Decoding method and related equipment
Patent Information
- Application Number
- CN202310520651.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-09
- Publication Date
- 2024-11-12
AI Technical Summary
[0003]目前的译码方案中,都存在计算复杂度较高的问题,例如,基于置信传播(BeliefPropagation,BP)的循环迭代译码方案,由于多次迭代译码和BP译码本身的高计算复杂度,导致该译码方案整体的计算复杂度较高;又例如,基于分类的深度神经网络译码方案,由于分类所导致的高计算复杂度和全连接结构的高计算量,导致该译码方案整体的计算复杂度较高
[0020]For the above-mentioned obtaining of the first noise information based on the initial noise information and the binarized noise estimator, in one case, the estimated noise information in the information to be decoded can be identified and removed according to actual needs or scenarios. Or at least a colored noise message. If we only consider identifying and removing the estimated noise information in the information to be decoded, obtaining the first noise information may include, for example: inputting the initial noise information into a binarized noise estimator, and the binarized noise estimator outputs the information corresponding to the information to be decoded. Estimating noise information, the output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information is the estimated noise information corresponding to the information to be decoded; such as If only considering identifying and removing at least one colored noise information in the information to be decoded, obtaining the first noise information may include, for example: inputting the initial noise information and the information to be decoded into a binarized noise estimator, and the binarized noise estimate The estimator outputs at least one colored noise information, and the output of the binarized noise estimator can be recorded as first noise information, that is, the first noise information includes at least one colored noise information. In another case, in order to improve the decoding accuracy, it is necessary to identify and remove all possible noise in the information to be decoded as much as possible. Then, obtaining the first noise information may include, for example: inputting the initial noise information and the information to be decoded. A binarized noise estimator outputs estimated noise information corresponding to the information to be decoded and at least one colored noise information. The output of the binarized noise estimator can be recorded as the first noise information, that is, , the first noise information includes estimated noise information corresponding to the information to be decoded and at least one colored noise information. In this case, the binarized noise estimator may include a main binarized noise estimation module and several binarized noise estimation modules. Then, the above-mentioned step of obtaining the first noise information may include: based on the initial noise information and the The main binarized noise estimation module in the binary noise estimator determines the to-be-translated The estimated noise of the code information itself; based on the initial noise information and the information to be decoded, obtain joint information; based on the joint information and at least one sub-binarized noise estimation module in the binarized noise estimator , determine at least one colored noise, and the sub-binarized noise estimation module corresponds to the colored noise. In this way, decoding accuracy can be improved while ensuring low computational complexity.
Smart Images

Figure CN118944681A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a decoding method and related equipment. Background Art
[0002] Coding technology and decoding technology play an important role in ensuring the effective transmission of information. Coding technology refers to the process by which the sender of information converts the information to be transmitted into information that can be transmitted, and decoding technology refers to the process by which the receiver of information restores the information received from the channel into the information that the sender hopes the receiver will obtain (i.e., the above-mentioned information to be transmitted).
[0003] The current decoding schemes all have the problem of high computational complexity. For example, the cyclic iterative decoding scheme based on belief propagation (BP) has a high overall computational complexity due to multiple iterative decoding and the high computational complexity of BP decoding itself. For another example, the deep neural network decoding scheme based on classification has a high overall computational complexity due to the high computational complexity caused by classification and the high computational amount of the fully connected structure.
[0004] Based on this, it is necessary to provide a decoding scheme that can reduce the computational complexity in the decoding process, so as to reduce the energy consumption of the decoding process. Summary of the invention
[0005] The present application provides a decoding method and related devices, which can achieve decoding with low computational complexity.
[0006] In the first aspect, the present application provides a decoding method, which is applied to a communication device having a function corresponding to the decoding method provided by the present application. In the decoding method, the communication device first processes the information to be decoded to obtain the first soft information (also referred to as probability or soft information probability) corresponding to the information to be decoded; then, the communication device decodes the first soft information based on the first binarized decoder to obtain the second soft information; then, the communication device can determine the decoding result corresponding to the information to be decoded based on the second soft information. Among them, the first binarized decoder is a decoder constructed based on a binarized neural network (Binarized Neural Networks, BNN). In this way, considering that BNN has the characteristics of low computational complexity, the present application constructs a binarized decoder based on BNN, and uses the binarized decoder to complete the decoding of the information to be decoded, so that the amount of information that needs to be stored and processed during the decoding process is greatly reduced, which can effectively reduce the computational complexity of decoding, thereby shortening the time used for decoding and improving the efficiency of decoding.
[0007] As an example, BNN includes an adder and a memory, which can be understood as a computing system composed of alternately connected adders and memories. Since BNN only binarizes weights and activation values, there is no need to change the structure of the network. Therefore, by designing a BNN that conforms to the decoding logic, the computational overhead of the entire decoder can be reduced while ensuring sufficient decoding accuracy. Compared with traditional decoding schemes such as BP-based iterative decoding schemes, decoding with lower computational complexity can be achieved, which can reduce the overhead in the decoding process.
[0008] In a possible implementation, the obtaining of the first soft information based on the information to be decoded may include: first, according to a preset corresponding relationship, converting each character included in the information to be decoded into a corresponding probability; then, based on the probability corresponding to each character included in the information to be decoded, obtaining the first soft information. In this way, obtaining the first soft information corresponding to the information to be decoded prepares for implementing the decoding method provided by the present application to realize decoding of the information to be decoded.
[0009] In another possible implementation, the obtaining of the first soft information based on the information to be decoded may include: first, removing the first noise information from the information to be decoded to obtain the first denoised information; then, according to a preset correspondence, converting each character included in the first denoised information into a corresponding probability; and then, based on the probability corresponding to each character included in the first denoised information, obtaining the first soft information. In this way, not only is it prepared to implement the decoding method provided by the present application to realize the decoding of the information to be decoded, but also, by considering the noise information in the information to be decoded and denoising before decoding, it is possible to obtain a higher-precision decoding result.
[0010] As an example, before removing the first noise information from the information to be decoded and obtaining the first denoised information, the implementation method may also include: obtaining the first noise information of the information to be decoded. In one case, the first noise information can be obtained based on a convolutional neural network (CNN). In another case, the first noise information can also be obtained based on a binarized noise estimator, wherein the binarized noise estimator is a model constructed based on a BNN for determining noise information. In this way, noise estimation is performed on the information to be decoded based on a BNN with low computational complexity, so that the computational complexity of the noise identification and denoising process before decoding is reduced, thereby achieving decoding with high decoding accuracy and low computational complexity.
[0011] As an example, obtaining the first noise information based on a binarized noise estimator and the information to be decoded may include: first, determining initial noise information based on a second binarized decoder and the information to be decoded, wherein the second binarized decoder is a decoder constructed based on a BNN; and then, obtaining the first noise information based on the initial noise information and the binarized noise estimator. The second binarized decoder may be understood as a primary decoder of the information to be decoded.
[0012] The above-mentioned determination of the initial noise information based on the second binarization decoder and the information to be decoded may include, for example: inputting the information to be decoded into the second binarization decoder and outputting a primary decoding result; then, calculating the difference between the information to be decoded and the primary decoding result to obtain the initial noise information.
[0013] In one case, for the above-mentioned method of obtaining the first noise information according to the initial noise information and the binarized noise estimator, the estimated noise information or at least one colored noise information in the information to be decoded can be identified and removed according to actual needs or scenarios. The estimated noise information can refer to the noise of the information to be decoded or the original information corresponding to the information to be decoded, rather than the noise generated due to signal transmission or other reasons; it can refer to the noise with an uneven power spectrum density function. The mutual influence of equipment in the park, equipment filtering, signal transmission in the channel, etc. are all possible factors that lead to the generation of colored noise information. If only the identification and removal of estimated noise information in the information to be decoded is considered, obtaining the first noise information may, for example, include: inputting the initial noise information into a binarized noise estimator, the binarized noise estimator outputs the estimated noise information corresponding to the information to be decoded, and the output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information is the estimated noise information corresponding to the information to be decoded; if only the identification and removal of at least one kind of colored noise information in the information to be decoded is considered, obtaining the first noise information may, for example, include: inputting the initial noise information and the information to be decoded into a binarized noise estimator, the binarized noise estimator outputs at least one kind of colored noise information, and the output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information includes at least one kind of colored noise information.
[0014] In another case, for obtaining the first noise information based on the initial noise information and the binarized noise estimator, in order to improve the decoding accuracy, it is necessary to identify and remove all possible noise in the information to be decoded as much as possible. Then, obtaining the first noise information may include, for example: inputting the initial noise information and the information to be decoded into the binarized noise estimator, the binarized noise estimator outputting the estimated noise information corresponding to the information to be decoded and at least one colored noise information, and the output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information includes the estimated noise information corresponding to the information to be decoded and at least one colored noise information. In this case, the binarization noise estimator may include a main binarization noise estimation module and several secondary binarization noise estimation modules, then, the step of obtaining the first noise information may include: determining the estimated noise of the information to be decoded itself based on the initial noise information and the main binarization noise estimation module in the binarization noise estimator; obtaining joint information based on the initial noise information and the information to be decoded; determining at least one first colored noise based on the joint information and at least one first binarization noise estimation module in the binarization noise estimator, the first binarization noise estimation module corresponding to the first colored noise. In this way, the decoding accuracy can be improved while ensuring low computational complexity.
