Variable-length channel coding method with feedback, electronic equipment and storage medium

Abstract

The invention provides a variable-length channel encoding method with feedback, electronic equipment and a storage medium, and the method comprises the steps: an encoding module encodes a to-be-sent bit stream to obtain first encoding output information, and the first encoding output information is superposed with forward channel additive Gaussian noise information and is input into a decoding module; the decoding module determines the confidence information of each element in the received information, judges whether the confidence information is greater than a preset confidence threshold, decodes the element if the confidence information is greater than the preset confidence threshold, and inputs the feedback information of the decoding module and the additive Gaussian noise information of the feedback channel into the coding module; and the encoding module encodes the combined information, superposes the output information with the forward channel additive Gaussian noise information and inputs the superposed information into the decoding module again, the encoding module determines confidence information corresponding to each element of the information, and the output information is synthesized and output until the confidence information of all the elements is greater than a preset confidence threshold. According to the invention, each bit block is decoded according to the size of noise carried by different bit blocks, so that adaptive code rate adjustment is realized.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a variable-length channel coding method with feedback, an electronic device, and a storage medium. Background Art

[0002] In the field of wireless communications, channel coding and decoding technology is the key to ensuring signal transmission quality. In the channel coding method with feedback, the classic feedback channel model transmits data from the transmitter to the receiver through a memoryless noise channel, and the receiver provides real-time feedback to perform coding through the forward channel, which significantly improves the coding reliability under limited length. However, in this method, the feedback coding cannot dynamically adapt to the feedback process and optimize the performance under changing channel conditions and receiver states. Therefore, a channel coding method based on a deep learning model is proposed in the prior art, which realizes the parallel input of bit blocks, iteratively uses the enhanced downlink channel to obtain feedback information for channel coding, so that the feedback coding cannot dynamically adapt to the feedback process, thereby reducing the bit error rate.

[0003] However, the deep learning model proposed in the prior art is limited to fixed-length encoding, which limits its adaptability, and therefore cannot fully utilize feedback information, and has low flexibility. In addition, the method has poor performance in a high bit rate range. Summary of the invention

[0004] The purpose of the present application is to provide a variable-length channel coding method with feedback, an electronic device and a storage medium to address the deficiencies in the prior art, so as to solve the problems of poor flexibility and adaptability in the prior art.

[0005] To achieve the above objectives, the technical solutions adopted in this application are as follows:

[0006] In a first aspect, the present application provides a variable length channel coding method with feedback, the method is applied to a variable length channel coding system with feedback, the variable length channel coding system with feedback comprises an encoding module and a decoding module, the transmission channel from the encoding module to the decoding module is a forward channel, and the transmission channel from the decoding module to the encoding module is a feedback channel, the method comprises:

[0007] The encoding module encodes the bit stream to be transmitted to obtain first encoding output information, and uses the sum of the first encoding output information and the first forward channel additive Gaussian noise information of the forward channel as the first decoding input information, wherein the bit stream to be transmitted includes multiple bit blocks;

[0008] The decoding module performs confidence processing on the first decoded input information to determine confidence information corresponding to each element in the first decoded input information, wherein each element corresponds to a bit block in the bit stream to be sent;

[0009] The decoding module sequentially determines whether the confidence information corresponding to each of the elements is greater than a preset confidence threshold, decodes the elements whose confidence information is greater than the preset confidence threshold, and uses the sum of the first decoded input information and the first feedback channel additive Gaussian noise information of the feedback channel as the second encoded input information;

[0010] The encoding module increases the encoding round by one, and performs encoding according to the second encoding input information, the first decoded identification information, the first encoding output information, and the bit stream to be sent to obtain second encoding output information, and takes the sum of the second encoding output information and the second forward channel additive Gaussian noise information of the forward channel as the second decoding input information, wherein the first decoded identification information is used to indicate the elements in the first decoding input information that have been successfully decoded;

[0011] The decoding module decodes according to the confidence information corresponding to each element in the first decoding input information and the second decoding input information, determines the confidence information corresponding to each element in the second decoding input information, and if the confidence information corresponding to all the elements is greater than a preset confidence threshold, the decoded elements are synthesized into output information and output.

[0012] Optionally, the process of generating the first decoded identification information includes:

[0013] The decoding module constructs initial identification information and initializes the value of each element in the initial identification information to a preset initial value, wherein the length of the initial identification information is the same as the length of the first decoding input information;

[0014] The decoding module traverses each element in the first decoding input information, and for the current element traversed, if the confidence information of the current element is greater than a preset confidence threshold, the value of the element at the position corresponding to the current element in the initial identification information is modified to a first value; otherwise, the value of the element at the position corresponding to the current element in the initial identification information is modified to a second value, wherein the first value is used to identify that it has been decoded, and the second value is used to identify that it has not been decoded.

[0015] Optionally, encoding is performed according to the second encoding input information, the first decoded identification information, the first encoding output information and the to-be-sent bit stream to obtain the second encoding output information, and the method includes:

[0016] Calculating a difference between the second coded input information and the first coded output information as first difference information;

[0017] Determine second combination information according to the position of the successfully decoded element in the first decoded identification information, the first difference information, the to-be-transmitted bit stream, and the first coded output information;

[0018] The second combined information is encoded to obtain second encoded output information.

[0019] Optionally, after determining the confidence information corresponding to each element in the second decoding input information, the method further includes:

[0020] If there is an element whose confidence information is less than or equal to the preset confidence threshold, it is determined whether the encoding round reaches the preset encoding round threshold. If so, each element in the second decoding input information is decoded and synthesized into output information and then output; if not, the second encoding input information is generated and encoding continues.

[0021] Optionally, the encoding module includes: an encoding input unit, an encoding feature extraction unit, a first encoding self-attention unit, a second encoding self-attention unit, an encoding feedforward network unit, an encoding head unit, and an encoding output unit;

[0022] The step of encoding the second combined information to obtain second coded output information includes:

[0023] The encoding input unit inputs the second combination information into the encoding feature extraction unit;

[0024] The encoding feature extraction unit determines a target extraction structure according to the encoding round, and uses the target extraction structure to extract features from the second combination information, and inputs the extracted encoding features into the first encoding self-attention unit;

[0025] The first encoding self-attention unit and the second encoding self-attention unit sequentially perform feature fusion on the extracted encoding features to obtain fused encoding features, and input the fused encoding features into the encoding feedforward network unit;

[0026] The encoding feedforward network unit performs feature enhancement on the fused encoding features to obtain enhanced encoding features, and inputs the enhanced encoding features into the encoding head unit;

[0027] The encoding head unit compresses the enhanced encoding features based on dual-objective optimization control to obtain compressed encoding features, and inputs the compressed encoding features into the encoding output unit;

[0028] The encoding output unit converts the compressed encoding feature into the second encoding output information.

