Decoder optimization method and device, electronic equipment, storage medium and product

Abstract

The invention provides a decoder optimization method and device, electronic equipment, a storage medium and a product, and relates to the technical field of data decoding. A target random sequence interval is determined according to a decoding evaluation model; and obtaining an optimal polarization code based on the target random sequence interval and the fixed sequence interval, and obtaining a decoder according to the sorting starting condition of the decoding nodes corresponding to the optimal polarization code and the distribution of the decoding nodes. Through the decoding evaluation model, the efficiency of obtaining the target random sequence interval is improved, and the optimization efficiency of the decoder is improved. The optimal polarization code is obtained through the target random sequence interval and the fixed sequence interval to obtain the distribution of the decoding nodes, so that the optimization of the distribution of the decoding nodes in the decoder is realized, and the decoding time delay of the decoder is reduced. And the sorting starting condition of the decoding nodes is optimized through the sorting entropy of the sorting operation of the decoding nodes, so that the decoding time delay of the decoder is further reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of data decoding, and particularly to an optimization method, device, electronic device, storage medium and product for a decoder. Background Art

[0002] As the standard code for 5G control channels, polar codes have attracted much attention due to their achievable channel capacity. The successive cancellation list (SCL) decoder of polar codes has attracted in-depth research in academia and industry due to its excellent error correction performance. However, this excellent performance advantage comes at the cost of latency, so the fast SCL (Fast-SCL) decoder based on special nodes has received extensive attention. The latency of this decoder mainly consists of two parts: node calculation and path sorting. The former depends on the processing latency of each special node, and the latter is related to the design and enabling of the sorting module. In addition, with the diversification of the expected scenarios of 6G, there are also differences in the requirements for key performance indicators (KPIs) such as the latency and frame error rate of the decoder in baseband signal processing, which requires the proposed Fast-SCL decoder that can flexibly respond to KPI requirements and is configurable.

[0003] Facing the requirements of future diverse mobile communication application scenarios, there is still redundancy in the decoding latency. Therefore, there is still room to reduce the decoding latency while not significantly sacrificing the decoding error correction performance.

[0004] The existing methods for reducing the decoding latency of decoders have the defect of being unable to flexibly respond to the error correction performance requirements of different scenarios, resulting in high decoding latency of existing decoders. Summary of the Invention

[0005] The present invention provides an optimization method, device, electronic device, storage medium and product for a decoder, so as to solve the defect of high decoding latency of the decoder in the prior art and achieve the reduction of the decoding latency of the decoder.

[0006] The present invention provides an optimization method for a decoder, including: obtaining a fixed sequence interval of a polar code and a plurality of random sequence intervals of the polar code based on a set code length and a set information bit length, where the fixed sequence interval is an interval in the polar code where the distribution of information bits and frozen bits is fixed, and the random sequence interval is an interval in the polar code where the distribution of information bits and frozen bits has a random change; inputting the plurality of random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; wherein, the decoding evaluation model determines the target random sequence interval with the optimal decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtaining an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtaining the distribution of decoding nodes of the optimal polar code; determining the sorting enabling situation of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtaining a decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes.

[0007] According to the optimization method for a decoder provided by the present invention, the decoding evaluation model is used to output a target random sequence interval: obtaining the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtaining the decoding negative evaluation of each random sequence interval based on the decoding delay and the corresponding decoding error rate; and taking the random sequence interval corresponding to the minimum decoding negative evaluation as the target random sequence interval.

[0008] According to the optimization method for a decoder provided by the present invention, the optimal polar code includes at least one decoding node, and determining the sorting enabling situation of the decoding nodes based on the sorting entropy of the decoding nodes includes: obtaining a sorting operation sequence corresponding to the optimal polar code according to the distribution of the decoding nodes; performing multiple sorting simulations on each sorting operation in the sorting operation sequence to obtain a plurality of sorting entropies of each sorting operation, and determining the average sorting entropy of each sorting operation based on the plurality of sorting entropies; re - sorting all the sorting operations in the sorting operation sequence from small to large according to the average sorting entropy to obtain a re - sorted sorting operation sequence; in the re - sorted sorting operation sequence, not enabling the first set number of sorting operations in the front, and enabling the remaining sorting operations to obtain the sorting enabling situation of all decoding nodes.

[0009] According to the optimization method for a decoder provided by the present invention, after obtaining a decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes, it further includes: when decoding a target polar code based on the decoder, determining the number of sorting operations of each target node corresponding to the target polar code based on the sorting enabling situation of the decoding nodes; the target nodes include SPC nodes, and the target nodes correspond to the decoding nodes one by one; determining the decoding delay of each target node based on the decoding parameters of the decoder, the number of information bits of the target node, and the number of sorting operations; and determining the decoding delay of the target polar code based on the decoding delay of the target nodes and the number of SPC nodes.