[0015] In some implementations, in order to further improve the decoding accuracy, the present application also designs a decoding architecture for parallel decoding of multiple binary decoders. Taking the decoding architecture as an example, in addition to the first binary decoder, the third binary decoder is also included. Assuming that the binary noise estimator also includes a second binary noise estimation module, the present application may also include: determining the second colored noise based on the joint information and the second binary noise estimation module; determining the second noise information based on the estimated noise and the second colored noise; removing the second noise information from the information to be decoded to obtain the second denoised information; decoding the third soft information corresponding to the second denoised information based on the third binary decoder to obtain the fourth soft information, and the third binary decoder is a decoder constructed based on BNN. Then, determining the decoding result of the information to be decoded may include, for example: determining the decoding result of the information to be decoded based on the aggregation result of the second soft information and the fourth soft information. In this way, by decoding multiple binary decoders in parallel, the expression ability of the decoding architecture can be improved, the decoding error rate can be reduced, and the decoding accuracy can be improved.
[0016] As an example, the input of the first binary decoder may also include the first lateral information of the third binary decoder, and / or the input of the third binary decoder may also include the second lateral information of the first binary decoder, the first lateral information is used to reflect the characteristics of the third binary decoder, and the second lateral information is used to reflect the characteristics of the first binary decoder. In this way, in a decoding scheme in which multiple binary decoders decode in parallel, the interaction of differences between different binary decoders is achieved by exchanging lateral information between multiple binary decoders, thereby increasing the learning resources of the binary decoders and improving the decoding accuracy.
[0017] In the second aspect, the present application also provides a decoding method, which is applied to a communication device having a function corresponding to the decoding method provided by the present application. In the decoding method, the communication device first obtains the first noise information corresponding to the information to be decoded based on the binary noise estimator; then, the communication device removes the first noise information from the information to be decoded to obtain denoised information; then, the communication device decodes the denoised information to obtain the decoding result of the information to be decoded. The binary noise estimator is a model constructed based on BNN for determining noise information. Thus, considering that BNN has the characteristic of low computational complexity, in order to design a decoding scheme with high decoding accuracy and low computational complexity, the present application adopts a binary noise estimator in the decoding process to complete the simple and convenient identification of noise information in the information to be decoded, and then decodes the denoised information obtained after removing the noise information from the information to be decoded, so that the amount of storage and information processed occupied by identifying the noise information in the decoding process are greatly reduced, which can effectively reduce the computational complexity of decoding, thereby shortening the time used for decoding and improving the efficiency of decoding; in addition, the decoding in the present application takes into account the noise information of the information to be decoded, and decodes the denoised information, which can improve the decoding accuracy.
[0018] In a possible implementation, the above-mentioned obtaining the first noise information of the information to be decoded based on the binarization noise estimator may, for example, include: determining initial noise information based on a first binarization decoder and the information to be decoded, wherein the first binarization decoder is a decoder constructed based on BNN; and obtaining the first noise information according to the initial noise information and the binarization noise estimator.
[0019] As an example, the above-mentioned determination of the initial noise information based on the first binarization decoder and the information to be decoded may include: inputting the information to be decoded into the first binarization decoder and outputting the initial decoding result; then, calculating the difference between the information to be decoded and the initial decoding result to obtain the initial noise information.
[0020] For the above-mentioned obtaining of the first noise information according to the initial noise information and the binarized noise estimator, in one case, the estimated noise information or at least one colored noise information in the information to be decoded can be identified and removed according to actual needs or scenarios. If only the identification and removal of the estimated noise information in the information to be decoded is considered, obtaining the first noise information can, for example, include: inputting the initial noise information into the binarized noise estimator, the binarized noise estimator outputs the estimated noise information corresponding to the information to be decoded, and the output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information is the estimated noise information corresponding to the information to be decoded; if only the identification and removal of at least one colored noise information in the information to be decoded is considered, obtaining the first noise information can, for example, include: inputting the initial noise information and the information to be decoded into the binarized noise estimator, the binarized noise estimator outputs at least one colored noise information, and the output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information includes at least one colored noise information. In another case, in order to improve the decoding accuracy, it is necessary to identify and remove all possible noises in the information to be decoded as much as possible. Then, obtaining the first noise information may include, for example: inputting the initial noise information and the information to be decoded into a binarized noise estimator, the binarized noise estimator outputs the estimated noise information corresponding to the information to be decoded and at least one colored noise information, the output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information includes the estimated noise information corresponding to the information to be decoded and at least one colored noise information. In this case, the binarized noise estimator may include a main binarized noise estimation module and several secondary binarized noise estimation modules, then, the above step of obtaining the first noise information may include: determining the estimated noise of the information to be decoded itself based on the initial noise information and the main binarized noise estimation module in the binarized noise estimator; obtaining joint information based on the initial noise information and the information to be decoded; determining at least one colored noise based on the joint information and at least one secondary binarized noise estimation module in the binarized noise estimator, the secondary binarized noise estimation module corresponding to the colored noise. In this way, the decoding accuracy can be improved while ensuring low computational complexity.
[0021] In a possible implementation, the above decoding of the denoised information to obtain the decoding result of the information to be decoded may include, for example: obtaining first soft information based on the denoised information; decoding the first soft information based on a second binarization decoder to obtain second soft information, wherein the second binarization decoder is a decoder constructed based on BNN; and determining the decoding result based on the second soft information. In this way, the decoder constructed based on BNN completes the decoding, and since BNN has the characteristic of low computational complexity, the computational complexity of the decoding scheme provided in the present application is further reduced.
[0022] In another possible implementation, the above-mentioned decoding of the denoised information to obtain the decoding result of the information to be decoded may include, for example: inputting the denoised information into a classification-based deep neural network decoder, and the classification-based deep neural network decoder outputs the decoding result. In this way, compared with the decoding scheme using the traditional deep neural network decoder + CNN, due to the low computational complexity of the binarized noise estimator, low computational complexity decoding is achieved, and because the binarized noise estimator can identify more comprehensive noise recognition, high-precision decoding is achieved.
[0023] In the first aspect or the second aspect above, the type of information to be decoded may include but is not limited to any one of the following types: low density parity check (LDPC) code, turbo code or polar code.
[0024] The method provided in the first aspect or the second aspect can be applied to a wireless communication device or a digital communication device, wherein the digital communication device may include, but is not limited to, a Field Programmable Gate Array (FPGA).
[0025] In a third aspect, the present application provides a communication device, which includes a processing unit, and the processing unit is used to implement methods corresponding to the first aspect, the second aspect and possible implementation methods thereof.
[0026] In a fourth aspect, the present application provides a communication device, comprising a processor and a memory; the processor is used to execute instructions stored in the memory so that the communication device implements the methods corresponding to the first aspect, the second aspect and possible implementation methods thereof.
[0027] In a fifth aspect, the present application further provides a storage medium, wherein the storage medium stores instructions, and when the instructions are executed on a processor, methods corresponding to the first aspect, the second aspect and possible implementations thereof are implemented.
[0028] In a sixth aspect, the present application further provides a program product, which includes a program, and when the program runs on a processor, implements the methods corresponding to the first aspect, the second aspect and their possible implementation methods.
[0029] In the seventh aspect, the present application provides a chip including a memory and a processor, the memory being used to store instructions, and the processor being used to call and execute the instructions from the memory to implement the methods corresponding to the first aspect, the second aspect and their possible implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0031] Figure 1 A possible structural diagram of a BNN in an embodiment of the present application;
[0032] Figure 2 1 is a flowchart of a decoding method 100 in an embodiment of the present application;
[0033] Figure 3 A possible structural diagram of the first binary decoder in the embodiment of the present application;
[0034] Figure 4 2 is a flowchart of another decoding method 200 according to an embodiment of the present application;
[0035] Figure 5 A possible structural diagram of a binary noise estimator in an embodiment of the present application;
[0036] Figure 6 A possible structural diagram of a decoding architecture in a communication device in an embodiment of the present application;
[0037] Figure 7 3 is a flowchart of another decoding method 300 in an embodiment of the present application;
[0038] Figure 8 A schematic diagram of noisy long code data transmission based on LDPC code in a wireless communication scenario in an embodiment of the present application;
[0039] Fig. 9 A schematic diagram of LDPC decoding on an FPGA in an embodiment of the present application;
[0040] Fig.10 This is a schematic diagram of the structure of a communication device 1000 according to an embodiment of the present application;
[0041] Fig.11 This is a schematic diagram of the structure of a communication device 1100 according to an embodiment of the present application;
[0042] Fig.12 1 is a schematic diagram of the structure of a communication device 1200 in an embodiment of the present application. DETAILED DESCRIPTION
[0043] Decoding technology is very important for the effective transmission of information. The current decoding technology has a relatively high computational complexity. For example, the BP-based iterative decoding scheme uses probability to determine whether each codeword has a transmission error. The logarithm likelihood rate (LLR) can usually be used to characterize the probability. In addition, multiple iterative calculations are required during the decoding process. It can be seen that due to the high computational complexity of BP decoding itself and the need for multiple iterative decoding, the overall computational complexity of the decoding scheme is relatively high. For another example, the deep neural network decoding scheme based on classification constructs a fully connected neural network and uses the idea of classification to achieve decoding. It can be seen that due to the high computational complexity caused by classification and the high computational amount of the fully connected structure, the overall computational complexity of the decoding scheme is relatively high.
[0044] Faced with the trend of increasing the amount of transmitted information, the current decoding scheme will lead to increasing decoding overhead, and the computing resources consumed by the decoding process need to continue to increase. However, simply increasing computing resources to solve the problem of increased decoding overhead is very limiting to the development of decoding technology. Therefore, this application aims to provide a decoding scheme with lower computational complexity.
[0045] In the embodiment of the present application, considering that BNN only uses two values of +1 and -1 to represent weights and activation values, and uses a simple calculation combination of XNOR (bitwise exclusive OR) and popcount (a function used to count the number of 1 bits in the binary representation of an integer), compared with the full-precision neural network using 32-bit floating-point data multiplication and accumulation to implement convolution calculations, BNN can save a lot of memory and computing resources. Therefore, the design of a decoder or noise estimator based on BNN can reduce the computational complexity in the decoding process.