[0029] Optionally, determining the target extraction structure according to the encoding round includes:

[0030] Determining whether the encoding round is greater than or equal to a preset structure round threshold;

[0031] If not, a perceptual structure consisting of a preset number of layers of linear layers based on a preset activation function is used as the target extraction structure;

[0032] If so, a linear layer with a preset activation function is added on the basis of the perception structure composed of a preset number of layers of linear layers based on the preset activation function to obtain a target extraction structure.

[0033] Optionally, the decoding module includes: a decoding input unit, a decoding feature extraction unit, a first decoding self-attention unit, a second decoding self-attention unit, a third decoding self-attention unit, a decoding feedforward network unit, a decoding head unit, and a decoding output unit;

[0034] The performing confidence processing on the first decoded input information to determine the confidence information corresponding to each element in the first decoded input information includes:

[0035] After receiving the first decoding input information, the decoding input unit inputs the first decoding input information into the decoding feature extraction unit;

[0036] The decoding feature extraction unit uses the target extraction structure to extract features from the first decoding input information, and inputs the extracted decoding features into the first decoding self-attention unit;

[0037] The first decoding self-attention unit, the second decoding self-attention unit, and the third decoding self-attention unit sequentially perform feature fusion on the extracted decoding features to obtain fused decoding features, and input the fused decoding features into the decoding feedforward network unit;

[0038] The decoding feedforward network unit performs feature enhancement on the fused decoding features to obtain enhanced decoding features, and inputs the enhanced decoding features into the decoding head unit;

[0039] The decoding head unit determines the confidence information of each element in the enhanced decoding feature based on the dual-objective optimization control, and outputs the confidence information corresponding to each element through the decoding output unit.

[0040] Optionally, the training process of the encoding module includes:

[0041] The initial encoding module performs encoding for multiple encoding rounds based on the sample information, and determines the loss value of each encoding round according to the confidence information corresponding to the encoding round at the end of each encoding round;

[0042] Determining the loss value of the initial encoding module according to the loss values ​​of all encoding rounds;

[0043] The initial encoding module is iteratively updated according to the loss value of the initial encoding module, and the initial encoding module at the end of the iteration is used as the encoding module.

[0044] In a second aspect, the present application provides a variable length channel coding system with feedback, the variable length channel coding system with feedback comprising an encoding module and a decoding module, the encoding module and the decoding module are used to execute the steps of the variable length channel coding method with feedback as described in the first aspect.

[0045] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the variable-length channel coding method with feedback as described in the first aspect are executed.

[0046] The beneficial effects of the present application are as follows: the encoding module encodes the bit stream to be sent to obtain the first encoding output information, and the sum of the first encoding output information and the first forward channel additive Gaussian noise information of the forward channel is used as the second encoding input information, the decoding module performs confidence processing on the first encoding input information, determines the confidence information corresponding to each element in the first decoding input information, judges in turn whether the confidence information corresponding to each element is greater than the preset confidence threshold, decodes the elements whose confidence information is greater than the preset confidence threshold, and the sum of the first decoding input information and the first feedback channel additive Gaussian noise information of the feedback channel is used as the second encoding input information, the encoding module adds one to the encoding round, and encodes according to the second encoding input information, the first decoded identification information, the first encoding output information and the bit stream to be sent to obtain the second encoding output information. In this embodiment, the second encoding input information, the first decoded identification information, the first encoding output information and the bit stream to be sent are encoded, so as to adjust the forward encoding strategy of the transmitting end using the feedback information. Then, the sum of the second coded output information and the second forward channel additive Gaussian noise information of the forward channel is used as the second decoding input information. The decoding module performs confidence processing according to the confidence information corresponding to each element in the first decoding input information and the second decoding input information, and determines the confidence information corresponding to each element in the second decoding input information. The decoding module can decode each bit block according to the noise size carried by the bit blocks corresponding to different elements, thereby realizing adaptive bit rate adjustment of the bit block granularity. If the confidence information corresponding to all elements is greater than the preset confidence threshold, the decoded elements are synthesized into output information and output. This embodiment iterates multiple rounds of communication through the forward channel and the downlink enhanced feedback channel between the encoding module and the decoding module, thereby improving the anti-noise performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 This is a schematic diagram of an application scenario of a variable-length channel coding method with feedback provided in an embodiment of the present application;

[0049] Figure 2 is a flow chart of a variable length channel coding method with feedback provided in this embodiment;

[0050] Figure 3 It is a schematic diagram of the architecture of a variable length channel coding system with feedback provided in an embodiment of the present application;

[0051] Figure 4 This is a schematic diagram of a process for obtaining second coded output information provided by an embodiment of the present application;

[0052] Figure 5 is a structural diagram of a coding module provided in an embodiment of the present application;

[0053] Figure 6 is another schematic diagram of a process for obtaining second coded output information provided by an embodiment of the present application;

[0054] Figure 7 It is a schematic diagram of a process for determining a target extraction structure provided by an embodiment of the present application;

[0055] Figure 8 is a structural diagram of a decoding module provided in an embodiment of the present application;

[0056] Fig. 9 It is a schematic diagram of a process for determining confidence information corresponding to each element in the first decoding input information provided by an embodiment of the present application;

[0057] Fig.10 It is a schematic diagram of a flow chart of encoding module training provided in an embodiment of the present application;

[0058] Fig.11 It is a schematic diagram comparing the bit error rates of a channel coding method with feedback and a variable length channel coding method with feedback under different channel conditions at different code rates provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0060] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0061] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0062] Among the channel coding methods with feedback, the feedback coding in the classic feedback channel model cannot dynamically adapt to the feedback process and optimize the performance under changing channel conditions and receiver states. The channel coding method based on the deep learning model is limited to fixed-length coding, which limits its adaptability and therefore cannot fully utilize the feedback information, and has low flexibility. In addition, the performance of this method is poor in the high bit rate range.

