s-1) _{th N out} LLR with

The LLR sequence obtained by the N _{out LLRs.}

The input LLR sequence of the first stage update unit 510 is

The second update output LLR sequence obtained from the second-stage update unit 520, the first-stage update unit 510 may be used to perform the first-stage update on the input LLR sequence to obtain the output LLR sequence of the parallel decoder 500

The output LLR sequence is used to provide the input LLR sequence of the next stage decoder.

For a Polar code with a code length of N, the factor graph is n+1 layers,

N is usually a power of 2, then n=log ₂ N. The input length of _{the parallel decoder is N in} , and the maximum number of decoding layers of the parallel decoder 500 can be n _in +1 layer, and n _in =log ₂ N _in . The length of the input sequence provided by the parallel decoder 500 to the next decoder is N _out , which can be obtained at the n _{out +1 layer of the parallel decoder 500}

LLR subsequences of length N _{out, where}

N _out is usually a power of 2, then n _out =log ₂ N _out . The parallel decoder 500 may include at least n _out +1 th to n _in +1 th decoding layers.

For the sth iteration, _{the N in} _{LLRs of the n in} +1th layer are assigned to the LLR sequence

_{The N in} LLRs of the n _out +1 layer are recorded as the LLR sequence

The LLR sequence that the first stage update unit 510 can input

N _in +1 from the layer begin to give a soft value updating n _out N _in number of nodes +1 LLR layer, the first n _out N _in a layer LLR +1 nodes as output LLR sequence

In a possible implementation, the first-stage update unit 510 obtains the LLR sequence as input

N _{_in} N _in number of nodes +1 LLR LLR for the layer sequence LLR N _in number, comprising a first stage update from n _in n to the layer of n _out +1 +1 direction _in -n _out secondary layer LLR soft value update, _{the calculation formula updated from the n in} +1 layer to the n _out +1 layer direction can refer to the formulas (10) and (11) of the L operation in the aforementioned parallel decoding algorithm, or, from top to bottom. Formula (14) or (15) for row calculation. Starting from the n _in +1 layer, update the LLR nodes layer by layer, and when updating to the n _out +1 layer, the obtained N _in LLR nodes are the output LLR sequence

It may include N _in /N _out LLR subsequences with a length of N _out.

In another possible implementation manner, the first-stage update unit 510 obtains the LLR sequence as input

N _{_in} N _in number of nodes +1 LLR LLR for the layer sequence LLR N _in number, the first stage of updating comprises the first layer of m + 1 + 1 layer n _out from the first direction and from n _in n _out +1 to update the first layer n _in +1 direction layer _{2 × m × (n in -n} out) times LLR soft values, and the layers from the n _in +1 direction of layer n _out +1 (n _in - n _out ) times of LLR soft value update, a total of (2×m+1)×(n _in -n _out ) times of LLR soft value update, m is an integer, m≥0. The calculation formula updated from the n _out +1th layer to the n _in +1th layer can refer to the formulas (12) and (13) of the R operation in the aforementioned parallel decoding algorithm. Repeat one or more times from _{layer n in} +1 to layer n _out +1 and from _{layer n out} +1 to layer n _in +1, layer by layer update, and then from layer n _in +1 The layer starts to update the LLR nodes layer by layer, and when the layer is _{updated to the n out} _{+1th layer, the obtained N in} LLR nodes are the output LLR sequence

It may include N _in /N _out LLR subsequences with a length of N _out. It should be noted that in each iteration of the parallel decoder 500, the value of m may be different.

Among them, the input LLR sequence of the first stage update unit 510

It may be the decoded input LLR sequence of the parallel decoder 500, or it may be the second updated output LLR sequence obtained by the second-stage update unit 520.

In a possible implementation, the first-stage update unit 510 may also be used to output LLR sequence from

Determine the input LLR sequence provided to the next-stage decoder. Wherein, the input LLR sequence provided to the next-stage decoder includes an output LLR sequence

There are N _out LLRs. Due to the output LLR sequence

It may include N _in /N _out LLR subsequences with a length of N _out , and the first-stage update unit 510 may start from

Determine an LLR subsequence as the input LLR sequence provided to the next-stage decoder. For example, for the sth iteration of the parallel decoder 500, the LLR subsequence is the output LLR sequence

The s-th subsequence in, may include the output LLR sequence

(S-1)×N _out + 1 LLR to s×N _out LLR,

The second-stage update unit 520 can be used to input the LLR sequence for the second update

To give n _{_in} N _in number of nodes LLR + 1 layer, the first n _{_in} N _in a layer of the +1 LLR node n _out +1 layer started from the second stage update

The LLR sequence is output as the second update.

In a possible implementation, the second-stage update unit 520 obtains the LLR sequence as input

N _out N _in number of nodes +1 LLR LLR for the layer sequence number N _in LLR, the second stage of updating comprises n _out +1 from layer to layer of n _in +1 direction n _in -n _out times The LLR soft value is updated, and the updated calculation formula can refer to the formulas (12) and (13) of the R operation in the aforementioned parallel decoding algorithm, or the formulas (14) or (15) calculated row by row from bottom to top. The second-stage update unit 520 _{updates the LLR nodes layer by layer starting from the n out} +1 layer, and when updating to the n _in +1 layer, _{the value of the N in} LLR nodes is the second update output LLR sequence

Its length is N _in .