[0046] The embodiment of the present application provides a decoding method, in which, first, the information to be decoded is processed to obtain the first soft information (also referred to as probability or soft information probability) corresponding to the information to be decoded; then, the first soft information is decoded based on the first binarization decoder to obtain the second soft information; then, based on the second soft information, the decoding result corresponding to the information to be decoded is determined. Among them, the first binarization decoder is a decoder constructed based on BNN. In this way, the binarization decoder is constructed based on BNN. Since BNN itself has the characteristic of low computational complexity, decoding the information to be decoded based on the binarization decoder can also reduce the computational complexity in the decoding process, shorten the time used for decoding, and improve the efficiency of decoding.
[0047] In addition, considering that there may be noise information in the information to be decoded in some cases, denoising is a necessary means to ensure the decoding accuracy in these cases. At present, denoising in the decoding process, for example, CNN can be used to identify the noise information in the information to be decoded. However, due to the high computational complexity of CNN, in order to reduce the computational complexity of denoising during the decoding process, a decoding method is also provided in an embodiment of the present application. In this method, first, the first noise information corresponding to the information to be decoded is obtained based on a binarized noise estimator; then, the first noise information is removed from the information to be decoded to obtain denoised information; then, the denoised information is decoded to obtain the decoding result of the information to be decoded. Among them, the binarized noise estimator is a model constructed based on BNN for determining noise information. It can be seen that, considering that BNN has the characteristic of low computational complexity, in order to design a decoding scheme with high decoding accuracy and low computational complexity, the embodiment of the present application uses a binary noise estimator designed based on BNN to perform noise identification on the information to be decoded, which can occupy less storage and processing resources to complete the identification of noise in the information to be decoded, and reduces the computational complexity of the decoding process in the decoding scheme that ensures decoding accuracy.
[0048] In addition, in the decoding method provided in the embodiment of the present application, a binary noise estimator and a binary decoder can also be combined to complete the decoding of the information to be decoded, which can not only identify and remove the noise information in the information to be decoded to ensure the decoding accuracy, but also, in the case of increasing amount of information transmitted, under the condition of limited resources such as storage and calculation, a binary noise estimator and a binary decoder designed based on BNN can be applied to the decoding process to achieve decoding with low computational complexity.
[0049] In order to facilitate understanding of the method provided in the embodiment of the present application, the BNN, the binarization decoder and the binarization noise estimator are briefly introduced below.
[0050] BNN can be understood as a computing system consisting of alternately connected adders and memories. The salient features of BNN are: first, the weights and activation values are binarized to 1 or -1, which significantly reduces the storage space required for using BNN; second, bit operations are used to replace the multiplication and addition operations in the network, which simplifies the operation method, reduces the computational complexity and shortens the operation time. Moreover, since BNN only binarizes the weights and activation values and does not change the structure of the network, by designing a BNN that conforms to the decoding logic and calculating the information matrix of the LLR, the computational overhead of the entire decoder is reduced while ensuring sufficient decoding accuracy. Compared with traditional decoding schemes such as the BP-based cyclic iterative decoding scheme, decoding with lower computational complexity is achieved, reducing the overhead in the decoding process. Figure 1 A possible structural diagram of BNN is shown in FIG. Figure 1As shown, BNN 1 may include, in sequence: an adder 11, a memory 21, an adder 12, a memory 22, ..., an adder 1N and a memory 2N, where N is an integer greater than 1.
[0051] A binary decoder is used to decode information to be decoded. For example, soft information 1 corresponding to the information to be decoded can be input into the binary decoder, and the binary decoder outputs soft information 2, which is used to directly obtain the decoding result of the information to be decoded. In one case, a binary decoder may include a BNN, and a communication device that implements the decoding method provided in an embodiment of the present application may include at least one binary decoder. If the communication device includes multiple binary decoders, then multiple binary decoders can be used for parallel decoding to enhance the network information processing capability. Moreover, multiple binary decoders are associated through lateral connections, and lateral information is exchanged between the binary decoders to improve the decoding accuracy. In another case, the communication device implementing the decoding method provided in the embodiment of the present application may include a binary decoder, each of which may include at least one BNN. If the binary decoder includes multiple BNNs, then the binary decoder may use multiple BNNs for parallel decoding to enhance the network information processing capability. Moreover, multiple BNNs may be associated through lateral connections, and lateral information may be exchanged between BNNs to improve the decoding accuracy. For other related descriptions of the binary decoder, please refer to Figure 2 , Figure 3 , Figure 6 and Figure 7 Related description.
[0052] The binary noise estimator is used to estimate the noise in the information to be decoded. For example, in the case where only the estimated noise information of the information to be decoded itself is considered, without considering various colored noise information, the input of the binary noise estimator can be the difference between the information to be decoded and the initial decoding result of the information to be decoded (which can be understood as the initial noise information), and the output of the binary noise estimator is the estimated noise information corresponding to the information to be decoded. Among them, the estimated noise information can refer to the noise of the information to be decoded or the original information corresponding to the information to be decoded itself, rather than the noise caused by signal transmission and other reasons; the colored noise information can refer to the noise with an uneven power spectrum density function. The mutual influence of equipment in the park, equipment filtering, and signal transmission in the channel are all possible factors that lead to the generation of colored noise information; the initial decoding result of the information to be decoded can be the output of any decoder (such as a binary decoder) after the information to be decoded is input into the decoder. For another example, for the case where only at least one colored noise information is considered without considering the estimated noise information, the binarized noise estimator may include a secondary binarized noise estimation module corresponding to each colored noise information in the at least one colored noise information, the input of each secondary binarized noise estimation module may be joint information obtained based on the information to be decoded and the initial noise information, and the output is the colored noise information corresponding to the secondary binarized noise estimation module. For another example, for the case where at least one colored noise information and the estimated noise information are considered, the binarized noise estimator may include: a main binarized noise estimation module and a secondary binarized noise estimation module corresponding to each colored noise information in the at least one colored noise information, the input of the binarized noise estimator is the information to be decoded and the initial noise information, and the output is the estimated noise information and at least one colored noise, wherein the input of the main binarized noise estimation module is the initial noise information, and the output is the estimated noise information; the input of each secondary binarized noise estimation module may be joint information obtained based on the information to be decoded and the initial noise information, and the output is the colored noise information corresponding to the secondary binarized noise estimation module. Each sub-binarized noise estimation module corresponds to a BNN and is applicable to a type of colored noise information. For other related descriptions of the binary noise estimator, see Figure 4 , Figure 5 , Figure 6 and Figure 7 Related description.
[0053] It should be noted that the decoding method provided in the embodiment of the present application is applicable to the types of information to be decoded, including but not limited to: low-density parity check (LDPC) code, turbo code or polar code, and the type of information to be decoded does not affect the implementation of the embodiment of the present application.
[0054] It should be noted that the decoding method provided in the embodiment of the present application can be implemented on a communication device, and the communication device can be, for example, a wireless communication device or a digital communication device (such as a Field Programmable Gate Array (FPGA)).
[0055] The following is a description of the possible technical solutions implemented by the decoding method provided in the embodiments of the present application in conjunction with the accompanying drawings.
[0056] Figure 2 FIG. 1 is a flow chart of a decoding method 100 provided in an embodiment of the present application. In the method 100, the communication device implementing the method 100 is used as the execution subject for relevant description. Figure 2 As shown, the method 100 may include, for example, the following S101 to S103:
[0057] S101, obtaining first soft information based on information to be decoded.
[0058] The information to be decoded includes multiple characters. The communication device can maintain the correspondence between characters and probabilities in advance. The correspondence can be maintained through a table or a model. Taking the maintenance of the correspondence through a table as an example, the correspondence maintained in the communication device can be shown in Table 1 below:
[0059] Table 1
[0060] character Probability A 0.1 B 0.6 C 0.2 …… ……
[0061] As an example, for the case where denoising of the information to be decoded is not considered, or for the case where there is no noise in the information to be decoded (the information to be decoded is denoised information or the information to be decoded itself does not include noise), S101 may include, for example: according to a preset corresponding relationship, converting each character included in the information to be decoded into a corresponding probability; based on the probability corresponding to each character included in the information to be decoded, obtaining the first soft information. Assume that the preset corresponding relationship in the communication device can refer to Table 1, the information to be decoded includes<A,C,…> , then, in this example, the first soft information can be expressed as <0.1, 0.2, …>.
[0062] As another example, if the information to be decoded includes noise and it is necessary to consider denoising the information to be decoded, then S101 may include: S101a, removing first noise information from the information to be decoded to obtain first denoised information; S101b, converting each character included in the first denoised information into a corresponding probability according to a preset corresponding relationship; S101c, obtaining the first soft information based on the probability corresponding to each character included in the first denoised information. In this way, it is possible to decode the information to be decoded after denoising, and preparations are made for the realization of high-precision decoding.
[0063] The first noise information represents all or part of the noise in the information to be decoded. Before S101a, the method 100 may also include: obtaining the first noise information in the information to be decoded. The communication device may obtain the first noise information in the information to be decoded in a manner that includes but is not limited to: a CNN noise estimator or a binarization noise estimator. The binarization noise estimator has lower computational complexity than the CNN noise estimator.
[0064] As an example, the communication device obtains the first noise information in the information to be decoded, which may include: the communication device obtains the first noise information in the information to be decoded based on a CNN noise estimator, and the CNN noise estimator may refer to a model constructed based on CNN for determining noise information.
[0065] As another example, the communication device obtains the first noise information in the information to be decoded, which may include: the communication device obtains the first noise information in the information to be decoded based on a binary noise estimator, and the binary noise estimator may refer to a model constructed based on BNN for determining noise information.
[0066] In the first case, the communication device may input information to be decoded into a binary noise estimator, and the binary noise estimator outputs first noise information.