[0010] According to the decoder optimization method provided by the present invention, a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code are obtained based on a set code length and a set information bit length, including: when the set code length is N and the information bit length is K, sorting N channels in descending order of channel reliability to obtain the sorted channels; selecting the first K channels from the sorted channels as information bit channels, and using the remaining N - K channels as frozen bit channels to obtain an initial polar code; in the initial polar code, randomly shuffling the information bits and frozen bits of the channels in the sequence interval [K - f + 1, K + f], where f is the set number of channels, to obtain multiple random sequence intervals; in the initial polar code, taking the sequence intervals outside the [K - f + 1, K + f] sequence interval as fixed sequence intervals.

[0011] According to the decoder optimization method provided by the present invention, the decoding evaluation model is determined based on the following steps: obtaining multiple sample random sequence intervals according to the sample code length and the sample information bit length; marking the sample random sequence intervals according to the decoding delay and the decoding error rate of the sample random sequence intervals to obtain training samples with labels; constructing a preset neural network based on the twin deep neural network of the self-attention mechanism; training the preset neural network based on the training samples until the errors of the output decoding error rate and the output decoding delay of the preset neural network both meet the set error to obtain a trained neural network; constructing an objective function based on the output decoding error rate of the trained neural network and the output decoding delay of the trained neural network to obtain the decoding evaluation model.

[0012] The present invention also provides an optimization device for a decoder, including: a construction module, configured to obtain a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length, where the fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code has undergone random changes; a selection module, configured to input the multiple random sequence intervals into the decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; wherein, the decoding evaluation model determines the target random sequence interval with the best decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; a first optimization module, configured to obtain an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtain the distribution of decoding nodes of the optimal polar code; a second optimization module, configured to determine the sorting enabling situation of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtain a decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the optimization method of any one of the above decoders is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the optimization method of any one of the above decoders is implemented.

[0015] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the optimization method of any one of the above decoders is implemented.

[0016] The optimization method, device, electronic device, storage medium, and product of the decoder provided by the present invention obtain a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length. The fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code changes randomly; input the multiple random sequence intervals into a decoding evaluation model, and obtain a target random sequence interval output by the decoding evaluation model; wherein, the decoding evaluation model determines the target random sequence interval with the optimal decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtain an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtain the distribution of decoding nodes of the optimal polar code; determine the enabling situation of the sorting of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtain a decoder based on the enabling situation of the sorting of the decoding nodes and the distribution of the decoding nodes. The efficiency of obtaining the target random sequence interval is improved through the decoding evaluation model, which is beneficial to improving the optimization efficiency of the decoder. The optimal polar code is obtained through the target random sequence interval and the fixed sequence interval to obtain the distribution of decoding nodes, realizing the optimization of the distribution of decoding nodes in the decoder, which is beneficial to reducing the decoding delay of the decoder. The enabling situation of the sorting of the decoding nodes is optimized through the sorting entropy of the sorting operation of the decoding nodes, which is beneficial to further reducing the decoding delay of the decoder. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 is a schematic flowchart of the optimization method of the decoder provided by the present invention.

[0019] Figure 2 It is a schematic structural diagram of an optimization device for a decoder provided by the present invention.

[0020] Figure 3 It is a schematic structural diagram of an electronic device provided by the present invention. Specific embodiments

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0022] The following combines Figures 1-3 to describe an optimization method, device, and electronic device for a decoder of the present invention.

[0023] Figure 1 It is a schematic flowchart of an optimization method for a decoder provided by the present invention. As Figure 1 shown, the method includes steps S100 to S400, and the specific steps are as follows.

[0024] S100: Obtain a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length.

[0025] The fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code undergoes random changes.

[0026] Adopt a 5G polar code construction method to construct multiple polar codes according to the set code length and the set information bit length. The polar code includes a fixed sequence interval and a random sequence interval. The fixed sequence interval and the random sequence interval are composed of multiple information bit channels and / or frozen bit channels. When the set code length and the set information bit length remain unchanged, the fixed sequence intervals of each polar code are the same (the distribution of the channels carrying information bits and the channels carrying frozen bits in the fixed sequence intervals of different polar codes is the same), and the random sequence intervals are different (the distribution of the channels carrying information bits and the channels carrying frozen bits in the random sequence intervals of different polar codes is different).

[0027] S200: Input the multiple random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model.

[0028] Among them, the decoding evaluation model determines the target random sequence interval with the best decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval.

[0029] The decoding evaluation model includes a trained neural network (including Neural Network 1 and Neural Network 2), for example, a trained dual deep neural network based on the self-attention mechanism.