In a possible implementation, the parallel decoder 500 can also be cascaded with another parallel decoder 500, for example, a first parallel decoder (N _in1 , N _out1 ) and a second parallel decoder ( N _in2 , N _out2 ) are cascaded, where the first parallel decoder is the upper-level decoder, the second parallel decoder is the next-level decoder, N _out1 = N _in2 , the second parallel decoder The decoding input LLR sequence of is derived from the input LLR sequence of the first parallel decoder, and the second parallel decoder is also used for the upper decoder, for example, to return one or more lengths to the first parallel decoder It is the decoded LLR sequence _{of N in2} or N _out1.

In a possible implementation manner, one or more parallel decoders 500 may be cascaded with a serial decoder supporting a smaller code length. As shown in Figure 6, the parallel decoder 620 (N, N _S ) and the serial decoder 630 supporting a code length of N _S are cascaded. As shown in Figure 8, the parallel decoder 810 (N _P , N ), the parallel decoder 620 (N, N _S ) and the serial decoder 630 supporting a code length of N _{S are cascaded.} Among them, N can be any of the following: 1024, 512, 256, 128, 64, 32, and Ns can be any of the following: 128, 64, 32, 16, 8, 4, 2, 1, N _p can be Value any of the following: 8192,4096,2048,1024,512,256,128. It should be noted that this is only an example and is not limited to this. For the convenience of description, hereinafter, the parallel decoder 620 is referred to as a second parallel decoder, and the parallel decoder 810 is referred to as a first parallel decoder.

The parallel decoder provided by the embodiment of the present invention can be used for cascaded decoding, which converts a larger code length into a smaller code length through parallel decoding and outputs it to the next-stage decoder, which can improve the throughput rate. It can also reduce the implementation overhead of the next-level decoder.

As shown in FIG. 6, it is a schematic structural diagram of a cascaded decoder 600 in which a parallel decoder 620 and a serial decoder 630 are cascaded according to an embodiment of the present invention, wherein the serial decoder 630 is a parallel decoder 620's next-level decoder. Among them, the parallel decoder 620 can use a BP decoding algorithm, an MS decoding algorithm or a DNN decoding algorithm, and the serial decoder 630 can use an SCL decoding algorithm, a CA-SCL decoding algorithm, and so on. The length of the input LLR sequence provided by the parallel decoder 620 to the serial decoder 630 is N _S. For example, the length of the input LLR sequence of the parallel decoder 620 is 8, and the length of the input LLR sequence provided to the serial decoder 630 is 4. In the parallel decoder 620, N _in =N and N _out =N _S. It should be noted that this is only an example and not a limitation.

Referring to FIG. 7, it is a flowchart of a decoding method of a cascade decoder according to an embodiment of the present invention. The decoding method of the cascaded decoder shown in FIG. 7 will be described below in conjunction with the cascaded decoder 600 of FIG. 6. Wherein, the input LLR sequence of the cascaded decoder 600 includes N input LLRs, and the input LLR sequence is subjected to a maximum of T iterative cascaded decoding, where the t-th iterative cascaded decoding, 0<t≤ T, including the method steps shown in Figure 7:

Step 710: The second parallel decoder 620 decodes the L(t) second input LLR sequences to obtain L(t) second output LLR sequences.

Among them, the length of each second input LLR sequence is N, expressed as

The length of each second output LLR sequence is N, expressed as

For each second input LLR sequence

The second parallel decoder 620 obtains the corresponding second output LLR sequence each iteration

Used to provide the third input LLR sequence of the serial decoder 630

(t). Among them, the third input LLR sequence

Include the second output LLR sequence

There are Ns LLRs, for example: cLLR _(t-1)·Ns+1 , cLLR _(t-1)·Ns+2 ,..., cLLR _t·Ns .

The second parallel decoder 620 responds to the second input LLR sequence

For the decoding process, refer to the description of the parallel decoder in the foregoing embodiment, which will not be repeated here.

In a possible implementation manner, the second parallel decoder 620 according to the second output LLR sequence

Determine the third input LLR sequence

And output to the serial decoder 630; in another possible implementation, the second parallel decoder 620 outputs the second LLR sequence

Output to the serial decoder 630, and the serial decoder 630 according to the second output LLR sequence

Determine the third input LLR sequence

(t).

When t=1, the first iteration, L(1)=1: the second parallel decoder 620 uses the initial input LLR sequence of the cascaded decoder as the second input LLR sequence

I.e. 1 second input LLR sequence

When the cascaded decoder is used for channel decoding, the initial input LLR sequence can be the LLR sequence obtained by demodulation and other processing after the receiving end device receives the signal

And information sequence

correspond.