[0067] In the second case, the communication device may first determine the initial noise information based on the second binary decoder and the information to be decoded, wherein the second binary decoder is a decoder constructed based on BNN; then, the communication device obtains the first noise information based on the initial noise information and the binary noise estimator. For the communication device to determine the initial noise information based on the second binary decoder and the information to be decoded, for example, it may include: the communication device inputs the information to be decoded into the second binary decoder, and the second binary decoder outputs the initial decoding result corresponding to the information to be decoded; then, the communication device calculates the difference between the information to be decoded and the initial decoding result to obtain the initial noise information. The second binary decoder can be understood as the initial decoder of the information to be decoded, and the structure of the second binary decoding can refer to the relevant description of the first binary decoder in S102 below. For a communication device to obtain first noise information based on initial noise information and a binarized noise estimator, in order to improve decoding accuracy, it is necessary to identify and remove all possible noise in the information to be decoded as much as possible. Obtaining the first noise information may include, for example: the communication device inputs the initial noise information and the information to be decoded into a binarized noise estimator, and the binarized noise estimator outputs estimated noise information corresponding to the information to be decoded and at least one colored noise information. The output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information includes the estimated noise information corresponding to the information to be decoded and at least one colored noise information. Alternatively, according to actual needs or scenarios, the estimated noise information or at least one colored noise information in the information to be decoded is identified and removed. If only the identification and removal of the estimated noise information in the information to be decoded is considered, obtaining the first noise information may include, for example: the communication device inputs the initial noise information into the binarized noise estimator, and the binarized noise estimator outputs the estimated noise information corresponding to the information to be decoded. The output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information includes the estimated noise information corresponding to the information to be decoded; if only the identification and removal of at least one colored noise information in the information to be decoded is considered, obtaining the first noise information may include, for example: the communication device inputs the initial noise information and the information to be decoded into the binarized noise estimator, and the binarized noise estimator outputs at least one colored noise information. The output of the binarized noise estimator can be recorded as the first noise information, that is, the first noise information includes at least one colored noise information. It should be noted that for the structure and working principle of the binarized noise estimator in various possible implementations, please refer to Figure 4 and Figure 5 Related instructions.
[0068] Based on S101, the first soft information corresponding to the information to be decoded is obtained, so that preparation is made for implementing the method 100 to realize decoding of the information to be decoded. Whether the first soft information is obtained based on the information to be decoded itself or based on the information after denoising the information to be decoded does not affect the implementation of the method 100.
[0069] S102: Decode the first soft information based on a first binarization decoder to obtain second soft information, where the first binarization decoder is a decoder constructed based on a BNN.
[0070] S103: Determine a decoding result of the information to be decoded based on the second soft information.
[0071] The binary decoder can be understood as a BNN composed of an adder and a memory. The adder in the binary decoder is used for the update operation of soft information. Each adder corresponds to a weight, which can be +1 or -1, representing the characteristics of the BNN. The memory in the binary decoder is used to store the data after the soft information is updated after each adder operation. In addition, it is also used to store the weight corresponding to the adder.
[0072] Figure 3 A possible structural diagram of the first binary decoder is shown in FIG. Figure 3 As shown, the first binary decoder 30 may include: an input layer 31, an intermediate layer 32 and an output layer 33, wherein the intermediate layer 32 may include a plurality of t e Layer and t E Layer, t e Layer and t E The number of layers is the same and they appear alternately. Figure 3 Only the middle layer 32 includes t e Layer 321 and t E Layer 322 is shown as an example. Input layer 31 is used to obtain the first soft information of the input and provide information to be used for BNN decoding; e Layer 321, used to calculate and update the information of the check node using the check node update equation; E The layer 322 is used to calculate and update the information of the variable node using the variable node update equation; the output layer 33 is used to output the second soft information, which is the basis for obtaining the decoding result of the information to be decoded.
[0073] As an example, S102 may include: the communication device inputs the first soft information into a first binarization decoder, and the first binarization decoder outputs the second soft information.
[0074] In some implementations, if the decoding architecture in the communication device does not include other binary decoders that decode in parallel with the first binary decoder, then the communication device in S103 can directly judge the second soft information to obtain the decoding result of the information to be decoded. The judgment of the second soft information can be understood as corresponding each probability in the second soft information to 0 or 1, and converting the array or matrix of the probability of the second soft information into an array or matrix composed of 0 and 1.
[0075] In other implementations, in order to improve the expressiveness of the decoding architecture within the communication device, reduce the decoding error rate, and improve the decoding accuracy, in addition to using the first binary decoder for decoding in the embodiment of the present application, other binary decoders can also be designed and used to perform decoding in parallel with the first binary decoder, thereby enhancing the communication device's ability to process information during the decoding process.
[0076] As an example, the decoding architecture of the communication device may further include at least one third binary decoder, the third binary decoder being another binary decoder constructed based on BNN, and if the number of the third binary decoders is multiple, each third binary decoder is used to decode its corresponding input, for example, the third binary decoder 1 is used to decode the soft information corresponding to the information obtained after the estimated noise and the colored noise information 1 are removed from the information to be decoded, the third binary decoder 2 is used to decode the soft information corresponding to the information obtained after the estimated noise and the colored noise information 2 are removed from the information to be decoded, and the third binary decoder 3 is used to decode the soft information corresponding to the information obtained after the estimated noise, the colored noise information 1 and the colored noise signal 2 are removed from the information to be decoded. The method 100 may further include: obtaining the third soft information corresponding to the information to be decoded, and obtaining the fourth soft information based on the third soft information and the third binary decoder; then, S103 may include: aggregating the second soft information and the fourth soft information to obtain an aggregation result, and determining the decoding result of the information to be decoded based on the aggregation result. In one case, the first soft information may be soft information corresponding to the information to be decoded itself, and the third soft information may be soft information corresponding to information obtained after removing estimated noise information from the information to be decoded, or the third soft information may be soft information corresponding to information obtained after removing at least one colored noise information from the information to be decoded, or the third soft information may be soft information corresponding to information obtained after removing estimated noise information and at least one colored noise information from the information to be decoded. In another case, the first soft information may be soft information corresponding to information obtained after the information to be decoded is removed from the estimated noise information and / or at least one colored noise information. Then, the third soft information may be soft information corresponding to the information to be decoded itself, or the third soft information may also be soft information corresponding to information obtained after the information to be decoded is removed from the estimated noise information and / or at least one colored noise information. The third soft information is different from the first soft information. For example, the first soft information is soft information corresponding to information obtained after the information to be decoded is removed from the estimated noise information. The third soft information may be soft information corresponding to information obtained after the information to be decoded is removed from the estimated noise information. Alternatively, the third soft information may be soft information corresponding to information obtained after the information to be decoded is removed from the estimated noise information and at least one colored noise information. Alternatively, the third soft information may also be soft information corresponding to information obtained after the information to be decoded is removed from the at least one colored noise information.
[0077] On the other hand, multiple binary decoders decode in parallel, and the multiple binary decoders can exchange lateral information, exchange differences between different binary decoders, improve learning resources of binary decoders, and improve decoding accuracy. Still taking the parallel decoding of the first binary decoder and at least one third binary decoder as an example, the input of the first binary decoder can also include first lateral information of at least one third binary decoder, and the input of the third binary decoder can also include second lateral information of the first binary decoder, the first lateral information is used to reflect the characteristics of the third binary decoder, and the second lateral information is used to reflect the characteristics of the first binary decoder. In this way, the binary decoder feeds back its own features to other binary decoders through lateral connections, and can also obtain the features of other binary decoders, so that the binary decoder can learn the differences between different binary decoders. In this way, the aggregation result of the soft information output by multiple binary decoders can more comprehensively and accurately translate the probability of each character in the decoding result, making it possible to achieve accurate decoding.
[0078] Taking the third binary decoder inputting the first lateral information into the first binary decoder as an example, the lateral information and t e The update equation of the check node of the layer and t E The variable node update equation of the layer is explained. In the first lateral information, the first lateral information weight can be expressed as The lateral input information can be expressed as E p,q , t e The update equation of the check node of the layer can be shown in the following formula (1), E The variable node update equation of the layer can be found in the following formula (2):
[0079]
[0080]
[0081] In formula (1) and formula (2), p and q represent each t e Layer and t E The subscript of the element in the layer, p' and q' represent the previous t e Layer and t E The subscript of the elements in the layer, i represents the i-th binarization decoder, j represents the j-th binarization decoder, and n is the total number of binarization decoders.
[0082] In formula (1), H[] is a possible activation function, W e Represents t e The weight of the layer, e p,q,j represents the jth binary decoder t e The element in the layer with the subscript (p,q), E′p,q,j Represents the last t of the j-th binary decoder E The elements in the layer with subscripts (p,q) have weights represents the i-th binary decoder and the j-th binary decoder (t+1) e Layer (ie the next t e The lateral weight between layers, E p,q,i represents the i-th binary decoder t E The element with the subscript (p,q) in the layer, q′∈L(p)\q represents the e Remove t from the set of element subscripts in the layer E The subscript of the element in the layer to get the previous t E The subscript of the element in the layer.
[0083] In formula (2), f[] is another possible activation function, W E Represents t E The weight of the layer, Represents the initial variables of the input layer, E′ p,q,j Represents the last t of the j-th binary decoder E The element in the layer with subscript (p,q), e p,q,j represents the jth binary decoder t e The elements in the layer with subscripts (p,q) have weights represents the i-th binary decoder and the j-th binary decoder (t+1) E Layer (ie the next t E The lateral weight between layers, e p,q,i represents the i-th binary decoder t e The element with the subscript (p,q) in the layer, (′∈M(q)\p represents the element from t E Remove t from the set of element subscripts in the layer e The subscript of the element in the layer to get the previous t e The subscript of the element in the layer.
[0084] It should be noted that the decoding architecture in the communication device can be found in Figure 6 And related instructions of method 300.