[0063] Based on this, the present application proposes a variable-length channel coding method with feedback, which is applied to the encoding module and the decoding module in the variable-length channel coding system with feedback. After the encoding module encodes the received information, the sum of the obtained encoded output information and the forward channel additive Gaussian noise information of the forward channel is used as the decoding input information. The decoding module performs confidence processing on the decoded input information to obtain the confidence information corresponding to each element in the decoded input information, and judges whether the confidence information of each element is greater than a preset confidence threshold in turn. If so, the element is decoded. Otherwise, the decoded input information and the first decoded identification information are summed with the noise information of the feedback channel as the encoding input information. Since then, the first round of encoding and decoding is completed, and the second round of encoding and decoding is started: the encoding module encodes according to the encoding input information, the first decoded identification information and the encoding output information of the previous round to obtain the encoding output information of the current round, and then the sum of the encoding output information of the current round and the noise information of the forward channel is used as the encoding input information of the current round. The encoding module performs confidence processing on the encoding input information to determine the confidence information corresponding to each element in each encoding input information, and continues to judge whether the confidence information is greater than the preset confidence threshold. If the confidence information corresponding to all elements is greater than the preset confidence threshold, all decoded elements are synthesized into output information and output. This application performs confidence processing / decoding on the elements based on the comparison result of the confidence information of each element in each decoding input information and the preset execution threshold, and feeds back the decoded input information to the encoding module based on the element position corresponding to the confidence information not greater than the preset execution threshold to continue encoding and decoding, so as to decode each bit block according to the noise size carried by different bit blocks, thereby realizing adaptive bit rate adjustment.

[0064] Next, refer to Figure 1 The application scenarios of the variable-length channel coding method with feedback in this application are introduced. Figure 1 Schematic diagram of an application scenario of a variable length channel coding method with feedback provided in an embodiment of the present application. Figure 1 As shown, the variable length channel coding method with feedback is applied to a variable length channel coding system with feedback, and the variable length channel coding system with feedback includes a coding module and a decoding module. The coding module can be deployed in a signal transmitting terminal, and the decoding module can be deployed in a base station, and the benchmark can realize signal reception and forwarding. The transmission channel from the terminal to the base station is a forward channel, and the transmission channel from the base station to the terminal is a feedback channel. This method can be applied to environments with poor channel conditions such as mines, and the signal coverage can be expanded by enhancing the downlink feedback method. It can also be applied to occasions such as mountains, forests and high-rise buildings where it is difficult to replace equipment, thereby providing low-power communication.

[0065] Next, refer to Figure 2The specific implementation process of the variable length channel coding method with feedback is introduced. Figure 2 It is a flowchart of a variable-length channel coding method with feedback provided in this embodiment.

[0066] S201, the encoding module encodes the bit stream to be sent to obtain first encoding output information, and uses the sum of the first encoding output information and the first forward channel additive Gaussian noise information of the forward channel as first decoding input information, and the bit stream to be sent includes multiple bit blocks.

[0067] Optionally, the bit stream to be sent may include multiple bit blocks, and the bit stream to be sent is encoded to obtain first coded output information, wherein the first coded output information includes information corresponding to multiple bits. It is worth noting that the bit stream to be sent is information received by the encoding module in the first encoding round. The encoding round is the communication round, and encoding and decoding are performed once in one communication round.

[0068] Specifically, the encoding module encodes the bit stream to be transmitted by the following formula (1):

[0069] x 1 =φ(b), b∈(0,1) m (1)

[0070] Among them, x 1 represents the first coded output information obtained by encoding in the first encoding round, wherein the first coded output information includes elements corresponding to each bit block, b represents the bit stream to be sent, φ represents the mathematical model of the encoding module, and m is the size of a single bit block.

[0071] It is worth mentioning that in formula (1) and the following formulas, the superscript symbol indicates the encoding round, and the subscript symbol indicates the corresponding position of the element in the information.

[0072] Optionally, the signal transmission environment is different, and the channel noise is also different. When the transmission environment is poor, the channel noise is greater. Before the signal transmission process, the noise of the forward channel and the feedback channel can be detected in advance to obtain the additive Gaussian noise information of the forward channel and the additive Gaussian noise information of the feedback channel. It is worth mentioning that in each coding round, the additive Gaussian noise information of the forward channel and the additive Gaussian noise information of the feedback channel may be different, but the variance of the additive Gaussian noise of the forward channel and the variance of the additive Gaussian noise of the feedback channel are the same.

[0073] Optionally, the sum of the first coded output information and the first forward channel additive Gaussian noise information of the forward channel as the first decoding input information can be expressed by the following formula (2):

[0074]

[0075] Among them, y 1 represents the first decoded input information, x 1 Represents the first coded output information, n 1 represents the additive Gaussian noise information of the first forward channel, n 1 obey The normal distribution of ff represents the variance of the additive Gaussian noise in the forward channel.

[0076] S202: The decoding module performs confidence processing on the first decoded input information to determine confidence information corresponding to each element in the first decoded input information, wherein each element corresponds to a bit block in the bit stream to be sent.

[0077] Optionally, during the first encoding round, an equal initialization strategy may be used to generate confidence information corresponding to each element.

[0078] S203, the decoding module determines in turn whether the confidence information corresponding to each element is greater than a preset confidence threshold, decodes the elements whose confidence information is greater than the preset confidence threshold, and uses the sum of the first decoded input information and the first feedback channel additive Gaussian noise information of the feedback channel as the second encoded input information.

[0079] Optionally, the decoding module obtains a preset confidence threshold in advance, and then determines in turn whether the confidence information corresponding to each element is greater than the preset confidence threshold. Specifically, it determines whether the element with the highest probability in the confidence information corresponding to each element is greater than the preset confidence threshold.

[0080] Optionally, after decoding the elements whose confidence elements are greater than a preset confidence threshold, the decoding module determines whether there is an element whose confidence information is not greater than the preset confidence threshold, and if so, the sum of the first decoding input information and the first feedback channel additive Gaussian noise information of the feedback channel is used as the second encoding input information. The first decoding input information can be used as feedback information of the first decoding module.

[0081] Optionally, the decoding module may also generate first decoded identification information according to the position of the decoded element in the first decoding input information, and send the first decoded identification information to the encoding module.

[0082] Optionally, the sum of the first decoded input information and the first feedback channel additive Gaussian noise information of the feedback channel as the second encoded input information can be expressed by the following formula (3):

[0083]

[0084] in, Represents the second coded input information, y1 represents the first decoding input information, represents the additive Gaussian noise information of the first feedback channel, obey The normal distribution of fb represents the variance of the additive Gaussian noise of the feedback channel.