The second parallel decoder 620 responds to the second input LLR sequence

Perform the first stage update to get the second output LLR sequence

When t>1, the tth iteration:

The second parallel decoder 620 obtains the M(t-1) third output LLR sequences of the previous iteration, that is, the t-1th iteration, from the serial decoder 630, where each third output LLR sequence Including N _S LLRs, denoted as

The second parallel decoder 620 obtains M(t-1) second output LLR sequences corresponding to the M(t-1) third output LLR sequences in the t-1th iteration

Since in the t-1th iteration, the second parallel decoder generates L(t-1) second output LLR sequences

The serial decoder 630 returns M(t-1) third output LLR sequences through path selection

The second parallel decoder 620 determines the sequence of LLRs with these third outputs

The second output LLR sequence corresponding to the parent path, and the corresponding M(t-1) second output LLR sequences are obtained

The second parallel decoder 620 converts M(t-1) third output LLR sequences

Replace the corresponding second output LLR sequence respectively

In the corresponding sequence number of LLR, cLLR _(t-2)·Ns+1 , cLLR _(t-2)·Ns+2 ,..., cLLR _(t-1)·Ns , get M(t-1) sequences

The second parallel decoder 620 pairs M(t-1) sequences

Perform the second stage update respectively to obtain M(t-1) second update output sequences, and the second parallel decoder 620 regards the M(t-1) second update output sequences as L(t) second Enter the LLR sequence

L(t)=M(t-1).

In another possible implementation manner, in order to achieve common mode with the LDPC decoder, the second parallel decoder 620 can also be used to determine the corresponding L(t) for the L(t) second input LLR sequences. LDPC check matrices; the second parallel decoder 620 decodes L(t) second input LLR sequences respectively based on the L(t) LDPC check matrices to obtain L(t) second output sequences .

The second parallel decoder 620 responds to the second input LLR sequence

Step 720: The serial decoder 630 decodes the L(t) third input LLR sequences to obtain M(t) decoding paths. Wherein, each of the third input LLR sequence

Include the corresponding second output LLR sequence respectively

Ns LLRs.

The maximum number of reserved decoding paths of the serial decoder 630 is M, and M is an integer greater than zero.

The serial decoder 630 serially decodes the L(t) third input LLR sequences to obtain M(t) decoding paths. The maximum reserved decoding path of the serial decoder 630 is M paths. Wherein, the M(t) decoding paths are the L(t)*2 ^k decoding paths generated by the serial decoder 630 and the M(t) decoding paths with the largest path metric value, or the The M(t) decoding paths are L(t)*2 generated by the serial decoder 630. Among the ^k decoding paths, the M(t) decoding paths have the largest path metric value and pass the CRC check, and k is Positive integer.

Among them, M(t) is the minimum value of M and L(t)*2 ^k. For example, M=8, L(t)*2 ^k =4, then M(t) is 4, another example: M=8, L(t)*2 ^k =16, M(t)=8.

Step 730: The serial decoder 630 determines to continue the next iteration process and executes step 740, or the serial decoder 630 determines to terminate the iteration process and executes step 750.

If the current number of iterations does not reach the maximum number of iterations and the iteration is not terminated early, for the cascaded decoder 800, t<T and the tth iteration does not meet the early termination condition, the serial decoder 630 executes step 740 and continues to the tth +1 iteration processing.

If the current number of iterations reaches the maximum number of iterations, or the current iteration t has met the termination condition, for the cascaded decoder 800, t=T or the t-th iteration meets the early termination condition, step 750 is executed.

Step 740: The serial decoder 630 obtains M(t) third output LLR sequences according to the M(t) decoding paths.

The serial decoder 630 judges the M(t) decoding paths to obtain M(t) third output LLR sequences, and each third output LLR sequence includes Ns LLR soft values, denoted as

Step 750: The serial decoder 630 obtains the decoding result according to the M(t) decoding paths, and terminates the iteration.

The serial decoder 630 makes a hard decision on one of the M(t) decoding paths with the largest path metric value or one decoding path with the largest path metric value and a successful CRC check, and obtains the information sequence

Corresponding information bits. Information sequence

It includes multiple information bits, or one or more information bits and one or more frozen bits. After the serial decoder 630 makes a hard decision, it only needs to output one or more information bits.

In the communication system, due to the large number of information bits, a large code length is usually generated. For traditional serial decoders, such as SCL decoders, when the input block length is very large, the decoding process of the SCL decoder is serial, and it is difficult to support a higher throughput rate. With the cascaded decoder provided by the embodiment of the application, a large number of calculations are completed by a parallel decoder that can be executed in parallel, and the serial decoder only needs to perform a decoding with a smaller length, for example, N=64 N _S = 8, the parallel decoder updates the input with a code length of 64 through the first stage, and obtains 8 sub-sequences with a length of 8, which are provided to the serial decoder for decoding. The serial decoder only needs every The decoding with a length of 8 is executed once. This method can greatly increase the decoding throughput, and can use the decoding performance of the serial decoder to compensate for the decoding performance of the parallel decoder, so that the overall decoding performance and throughput are improved. 10, when the cascaded decoder and decoding method provided by the embodiments of this application decode Polar codes with a code length of N=64, and SC decoder, SCL decoder (number of decoding paths) Is 8, represented by SCL8) and the decoding performance of the BP decoder. Among them, the number of information bits is 21, the number of CRC bits is 11, and the code length is 64. In an additive white gaussian noise (AWGN) environment, each decoder has a different block error rate (block error rate, Under BLER), the decoding performance is represented by energy per symbol to noise density (Es/N0). The performance curve of the SC decoding algorithm is represented by a diamond curve, the performance curve of the SCL8 decoding algorithm is represented by a square curve, and the performance curve of the BP decoding algorithm is represented by an X-shaped curve. The performance curve is represented by a circular curve. It can be seen from Figure 10 that the path data of the serial decoder in the cascaded decoder is also 8, and its decoding performance is basically the same as that of SCL8, which is better than SC and BP/MS by more than 1dB. However, the throughput rate of the cascade decoder is higher than that of SCL8 due to the parallel processing of the parallel decoder.