[0030] Neural Network 1 outputs the decoding delay of each random sequence interval, and Neural Network 2 outputs the decoding error rate of each random sequence interval.

[0031] The decoding evaluation model evaluates the decoding situation of each random sequence interval according to the decoding delay and the decoding error rate, and takes the random sequence interval with the best decoding as the target random sequence interval.

[0032] S300: Obtain the optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtain the distribution of the decoding nodes of the optimal polar code.

[0033] The target random sequence interval and the fixed sequence interval are spliced to obtain the optimal polar code. The optimal polar code contains the optimal decoding method of the polar code for the set code length and the set information bit length. The distribution of all decoding nodes of the decoder is constructed according to the optimal polar code.

[0034] S400: Determine the enabling situation of the sorting of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtain the decoder based on the enabling situation of the sorting of the decoding nodes and the distribution of the decoding nodes.

[0035] The decoder of the present invention includes a Fast Successive Cancellation List (Fast-SCL) decoder. During the decoding process of each decoding node of the Fast-SCL decoder, a set number of candidates (for example, 2L) of candidate codewords are generated, corresponding to 2L Path Metrics (PMs) for evaluating the possibility that different candidate codewords are correct codewords. The smaller |PM| is, the greater the possibility that the candidate codeword is correct. Sorting is to select the candidate codewords corresponding to the first L smallest |PM|s from the |PM|s of the 2L candidate codewords.

[0036] During the decoding process of the decoder, some decoding nodes correspond to at least one sorting operation. According to the sorting entropy (such as the average sorting entropy) of the sorting operation of the decoding nodes, evaluate the necessity of the sorting operation, and then determine the enabling situation of each sorting operation of the decoding nodes. If the necessity of the sorting operation is low, the sorting operation is not enabled (the sorting operation is not turned on). If the necessity of the sorting operation is high, the sorting operation is enabled (the sorting operation is turned on).

[0037] Obtain the sorting enable status of the decoding nodes based on the enabling status of all sorting operations of the decoding nodes. Obtain the decoder based on the sorting enable status of all decoding nodes and the distribution of the decoding nodes.

[0038] The method for optimizing a decoder provided by an embodiment of the present invention obtains a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length. The fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code undergoes random changes. Input the multiple random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model. Among them, the decoding evaluation model determines the target random sequence interval with the best decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval. Obtain the optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtain the distribution of the decoding nodes of the optimal polar code. Determine the sorting enable status of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtain the decoder based on the sorting enable status of the decoding nodes and the distribution of the decoding nodes. The efficiency of obtaining the target random sequence interval is improved through the decoding evaluation model, which is beneficial to improving the optimization efficiency of the decoder. The optimal polar code is obtained through the target random sequence interval and the fixed sequence interval to obtain the distribution of the decoding nodes, realizing the optimization of the distribution of the decoding nodes in the decoder, which is beneficial to reducing the decoding delay of the decoder. The sorting enable status of the decoding nodes is optimized through the sorting entropy of the sorting operation of the decoding nodes, which is beneficial to further reducing the decoding delay of the decoder.

[0039] Based on the above embodiment, the decoding evaluation model is used to output a target random sequence interval: obtain the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtain the decoding negative evaluation of each random sequence interval based on the decoding delay and the corresponding decoding error rate; use the random sequence interval corresponding to the minimum decoding negative evaluation as the target random sequence interval.

[0040] The decoding evaluation model includes a trained dual deep neural network based on the self-attention mechanism (including neural network 1 and neural network 2).

[0041] The decoding evaluation model obtains the decoding delay of the random sequence interval according to one sub-neural network in the trained dual deep neural network based on the self-attention mechanism, and obtains the decoding error rate of the random sequence interval according to the other sub-neural network in the trained dual deep neural network based on the self-attention mechanism.

[0042] The logarithmic form of the decoding delay of the random sequence interval and the logarithmic form of the decoding error rate of the random sequence interval are weighted and summed according to the objective function of the decoding evaluation model to obtain the negative decoding evaluation of the random sequence interval. The calculation formula for the negative decoding evaluation is as follows.

[0043] ; Among them, is the negative decoding evaluation, is the logarithmic form of the decoding delay of the random sequence interval, is the weight, is the logarithmic form of the decoding error rate of the random sequence interval.

[0044] The negative decoding evaluation characterizes the decoding effect of each random sequence interval. If the negative decoding evaluation is larger, the decoding effect of the random sequence interval is worse. If the negative decoding evaluation is smaller, the decoding effect of the random sequence interval is better. The random sequence interval corresponding to the minimum negative decoding evaluation is used as the target random sequence interval.