[0085] S204, the encoding module increases the encoding round by one, and encodes according to the second encoding input information, the first decoded identification information, the first encoding output information and the bit stream to be sent to obtain the second encoding output information, and takes the sum of the second encoding output information and the second forward channel additive Gaussian noise information of the forward channel as the second decoding input information, wherein the first decoded identification information is used to indicate the elements in the first decoding input information that have been successfully decoded.

[0086] Optionally, after receiving the second coding input information, the coding module increases the coding round by 1. It is worth noting that the coding module increases the coding round by 1 each time it receives coding input information sent by a decoding module.

[0087] Optionally, in the second encoding round, the encoding module encodes according to the second encoding input information, the first decoded identification information, the first encoding output information and the bit stream to be sent to obtain the second encoding output information. In the Tth encoding round, the encoding module encodes according to the Tth encoding output information, all encoding output information of the previous encoding round, the Tth decoded identification information, all encoding output information of the previous encoding round and the bit stream to be sent to obtain the second encoding output information. The encoding process of the τth encoding round is introduced below through formula (4) and formula (5). Wherein, τ is a number greater than 1 and less than T.

[0088]

[0089] x τ =φ(q τ )(5)

[0090] in, Indicates that the encoding module receives the element at the kth position in the τth encoding round, b k is the k-th bit block of the bit stream to be sent, represents the element at the kth position in the first coded output information, represents the element at the kth position in the output information of the τ-1th encoding round, represents the element at position k in the encoded input information, q τ is the τth combination information of all positions at the input end of the τth encoding module, x τ Encode the output information for the τth encoding round.

[0091] Optionally, the first decoded identification information can indicate the position of the successfully decoded element in the first decoding input information through the value of each element in the information, and the position can correspond to the T-th encoded output information, all the encoded output information of the previous encoding round, the T-th decoded identification information, all the encoded output information of the previous encoding round, and the position of each element in the bit stream to be sent. Specifically, the first decoded identification information can be used to make the element at the position corresponding to the decoded bit block in the information received by the encoding module no longer participate in forward transmission and feedback, and at the same time terminate the update of the encoded output information and the decoded output information at this position in the next encoding round.

[0092] S205. The decoding module performs confidence processing according to the confidence information corresponding to each element in the first decoding input information and the second decoding input information, and determines the confidence information corresponding to each element in the second decoding input information. If the confidence information corresponding to all elements is greater than a preset confidence threshold, the decoded elements are synthesized into output information and output.

[0093] Specifically, in the second encoding round, confidence processing is performed based on the confidence information corresponding to each element in the first decoding input information and the second decoding input information, and the confidence information corresponding to each element in the second decoding input information is determined. In the τth encoding round, confidence processing is performed based on the confidence information corresponding to each element in the τ-1th decoding input information and the τth decoding input information, and the confidence information corresponding to each element in the τth decoding input information is determined. Exemplarily, the process of determining the confidence information corresponding to each element in the τth decoding input information can be represented by formula (6), formula (7) and formula (8).

[0094]

[0095] in, represents the confidence information corresponding to the kth element in the τth decoding input information, Indicates the confidence of the i-th element class corresponding to the k-th element, represents the object encoded by the decoding module in the second encoding round, and ψ represents the mathematical model of the decoding module.

[0096] Optionally, if the confidence information corresponding to all elements is greater than a preset confidence threshold, the decoded elements are synthesized into output information and output. If there is an element whose confidence information corresponding to the element has the largest probability element that is not greater than the preset confidence threshold, the sum of the second decoded input information and the second feedback channel additive Gaussian noise information of the feedback channel is continued to be used as the third coding input information, the coding module increases the coding round by one, enters the third coding round, and encodes according to the third coding input information, the second coding input information, the second decoded identification information, the first coding output information, the second coding output information and the bit stream to be sent to obtain the third coding output information, and the sum of the third coding output information and the third forward channel additive Gaussian noise information of the forward channel is used as the third decoding input information. The decoding module performs confidence processing according to the confidence information corresponding to each element in the second decoding input information and the third decoding input information, determines the confidence information corresponding to each element in the third decoding input information, and determines whether the confidence information corresponding to each element is greater than the preset confidence threshold. The above is the complete encoding and decoding process of the third encoding round. When the confidence information corresponding to all elements is greater than the preset confidence threshold, the decoded elements are synthesized into output information and output. If there is an element whose corresponding confidence information is not greater than the preset confidence threshold, the encoding and decoding of the fourth encoding round is entered until the confidence information corresponding to all elements is greater than the preset confidence threshold. The decoded elements are synthesized into output information and output, thereby completing the adaptive bit rate adjustment.

[0097] The following is the encoding and decoding process of the fourth encoding round: if in the third encoding round, the decoding module determines that the maximum probability element in the confidence information corresponding to the element is not greater than the preset confidence threshold, then the sum of the third decoding input information and the third feedback channel additive Gaussian noise information of the feedback channel is continued as the fourth encoding input information, the encoding module adds one to the encoding round, enters the fourth encoding round, and encodes according to the fourth encoding input information, the third encoding input information, the second encoding input information, the third decoded identification information, the first encoding output information, the second encoding output information, the third encoding output information and the bit stream to be sent to obtain the fourth encoding output information, and the sum of the fourth encoding output information and the fourth forward channel additive Gaussian noise information of the forward channel is used as the fourth decoding input information. The decoding module performs confidence processing according to the confidence information corresponding to each element in the third decoding input information and the fourth decoding input information, determines the confidence information corresponding to each element in the fourth decoding input information, and determines whether the confidence information corresponding to each element is greater than the preset confidence threshold. The above is the complete encoding and decoding process of the fourth encoding round. When the confidence information corresponding to all elements is greater than the preset confidence threshold, the decoded elements are synthesized into output information and output. If there is an element whose corresponding confidence information is not greater than the preset confidence threshold, the encoding and decoding of the fifth encoding round is entered until the confidence information corresponding to all elements is greater than the preset confidence threshold.