[0085] It can be seen that by using the method 100, a binary decoder is constructed based on BNN. Since BNN itself has the characteristic of low computational complexity, decoding the information to be decoded based on the binary decoder can reduce the computational complexity in the decoding process, shorten the decoding time, and improve the decoding efficiency. In addition, since BNN is composed of a simple adder and a memory connection, the binary decoder has the advantages of simple design and easy deployment, and decoding based on the method 100 is more convenient.
[0086] Figure 4 FIG. 2 is a flow chart of another decoding method 200 provided in an embodiment of the present application. In the method 200, the communication device implementing the method 200 is used as the execution subject for related description. Figure 4 As shown, the method 200 may include, for example, the following S201 to S203:
[0087] S201, obtaining first noise information corresponding to information to be decoded based on a binary noise estimator, wherein the binary noise estimator is a model constructed based on a BNN and used to determine noise information.
[0088] The binary noise estimator is used to identify first noise information in the information to be decoded. The first noise information includes estimated noise information existing in the information to be decoded itself and / or at least one kind of colored noise information in the information to be decoded.
[0089] In the first example, S201 may include, for example: the communication device inputs the information to be decoded into the binarized noise estimator, and the binarized noise estimator outputs the first noise information. In this example, the binarized noise estimator may include not only the binarized noise estimation unit, but also the primary decoding unit and the initial noise determination unit. The communication device may first input the information to be decoded into the primary decoding unit, and the primary decoding unit outputs the primary decoding result. The initial noise determination unit calculates the difference between the information to be decoded and the primary decoding result as the initial noise information, and then inputs the initial noise information into the binarized noise estimation unit, and the output of the binarized noise estimation unit is the first noise information. Among them, the primary decoding unit can be implemented by a binarized decoder, and the primary decoding unit can refer to the relevant description of the second binarized decoder in the second example below. The binarized noise estimation unit can refer to the relevant description of the binarized noise estimator in the second example below.
[0090] In the second example, S201 may include, for example: S2011, obtaining a first decoding result of the information to be decoded, S2012, taking out the first decoding result from the information to be decoded to obtain initial noise information, and S2013, inputting the initial noise information into a binarization noise estimator to obtain first noise information. Wherein, S2011 may include, for example: inputting the soft information corresponding to the information to be decoded into a second binarization decoder, and obtaining a first decoding result based on the output of the second binarization decoder. The implementation of S2013 and the content of the first noise information can be determined according to the structure in the binarization noise estimator and the actual decoding requirements. For example, the binarization noise estimator only includes a functional module for identifying the estimated noise information of the information to be decoded itself (i.e., the main binarization noise estimation module in the present application), then S2013 may include: inputting the initial noise information into the main binarization noise estimation module in the binarization noise estimator, the main binarization noise estimation module outputs the estimated noise information, and the estimated noise information is recorded as the first noise information; for another example, the binarization noise estimator only includes a functional module for identifying the colored noise information of the information to be decoded (i.e., the secondary binarization noise estimation module in the present application), then S2013 may include: determining joint information based on the initial noise information and the information to be decoded, inputting the joint information into each secondary binarization noise estimation module in the binarization noise estimator For example, the binarization noise estimator includes a main binarization noise estimation module and a sub-binarization noise estimation module, then S2013 may include: determining joint information based on initial noise information and information to be decoded, inputting the initial noise information into the main binarization noise estimation module in the binarization noise estimator, inputting the joint information into each sub-binarization noise estimation module in the binarization noise estimator, the main binarization noise estimation module outputs estimated noise information, each sub-binarization noise estimation module outputs colored noise information corresponding to the function of the sub-binarization noise estimation module, and the first noise information includes estimated noise information and colored noise information output by each sub-binarization noise estimation module.
[0091] In one case, determining the joint information in S2013 may include: adding the corresponding values in the initial noise information and the information to be decoded to obtain the joint information. Then, the joint information, the initial noise information and the information to be decoded have the same scale (such as 4*1 matrices). It should be noted that the mathematical expressions of the initial noise information and the information to be decoded may be arrays or matrices composed of complex numbers. The joint information may be obtained by adding the real part and the imaginary part of the initial noise information and the information to be decoded, respectively, and then combining the real part and the imaginary part obtained by the addition into the joint information. For example, the initial noise information is (a1+b1i, a2+b2i, a3+b3i, a4+b4i)T , the information to be decoded is (c1+d1i,c2+d2i,c3+d3i,c4+d4i) T , then the joint information can be expressed as: ((a1+c1)(b1+d1)i,(a2+c2)(b2+d2)i,(a3+c3)(b3+d3)i,(a4+c4)(b4+d4)i) T .
[0092] In another case, determining the joint information in S2013 may include: concatenating the initial noise information and the information to be decoded to obtain the joint information. Then, the number of elements included in the joint information is the sum of the number of elements included in the initial noise information and the number of elements included in the information to be decoded. If the initial noise information and the information to be decoded are both 4*1 matrices, the joint information may be a 4*2 matrix. Still taking the initial noise information as (a1+b1i, a2+b2i, a3+b3i, a4+b4i) T , the information to be decoded is (c1+d1i,c2+d2i,c3+d3i,c4+d4i) T For example, the joint information can be expressed as:
[0093] The manner of determining the joint information does not affect the scale of the colored noise information output by the secondary binarization noise estimation module. The scales of the colored noise information, the estimated noise information, the information to be decoded and the initial noise information may be consistent, for example, all are 4*1 matrices.
[0094] Figure 5 A schematic diagram showing a possible structure of a binary noise estimator is shown in FIG. Figure 5As shown, the binarization noise estimator 50 may include: a main binarization noise estimation module 51, a secondary binarization noise estimation module 521, a secondary binarization noise estimation module 522, ..., a secondary binarization noise estimation module 52n, where n is an integer greater than 1, wherein the main binarization noise estimation module 51 is used to identify the estimated noise information of the information to be decoded itself from the initial noise information, and the secondary binarization noise estimation module 52i (i=1, 2, ..., n) is used to identify the corresponding colored noise information included in the information to be decoded from the joint information of the initial noise information and the information to be decoded. It should be noted that the number of colored noise information included in the first noise information may be less than or equal to n. When the colored noise information in the first noise information is obtained, the joint information may be input into all sub-binarization noise estimation modules in the binarization noise estimator 50, and all the colored noise information output by each sub-binarization noise estimation module may be included in the first noise information, or, the colored noise information that meets the preset conditions in the colored noise information output by each sub-binarization noise estimation module may be included in the first noise information; or, if the target types (the number of target types) of colored noise information to be identified and removed are predetermined, It may be at least one), the target type may be, for example, a type of colored noise information that may be included in the information to be decoded or a type of colored noise information that needs to be paid attention to. When the colored noise information in the first noise information is obtained, the joint information may also be input into a sub-binarization noise estimation module corresponding to the target type in the binarization noise estimator 50, and all the colored noise information output by the sub-binarization noise estimation module corresponding to the target type is included in the first noise information, or, in the colored noise information output by the sub-binarization noise estimation module corresponding to the target type, the colored noise information that meets the preset conditions is included in the first noise information.
[0095] It should be noted that the mathematical expression of the initial noise information and the joint information may be an array or matrix composed of complex numbers. The binary noise estimator can split the real and imaginary parts of the complex number, perform iterative addition and subtraction operations on the real and imaginary parts through the binary weights of the BNN, and then combine the output results corresponding to the real and imaginary parts to obtain the complex number processing result. Taking the main binary noise estimator as an example, the initial noise information (a1+b1i, a2+b2i, a3+b3i, a4+b4i) can be firstly T Split into real part information 1 (ie (a1, a2, a3, a4) T ) and imaginary information 1 (i.e. (b1, b2, b3, b4) T ); Then, the main binary noise estimator processes the real information 1 to obtain the real information 2, which may be (a5, a6, a7, a8) T ), the imaginary information 1 is processed to obtain the imaginary information 2, and the imaginary information 2 may be, for example, (b5, b6, b7, b8) T); Then, the main binary noise estimator obtains estimated noise information according to the real information 2 and the imaginary information 2. The estimated noise information can be expressed as (a5+b5i, a6+b6i, a7+b7i, a8+b8i) T .
[0096] S202: Remove the first noise information from the information to be decoded to obtain denoised information.
[0097] S203: Decode the denoised information to obtain a decoding result of the information to be decoded.
[0098] As an example, S202 may include: subtracting the first noise information from the information to be decoded to obtain denoised information, where the value of each element in the denoised information is the value of the element at the corresponding position in the information to be decoded minus the value of the element at the corresponding position in the first noise information, and the denoised information, the information to be decoded and the first noise information have the same scale, for example, all are 4*1 matrices.
[0099] As an example, S203 may include: inputting the denoising information into a classification-based deep neural network decoder, and the classification-based deep neural network decoder outputting a decoding result.
[0100] As another example, in order to further reduce the computational complexity of decoding, S203 may include: determining a decoding result according to the denoising information and the first binarization decoder. For example, determining the decoding result according to the denoising information and the first binarization decoder may include: obtaining fifth soft information based on the denoising information; inputting the fifth soft information into the first binarization decoder, and the first binarization decoder outputting sixth soft information; and determining the decoding result based on the sixth soft information.
[0101] It should be noted that the binary decoder in the embodiment of the present application can refer to the relevant description of method 100.
[0102] It can be seen that through this method 200, the binary noise estimator designed based on BNN performs noise identification on the information to be decoded, fully utilizing the low computational complexity of BNN, and can occupy fewer storage and processing resources to complete the identification of noise in the information to be decoded. In the decoding scheme that ensures decoding accuracy, the computational complexity of the decoding process is reduced.
[0103] The embodiment of the present application also provides a decoding method based on a binary decoder + binary noise estimator architecture. Figure 6 The decoding architecture shown is Figure 7 Another decoding method 300 provided by an embodiment of the present application is illustrated.