[0045] The decoding evaluation model of the embodiments of the present invention determines the negative decoding evaluation according to the decoding delay and the corresponding decoding error rate, realizes the accurate evaluation of the decoding of the random sequence interval, and is beneficial to improving the accuracy and efficiency of determining the target random sequence interval.

[0046] Based on the above embodiments, the optimal polar code includes at least one decoding node. The sorting enabling situation of the decoding node is determined based on the sorting entropy of the decoding node, including: obtaining the sorting operation sequence corresponding to the optimal polar code according to the distribution of the decoding nodes; performing multiple sorting simulations on each sorting operation in the sorting operation sequence to obtain multiple sorting entropies of each sorting operation, and determining the average sorting entropy of each sorting operation based on the multiple sorting entropies; re - sorting all the sorting operations in the sorting operation sequence from small to large according to the average sorting entropy to obtain the re - sorted sorting operation sequence; in the re - sorted sorting operation sequence, the first set number of sorting operations are not enabled, and the remaining sorting operations are enabled to obtain the sorting enabling situation of all decoding nodes.

[0047] According to the distribution of the decoding nodes of the optimal polar code, the distribution of the decoding nodes of the decoder is obtained (for example, decoding node A, decoding node B, and decoding node C). Each decoding node of the decoder includes at least one sorting operation. For example, decoding node A includes the first sorting operation, the second sorting operation, and the third sorting operation. According to the distribution of the decoding nodes, the sorting operation sequence of all the decoding nodes of the decoder is obtained. For example, the sorting operation sequence is {the first sorting operation, the second sorting operation, the third sorting operation, the fourth sorting operation, the fifth sorting operation}.

[0048] Perform multiple rounds of sorting simulations on each sorting operation in the sorting operation sequence, and obtain multiple sorting entropies for each sorting operation according to the results of each round of sorting simulation.

[0049] For example, in the first round of sorting simulation, for the th sorting operation in the sorting operation sequence, the corresponding input sequence is {1.1, 2.4, 5.6, 2.2, 4.1, 6.7, 9.5, 10.8}, and according to the th sorting operation, sort each input data in the input sequence from smallest to largest. Obtain the sequence of sorted order numbers of the input data according to the input sequence as {1 (the sorted order number of input data 1.1 is 1), 3 (the sorted order number of input data 2.4 is 3), 5, 2, 4, 6, 7, 8}. Then the calculation formula for the sorting entropy of the th sorting operation in the first round of sorting simulation is as follows.

[0050] ; Among them, is the decoding parameter, is the sorted order number of the input data in the input sequence, is the th sorting entropy of the sorting operation, is the temperature introduced to define the sorting entropy, which is a fixed constant, is the th sorting energy difference of the sorting operation.

[0051] Perform multiple rounds of sorting simulations on the sorting operation sequence to obtain multiple sorting entropies for each sorting operation. Calculate the average sorting entropy for each sorting operation. The smaller is, the smaller the loss caused by not enabling the sorting of the

[0052] th sorting operation.

[0053] Determine the set number according to the dichotomy method. In the sorting operation sequence, do not enable the sorting operations ranked in the front, and enable the sorting operations ranked after the set number to obtain the sorting enabling situation of all decoding nodes.

[0054] Set number It is determined based on the following steps: For each sorting operation, the number of input data (l) in the input sequence of the decoder is fixed. For example, l = 8. According to the dichotomy method, a set number , that is, the sorting operations of the first 4 input data in the input sequence are not enabled.

[0055] Do not enable the sorting operations of the set number of preceding ones, that is, turn off the sorting operations of the set number of preceding ones and do not perform the sorting corresponding to this sorting operation. Enable the sorting operations after the set number of sorting operations, that is, turn on the sorting operations after the set number and perform the sorting corresponding to this sorting operation. For example, the sorting operation sequence corresponding to the decoding node A is {the first sorting operation, the second sorting operation, the third sorting operation, the fourth sorting operation, the fifth sorting operation}, and the set number is 4, then the first sorting operation, the second sorting operation, the third sorting operation, and the fourth sorting operation are not enabled. Only the fifth sorting operation is enabled.

[0056] The present invention determines the average sorting entropy of the sorting operation through the sorting simulation results of each sorting operation, realizing the accurate calculation of the average sorting entropy. Reorder the sorting operation sequence according to the average sorting entropy, and then determine the sorting operations that are not enabled for execution, realizing that when the decoder deactivates the set number of least important sorting operations, the error correction performance of the decoder will not be significantly reduced.