[0098] In this embodiment, the encoding module encodes the bit stream to be sent to obtain the first encoding output information, and takes the sum of the first encoding output information and the first forward channel additive Gaussian noise information of the forward channel as the second encoding input information. The decoding module performs confidence processing on the first encoding input information, determines the confidence information corresponding to each element in the first decoding input information, judges in turn whether the confidence information corresponding to each element is greater than a preset confidence threshold, decodes the elements whose confidence information is greater than the preset confidence threshold, and takes the sum of the first decoding input information and the first feedback channel additive Gaussian noise information of the feedback channel as the second encoding input information. The encoding module adds one to the encoding round, and encodes according to the second encoding input information, the first decoded identification information, the first encoding output information, and the bit stream to be sent to obtain the second encoding output information. In this embodiment, the second encoding input information, the first decoded identification information, the first encoding output information, and the bit stream to be sent are encoded, so as to adjust the forward encoding strategy of the transmitting end using feedback information. Then, the sum of the second coded output information and the second forward channel additive Gaussian noise information of the forward channel is used as the second decoding input information. The decoding module performs confidence processing according to the confidence information corresponding to each element in the first decoding input information and the second decoding input information, and determines the confidence information corresponding to each element in the second decoding input information. The decoding module can decode each bit block according to the noise size carried by the bit blocks corresponding to different elements, thereby realizing adaptive bit rate adjustment of the bit block granularity. If the confidence information corresponding to all elements is greater than the preset confidence threshold, the decoded elements are synthesized into output information and output. This embodiment iterates multiple rounds of communication through the forward channel and the downlink enhanced feedback channel between the encoding module and the decoding module, thereby improving the anti-noise performance.

[0099] Next, the process of generating the first decoded identification information in the above step S204 is introduced.

[0100] Optionally, the decoding module constructs initial identification information and initializes the value of each element in the initial identification information to a preset initial value, and the length of the initial identification information is the same as the length of the first decoding input information.

[0101] Optionally, the preset initial value may be, for example, 0, and the value of each element in the initial identification information is 0.

[0102] Optionally, the decoding module traverses each element in the first decoding input information, and for the current element traversed, if the confidence information of the current element is greater than a preset confidence threshold, the value of the element at the position corresponding to the current element in the initial identification information is modified to a first value; otherwise, the value of the element at the position corresponding to the current element in the initial identification information is modified to a second value, wherein the first value is used to identify what has been decoded, and the second value is used to identify what has not been decoded.

[0103] Exemplarily, if the first decoded input information includes 3 elements, and the maximum probability element in the confidence information corresponding to the third element is greater than the preset confidence threshold, the value of the third element in the initial identification information is modified to the first value, and the values ​​of the first element and the second element are modified to the second value. Exemplarily, the first value may be 1, for example, and the second value may be 0, for example. Then the first decoded identification information may be (0, 0, 1).

[0104] Optionally, after the decoding module generates the first decoded identification information, the first decoded identification information and the first decoding output information can be sent to the encoding module through the feedback channel, and in this process, the sum of the first decoding input information and the first feedback channel additive Gaussian noise information of the feedback channel is used as the second encoding input information.

[0105] In this embodiment, by constructing initial identification information, and modifying the value of the corresponding position of the initial identification information according to the comparison between the confidence information corresponding to each element in the decoded input information and the preset confidence threshold, the element in the information received by the encoding module corresponding to the corresponding position of the element with the first value in the first decoded identification information no longer participates in forward transmission and feedback, and at the same time, the update of the encoded output information and the decoded output information at this position in the next encoding round is terminated.

[0106] As an optional implementation, Figure 3 Schematic diagram of the architecture of a variable length channel coding system with feedback provided in an embodiment of the present application. Figure 3 As shown, the encoding module is connected to the decoding module through a forward channel and a feedback channel. The encoding module receives the τth combination information q of all positions at the input end of the τth encoding module. τ After encoding, output the encoded output information x τ , and then encode the output information x τ and the additive Gaussian noise information n of the τth forward channel in the forward channel τ The sum of is used as the τth decoding input information y τ The decoding module decodes the confidence information corresponding to each element in the input information based on the τ-1th decoding And the τth decoded input information y τ Perform confidence processing and output the confidence information corresponding to each element after decoding Combined with the preset signal threshold, a decision is made to generate the decoded identification information The decoded identification information and the τth decoded input information y τ The encoder is fed through the feedback channel, during which the τth decoded input information y τ and the τth feedback channel additive Gaussian noise information The sum of is used as the τ+1th coded input information

[0107] Further, refer to Figure 4 The specific process of encoding the second coded input information, the first decoded identification information, the first coded output information and the bit stream to be sent in the above step S204 to obtain the second coded output information is introduced. Figure 4 It is a flow chart of obtaining second coded output information provided by an embodiment of the present application.

[0108] S401. Calculate the difference between second coded input information and first coded output information as first difference information.

[0109] Specifically, the difference between the encoding input information of the current encoding round and the encoding output information of the previous encoding round is used as the first difference information.

[0110] S402: Determine second combination information according to the position of the successfully decoded element in the first decoded identification information, the first difference information, the bit stream to be sent, and the first encoded output information.

[0111] Specifically, whether the bit blocks corresponding to the elements in the first difference information, the bit stream to be sent and the first encoded output information are decoded is determined according to the values ​​of the elements in the first decoded identification information; if the value of an element in the first decoded identification information is the first value, the elements in the corresponding positions in the first difference information, the bit stream to be sent and the first encoded output information are no longer encoded and decoded.

[0112] S403: Encode the second combined information to obtain second encoded output information.

[0113] Optionally, taking the third encoding round as an example, the difference between the third encoding input information and the second encoding output information is calculated as the second difference information, and the difference between the second encoding input information and the first encoding output information is calculated as the first difference information, and the third combination information is determined according to the position of the successfully decoded elements in the second decoded identification information, the first difference information, the second difference information, the bit stream to be sent, the first encoding output information and the second encoding output information, and the third combination information is encoded to obtain the third encoding output information.

[0114] In this embodiment, the first difference information is calculated, and the second combination information is determined according to the position of the successfully decoded element in the first decoded identification information, the first difference information, the bit stream to be sent, and the first encoded output information, so that the second combination information is encoded to obtain the second encoded output information, thereby using the value of each element in the decoded identification information to stop the forward transmission and feedback of the elements corresponding to the decoded bit block in the information.

[0115] As an optional implementation, after determining the confidence information corresponding to each element in the second decoding input information in the above step S205, the following steps may also be performed.

[0116] Optionally, if there is an element whose confidence information is less than or equal to a preset confidence threshold, it is determined whether the encoding round reaches the preset encoding round threshold, and if so, each element in the second decoding input information is decoded and synthesized into output information and then outputted. If not, the second encoding input information is generated and encoding is continued.