[0104] like Figure 6As shown, the decoding architecture in the communication device may include: a binary decoder 0, a binary noise estimator 50, a binary decoder 1, a binary decoder 2, ..., a binary decoder m and an aggregation module 60, where m may be an integer greater than or equal to 1. Among them, the binary decoder 0, as the decoder used for the initial decoding, may correspond to the second binary decoder in the above-mentioned method 100 and method 200. The binary decoder 1 to the binary decoder m may correspond to the first binary decoder or the third binary decoder in the above-mentioned method 100 and method 200. The binary noise estimator 50 may correspond to the binary noise estimator 50 in method 200. The aggregation module 60 is used to aggregate the output results of multiple binary decoders. The size relationship and specific values of m and n are not limited in the embodiments of the present application, and can be flexibly designed based on requirements.
[0105] like Figure 7 As shown, the method 300 provided in the embodiment of the present application may include, for example, the following S301 to S305:
[0106] S301, the communication device inputs the information to be decoded Y into the binary decoder 0, and the binary decoder 0 outputs the initial decoding result Y0.
[0107] S302: The communication device calculates the difference between Y and Y0 as initial noise information N0.
[0108] S303, the communication device inputs Y and N0 to the binary noise estimator 50, and the binary noise estimator 50 outputs estimated noise information N1, colored noise information C1, colored noise information C2, ..., colored noise information Cn.
[0109] As an example, S303 may include: inputting N0 into the main binarization noise estimation module 51, and the main binarization noise estimation module 51 outputs estimated noise information N1; calculating the joint information L of Y and N0; inputting L into the secondary binarization noise estimation module 521, the secondary binarization noise estimation module 522, ..., the secondary binarization noise estimation module 52n, and the secondary binarization noise estimation module 521 to the secondary binarization noise estimation module 52n respectively output colored noise information C1 to Cn.
[0110] S304, the communication device inputs the difference between Y and any number of combinations of (N1, C1, C2, ..., Cn) to the corresponding binary decoder among binary decoders 1 to m to obtain the corresponding output.
[0111] In specific implementation, the number of binary decoders and the function of each binary decoder can be designed based on the requirements of the decoding architecture (such as the accuracy requirements of decoding, the computational complexity limits of decoding, etc.). Binarized decoders with different functions correspond to different inputs, and the connection relationship in the decoding architecture can also be adaptively adjusted. The function of the binary decoder can be determined according to the input requirements of the binary decoder. For example, the input of the binary decoder 1 is (Y-N1), and the input of the binary decoder 2 is (Y-(N1+C1)). Then, it can be considered that the functions of the binary decoder 1 and the binary decoder 2 are different.
[0112] Figure 6 A possible implementation of S304 is shown as an example. Figure 6 As shown, S304 may include: inputting (Y-N1) into binary decoder 1, and binary decoder 1 outputs Y1; inputting (Y-(N1+C1)) into binary decoder 2, and binary decoder 2 outputs Y2; inputting (Y-(N1+C2)) into binary decoder 3, and binary decoder 3 outputs Y3; inputting (Y-(N1+Cn)) into binary decoder m, and binary decoder m outputs Ym.
[0113] The input of the binary decoder can be any one of N1, C1, C2, ..., Cn, or the sum of any multiple of N1, C1, C2, ..., Cn.
[0114] It should be noted that the inputs of the binary decoders 1 to m may also include lateral information from other binary decoders. Figure 6 Taking the scenario shown as an example, the input of the binary decoder 1 may also include lateral information a21 from the binary decoder 2, the input of the binary decoder 2 may also include lateral information a12 from the binary decoder 1 and lateral information a32 from the binary decoder 3, and the input of the binary decoder 3 may also include lateral information a23 from the binary decoder 2.
[0115] S305 , the communication device aggregates the outputs of the binary decoders 1 to m using the aggregation module 60 to obtain a decoding result X.
[0116] The aggregation module 60 may include an aggregation algorithm or aggregation rule. S305 may include, for example: using the aggregation algorithm or aggregation rule indicated in the aggregation module 60 to aggregate Y1-Ym to obtain X. The aggregation rule may be, for example, summation, weighted summation, or weighted average. The aggregation algorithm may be any algorithm that can achieve the purpose of aggregating multiple information.
[0117] It should be noted that in the embodiments of the present application, it is not explicitly stated whether each information is soft information or a codeword signal, nor is the process of converting each codeword signal into soft information and the process of obtaining a decoding result by judging the soft information described in detail. In the specific implementation, the soft information and the codeword signal can be converted to each other based on the characteristics of information processing in each part of the decoding architecture, and the embodiments of the present application do not limit the mutual conversion between soft information and codeword signals. For example, the binary decoder 0 needs to process soft information. In one case, Y can be understood as the codeword signal corresponding to the information to be decoded. Then, before the codeword signal Y is input into the binary decoder 0, Y can be converted into the corresponding soft information, and then the converted soft information is input into the binary decoder 0; in another case, Y can be directly the soft information corresponding to the information to be decoded, so Y can be directly input into the binary decoder 0.
[0118] It can be seen that through this method 300, the decoding of the information to be decoded is completed by combining the binarization noise estimator and the binarization decoder. The binarization decoder can effectively reduce the amount of calculation in the decoding process, reduce the hardware energy consumption, and realize decoding with low computational complexity; multiple binarization decoders decode in parallel, which can improve the expression ability of the decoding architecture and ensure high decoding accuracy while reducing the decoding computational complexity; the binarization noise estimator can identify and remove various noise information in the information to be decoded while consuming less computing resources and storage resources, thereby improving the decoding accuracy. In other words, in the case of increasing amount of information transmitted, under the condition of limited storage and computing resources, the binarization noise estimator and the binarization decoder can be designed based on BNN in the embodiment of the present application, and decoding is performed based on the binarization noise estimator and at least one binarization decoder, thereby realizing decoding with low computational complexity and high accuracy.
[0119] Taking the decoding process of LDPC code in different scenarios as an example, the scenario implementation scheme of the method provided in the embodiment of the present application is introduced.
[0120] The method provided in the embodiment of the present application can be applied in the wireless communication scenario of the park. In the wireless communication scenario of the park, it has the following characteristics: Feature 1, the transmission performance of the long code of the LDPC code in the park is better than that of the short code; Feature 2, since there are usually a variety of electronic devices in the park, the different electronic devices themselves and the mutual interference between them will cause the LDPC code in the park to be mixed with a large amount of colored noise during the transmission process. Then, in order to overcome the problems of high computational complexity and low decoding accuracy due to the lack of comprehensive consideration of denoising in the current deep neural network decoding technology in the decoding of the park LDPC code, the method provided in the embodiment of the present application can be used to decode the LDPC code of the park.
[0121] As an example, the embodiments of the present application are applicable to noisy long code data transmission based on LDPC codes in wireless communication scenarios. Figure 8 FIG. 4 shows a schematic diagram of noisy long code data transmission based on LDPC code in a wireless communication scenario, such as Figure 8 As shown, the noisy long code data transmission process of the LDPC code may include: the information to be decoded including the colored noise interference is sequentially passed through low noise amplifier 801, down conversion 802, analog to digital converter (Analog To Digital Converter, ADC) 803, synchronization processing 804, serial to parallel conversion 805, fast Fourier transform (fast Fourier transform, FFT) 806, channel estimation and equalization 807, deinterleaving 808, binary decoder 809, mapping 810, binary noise estimator 811, demodulation 812, binary decoder 813 and descrambler 814 to obtain the decoding result of the information to be decoded, and the input of the binary noise estimator 811 includes the output of the down conversion 802 in addition to the output of the mapping 810. The above processing process can occur in a receiver of a wireless communication system.
[0122] It should be noted that Figure 8 In the scenario shown, the signal received by the receiver of the wireless communication system has two significant characteristics: noisy and long code. Based on this, in order to overcome the problem of high decoding calculation complexity and low decoding accuracy of the current traditional decoding scheme, the decoding part of the park's LDPC code on the receiver of the wireless communication system can use the binary decoder and binary noise estimator designed in the embodiment of the present application to replace the current traditional decoder.
[0123] The method provided in the embodiment of the present application can be applied to digital hardware devices to improve the decoding performance of digital hardware devices and reduce the requirements for hardware specifications. Digital hardware devices include but are not limited to: FPGA, XILINX, ALTERA, etc. The embodiment of the present application takes the deployment of a binary noise estimator and a binary decoder in an FPGA as an example for explanation. In the FPGA, the binary decoder and the binary noise estimator complete the decoding by interacting with the storage unit, control module, etc. in the FPGA.
[0124] As an example, the embodiment of the present application is applicable to decoding of LDPC codes on FPGA. Fig. 9 A schematic diagram of LDPC decoding on an FPGA is shown, Fig. 9As shown, the decoding process of the LDPC code on the FPGA may include: the information to be decoded is stored in the initial decoding data storage unit 902 through the denoising module 901, at this time, the initial decoding data storage unit 902 stores the chaotic and undecoded codeword information to be decoded, and the initial decoding data storage unit 902 passes the codeword to be decoded through the initialization module 903, the address update module 904 and the weight bias calculation module 905 respectively; the initialization module 903 inputs the output probability soft information into the binarization decoder 906, the binarization decoder 906 inputs the output decoding result into the decision module 907, the binarization decoder 906 inputs the output loss function data into the weight bias calculation module 905, and the input of the binarization decoder 906 also includes the binarization decoding (BD) network deployment address output by the address update module 904 and the trigger condition of reaching the maximum number of iterations output by the control module 908; the address update module 904 also outputs the binarization noise estimation (Binarized The address of the network deployment of the network is input into the binary noise estimator 909, and the address update module 904 starts and ends the calculation under the control of the control module 908. The address update module 904 also provides the function of reading data for the random access memory (RAM) 910 and receives the return data of the RAM 910; the weight bias calculation module 905 also performs the operation of writing address and data to the RAM 910, and the weight bias calculation module 905 starts and ends the calculation under the control of the control module 908. The decision module 907 transmits the output codeword to the information bit codeword register 911, and the decision module 907 also transmits the output codeword to the mapping module 912, the complex / real conversion module 913, the binary noise estimator 909 and the real / complex conversion module 914 for processing, and reaches the denoising module 901. The binary noise estimator 909 starts and ends the calculation under the control of the control module 908.
