Directional neuroevolution channel coding method, storage medium and device
By introducing directional neural evolution algorithms into channel coding, using neural evolution algorithms to train noise tables and assist channel decoding, the problem of insufficient error correction ability when channel noise is severe is solved, the system's noise immunity and reliability are improved, and the system's noise immunity and reliability are reduced. delay.
Patent Information
- Application Number
- CN202111651069.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-30
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-12-30
AI Technical Summary
When the channel noise is serious, the error correction capability exceeds the error correction code itself, resulting in a decrease in system reliability, and it is difficult for existing channel encoding to effectively deal with this situation.
The channel compilation and decoding method of directional neural evolution is adopted. By adding random perturbation noise and directional noise respectively on the sending and receiving ends, the directional noise table is trained using the neural evolution algorithm to assist channel decoding and improve error correction capabilities.
It improves the system's error correction ability in a noisy environment, reduces the system delay, and ensures the direction of noise, avoiding the increase in decoding complexity.
Smart Images

Figure CN114301472B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of channel coding technology, and in particular to a directional neuro-evolved channel coding method, storage medium and device. Background Art
[0002] Channel coding, also known as error control coding, is the cornerstone of all modern communication systems. In the process of signal transmission, the transmitted data often has errors due to the inevitable interference of the environment and the physical defects of the transmission medium. Channel coding can improve the system's error correction and anti-interference capabilities by generating additional redundancy for the information to be sent in exchange for the reliability of the received information. In recent years, LDPC codes (Low-Density Parity-Check codes) that are close to the Shannon limit and polar codes (Polar codes) that are up to the capacity limit have become the focus of widespread attention due to their stronger error correction performance, and have been adopted in application scenarios such as 5G.
[0003] Usually, the noise in the system is considered a negative phenomenon, and corresponding measures need to be taken to eliminate it. Generally, channel coding is used to offset the interference of noise. Although channel coding can help improve the anti-interference ability of the transmission system, when the channel noise is serious, it may cause the effective information to be unable to be recovered during decoding, exceeding the error correction capability of the error correction code itself, and seriously reducing the reliability of the system. On the other hand, from the perspective of random resonance, specific noise may help channel decoding, so that the channel receiving signal can be corrected with the help of noise, thereby obtaining the correct message. Summary of the invention
[0004] In view of the situation that the channel noise seriously exceeds the error correction capability of the error correction code itself, the present invention proposes a directional neural evolutionary channel coding method, storage medium and device.
[0005] In a first aspect, a directional neuroevolutionary channel coding method comprises the following steps:
[0006] The transmitting end performs channel coding on the information bits to be sent, and after modulation, enters the channel for transmission to obtain a first channel output sequence; the receiving end performs channel decoding on the first channel output sequence, and ends the decoding if the decoding result is correct, otherwise randomly adds disturbance noise; decodes the second channel output sequence after adding the disturbance noise, and outputs the decoding result if the decoding result is correct; otherwise, continues to add disturbance noise until the number of additions reaches a first threshold value; calls the table function, loads the directional noise corresponding to the signal-to-noise ratio in the directional noise table; adds the directional noise to the first channel output sequence and performs decoding; if the decoding result is correct, ends the decoding, otherwise, repeatedly adds directional noise until the number of additions reaches a second threshold value.
[0007] Furthermore, the formation process of the directional noise table is as follows:
[0008] S41, the transmitting end performs channel coding on the randomly generated information bits, modulates the information bits and transmits them through the channel to obtain a third channel output sequence;
[0009] S42, the receiving end performs channel decoding on the third channel output sequence, and discards the decoding result if it is correct, and repeats S41 until an incorrect decoding is obtained;
[0010] S43, using the wrong decoding to establish and train a directional noise table, specifically including:
[0011] S431, initializing the population; generating a number of additive Gaussian white noise vectors as the initial population; members of the initial population exchange noise sequence sub-parts to form new offspring; then mutating the random bits of the population members to generate new random noise at the mutation point;
[0012] S432, calculating the fitness of each member in the population except the initial population, and the member with a larger fitness value has a greater probability of being selected; the selected members in the population exchange the noise sequence sub-part to form a new offspring; the random bits of the selected members in the population mutate to generate new random noise at the mutation point;
[0013] S433, adding the new generation population to the third channel output sequence once, and performing channel decoding again;
[0014] S434, if the decoding result passes the verification, it means that the decoding is successful, the decoding is completed, and the current noise is recorded as directional noise in the directional noise table; otherwise, the next generation population is generated;
[0015] S435 , repeat steps S432 to S434 until the amount of directional noise reaches a threshold.