[0117] Optionally, if there are elements whose confidence information is less than or equal to a preset confidence threshold, it means that the bit block has not been fully decoded. Then it is determined whether the coding round reaches the preset coding round threshold. If so, although the current bit rate adjustment of the bit block cannot achieve the most perfect effect, the adjusted bit block is still decoded and output, thereby ensuring signal transmission efficiency. If the coding round does not reach the preset coding round threshold, the sum of the second decoding input information and the second feedback channel additive Gaussian noise information is used as the second coding input information, the coding module adds one to the coding round, and enters the next round of coding and decoding until all elements have confidence information greater than the preset confidence threshold, or the coding round reaches the preset coding round threshold.

[0118] In this embodiment, the signal transmission efficiency is ensured by judging whether the encoding round reaches a preset encoding round threshold.

[0119] Next, the structure of the encoding module and decoding module and the specific encoding and decoding process are introduced.

[0120] Figure 5 is a schematic diagram of the structure of a coding module provided in an embodiment of the present application. Figure 5 The structure of the encoding module is introduced.

[0121] Optionally, the encoding module includes an encoding input unit, an encoding feature extraction unit, a first encoding self-attention unit, a second encoding self-attention unit, an encoding feedforward network unit, an encoding head unit and an encoding output unit.

[0122] Next, refer to Figure 6 The specific steps of encoding the second combined information in the above step S403 to obtain the second coded output information are introduced. Figure 6 It is another flowchart of obtaining second coded output information provided in an embodiment of the present application.

[0123] S601. The coding input unit inputs the second combination information into the coding feature extraction unit.

[0124] S602. The coding feature extraction unit determines a target extraction structure according to the coding round, uses the target extraction structure to perform feature extraction on the second combination information, and inputs the extracted coding features into the first coding self-attention unit.

[0125] As an optional implementation, the coding feature extraction unit determines the target extraction structure according to the current coding round and a preset structure round threshold, wherein the target extraction structure can change depth according to the current coding round.

[0126] Optionally, the coding feature extraction unit extracts features of the second combined information and projects the feature matrix to a higher dimension to form extracted coding features.

[0127] S603. The first encoding self-attention unit and the second encoding self-attention unit successively fuse the extracted encoding features to obtain fused encoding features, and input the fused encoding features into the encoding feedforward network unit.

[0128] Optionally, the first encoding self-attention unit and the second encoding self-attention unit can fuse feature representations between bit blocks and achieve parallel processing.

[0129] Optionally, the first encoding self-attention unit and the second encoding self-attention unit can map the fused encoding features to three features with the same dimension d model Different projection spaces of produce three sets of characteristic matrices, denoted as Q, K and V respectively, then the characteristic matrix satisfies They represent the query, key and value of the attention mechanism respectively.

[0130] Exemplarily, the first encoding self-attention unit and the second encoding self-attention unit can be expressed by the following formula (9):

[0131]

[0132] S604, the encoding feedforward network unit performs feature enhancement on the fused encoding features to obtain enhanced encoding features, and inputs the enhanced encoding features into the encoding head unit.

[0133] Optionally, the encoded feed-forward network unit can project the fused encoded features into a higher dimensional nonlinear space, thereby enhancing the robustness of the error correction code.

[0134] Optionally, the encoding feedforward network unit may be composed of two linear layers with large intrinsic dimensions and an activation function, wherein the activation function may be a Gelu activation function.

[0135] Optionally, the encoding feedforward network unit can be expressed by the following formula (10):

[0136]

[0137] Among them, FFN(X) is the enhanced encoded feature, f represents an upsampled linear projection, and f ′ Represents a downsampled linear projection.

[0138] S605. The encoding head unit compresses the enhanced encoding features based on dual-objective optimization control to obtain compressed encoding features, and inputs the compressed encoding features into the encoding output unit.

[0139] Optionally, the encoding head unit may be composed of two linear layers and a power control layer, so as to compress the hidden layer features of the encoded feedforward network unit into a compressed encoded feature.

[0140] The dual-objective optimization control can be expressed as the following formula (11) and formula (12):

[0141]

[0142] Among them, P e express The ratio of b is the bit stream to be sent. To output information, represents the average coding rounds of all bit blocks, and formula (12) is the constraint condition, Indicates that power control is performed on all coded transmitted information.

[0143] S606: The encoding output unit converts the compressed encoding features into second encoding output information.

[0144] Optionally, the encoding output unit converts the compressed encoding features into second encoding output information for transmission on a forward channel.

[0145] In this embodiment, the bit stream or combined information to be sent is encoded through the encoding input unit, the encoding feature extraction unit, the first encoding self-attention unit, the second encoding self-attention unit, the encoding feedforward network unit, the encoding head unit and the encoding output unit in the encoding module, thereby realizing variable length coding.

[0146] Further, refer to Figure 7 The specific steps of determining the target extraction structure according to the encoding round in the above step S602 are introduced. Figure 7 It is a schematic diagram of a process for determining a target extraction structure provided in an embodiment of the present application.

[0147] S701: Determine whether the encoding round is greater than or equal to a preset structure round threshold.

[0148] Optionally, the following conclusion can be drawn based on the DeepCode experimental results: As the number of encoding rounds increases, the nonlinear relationship between the input and output of the encoding module becomes stronger and stronger. Specifically, if the encoding round is less than the preset structural round threshold, the nonlinear relationship is not strong, so a shallower neural network can be used for fitting. If the encoding round is greater than or equal to the preset structural round threshold, the nonlinear relationship is strong, so a deeper neural network is used for fitting. Based on this, perform the following steps.

[0149] S702: If not, taking a perceptual structure of a preset number of layers composed of linear layers based on a preset activation function as a target extraction structure.

[0150] Specifically, if the encoding round is less than a preset structure round threshold, a perceptual structure of a preset number of layers composed of linear layers based on a preset activation function is used as a target extraction structure. The activation function may be a ReLU activation function, and the perceptual structure may be a multilayer perceptron (MLP). That is, an MLP of a preset number of layers composed of linear layers with a ReLU activation function is used as a target extraction structure.

[0151] S703: If yes, then add a linear layer with a preset activation function on the basis of the perception structure composed of a preset number of layers of linear layers based on a preset activation function to obtain a target extraction structure.