[0057] Based on the above embodiments, after obtaining the decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes, it further includes: when decoding the target polar code based on the decoder, determining the number of sorting operations of each target node corresponding to the target polar code based on the sorting enabling situation of the decoding nodes; the target nodes include SPC nodes, and the target nodes correspond to the decoding nodes one by one; determining the decoding delay of each target node based on the decoding parameters of the decoder, the number of information bits of the target node, and the number of sorting operations; determining the decoding delay of the target polar code based on the decoding delay of the target node and the number of SPC nodes.

[0058] When decoding the target polar code according to the decoder, calculate the decoding delay of the decoder. The target nodes include Rate-0 (code rate - 0) nodes, Rate-1 (code rate - 1) nodes, Rep (Repetition) nodes, and SPC (Single Parity Check) nodes.

[0059] Specifically, for a target node containing 2 p bits, the structures corresponding to each bit are as follows. For each target node, without considering the sorting operation, the decoding delays of each target node are as follows.

[0060] All bits in the Rate-0 node are frozen bits, and decoding requires 1 time step (decoding delay).

[0061] All bits in the Rate-1 node are information bits, and decoding requires time steps, where is the decoding parameter of the decoder, and is the number of information bits of the target node.

[0062] In the Rep node, except for the last 1 bit which is an information bit, the remaining 2 p - 1 bits are all frozen bits, and decoding requires 2 time steps.

[0063] In the SPC node, except for the first bit which is a frozen bit, the remaining 2 p - 1 bits are all information bits, and decoding requires time steps.

[0064] For example, the target polar code is {0, 1, 0, 0, 0, 0, 1, 1}, and this target polar code is composed of a Rep node with a length of 2, a Rate-0 node with a length of 2, a Rate-0 node with a length of 2, and a Rate-1 node with a length of 2. Therefore, given the length of the polar code and the length of the information bits, by reasonably selecting the construction of the polar code, the distribution of the target nodes in the codeword can be changed. Different target nodes consume different delays in the decoder. The target nodes correspond one-to-one with the decoding nodes, and the decoder decodes the corresponding target nodes through the decoding nodes.

[0065] During the decoding process, some target nodes involve multiple sorting operations. For example, the Rate-0 node has no sorting operation. The Rep node requires 1 sorting operation. The Rate-1 node requires sorting operations. The SPC node requires sorting operations.

[0066] The present invention determines the enabling situation of each sorting operation through the average sorting entropy of each sorting operation, realizes disabling some sorting operations of the decoding nodes, and then obtains the sorting enabling situation of each decoding node. For each decoding node, according to the determined sorting enabling situation of the decoding node, determine the number of sorting operations of the th target node of the target polar code .

[0067] Considering the sorting operation, the decoding delays of the respective target nodes of the target polar code are as follows. The Rate-0 node requires 1 time step (decoding delay). The Rate-1 node requires time steps, where is the decoding parameter, is the number of information bits of the target node, is the number of sorting operations. The Rep node requires time steps. The SPC node requires time steps. The calculation formula for the decoding delay of the target polar code is as follows.

[0068]

[0069] Among them, is the construction sequence of the target polar code of the th target node's decoding delay, is the decoding parameter of the decoder, is the construction sequence of the target polar code of the th target node's number of information bits, is the th target node's number of sorting operations (determined by the sorting enabling situation of the decoding node corresponding to the target node), is the decoding delay of the target polar code , is the number of target nodes in the construction sequence of the target polar code , is the number of SPC nodes in the target polar code.

[0070] According to the sorting enabling situation of the decoding node, the present invention determines the number of sorting operations of the target node, and realizes the regulation of the decoding delay of the target polar code without affecting the performance of the decoder. According to the number of sorting operations, decoding parameters, the number of information bits of the target node, and the number of SPC nodes, the decoding delay of the target polar code is determined, and the accurate calculation of the decoding delay of the target polar code is realized.

[0071] Based on the above embodiments, obtaining the fixed sequence interval of the polar code and multiple random sequence intervals of the polar code based on the set code length and the set information bit length includes: when the set code length is N and the information bit length is K, sorting the N channels in descending order of channel reliability to obtain the sorted channels; selecting the first K channels from the sorted channels as the information bit channels, and using the remaining N - K channels as the frozen bit channels to obtain the initial polar code; in the initial polar code, randomly shuffling the information bits and frozen bits of the channels in the sequence interval of [K - f + 1, K + f], where f is the set number of channels; in the initial polar code, using the sequence interval outside the [K - f + 1, K + f] sequence interval as the fixed sequence interval.

[0072] The construction of the random sequence intervals of the present invention is all based on the 5G polar code Beta (β function) expansion construction method and obtained through fine-tuning.