Further, in addition to supporting Polar codes, communication systems that also need to support other codes (such as LDPC codes, Turbo codes, etc.), for example, 5G communication systems support both Polar codes and LDPC codes, using DNN, BP, MS and other parallel translation The decoder of the code algorithm is a general architecture, which can support Polar code decoding and LDPC code decoding. The cascaded decoder of the embodiment of this application can share the parallel decoding operation of LDPC code. Unit, which can save hardware implementation overhead and avoid waste.

As shown in FIG. 8, the first parallel decoder 810 (Np, N), the second parallel decoder 620 (N, Ns) and the serial decoder 630 are cascaded according to another embodiment of this application. The first-level decoder is a parallel decoder 810, the second-level decoder is a parallel decoder 620, and the third-level decoder 630 is a serial decoder. It can also be regarded as the parallel decoder 810 and the cascaded decoder 600 shown in FIG. 6 are cascaded, the parallel decoder 810 performs the first stage iterative decoding, and the cascade decoder 600 performs the second stage iterative decoding. code. Among them, the first parallel decoder 810 may use a BP decoding algorithm, an MS decoding algorithm or a DNN decoding algorithm.

The first parallel decoder 810, N _in =N _p , N _out =N, its input sequence length is N _p , which is a positive integer greater than N, which is provided to the next stage decoder, the cascaded decoder 600 The length of the input LLR sequence of, is N, or it can be said that the length of the input LLR sequence provided to the second parallel decoder 620 is N. The decoding process of the first parallel decoder 810 for the input sequence is similar to the decoding process of the second parallel decoder 620. You can refer to the description of the foregoing embodiment. The difference is that the second parallel decoder 620 serves as the next stage. The decoder also needs to return to its upper level decoder, the first parallel decoder 810, to decode the LLR sequence.

For the first parallel decoder 810, the maximum number of iterations I=N _p /N, that is, the maximum number of iterations of the first-level iteration is I, and the input of the first parallel decoder 810 is K(i) The first input LLR sequence, the length is N _p , expressed as

For the cascaded decoder 600, the second-level iteration is performed, and the initial input LLR sequence is an LLR sequence of length N obtained based on the first output LLR sequence of the previous-level decoder 810, and the maximum number of iterations is T Times, T=N/Ns, the maximum number of iterations of the second level iteration is T.

The maximum number of iterations for the multi-cascade decoder to complete the decoding of the input LLR sequence is T·I=N _p /Ns times.

The cascaded decoder 600 includes a second parallel decoder 620 and a serial decoder 630 cascaded, for each second input LLR sequence

Decode to get the second output LLR sequence

The process can refer to the method steps described in Figure 7. The difference is that the second parallel decoder 620 also needs to return the decoded LLR sequence to the first parallel decoder, and the serial decoder 630 needs to iterate according to the first stage, the first The output is different whether the level 2 iteration is terminated or not.

The decoding process of the multi-cascade decoder is described below in conjunction with Fig. 8 and Fig. 9, where the i-th iteration includes:

Step 910: The first parallel decoder 810 performs the first stage decoding on the K(i) first input LLR sequences to obtain K(i) first output LLR sequences.

Among them, the length of each first input LLR sequence is N _p , expressed as

The length of each first output LLR sequence is N _p , expressed as

i is the number of iterations of the first level, and t is the number of iterations of the second level.

For each first input LLR sequence

The first parallel decoder 810 obtains the corresponding first output LLR sequence each iteration

Used to provide the second input LLR sequence of the cascaded decoder 600

Among them, the second input LLR sequence

Include the corresponding first output LLR sequence

There are N LLRs, for example: eLLR _(i-1)·N+1 , eLLR _(i-1)·N+2 ,..., eLLR _i·N . The second input LLR sequence can also be expressed as

In order to simplify the description, it is consistent with the expression in Figure 7, and the second input LLR sequence is expressed as

Understandable, that is

The first parallel decoder 810 responds to the first input LLR sequence

For the decoding process, refer to the description of the parallel decoder decoding process in the foregoing embodiment, which is not repeated here.

In a possible implementation manner, the first parallel decoder 810 according to the first output LLR sequence

Determine the second input LLR sequence

And output to the next-level decoder: the cascade decoder 600 or the second parallel decoder 620; in another possible implementation manner, the first parallel decoder 810 outputs the first LLR sequence

Output to the next stage decoder: cascade decoder 600, or second parallel decoder 620, the next stage decoder according to the first output LLR sequence

Determine the second input LLR sequence

When i=1, the first iteration: K(i)=1, the first parallel decoder 810 uses the initial input LLR sequence of the multi-cascade decoder as the first input LLR sequence

I.e. 1 first input LLR sequence

When a multi-cascade decoder is used for channel decoding, the initial input LLR sequence can be the LLR sequence obtained by demodulation and other processing after the receiving end device receives the signal

And information sequence

correspond.

The first parallel decoder 810 responds to the first input LLR sequence

Perform the first stage update to get the first output LLR sequence

When i>1, the i-th iteration:

The first parallel decoder 810 obtains the previous iteration, that is, the M(i-1,t) of the i-1th iteration from the next-stage decoder, as shown in FIG. 8 The second decoded LLR sequence

Each second decoding LLR sequence includes N LLRs, and the corresponding first output LLR sequence of the i-1th iteration

The (i-2)·N+1th LLR to the (i-1)·Nth LLR correspond.