[0152] Specifically, if the encoding round is greater than or equal to the preset structural round threshold, a linear layer with a preset activation function is added on the basis of the perceptual structure of a preset number of layers composed of linear layers based on a preset activation function. That is, on the basis of the MLP of a preset number of layers composed of linear layers with ReLU activation functions, a linear layer with ReLU activation function is added, and this structure is used as the target extraction structure.

[0153] Exemplarily, determining the target extraction structure can be expressed by the following formula (13) and formula (14):

[0154]

[0155] in, represents the lth linear layer, σ represents the activation function, represents the number of intermediate neural network layers at the τth encoding round. When the encoding round is less than the preset structural round threshold, the number of layers is n. When the encoding round is greater than or equal to the preset structural round threshold, the number of layers is n+1. ξ represents the upsampling projection, stands for downsampled projection. Represents the target extraction structure.

[0156] In this embodiment, the target extraction structure is determined according to the encoding round and the preset structure round threshold, so that the nonlinearity of the target extraction structure is adjusted according to the encoding round, thereby more accurately capturing the characteristics of the combined information.

[0157] Next, refer to Figure 8 Introduce the structure of the decoding module. Figure 8 Schematic diagram of a decoding module provided in an embodiment of the present application. Figure 8 As shown, the decoding module includes: a decoding input unit, a decoding feature extraction unit, a first decoding self-attention unit, a second decoding self-attention unit, a third decoding self-attention unit, a decoding feedforward network unit, a decoding head unit and a decoding output unit.

[0158] based on Figure 8 The structure of the decoding module shown in FIG. Fig. 9 The specific process of performing confidence processing on the first decoded input information in the above step S202 to determine the confidence information corresponding to each element in the first decoded input information is introduced. Fig. 9 It is a flowchart of determining confidence information corresponding to each element in the first decoding input information provided by an embodiment of the present application.

[0159] S901. After receiving first decoding input information, the decoding input unit inputs the first decoding input information into a decoding feature extraction unit.

[0160] S902. The decoding feature extraction unit uses the target extraction structure to extract features from the first decoding input information, and inputs the extracted decoding features into the first decoding self-attention unit.

[0161] Optionally, the decoding feature extraction unit uses the target extraction structure to extract features from the first decoding input information and projects it to a higher dimension to generate extracted decoding features.

[0162] S903, the first decoding self-attention unit, the second decoding self-attention unit and the third decoding self-attention unit successively fuse the extracted decoding features to obtain fused decoding features, and input the fused decoding features into the decoding feedforward network unit.

[0163] Optionally, the first decoding self-attention unit, the second decoding self-attention unit, and the third decoding self-attention unit aggregate features between elements corresponding to each bit block to generate extracted decoding features.

[0164] S904, the decoding feedforward network unit performs feature enhancement on the fused decoding features to obtain enhanced decoding features, and inputs the enhanced decoding features into the decoding head unit.

[0165] S905. The decoding head unit determines the confidence information of each element in the enhanced decoding feature based on the dual-objective optimization control, and outputs the confidence information corresponding to each element through the decoding output unit.

[0166] Optionally, the decoding head unit may be a confidence classification head unit, which may output a confidence information for the element corresponding to each bit block. When the confidence information corresponding to an element is greater than a preset confidence threshold, the element is decoded.

[0167] In this embodiment, decoding and confidence judgment of decoding input information are achieved through a decoding input unit, a decoding feature extraction unit, a first decoding self-attention unit, a second decoding self-attention unit, a third decoding self-attention unit, a decoding feedforward network unit, a decoding head unit and a decoding output unit in a decoding module.

[0168] Next, refer to Fig.10 The training process of the encoding module is introduced. Fig.10 It is a flowchart of a coding module training provided in an embodiment of the present application.

[0169] S1001. An initial encoding module performs encoding for multiple encoding rounds based on sample information, and determines a loss value of the encoding round according to confidence information corresponding to the encoding round at the end of each encoding round.

[0170] Optionally, during the training process, the additive white Gaussian noise component forward channel environment under different signal-to-noise ratio environments may be simulated, and the feedback channel adopts a noise-free channel environment.

[0171] Optionally, m*K integer values ​​may be uniformly randomly sampled from 0 and 1, and divided into K bit blocks to form a training sample, wherein m and K may be random numbers.

[0172] S1002. Determine the loss value of the initial encoding module according to the loss values ​​of all encoding rounds.

[0173] Specifically, the sum of the loss values ​​of all encoding rounds is taken as the loss value of the initial encoding module.

[0174] S1003, iteratively updating the initial encoding module according to the loss value of the initial encoding module, and using the initial encoding module at the end of the iteration as the encoding module.

[0175] Specifically, the loss function can be expressed by the following formula (15):

[0176]

[0177] Among them, L is the loss function, T is the preset encoding round threshold, γ τ-1is an exponential weight coefficient, which increases exponentially with the increase of encoding rounds.

[0178] Optionally, the training process of the decoding module is the same as the training process of the encoding module, which will not be repeated here.

[0179] In this embodiment, the loss function is defined as a weighted sum polynomial of confidence cross entropy loss exponents of multiple coding rounds, thereby being more suitable for channel coding optimization.

[0180] in addition, Fig.11 FIG. 1 is a schematic diagram showing a comparison of the bit error rates of a general block attention feedback code method with feedback under different channel conditions and a variable length channel coding method with feedback under different code rates provided by an embodiment of the present application. Fig.11 As shown, under the channel conditions of SNR of 0dB and 1dB, the bit error rate obtained by the variable length channel coding method with feedback in the embodiment of the present application is significantly lower than that of the traditional coding method, which reflects that the method improves the short code performance of the feedback code.

[0181] An embodiment of the present application further provides a variable length channel coding system with feedback, the variable length channel coding system with feedback comprising a coding module and a decoding module, the coding module and the decoding module being used to execute the steps of the variable length channel coding method with feedback as described above.

[0182] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the variable-length channel coding method with feedback are executed.

[0183] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the method embodiment, and will not be repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0185] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or part of the technical solution that contributes to the prior art or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), disk or optical disk and other media that can store program code.