[0073] When the code length is set to N and the information bit length is K, the N channels are sorted in descending order of channel reliability to obtain the sorted channel order (IdR). The first K channels selected from the sorted channels are used as information bit channels, and the remaining N - K channels are used as frozen bit channels to obtain the initial polar code. For example, the initial polar code is A = {a1,..., a K , a K+1 ,..., a N}, where for the index i such that the reliability order IdR i ≤K, then a i = 1; for the index i such that the reliability order IdR i >K, then ai = 0..

[0074] Based on the initial polar code, for i corresponding to K - f + 1 ≤ IdR i ≤K + f, random shuffling is performed to obtain multiple random sequence intervals. Where f is the set number of channels.

[0075] The present invention enriches the construction form of the random sequence intervals by randomly shuffling the allocation of information bits and frozen bits in the sequence intervals corresponding to the set of indices i in the initial polar code where K - f + 1 ≤ IdR i ≤K + f, which is beneficial to obtaining a construction with almost lossless decoding error correction performance and lower decoding delay of the polar code.

[0076] Based on the above embodiments, the decoding evaluation model is determined based on the following steps: obtaining multiple sample random sequence intervals according to the sample code length and the sample information bit length; marking the sample random sequence intervals according to the decoding delay and the decoding error rate of the sample random sequence intervals to obtain the training samples with labels; constructing a preset neural network based on the dual deep neural network with self-attention mechanism; training the preset neural network based on the training samples until the errors of the output decoding error rate and the output decoding delay of the preset neural network both meet the set errors to obtain the trained neural network; constructing an objective function based on the output decoding error rate of the trained neural network and the output decoding delay of the trained neural network to obtain the decoding evaluation model.

[0077] Construct a sample initial polar code according to the sample code length and the sample information bit length. For the sample initial polar code where K - f + 1 ≤ IdR iRandomly shuffle the information bits and frozen bits of the channel in the sequence interval corresponding to the set of numbers i that are ≤ K + f to obtain multiple sample random sequence intervals. Obtain the decoding delay of each sample random sequence interval and the decoding error rate of the sample random sequence interval as labels. Mark the sample random sequence intervals according to the labels to obtain the training samples carrying labels.

[0078] Train a preset neural network constructed by a twin deep neural network with self-attention mechanism according to the training samples carrying labels to obtain a trained neural network. The trained neural network can relatively accurately estimate the decoding error rate and decoding delay of the sample random sequence intervals.

[0079] Construct an objective function according to the output decoding error rate and output decoding delay of the trained neural network to obtain a decoding evaluation model.

[0080] The present invention realizes the accurate calculation of the decoding error rate and decoding delay of the decoding model for the random sequence interval by training a twin deep neural network based on the self-attention mechanism. By constructing an objective function through the output decoding error rate and output decoding delay, the accurate evaluation of the decoding of the random sequence interval is realized.

[0081] The optimization device of the decoder provided by the present invention will be described below. The optimization device of the decoder described below can be mutually corresponding and referred to the optimization method of the decoder described above.

[0082] As Figure 2 shown, the present invention provides an optimization device of a decoder, including: a construction module 201, configured to obtain a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length, where the fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code has undergone random changes.

[0083] A selection module 202, configured to input multiple random sequence intervals into the decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; wherein, the decoding evaluation model determines the target random sequence interval with the optimal decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval.

[0084] A first optimization module 203, configured to obtain an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtain the distribution of decoding nodes of the optimal polar code.

[0085] A second optimization module 204, configured to determine the sorting enabling situation of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtain a decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes.

[0086] The optimization device of the decoder provided by the embodiment of the present invention obtains a fixed sequence interval of the polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length. The fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code has random changes; input the multiple random sequence intervals into the decoding evaluation model, and obtain the target random sequence interval output by the decoding evaluation model; wherein, the decoding evaluation model determines the target random sequence interval with the best decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtain the optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtain the distribution of the decoding nodes of the optimal polar code; determine the sorting enabling situation of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtain the decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes. The efficiency of obtaining the target random sequence interval is improved through the decoding evaluation model, which is beneficial to improving the optimization efficiency of the decoder. The optimal polar code is obtained through the target random sequence interval and the fixed sequence interval to obtain the distribution of the decoding nodes, realizing the optimization of the distribution of the decoding nodes in the decoder, which is beneficial to reducing the decoding delay of the decoder. The sorting enabling situation of the decoding nodes is optimized through the sorting entropy of the sorting operation of the decoding nodes, which is beneficial to further reducing the decoding delay of the decoder.

[0087] In one embodiment, the selection module 202 is used to: obtain the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtain the decoding negative evaluation of each random sequence interval based on the decoding delay and the corresponding decoding error rate; use the random sequence interval corresponding to the minimum decoding negative evaluation as the target random sequence interval.