[0125] Among them, the binary decoder 906 is controlled by the control module 908 and receives different information, which can correspond to the initial decoding and at least one decoding process of the denoised information in the decoding method provided in the embodiment of the present application. During the decoding process, the control module 908 is used to manage the flow of decoding-related information, for example, according to the address information transmitted by the address update module 904 and the weight information transmitted by the weight bias calculation module 905, the parameters and information of the binary decoder 906 are updated. Fig. 9The step of transmitting the loss function data from the binarization decoder 906 to the weight bias calculation module 905 is optional, and the loss function data is used to feed back the correction weight. The decoding related information involved in the decoding process can be temporarily stored in registers such as RAM. It should be noted that the binarization noise estimator 909 is also an optional device in the FPGA. The binarization noise estimator 909 is embedded in the FPGA, and the decoding scheme provided in the embodiment of the present application is implemented in combination with the binarization decoder 906, which can achieve better decoding results.
[0126] As an example, in the embodiment of the present application, since the circuit structure of the FPGA can be used to construct the adder and memory in the BNN, the register unit in the FPGA can also achieve the effect of the adder and memory. Therefore, the circuit structure and components of the FPGA can be used to build a binary decoder and a binary noise estimator, and the FPGA circuit structure is adjusted according to the technical solution corresponding to the embodiment of the present application to combine into a decoding architecture based on the binary decoder and the binary noise estimator. In this way, through the adjustment and design of the results in the FPGA itself, the decoding architecture provided by the embodiment of the present application is introduced into the FPGA, so that the FPGA has the function of realizing low computational complexity decoding.
[0127] As another example, in the embodiments of the present application, a binary decoder and a binary noise estimator can also be constructed by other means, and the designed binary decoder and binary noise estimator can be externally connected to the FPGA. Through the external binary decoder and binary noise estimator, low computational complexity decoding of the LDPC code on the FPGA can be achieved.
[0128] It can be seen that the decoding method provided in the embodiment of the present application can be applied to the decoding of LDPC codes in different application scenarios.
[0129] It should be noted that the decoding method provided in the embodiment of the present application is also applicable to decoding of other types of information to be decoded, for example, decoding of turbo codes or polar codes. In specific implementation, the decoder in the decoding architecture corresponding to other types of information to be decoded can be replaced with the binary decoder in the embodiment of the present application, and the binary noise estimator in the embodiment of the present application can be added to the decoding architecture corresponding to other types of information to be decoded to achieve decoding with lower computational complexity.
[0130] Accordingly, the embodiment of the present application further provides a communication device 1000, such as Fig.10 The communication device 1000 may include: a first obtaining unit 1001, a first decoding unit 1002 and a first determining unit 1003. Wherein:
[0131] The first obtaining unit 1001 is configured to obtain first soft information based on the information to be decoded. The first obtaining unit 1001 may be executed Figure 2 S101 shown.
[0132] The first decoding unit 1002 is used to decode the first soft information based on a first binary decoder to obtain second soft information, wherein the first binary decoder is a decoder constructed based on BNN. The first decoding unit 1002 can perform Figure 2 S102 shown.
[0133] The first determining unit 1003 is configured to determine a decoding result of the information to be decoded based on the second soft information. The first determining unit 1003 may be executed Figure 2 S103 shown.
[0134] Optionally, the first obtaining unit 1001 is specifically configured to: convert each character included in the information to be decoded into a corresponding probability according to a preset corresponding relationship; and obtain the first soft information based on the probability corresponding to each character included in the information to be decoded.
[0135] Optionally, the first obtaining unit 1001 is specifically used to: remove first noise information from the information to be decoded to obtain first denoised information; convert each character included in the first denoised information into a corresponding probability according to a preset correspondence; and obtain the first soft information based on the probability corresponding to each character included in the first denoised information.
[0136] Optionally, the communication device 1000 may further include a second obtaining unit. The second obtaining unit is configured to obtain the first noise information of the information to be decoded.
[0137] Optionally, the second obtaining unit is specifically used to obtain the first noise information based on a binarized noise estimator and the information to be decoded, wherein the binarized noise estimator is a model constructed based on BNN and used to determine noise information.
[0138] Optionally, the second obtaining unit may include: a first determining subunit and a first obtaining subunit. The first determining subunit is used to determine the initial noise information based on the second binarization decoder and the information to be decoded, wherein the second binarization decoder is a decoder constructed based on BNN; and the first obtaining subunit is used to obtain the first noise information according to the initial noise information and the binarization noise estimator.
[0139] Optionally, the first obtaining subunit is specifically configured to obtain the first noise information according to the initial noise information, the information to be decoded and the binarization noise estimator.
[0140] Optionally, the first obtaining subunit is specifically used to: determine the estimated noise of the information to be decoded itself based on the initial noise information and the main binarization noise estimation module in the binarization noise estimator; obtain joint information based on the initial noise information and the information to be decoded; determine at least one first colored noise based on the joint information and at least one first binarization noise estimation module in the binarization noise estimator, the first binarization noise estimation module corresponds to the first colored noise; wherein the first noise information includes the estimated noise and the at least one first colored noise.
[0141] Optionally, the binarization noise estimator also includes a second binarization noise estimation module, and the communication device 1000 may also include: a second determination unit, a third determination unit, a third acquisition unit and a second decoding unit. Among them, the second determination unit is used to determine the second colored noise based on the joint information and the second binarization noise estimation module; the third determination unit is used to determine the second noise information based on the estimated noise and the second colored noise; the third acquisition unit is used to remove the second noise information from the information to be decoded to obtain second denoised information; the second decoding unit is used to decode the third soft information corresponding to the second denoised information based on the third binarization decoder to obtain fourth soft information, and the third binarization decoder is a decoder constructed based on BNN. Then, the first determination unit 1003 is specifically used to: determine the decoding result of the information to be decoded based on the aggregation result of the second soft information and the fourth soft information. This implementation method can be found in Figure 7 A related description of method 300 is shown.
[0142] Optionally, the input of the first binarization decoder also includes first lateral information of the third binarization decoder, and / or the input of the third binarization decoder also includes second lateral information of the first binarization decoder, the first lateral information is used to reflect the characteristics of the third binarization decoder, and the second lateral information is used to reflect the characteristics of the first binarization decoder.
[0143] Optionally, the type of the information to be decoded includes any one of the following types: LDPC code, turbo code or polar code.
[0144] Optionally, the communication device may be a wireless communication device or a digital communication device.
[0145] It should be noted that various specific embodiments of the communication device 1000 can be found in Figure 2 The corresponding introduction of method 100 will not be repeated in this embodiment.
[0146] Accordingly, the embodiment of the present application further provides a communication device 1100, such as Fig.11 The communication device 1100 may include: an acquisition unit 1101, a denoising unit 1102 and a decoding unit 1103. Wherein:
[0147] The obtaining unit 1101 is configured to obtain first noise information corresponding to the information to be decoded based on a binary noise estimator, wherein the binary noise estimator is a model constructed based on a BNN and used to determine noise information. The obtaining unit 1101 may be executed Figure 4 S201 shown.
[0148] The denoising unit 1102 is configured to remove the first noise information from the information to be decoded to obtain denoised information. The denoising unit 1102 may perform Figure 4 S202 shown.
[0149] The decoding unit 1103 is used to decode the denoised information to obtain a decoding result of the information to be decoded. The decoding unit 1103 can perform Figure 4 S203 shown.
[0150] Optionally, the obtaining unit 1101 may include: a determining subunit and an obtaining subunit. The determining subunit is used to determine the initial noise information based on a first binarization decoder and the information to be decoded, wherein the first binarization decoder is a decoder constructed based on a BNN; and the obtaining subunit is used to obtain the first noise information according to the initial noise information and the binarization noise estimator.
[0151] Optionally, the obtaining subunit is specifically used to: determine the estimated noise of the information to be decoded itself based on the initial noise information and a main binarization noise estimation module in the binarization noise estimator; obtain joint information based on the initial noise information and the information to be decoded; determine at least one colored noise based on the joint information and at least one secondary binarization noise estimation module in the binarization noise estimator, the secondary binarization noise estimation module corresponding to the colored noise; wherein the first noise information includes the estimated noise and the at least one colored noise.
[0152] Optionally, the decoding unit 1103 is specifically used to: obtain first soft information based on the denoising information; decode the first soft information based on a second binarization decoder to obtain second soft information, where the second binarization decoder is a decoder constructed based on BNN; and determine the decoding result based on the second soft information.
[0153] Optionally, the type of the information to be decoded includes any one of the following types: LDPC code, turbo code or polar code.
[0154] Optionally, the communication device may be a wireless communication device or a digital communication device.
[0155] It should be noted that various specific embodiments of the communication device 1100 can be found in Figure 4 The corresponding introduction of method 200 will not be repeated in this embodiment.
[0156] See also Fig.12 , an embodiment of the present application provides a communication device 1200. The communication device 1200 may be an execution subject in any of the above embodiments. The communication device 1200 may implement the functions of the above embodiments. The communication device 1200 includes at least one processor 1201, a bus system 1202, a memory 1203, and at least one communication interface 1204.