The first parallel decoder 810 obtains M(i-1,t) first output LLR sequences corresponding to the M(i-1,t) second decoded LLR sequences in the i-1th iteration

Since in the i-1th iteration, the first parallel decoder 810 generates K(i-1) first output LLR sequences

After t iterations, the cascaded decoder 600 obtains M(i-1, t) second decoded LLR sequences

The parent path where each second decoded LLR sequence is located corresponds to a first output LLR sequence.

The first parallel decoder 810 decodes M(i-1, t) second decoded LLR sequences

Replace the corresponding first output LLR sequence respectively

In the corresponding sequence number LLR, eLLR _(i-2)·N+1 , eLLR _(i-2)·N+2 ,..., eLLR _(i-1)·N get the sequence

The first parallel decoder 810 pairs M(i-1, t)

Perform the second stage update respectively to obtain M(i-1, t) second update output sequences, and the first parallel decoder 810 regards the M(i-1, t) second update output sequences as K(i-1, t) second update output sequences. ) First input LLR sequence

The first parallel decoder 810 responds to the first input LLR sequence

Step 920: The cascade decoder 600 performs the second stage iterative decoding on the K(i) second input LLR sequences.

Wherein, the iterative process of the cascade decoder 600 decoding the K(i) second input LLR sequences is the second-level iteration, and the maximum number of iterations is T times.

For the process of the tth iteration, refer to step 710 to step 750. For the multi-cascade decoder 800, the maximum number of iterations is the product of the number of iterations of the two-stage decoder, T·I. For step 730, the current iteration When the number of iterations reaches the maximum number of iterations, i=I and t=T. That is to say, i=I and t=T or the current iteration t has met the termination condition, step 750 is executed; if the current iteration number does not reach the above-mentioned maximum iteration number and the iteration is not terminated early, step 740 is executed to continue the next iteration process.

Among them, the tth iteration includes the following steps:

Step 9201: The second parallel decoder 620 decodes the L(i,t) second input LLR sequences to obtain L(i,t) second output LLR sequences.

See step 710 in the foregoing embodiment, which is not repeated here.

Among them, the length of each second input LLR sequence is N, expressed as

The length of each second output LLR sequence is N, expressed as

When t=1, L(i,t)=K(i), the second input LLR sequence is

The second parallel decoder 620 responds to each second input LLR sequence

Perform the first stage update separately to get the second output LLR sequence

Used to provide the third input LLR sequence of the serial decoder 630

When t>1, the tth iteration:

The second parallel decoder 620 obtains the M(i, t-1) third output LLR sequences of the previous iteration, that is, the t-1th iteration, from the serial decoder 630, where each third output The LLR sequence includes N _S LLRs, denoted as

The second parallel decoder 620 obtains M(i, t-1) second output LLR sequences corresponding to the M(i, t-1) third output LLR sequences in the t-1th iteration

Since in the t-1th iteration, the second parallel decoder generates L(i, t-1) second output LLR sequences

The serial decoder 630 returns M(i, t-1) third output LLR sequences through path selection

The second output LLR sequence corresponding to the parent path, and the corresponding M(i, t-1) second output LLR sequences are obtained

The second parallel decoder 620 converts M(i, t-1) third output LLR sequences

Replace the corresponding second output LLR sequence respectively

In the corresponding sequence number of LLR, cLLR _(t-2)·Ns+1 , cLLR _(t-2)·Ns+2 ,..., cLLR _(t-1)·Ns , get M(i,t-1) sequence

The second parallel decoder 620 pairs M(i, t-1) sequences

Perform the second-stage update respectively to obtain M(i,t-1) second update output sequences, and the second parallel decoder 620 regards the M(i,t-1) second update output sequences as L(i , T) second input LLR sequence

L(i,t)=M(i,t-1).

The second parallel decoder 620 responds to the second input LLR sequence

Step 9202: The serial decoder 630 decodes the L(i,t) third input LLR sequences to obtain M(i,t) decoding paths.

See step 720 in the foregoing embodiment.

Step 9203: The serial decoder 630 determines to continue the next iterative process and executes step 9204, or the serial decoder 630 determines to terminate the second-level iterative process but does not terminate the first-level iterative process, and executes steps 9204 to 9205, or The serial decoder 630 determines to terminate the first-stage iterative process, and executes step 9206.

If the current number of iterations t does not reach the maximum number of iterations of the cascaded decoder 600 and the second-level iteration does not terminate early, for the cascaded decoder 800, t<T and the iteration does not terminate early, the serial decoder 630 executes Step 9204, continue the t+1th level 2 iterative processing.

If the current iteration number t reaches the maximum iteration number of the cascaded decoder 600, t=T, or the second-level iteration meets the early termination condition, and the first-level iteration does not terminate early,

steps

9204 and 9205 are executed.

If the number of iterations t of the second stage reaches the maximum number of iterations of the cascade decoder 800, t=T, and the number of iterations i of the first stage reaches the maximum number of iterations of the multi-cascade decoder, i=I, or the first stage The iteration meets the early termination condition, and step 9206 is executed.

Step 9204: The serial decoder 630 obtains M(i,t) third output LLR sequences according to the M(i,t) decoding paths.