[0088] In one embodiment, the optimal polar code includes at least one decoding node. The second optimization module 204 is used to: obtain the sorting operation sequence corresponding to the optimal polar code according to the distribution of the decoding nodes; perform multiple sorting simulations on each sorting operation in the sorting operation sequence to obtain multiple sorting entropies of each sorting operation, and determine the average sorting entropy of each sorting operation based on the multiple sorting entropies; re-sort all the sorting operations in the sorting operation sequence from small to large according to the average sorting entropy to obtain the re-sorted sorting operation sequence; in the re-sorted sorting operation sequence, do not enable the first set number of sorting operations in the ranking, and enable the remaining sorting operations to obtain the sorting enabling situation of all decoding nodes.

[0089] In one embodiment, the optimization device of the decoder further includes a decoding module, and the decoding module is configured to: when decoding a target polar code based on the decoder, determine the number of sorting operation times of each target node corresponding to the target polar code based on the sorting enabling situation of decoding nodes; the target nodes include SPC nodes, and the target nodes and the decoding nodes are in one-to-one correspondence; determine the decoding delay of each target node based on the decoding parameters of the decoder, the number of information bits of the target node, and the number of sorting operation times; determine the decoding delay of the target polar code based on the decoding delay of the target node and the number of SPC nodes.

[0090] In one embodiment, the construction module 201 is configured to: when the set code length is N and the information bit length is K, sort N channels in descending order of channel reliability to obtain the sorted channels; select the first K channels from the sorted channels as information bit channels, and use the remaining N - K channels as frozen bit channels to obtain an initial polar code; in the initial polar code, randomly shuffle the information bits and frozen bits of the channels in the sequence interval [K - f + 1, K + f], where f is the set number of channels, to obtain a plurality of random sequence intervals; in the initial polar code, use the sequence intervals outside the [K - f + 1, K + f] sequence interval as fixed sequence intervals.

[0091] In one embodiment, the selection module 202 is configured to: obtain a plurality of sample random sequence intervals according to the sample code length and the sample information bit length; label the sample random sequence intervals according to the decoding delay and the decoding error rate of the sample random sequence intervals to obtain training samples with labels; construct a preset neural network based on the twin deep neural network of the self-attention mechanism; train the preset neural network based on the training samples until the error of the output decoding error rate and the error of the output decoding delay of the preset neural network both meet the set error to obtain a trained neural network; construct an objective function based on the output decoding error rate of the trained neural network and the output decoding delay of the trained neural network to obtain a decoding evaluation model.

[0092] Figure 3 An entity structure diagram of an electronic device is illustrated, such as Figure 3As shown in the figure, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340. Among them, the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logical instructions in the memory 330 to execute the optimization method of the decoder. The method includes: obtaining a fixed sequence interval of the polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length. The fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code undergoes random changes; inputting the multiple random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; where the decoding evaluation model determines the target random sequence interval with the optimal decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtaining an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtaining the distribution of decoding nodes of the optimal polar code; determining the sorting enabling situation of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtaining a decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes.

[0093] In addition, when the logical instructions in the above-mentioned memory 330 are implemented in the form of software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0094] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the optimization method of the decoder provided by the above-mentioned various methods. The method includes: obtaining a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length. The fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code undergoes random changes; inputting the multiple random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; wherein, the decoding evaluation model determines the target random sequence interval with the optimal decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtaining an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtaining the distribution of decoding nodes of the optimal polar code; determining the sorting enabling situation of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtaining a decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes.

[0095] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the optimization method of the decoder provided by the above-mentioned various methods. The method includes: obtaining a fixed sequence interval of a polar code and multiple random sequence intervals of the polar code based on a set code length and a set information bit length. The fixed sequence interval is an interval where the distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval where the distribution of information bits and frozen bits in the polar code undergoes random changes; inputting the multiple random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; wherein, the decoding evaluation model determines the target random sequence interval with the optimal decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; obtaining an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtaining the distribution of decoding nodes of the optimal polar code; determining the sorting enabling situation of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtaining a decoder based on the sorting enabling situation of the decoding nodes and the distribution of the decoding nodes.

[0096] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0097] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A decoder optimization method, characterized in that: include: acquiring, based on a set code length and a set information bit length, a fixed sequence interval of the polar code and a plurality of random sequence intervals of the polar code, wherein the fixed sequence interval is an interval in which distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval in which distribution of information bits and frozen bits in the polar code changes randomly; Inputting the multiple random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; wherein the decoding evaluation model determines the target random sequence interval with the best decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; acquiring an optimal polar code based on the target random sequence interval and the fixed sequence interval, and acquiring distribution of decoding nodes of the optimal polar code; The ordering entropy of the ordering operation of the decoding nodes is used to determine the ordering activation status of the decoding nodes, and the decoder is obtained based on the ordering activation status of the decoding nodes and the distribution of the decoding nodes.