[0157] The communication device 1200 is a hardware structure device that can be used to implement Fig.10 The functional modules in the communication device 1000 are shown. For example, those skilled in the art may think of Fig.10 The first obtaining unit 1001, the first decoding unit 1002 and the first determining unit 1003 in the communication device 1000 shown in the figure can be implemented by the at least one processor 1201 calling the code in the memory 1203. For another example, those skilled in the art can imagine Fig.11 The obtaining unit 1101 , the denoising unit 1102 and the decoding unit 1103 in the communication device 1100 shown can be implemented by the at least one processor 1201 calling the code in the memory 1203 .
[0158] Optionally, the communication device 1200 may be a network device or a control entity implementing an embodiment of the present application.
[0159] Optionally, the processor 1201 may be a general-purpose central processing unit (CPU), a network processor (NP), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the present application.
[0160] The bus system 1202 may include a path for transmitting information between the components.
[0161] The communication interface 1204 is used to communicate with other devices or communication networks.
[0162] The memory 1203 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may exist independently and be connected to the processor via a bus. The memory may also be integrated with the processor.
[0163] The memory 1203 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 1201. The processor 1201 is used to execute the application code stored in the memory 1203, so as to realize the functions in the method of the present application.
[0164] In a specific implementation, as an embodiment, the processor 1201 may include one or more CPUs, such as Fig.12 CPU0 and CPU1 in.
[0165] In a specific implementation, as an embodiment, the communication device 1200 may include multiple processors, such as Fig.12 1 and 1207. Each of these processors may be a single-CPU processor or a multi-CPU processor. A processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).
[0166] It should be understood that the communication devices or communication equipment in the above-mentioned various product forms respectively have any functions implemented by the execution subject in the above-mentioned method embodiment, which will not be repeated here.
[0167] The embodiment of the present application further provides a chip, including a processor and an interface circuit, the interface circuit is used to receive instructions and transmit them to the processor; the processor, for example, can be a specific implementation form of the message processing device in the embodiment of the present application, and can be used to execute the above method 100, method 200 or method 300. The processor is coupled to a memory, and the memory is used to store programs or instructions. When the program or instruction is executed by the processor, the chip system implements the method in any of the above method embodiments.
[0168] Optionally, the processor in the chip system may be one or more. The processor may be implemented by hardware or by software. When implemented by hardware, the processor may be a logic circuit, an integrated circuit, etc. When implemented by software, the processor may be a general-purpose processor implemented by reading software code stored in a memory.
[0169] Optionally, the memory in the chip system may be one or more. The memory may be integrated with the processor or may be separately provided with the processor, which is not limited in the present application. Exemplarily, the memory may be a non-transient processor, such as a read-only memory ROM, which may be integrated with the processor on the same chip or may be provided on different chips. The present application does not specifically limit the type of memory and the arrangement of the memory and the processor.
[0170] Exemplarily, the chip system can be a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD) or other integrated chips.
[0171] In addition, an embodiment of the present application further provides a storage medium, in which program code or instructions are stored. When the storage medium is run on a processor, the processor executes a method in any one of the implementation modes in the above embodiments.
[0172] In addition, an embodiment of the present application also provides a program product, which, when executed on a processor, enables the processor to execute a method of any one of the implementation modes of the aforementioned method 100, method 200 or method 300.
[0173] It should be understood that "determining B based on A" mentioned in the embodiments of the present application does not mean determining B only based on A, but B can also be determined based on A and / or other information.
[0174] It should be understood that the network architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.
[0175] In this application, ordinal numbers such as "1", "2", "3", "first", "second" and "third" are used to distinguish multiple objects but are not used to limit the order of multiple objects.
[0176] “A and / or B” mentioned in this application should be understood to include the following situations: only A, only B, or both A and B.
[0177] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.
[0178] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for system embodiments and device embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, in which the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0179] The above is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the present application, and these improvements and modifications should also be regarded as the protection scope of the present application.
Claims
1. A decoding method, characterized in that: include: Obtaining first soft information based on the information to be decoded; Decoding the first soft information based on a first binarization decoder to obtain second soft information, wherein the first binarization decoder is a decoder constructed based on a binarization neural network; Based on the second soft information, a decoding result of the information to be decoded is determined.
2. The method according to claim 1, characterized in that The obtaining of first soft information based on the information to be decoded includes: According to a preset corresponding relationship, each character included in the information to be decoded is converted into a corresponding probability; The first soft information is obtained based on the probability corresponding to each character included in the information to be decoded.
3. The method according to claim 1, characterized in that: The obtaining of first soft information based on the information to be decoded includes: Removing first noise information from the information to be decoded to obtain first denoised information; According to a preset corresponding relationship, convert each character included in the first denoising information into a corresponding probability; The first soft information is obtained based on the probability corresponding to each character included in the first denoising information.
4. The method according to claim 3, characterized in that The method further comprises: The first noise information of the information to be decoded is obtained.
5. The method according to claim 4, characterized in that The obtaining the first noise information of the information to be decoded includes: The first noise information is obtained based on a binarized noise estimator and the information to be decoded, wherein the binarized noise estimator is a model constructed based on a binarized neural network and used to determine noise information.
6. The method according to claim 5, characterized in that The step of obtaining the first noise information based on the binarized noise estimator and the information to be decoded includes: Determining initial noise information based on a second binarization decoder and the information to be decoded, wherein the second binarization decoder is a decoder constructed based on a binarization neural network; The first noise information is obtained according to the initial noise information and the binarization noise estimator.
7. The method according to claim 6, characterized in that The step of obtaining the first noise information according to the initial noise information and the binarized noise estimator includes: The first noise information is obtained according to the initial noise information, the information to be decoded and the binarization noise estimator.
8. The method according to claim 7, characterized in that The step of obtaining the first noise information according to the initial noise information, the information to be decoded and the binary noise estimator includes: Determining the estimated noise of the information to be decoded itself based on the initial noise information and a main binarization noise estimation module in the binarization noise estimator; Based on the initial noise information and the information to be decoded, obtaining joint information; Determine at least one first colored noise based on the joint information and at least one first binarization noise estimation module in the binarization noise estimator, the first binarization noise estimation module corresponding to the first colored noise; The first noise information includes the estimated noise and the at least one first colored noise.
9. The method according to claim 8, characterized in that The binarization noise estimator further includes a second binarization noise estimation module, and the method further includes: Determine a second colored noise based on the joint information and the second binarization noise estimation module; determining second noise information based on the estimated noise and the second colored noise; Removing the second noise information from the information to be decoded to obtain second denoised information; Decoding the third soft information corresponding to the second denoising information based on a third binarization decoder to obtain fourth soft information, wherein the third binarization decoder is a decoder constructed based on a binarization neural network; The determining, based on the second soft information, a decoding result of the information to be decoded includes: The decoding result of the information to be decoded is determined based on an aggregation result of the second soft information and the fourth soft information.
10. The method according to claim 9, characterized in that The input of the first binary decoder also includes the first lateral information of the third binary decoder, and / or the input of the third binary decoder also includes the second lateral information of the first binary decoder, the first lateral information is used to reflect the characteristics of the third binary decoder, and the second lateral information is used to reflect the characteristics of the first binary decoder.
11. The method according to any one of claims 1 to 10, characterized in that: The type of the information to be decoded includes any one of the following types: low-density parity check LDPC code, turbo code or polar code.
12. The method according to any one of claims 1 to 11, characterized in that: The method is applied to a wireless communication device or a digital communication device.
13. A decoding method, characterized in that: include: Obtaining first noise information corresponding to the information to be decoded based on a binarized noise estimator, wherein the binarized noise estimator is a model constructed based on a binarized neural network and used to determine noise information; Removing the first noise information from the information to be decoded to obtain denoised information; The denoised information is decoded to obtain a decoding result of the information to be decoded.
14. The method according to claim 13, characterized in that The step of obtaining the first noise information of the information to be decoded based on the binary noise estimator includes: Determining initial noise information based on a first binarization decoder and the information to be decoded, wherein the first binarization decoder is a decoder constructed based on a binarization neural network; The first noise information is obtained according to the initial noise information and the binarization noise estimator.
15. The method according to claim 14, characterized in that The step of obtaining the first noise information according to the initial noise information and the binarized noise estimator includes: Determining the estimated noise of the information to be decoded itself based on the initial noise information and a main binarization noise estimation module in the binarization noise estimator; Based on the initial noise information and the information to be decoded, obtaining joint information; determining at least one colored noise based on the joint information and at least one sub-binarized noise estimation module in the binarized noise estimator, the sub-binarized noise estimation module corresponding to the colored noise; The first noise information includes the estimated noise and the at least one colored noise.
16. The method according to any one of claims 13 to 15, characterized in that: The step of decoding the denoised information to obtain a decoding result of the information to be decoded includes: Obtaining first soft information based on the denoised information; Decoding the first soft information based on a second binarization decoder to obtain second soft information, wherein the second binarization decoder is a decoder constructed based on a binarization neural network; The decoding result is determined based on the second soft information.
17. The method according to any one of claims 13 to 16, characterized in that: The type of the information to be decoded includes any one of the following types: low-density parity check LDPC code, turbo code or polar code.
18. The method according to any one of claims 13 to 17, characterized in that: The method is applied to a wireless communication device or a digital communication device.
19. A communication device, characterized in that: The communication device comprises a processing unit; The processing unit is used to execute the method according to any one of claims 1 to 18.
20. A communication device, characterized in that: The communication device includes a processor and a memory; The processor is configured to execute instructions stored in the memory, so that the communication device executes the method according to any one of claims 1 to 18.
21. A storage medium, characterized in that: The storage medium stores instructions, and when the instructions are executed on the processor, the method according to any one of claims 1 to 18 is executed.
22. A program product, characterized in that The program product comprises a program, and when the program is run on a processor, the method according to any one of claims 1 to 18 is executed.