The serial decoder 630 determines the M(i,t) decoding paths to obtain M(i,t) third output LLR sequences, and each third output LLR sequence includes Ns LLR soft values, denoted as

Step 9205: The second parallel decoder 620 obtains the second decoded LLR sequence according to the M(i,t) third output LLR sequences, and terminates the second stage iteration.

When the second parallel decoder 620 determines that t=T or the t-th iteration meets the early termination condition:

The second parallel decoder 620 obtains the LLR sequence with the M(i,t) third output

The corresponding M(i,t) second output LLR sequence of the tth iteration

The second parallel decoder 620 according to the M(i,t) third output LLR sequences

And the M(i,t) second output LLR sequence of the tth iteration

Get the second updated input LLR sequence of M(i,t) iteration t

The second parallel decoder 620 updates the M(i,t) second update input LLR sequences

Perform the second stage update to get M(i,t) second decoded LLR sequences

The second parallel decoder 620 outputs the M(i,t) second decoded LLR sequences to the first parallel decoder 810

And terminate the second iteration.

Here, M(i,t) second decoded LLR sequences

It can also be regarded as the output of the cascaded decoder 600 to the upper-level decoder, the first parallel decoder 810.

Step 9206: The serial decoder 630 obtains the decoding result according to the M(i,t) decoding paths, and terminates the multi-cascade decoding iteration.

The serial decoder 630 makes a hard decision on M(i,t) decoding paths to obtain the information sequence

Corresponding information bits. Information sequence

It includes multiple information bits, or one or more information bits and one or more frozen bits, and the serial decoder 630 only needs to output the information bits after making a hard decision.

Since LDPC decoders usually use parallel decoding algorithms for decoding, such as BP, MS, DNN, etc., the multi-cascade decoder provided by the embodiments of the present invention can share part of the parallel decoding units of the LDPC decoder, saving system overhead . For example, the second parallel decoder 620, in this case, can convert the factor graph of the second input LLR sequence, determine the corresponding LDPC check matrix, and perform LDPC decoding through a small block-length serial The decoder performs path selection. This method not only improves the throughput of decoding, improves the decoding performance of parallel decoders, but also provides a basis for common mode with decoders of other codes such as LDPC and saves system overhead.

It should be understood that the cascaded decoding method provided in the embodiments of the present application may be executed by a decoding device or a chip in a decoding device in various network equipment or terminal equipment.

An embodiment of the present application also provides a decoding device. The decoding device may adopt the structure of FIG. 6 or FIG. 8 to execute the decoding method shown in FIG. 7 or FIG. 9. Some or all of these decoding methods can be implemented by hardware or software. The decoding device can include: an input interface circuit for obtaining the LLR sequence corresponding to the information sequence; and a logic circuit for implementing FIG. 7 Or the decoding method shown in Figure 9; output interface circuit for outputting information bits.

Optionally, the decoding device may be a chip or an integrated circuit in specific implementation.

An embodiment of the present application also provides a decoding device. The decoding device may adopt the structure of FIG. 6 or FIG. 8 to execute the decoding method shown in FIG. 7 or FIG. 9. Some or all of these decoding methods can be implemented by hardware or software. The decoding device can include: a memory for storing a program; a processor for executing a program stored in the memory. When the program is executed At this time, the decoding device can realize the decoding method shown in FIG. 7 or 9.

Optionally, the decoding device may also only include a processor. The memory for storing the program is located outside the decoding device, and the processor is connected to the memory through a circuit/wire for reading and executing the program stored in the memory.

The processor may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP.

The processor may further include a hardware chip. The aforementioned hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The above-mentioned PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.

The memory may include volatile memory (volatile memory), such as random-access memory (RAM); the memory may also include non-volatile memory (non-volatile memory), such as flash memory (flash memory) , Hard disk drive (HDD) or solid-state drive (solid-state drive, SSD); the memory may also include a combination of the foregoing types of memory.

The embodiment of the present application also provides a computer storage medium storing a computer program, and the computer program includes a decoding method for executing the decoding method provided in the foregoing method embodiment.

The embodiments of the present application also provide a computer program product containing instructions, which when run on a computer, cause the computer to execute the decoding method provided by the foregoing method embodiments.

Any decoding device provided in the embodiments of the present application may also be a chip.

Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, this application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

This application is described with reference to flowcharts and/or block diagrams of methods, devices (systems), and computer program products according to embodiments of this application. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing equipment are generated It is a device that realizes the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device. The device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to produce computer-implemented processing, so as to execute on the computer or other programmable equipment. The instructions provide steps for implementing the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. In this way, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

A cascaded decoder, the cascaded decoder includes a second parallel decoder and a serial decoder, the input LLR sequence is iteratively decoded a maximum of T times, and the second parallel decoder is used for Decode one or more LLR sequences of length N. The serial decoder is used to decode one or more LLR sequences of length N S , where N S <N, where the t th Sub-iteration decoding, including:

The second parallel decoder decodes L(t) second input LLR sequences to obtain L(t) second output LLR sequences, each second input LLR sequence has a length of N, and each second input LLR sequence has a length of N. The length of the output LLR sequence is N;

The serial decoder decodes the L(t) third input LLR sequences to obtain M(t) decoding paths, and each of the third input LLR sequences includes the corresponding second output LLR sequence. Ns LLRs;