2. The decoder optimization method according to claim 1, characterized in that: The decoding evaluation model is used to output the target random sequence interval: Obtaining a decoding delay of each of the random sequence intervals and a decoding error rate of each of the random sequence intervals; Based on the decoding delay and the corresponding decoding error rate, a negative decoding evaluation of each random sequence interval is obtained; The random sequence interval corresponding to the minimum negative decoding evaluation is used as the target random sequence interval.

3. The decoder optimization method according to claim 1, characterized in that: The optimal polar code includes at least one of the decoding nodes, and determining the ordering activation status of the decoding nodes based on the ordering entropy of the decoding nodes includes: Acquire a sorting operation sequence corresponding to the optimal polar code according to the distribution of the decoding nodes; Performing multiple sorting simulations on each sorting operation in the sorting operation sequence to obtain multiple sorting entropies of each sorting operation, and determining an average sorting entropy of each sorting operation based on the multiple sorting entropies; Rearrange all the sorting operations in the sorting operation sequence from small to large according to the average sorting entropy to obtain a reordered sorting operation sequence; In the reordered sorting operation sequence, a set number of sorting operations are not enabled, and the remaining sorting operations are enabled, so as to obtain the sorting activation status of all the decoding nodes.

4. The decoder optimization method according to claim 1, characterized in that: After obtaining the decoder based on the ordering and activation of the decoding nodes and the distribution of the decoding nodes, the method further includes: When decoding a target polar code based on the decoder, determining the number of sorting operations of each target node corresponding to the target polar code based on the sorting activation status of the decoding node; the target node includes an SPC node, and the target node corresponds to the decoding node one by one; Determining a decoding delay of each of the target nodes based on a decoding parameter of the decoder, the number of information bits of the target node, and the number of sorting operations; The decoding delay of the target polar code is determined based on the decoding delay of the target node and the number of SPC nodes.

5. The decoder optimization method according to claim 1, characterized in that: The acquiring, based on a set code length and a set information bit length, a fixed sequence interval of a polarization code and a plurality of random sequence intervals of the polarization code comprises: When the set code length is N and the information bit length is K, sorting the N channels in descending order of channel reliability to obtain sorted channels; Selecting the top K channels from the sorted channels as information bit channels, and using the remaining NK channels as frozen bit channels to obtain an initial polarization code; In the initial polar code, information bits and frozen bits are randomly shuffled for channels in a [K-f+1, K+f] sequence interval to obtain the multiple random sequence intervals, where f is a set number of channels; In the initial polar code, a sequence interval other than the [K-f+1, K+f] sequence interval is used as the fixed sequence interval.

6. The decoder optimization method according to claim 1, characterized in that: The decoding evaluation model is determined based on the following steps: Obtain multiple sample random sequence intervals according to the sample code length and the sample information bit length; Marking the sample random sequence interval according to the decoding delay of the sample random sequence interval and the decoding error rate of the sample random sequence interval to obtain a training sample carrying the label; The twin deep neural network based on the self-attention mechanism builds a preset neural network; The preset neural network is trained based on the training samples until the error of the output decoding error rate and the error of the output decoding delay of the preset neural network both meet the set error, thereby obtaining a trained neural network; Based on the output decoding error rate of the trained neural network and the output decoding delay of the trained neural network, an objective function is constructed to obtain the decoding evaluation model.

7. A decoder optimization device, characterized in that: include: a construction module, configured to acquire, based on a set code length and a set information bit length, a fixed sequence interval of a polar code and a plurality of random sequence intervals of the polar code, wherein the fixed sequence interval is an interval in which distribution of information bits and frozen bits in the polar code is fixed, and the random sequence interval is an interval in which distribution of information bits and frozen bits in the polar code changes randomly; A selection module is used to input the multiple random sequence intervals into a decoding evaluation model to obtain a target random sequence interval output by the decoding evaluation model; wherein the decoding evaluation model determines the target random sequence interval with the best decoding based on the decoding delay of each random sequence interval and the decoding error rate of each random sequence interval; a first optimization module, configured to obtain an optimal polar code based on the target random sequence interval and the fixed sequence interval, and obtain a distribution of decoding nodes of the optimal polar code; The second optimization module is used to determine the sorting activation status of the decoding nodes based on the sorting entropy of the sorting operation of the decoding nodes, and obtain a decoder based on the sorting activation status of the decoding nodes and the distribution of the decoding nodes.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the decoder optimization method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the decoder optimization method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the decoder optimization method according to any one of claims 1 to 6 is implemented.

Images