[0016] Furthermore, step S431 is specifically to use a noise generator to generate a number of noises with a mean of 0 and a variance of Additive white Gaussian noise vector of length N.
[0017] Furthermore, the fitness F c It is obtained by the following formula
[0018] F c =1 / sum([uv p(i) ] 2 )
[0019] Among them, u is a randomly generated information bit, v p(i) is the channel decoding result after adding the i-th disturbance noise in the population.
[0020] Furthermore, the roulette wheel selection method is used to select population members, and members with large fitness values have a high probability of being selected.
[0021] Furthermore, the population member exchange sub-part is specifically, the crossover probability p c =0.8, take a random number a∈[0,1), compare it with the crossover probability, and when a is satisfied <p c , then the cross point is determined as cross_point=round(a*codelength), where codelength is the sequence length and round() is a rounding function.
[0022] Furthermore, the random bits of the population members are mutated specifically with a mutation probability p m =0.1, take a random number a∈[0,1), if a <p m , then the mutation condition is met, and a new variance of σ is generated at the mutation point mutation_point = round (a*codelength) p 2 The noise value.
[0023] According to a second aspect of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the method described in any one of the above technical solutions when running.
[0024] According to a third aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the method described in any one of the above technical solutions.
[0025] The present invention has the following beneficial effects: applying neuroevolutionary learning to the channel coding algorithm to ensure the directionality of noise, and the generation of directional noise adopts offline operation, which will not increase the decoding complexity; adding random noise and adding table directional noise will only be introduced after the previous level decoding fails, reducing the delay of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0027] Figure 1A schematic diagram of the principle of the error correction space used in the channel coding and decoding method of directional neural evolution according to an embodiment of the present invention;
[0028] Figure 2 Schematic diagram of vectors used to train directional noise for neuroevolution algorithms;
[0029] Figure 3 Flow chart of the channel coding and decoding method of directional neural evolution according to an embodiment of the present invention;
[0030] Figure 4 for Figure 3 Flowchart of the construction and training process of the directional noise table in the embodiment. DETAILED DESCRIPTION
[0031] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution 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. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.
[0032] It should be noted that the terms "comprises" and "has" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatus.
[0033] As a non-gradient optimization algorithm, the neuroevolution algorithm can find the global minimum of the loss surface by means of crossover and mutation operations given a sufficient number of iterations. Based on this optimization algorithm, after the channel decoding fails, the error codewords are assisted by adding trained noise with a specific direction to approach the error-correctable area, thereby achieving successful decoding. The method proposed in the present invention can be applied to a variety of channel coding methods, such as LDPC code, Polar code, etc.
[0034] Figure 1 The figure shows the principle diagram of the error correction space provided by the present invention. The error-correctable areas of the M codewords are respectively denoted as a 1 ,a 2 ,a 3 ,…,a M . Assume that a codeword is sent as c 2, due to the noise interference during the retransmission process, the position it falls in the decoding space is random. If the error is within the correctable range, it can be correctly decoded according to the decoding rules to obtain the correct information. If the error exceeds the correctable range, it means that the channel interference has a great impact on the signal and the correct codeword cannot be decoded. Adding a secondary independent random noise n to the received signal causes the signal to move in the decoding space and enter the error correction area.