[0186] The above are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A variable length channel coding method with feedback, characterized in that: The method is applied to a variable length channel coding system with feedback, the variable length channel coding system with feedback comprises a coding module and a decoding module, the transmission channel from the coding module to the decoding module is a forward channel, and the transmission channel from the decoding module to the coding module is a feedback channel, the method comprises: The encoding module encodes the bit stream to be transmitted to obtain first encoding output information, and uses the sum of the first encoding output information and the first forward channel additive Gaussian noise information of the forward channel as the first decoding input information, wherein the bit stream to be transmitted includes multiple bit blocks; The decoding module performs confidence processing on the first decoded input information to determine confidence information corresponding to each element in the first decoded input information, wherein each element corresponds to a bit block in the bit stream to be sent; The decoding module sequentially determines whether the confidence information corresponding to each of the elements is greater than a preset confidence threshold, decodes the elements whose confidence information is greater than the preset confidence threshold, and uses the sum of the first decoded input information and the first feedback channel additive Gaussian noise information of the feedback channel as the second encoded input information; The encoding module increases the encoding round by one, and performs encoding according to the second encoding input information, the first decoded identification information, the first encoding output information, and the bit stream to be sent to obtain second encoding output information, and takes the sum of the second encoding output information and the second forward channel additive Gaussian noise information of the forward channel as the second decoding input information, wherein the first decoded identification information is used to indicate the elements in the first decoding input information that have been successfully decoded; The decoding module performs confidence processing based on the confidence information corresponding to each element in the first decoded input information and the second decoded input information, determines the confidence information corresponding to each element in the second decoded input information, and if the confidence information corresponding to all the elements is greater than a preset confidence threshold, the decoded elements are synthesized into output information and output.

2. The variable length channel coding method with feedback according to claim 1, characterized in that: The process of generating the first decoded identification information includes: The decoding module constructs initial identification information and initializes the value of each element in the initial identification information to a preset initial value, wherein the length of the initial identification information is the same as the length of the first decoding input information; The decoding module traverses each element in the first decoding input information, and for the current element traversed, if the confidence information of the current element is greater than a preset confidence threshold, the value of the element at the position corresponding to the current element in the initial identification information is modified to a first value; otherwise, the value of the element at the position corresponding to the current element in the initial identification information is modified to a second value, wherein the first value is used to identify that it has been decoded, and the second value is used to identify that it has not been decoded.

3. The variable length channel coding method with feedback according to claim 1, characterized in that: The method of encoding according to the second coded input information, the first decoded identification information, the first coded output information and the to-be-sent bit stream to obtain the second coded output information comprises: Calculating a difference between the second coded input information and the first coded output information as first difference information; Determine second combination information according to the position of the successfully decoded element in the first decoded identification information, the first difference information, the to-be-transmitted bit stream, and the first coded output information; The second combined information is encoded to obtain second encoded output information.

4. The variable length channel coding method with feedback according to claim 1, characterized in that: After determining the confidence information corresponding to each element in the second decoding input information, the method further includes: If there is an element whose confidence information is less than or equal to the preset confidence threshold, it is determined whether the encoding round reaches the preset encoding round threshold. If so, each element in the second decoding input information is decoded and synthesized into output information and then output; if not, the second encoding input information is generated and encoding continues.

5. The variable length channel coding method with feedback according to claim 3, characterized in that: The encoding module includes: an encoding input unit, an encoding feature extraction unit, a first encoding self-attention unit, a second encoding self-attention unit, an encoding feedforward network unit, an encoding head unit and an encoding output unit; The step of encoding the second combined information to obtain second coded output information includes: The encoding input unit inputs the second combination information into the encoding feature extraction unit; The encoding feature extraction unit determines a target extraction structure according to the encoding round, and uses the target extraction structure to extract features from the second combination information, and inputs the extracted encoding features into the first encoding self-attention unit; The first encoding self-attention unit and the second encoding self-attention unit sequentially perform feature fusion on the extracted encoding features to obtain fused encoding features, and input the fused encoding features into the encoding feedforward network unit; The encoding feedforward network unit performs feature enhancement on the fused encoding features to obtain enhanced encoding features, and inputs the enhanced encoding features into the encoding head unit; The encoding head unit compresses the enhanced encoding features based on dual-objective optimization control to obtain compressed encoding features, and inputs the compressed encoding features into the encoding output unit; The encoding output unit converts the compressed encoding feature into the second encoding output information.

6. The variable length channel coding method with feedback according to claim 5, characterized in that: The step of determining the target extraction structure according to the encoding rounds comprises: Determining whether the encoding round is greater than or equal to a preset structural round threshold; If not, a perceptual structure consisting of a preset number of layers of linear layers based on a preset activation function is used as the target extraction structure; If so, a linear layer with a preset activation function is added on the basis of the perception structure composed of a preset number of layers of linear layers based on the preset activation function to obtain a target extraction structure.

7. The variable length channel coding method with feedback according to claim 5, characterized in that: The decoding module includes: a decoding input unit, a decoding feature extraction unit, a first decoding self-attention unit, a second decoding self-attention unit, a third decoding self-attention unit, a decoding feedforward network unit, a decoding head unit and a decoding output unit; The performing confidence processing on the first decoded input information to determine the confidence information corresponding to each element in the first decoded input information includes: After receiving the first decoding input information, the decoding input unit inputs the first decoding input information into the decoding feature extraction unit; The decoding feature extraction unit uses the target extraction structure to extract features from the first decoding input information, and inputs the extracted decoding features into the first decoding self-attention unit; The first decoding self-attention unit, the second decoding self-attention unit, and the third decoding self-attention unit sequentially perform feature fusion on the extracted decoding features to obtain fused decoding features, and input the fused decoding features into the decoding feedforward network unit; The decoding feedforward network unit performs feature enhancement on the fused decoding features to obtain enhanced decoding features, and inputs the enhanced decoding features into the decoding head unit; The decoding head unit determines the confidence information of each element in the enhanced decoding feature based on the dual-objective optimization control, and outputs the confidence information corresponding to each element through the decoding output unit.

8. The variable length channel coding method with feedback according to claim 1, characterized in that: The training process of the encoding module includes: The initial encoding module performs encoding for multiple encoding rounds based on the sample information, and determines the loss value of each encoding round according to the confidence information corresponding to the encoding round at the end of each encoding round; Determining the loss value of the initial encoding module according to the loss values ​​of all encoding rounds; The initial encoding module is iteratively updated according to the loss value of the initial encoding module, and the initial encoding module at the end of the iteration is used as the encoding module.

9. A variable length channel coding system with feedback, characterized in that: The variable length channel coding system with feedback comprises a coding module and a decoding module, and the coding module and the decoding module are used to execute the steps of the variable length channel coding method with feedback as claimed in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the variable-length channel coding method with feedback as claimed in any one of claims 1 to 8 are executed.

Images