The serial decoder determines to continue the iteration, or the serial decoder determines to terminate the iteration.
The cascaded decoder according to claim 1, wherein each of the third input LLR sequences includes (t-1)×N S +1th LLR to t×N S in the corresponding second output LLR sequence LLR.
The cascaded decoder according to claim 1 or 2, wherein the second parallel decoder decodes L(t) second input LLR sequences to obtain L(t) second output LLR sequences, including :

The second parallel decoder performs a first stage update on the L(t) second input LLR sequences respectively to obtain the L(t) second output LLR sequences.
The cascaded decoder according to any one of claims 1 to 3, t=1, L(t)=1, and the second input LLR sequence is the input LLR sequence of the cascaded decoder.
The cascaded decoder according to any one of claims 1 to 4, t>1, and the t-1th iterative decoding further comprises:

The serial decoder obtains M(t-1) third output LLR sequences according to the M(t-1) decoding paths, and each third output LLR sequence includes Ns LLRs.
The cascaded decoder according to claim 5, wherein the t-th iteration further comprises:

Acquiring, by the second parallel decoder, M(t-1) second output LLR sequences corresponding to the M(t-1) third output LLR sequences of the t-1th iteration;

The second parallel decoder obtains L(t) according to the M(t-1) third output LLR sequences and the M(t-1) second output LLR sequences of the t-1th iteration The second updated input LLR sequence of the t-th iteration, L(t)=M(t-1);

The second parallel decoder performs a second stage update on the L(t) second updated input LLR sequences to obtain the L(t) second input LLR sequences.
The cascaded decoder according to claim 6, wherein the second parallel decoder includes at least n s +1 th to n+1 th layer nn s decoding layers,
Each layer includes N LLR nodes;

The second parallel decoder performs a first stage update on each second input LLR sequence to obtain a corresponding second output LLR sequence, including:

The second parallel decoder updates the N LLR nodes of the n+1th layer to N LLRs in the second input LLR sequence,

Said second decoder performs parallel from the first layer n + 1 to the first direction of the soft layer n s +1 to obtain the updated value of n s N + 1 th layer LLR nodes, the corresponding first The second output LLR sequence includes the N LLR nodes of the n s +1 th layer;

The second parallel decoder performs a second stage update on each second update input LLR sequence to obtain the corresponding second input LLR sequence, including:

The second parallel decoder updates the N LLR nodes of the n s +1 th layer to the N LLRs in the second updated input LLR sequence,

The second parallel decoder performs a soft value update from the n s +1 th layer to the n+1 th layer to obtain the N LLR nodes of the n+1 th layer, and the corresponding second The input LLR sequence includes N LLR nodes of the n+1th layer.
The cascaded decoder according to any one of claims 1 to 7, wherein the serial decoder decodes the L(t) third input LLR sequences to obtain M(t) decoding paths ,include:

The serial decoder decodes the L(t) third input LLR sequences to obtain L(t)*2 k decoding paths, where k is a positive integer;

The M(t) decoding paths are M(t) decoding paths with the largest path metric among the L(t)*2 k decoding paths, or the M(t) decoding paths Among the L(t)*2 k decoding paths, M(t) decoding paths have the largest path metric and pass the CRC check.
The cascaded decoder according to any one of claims 1 to 6, wherein the second parallel decoder is configured to determine corresponding LDPC check matrices for the L(t) second input LLR sequences respectively;

The second parallel decoder decodes the L(t) second input LLR sequences based on the LDPC check matrix to obtain L(t) second output sequences.
The cascaded decoder according to any one of claims 1 to 9, wherein the second parallel decoder includes one or more of the following: a BP decoder or an MS decoder or a DNN decoder, so The serial decoder includes an SCL decoder or a CA-SCL decoder.
The cascaded decoder according to any one of claims 1 to 10, N is any one of the following: 1024, 512, 256, 128, 64, 32; Ns is any one of the following: 128, 64, 32, 16,8,4,2,1.
The cascaded decoder according to any one of claims 1 to 11, wherein the serial decoder determines to continue iteration, comprising:

The serial decoder determines that t<T and the t-th iteration does not meet the early termination condition;

The t-th iteration further includes:

The serial decoder obtains M(t) third output LLR sequences according to the M(t) decoding paths.
The cascaded decoder according to any one of claims 1 to 12, wherein the serial decoder determines to terminate the iteration, comprising:

The serial decoder determines that t=T, or the t-th iteration satisfies an early termination condition;

The t-th iteration further includes:

The serial decoder obtains the decoding result according to the M(t) decoding paths, and terminates iteration.
The cascaded decoder according to claim 13, wherein the input LLR sequence of the cascaded decoder is an information sequence
A corresponding LLR sequence, where the information sequence includes multiple information bits, or one or more information bits and one or more frozen bits;

The serial decoder obtains a decoding result according to the M(t) decoding paths, including:

The serial decoder performs a hard decision on the M(t) decoding paths with the largest path metric value or one decoding path with the largest path metric value and a successful CRC check, to obtain the information sequence Each information bit.
A multi-cascade decoder, comprising the cascade decoder according to any one of claims 1 to 11 and a first parallel decoder, wherein the cascade decoder is the first parallel decoder The first parallel decoder is used to decode one or more LLR sequences of length N p , and the first parallel decoder performs a maximum of 1 iterations on the initial LLR sequence, I=N p /N, the i-th iteration, i<I, including:

The first parallel decoder performs the first stage iterative decoding on K(i) first input LLR sequences to obtain K(i) first output LLR sequences, and the length of each first input LLR sequence is N p , The length of each first output LLR sequence is N p ;

The cascaded decoder performs a second stage iterative decoding on K(i) second input LLR sequences, and each of the second input LLR sequences includes N LLRs in the corresponding first output LLR sequence.
The multi-cascade decoder according to claim 15, wherein each of the second input LLR sequences includes the (i-1)×N+1th LLR to the i×th LLR in the corresponding first output LLR sequence N LLRs.
The multi-cascade decoder according to claim 16, wherein the first parallel decoder decodes the K(i) first input LLR sequences to obtain K(i) first output LLR sequences, include:

The first parallel decoder performs a first stage update on the K(i) first input LLR sequences respectively to obtain the K(i) first output LLR sequences.
The multi-cascade decoder according to claim 16 or 17, i=1, K(i)=1, and the first input LLR sequence is an initial LLR sequence.
The multi-cascade decoder according to any one of claims 16 to 18, i>1, and the i-1th iterative decoding further comprises:

The cascaded decoder outputs M(i-1,t) second decoded LLR sequences to the first parallel decoder, and the second decoded LLR sequence includes N LLRs.
The multi-cascade decoder according to claim 19, wherein the i-th iteration further comprises:

The first parallel decoder obtains M(i-1,t) corresponding to the i-1th iteration of M(i-1,t) second decoding LLR sequences of the i-1th iteration ) First output LLR sequence;

The first parallel decoder according to the M(i-1,t) second decoding LLR sequences and the M(i-1,t) first output LLR sequences of the i-1th iteration Obtain K(i) the second updated input LLR sequence of the i-th iteration, K(i)=M(i-1,t);

The first parallel decoder performs a second stage update on the K(i) second updated input LLR sequences to obtain the L(i, 1) first input LLR sequences.
The multi-cascade decoder according to claim 20, wherein the first parallel decoder includes at least n+1th to np +1th layers n p -n decoding layers,
Each layer includes N p LLR nodes;

The first parallel decoder performs a first stage update on each first input LLR sequence to obtain the corresponding first output LLR sequence, including:

Parallel to said first coder p +1 said first layer of N p n of nodes updated LLR LLR input to the first sequence of N p LLR,

The first parallel decoder performs a soft value update from the np +1th layer to the n+1th layer to obtain the Np LLR nodes of the n+1th layer, and the corresponding An output LLR sequence includes N p LLR nodes of the n+1th layer;

The first parallel decoder performs a second stage update on each second update input LLR sequence to obtain the corresponding first input LLR sequence, including:

The first parallel decoder LLR the nodes of N p n + 1, the second layer updates the LLR updated input sequence of N p LLR,

The first parallel decoder performs a soft value update from the n+1th layer to the np +1th layer to obtain the Np LLR nodes of the np +1th layer, and the corresponding The first input LLR sequence includes N p LLR nodes of the n p +1 th layer.
The multi-cascade decoder according to any one of claims 15 to 21, wherein the serial decoder determines to continue iteration, comprising:

The serial decoder determines that t<T and the tth iteration does not meet the early termination condition, or i<I and the i-th previous iteration does not meet the early termination condition of the multi-cascade decoder;

The t-th iteration of the cascaded decoder includes:

The serial decoder obtains M(i,t) third output LLR sequences according to the M(i,t) decoding paths.
The multi-cascade decoder according to claim 22, the second parallel decoder is further configured to determine t=T, or the t-th iteration satisfies an early termination condition,

Acquiring, by the second parallel decoder, M(i,t) second output LLR sequences corresponding to the M(i,t) third output LLR sequences of the t-th iteration;

The second parallel decoder obtains M(i,t) according to the M(i,t) third output LLR sequences and the M(i,t) second output LLR sequences of the tth iteration The second updated input LLR sequence of the t-th iteration;

The second parallel decoder performs a second stage update on the M(i,t) second update input LLR sequences to obtain M(i,t) second decoded LLR sequences;

The second parallel decoder outputs the M(i,t) second decoded LLR sequences to the first parallel decoder.
The multi-concatenated decoder according to any one of claims 16 to 22, wherein the serial decoder determines to terminate the iteration, comprising:

The serial decoder determines that i=I and t=T, or the tth iteration satisfies the early termination condition of the cascade decoder and the i-th previous iteration satisfies the multi-cascade translation Conditions for early termination of the encoder;

The t-th iteration of the cascaded decoder includes:

The serial decoder obtains the decoding result according to the M(i,t) decoding paths, and terminates the iteration.
The multi-cascade decoder according to claim 24, wherein the initial LLR sequence is an information sequence
A corresponding LLR sequence, where the information sequence includes multiple information bits, or one or more information bits and one or more frozen bits;

The serial decoder obtains a decoding result according to the M(t) decoding paths, including:

The serial decoder performs a hard decision on the M(t) decoding paths with the largest path metric value or one decoding path with the largest path metric value and a successful CRC check, to obtain the information sequence Each information bit.
The multi-cascade decoder according to any one of claims 16 to 25, the first parallel decoder includes one or more of the following: a BP decoder, an MS decoder, or a DNN decoder.
According to the multi-cascade decoder according to any one of claims 16 to 26, the value of N p includes any one of the following: 8192, 4096, 2048, 1024, 512, 256, 128.