[0035] Figure 2 The figure shows a vector diagram of directional noise training by the neuroevolutionary algorithm. Adding random noise makes the landing point of the received signal random in the decoding space, and the correct decoding of the channel received signal cannot be guaranteed. With the help of the neuroevolutionary algorithm, the noise is analogized to the weights in the neural network, and the evolutionary algorithm is used to update the random noise in the direction of auxiliary decoding. Through training, a table storing directional noise is obtained, and the corresponding function is designed to read the directional noise under the corresponding signal-to-noise ratio during the transmission of long code information, so as to avoid the increase of decoding delay caused by the table being too large. During the information transmission process, if the decoding fails to be correct after adding random noise, the directional noise in the table is called for further auxiliary decoding.
[0036] Based on the above principles, this embodiment provides a directional neural evolution channel coding method, such as Figure 3 As shown, the following steps are included:
[0037] S1, the transmitting end performs channel coding on the information bits to be sent, modulates the information bits and transmits them through the channel to obtain a first channel output sequence;
[0038] In this step, the information bit is a randomly generated information bit u, which is sent to the channel for transmission after BPSK modulation, and the first channel output sequence is set to y.
[0039] S2, the receiving end performs channel decoding on the first channel output sequence y to obtain a decoding result v. If the decoding result is correct, the decoding is terminated; otherwise, a disturbance noise n is randomly added;
[0040] S3, decoding the second channel output sequence y'=y+n after adding the disturbance noise, and outputting the decoding result if the decoding result is correct; otherwise, continue to add disturbance noise until the number of additions reaches the first threshold value T 1 ;
[0041] S4. Call the table function to load the directional noise corresponding to the signal-to-noise ratio in the directional noise table; the directional noise is a vector n p ={n 1 ,n 2 ,…,n T};
[0042] S5, adding the loaded directional noise to the first channel output sequence for decoding;
[0043] Specifically, directional noise is added to the first channel output sequence independently in sequence to form y i =y+n i , i = 1, 2, ..., T. The decoding result is recorded as v i .
[0044] S6. If the decoding is correct, the decoding is terminated. Otherwise, the directional noise is repeatedly added until the number of additions reaches the second threshold value T. 2 .
[0045] The construction and training process of the directional noise table is as follows: Figure 4 As shown, specifically including,
[0046] S41, the transmitting end performs channel coding on the randomly generated information bits, modulates them through BPSK and transmits them through the channel, and obtains the third channel output sequence y 3 ;
[0047] S42, the receiving end outputs sequence y to the third channel 3 Perform channel decoding, discard the decoding result if it is correct, and repeat S41 until an incorrect decoding is obtained;
[0048] S43, using the wrong decoding to establish and train a directional noise table, specifically including:
[0049] S431, initializing the population; generating a number of additive Gaussian white noise vectors as the initial population;
[0050] In this embodiment, it is assumed that a noise generator is used to generate 10 noises with a mean of 0 and a variance of σ. p 2 , additive Gaussian white noise vector n of length N 1 ,n 2 ,...,n 10 , and take it as the initial population n p1 =[n 1 ,n 2 ,...,n 10 ], at this time noise noise_count = 0.
[0051] The members of the initial population exchange noise sequence sub-parts to form new offspring; specifically, let the crossover probability p c =0.8, take a random number a∈[0,1), compare it with the crossover probability, and when a is satisfied <p c, then the cross point is determined as cross_point=round(a*codelength), where codelength is the sequence length and round() is a rounding function.
[0052] Then, the random bits of the population members are mutated to generate new random noise at the mutation point. Specifically, the mutation probability p m =0.1, take a random number a∈[0,1), if a <p m , then the mutation condition is met, and a new variance of σ is generated at the mutation point mutation_point = round (a*codelength) p 2 The noise value.
[0053] S432, calculating the fitness of each member in the population except the initial population, evaluating the fitness, the member with a larger fitness value has a greater probability of being selected; the selected members in the population exchange noise sequence sub-parts to form new offspring; the random bits of the selected members in the population mutate to generate new random noise at the mutation point.
[0054] Specifically, the fitness F c It is obtained by the following formula
[0055] F c =1 / sum([uv p(i) ] 2 )
[0056] Among them, u is a randomly generated information bit, v p(i) is the channel decoding result after adding the i-th disturbance noise in the population.
[0057] The process of exchanging noise sequence sub-parts and mutating random bits by the selected members in the population can be referred to the corresponding operations of the members in the initial population, and will not be repeated here.
[0058] Usually, the method of selecting population members adopts the roulette wheel selection method, and members with large fitness values have a high probability of being selected.
[0059] S433, the new generation population n p(i) Once added to the third channel output sequence, forming y p(i) =y+n p(i) , i=1,2,…,N, perform channel decoding again;
[0060] S434, if the decoding result passes the verification, it means that the decoding is successful, the decoding is completed, and the current noise is recorded as directional noise in the directional noise table; otherwise, the next generation population is generated;
[0061] S435 , repeat steps S432 to S434 until the amount of directional noise reaches a threshold.
[0062] At this point, the directional noise table is constructed and trained, and the corresponding table function is designed for the main program to call.
[0063] This embodiment further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the method described in any one of the above technical solutions when running.
[0064] The computer-readable storage medium may include random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field.
[0065] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the method described in any one of the above technical solutions.
[0066] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A directional neuroevolutionary channel coding method, It is characterized in that The following steps are involved: The transmitting end performs channel coding on the information bits to be transmitted, modulates the information bits and transmits them through the channel to obtain a first channel output sequence. The receiving end performs channel decoding on the first channel output sequence, and ends the decoding if the decoding result is correct, otherwise randomly adds disturbance noise; Decoding the second channel output sequence after adding the disturbance noise, and outputting the decoding result if the decoding result is correct; Otherwise, continue to add disturbance noise until the number of additions reaches the first threshold value; Call the table function to load the directional noise corresponding to the signal-to-noise ratio into the directional noise table; Adding the directional noise to the first channel output sequence for decoding; If the decoding result is correct, the decoding is terminated; otherwise, the directional noise is repeatedly added until the number of additions reaches a second threshold value.
2. The channel coding method according to claim 1, It is characterized in that The formation process of the directional noise table is as follows: S41, the transmitting end performs channel coding on the randomly generated information bits, modulates the information bits and transmits them through the channel to obtain a third channel output sequence; S42, the receiving end performs channel decoding on the third channel output sequence, and discards the decoding result if it is correct, and repeats S41 until an incorrect decoding is obtained; S43, using the wrong decoding to establish and train a directional noise table, specifically including: S431, initializing the population; Generate several additive Gaussian white noise vectors as the initial population; members of the initial population exchange noise sequence sub-parts to form new offspring; then mutate the random bits of the population members to generate new random noise at the mutation point; S432, calculating the fitness of each member in the population except the initial population, and the member with a larger fitness value has a greater probability of being selected; the selected members in the population exchange the noise sequence sub-part to form a new offspring; the random bits of the selected members in the population mutate to generate new random noise at the mutation point; S433, adding the new generation population to the third channel output sequence once, and performing channel decoding again; S434, if the decoding result passes the verification, it means that the decoding is successful, the decoding is completed, and the current noise is recorded as directional noise in the directional noise table; otherwise, the next generation population is generated; S435 , repeat steps S432 to S434 until the amount of directional noise reaches a threshold.
3. The channel coding method according to claim 2, It is characterized in that The specific step S431 is to use a noise generator to generate a number of noises with a mean of 0 and a variance of Additive white Gaussian noise vector of length N.
4. The channel coding method according to claim 2, It is characterized in that The fitness F c It is obtained by the following formula F c =1 / sum([u-v p(i) ] 2 ) Among them, u is a randomly generated information bit, v p(i) is the channel decoding result after adding the i-th disturbance noise in the population.
5. The channel coding method according to claim 2, It is characterized in that The roulette wheel selection method is used to select population members, and members with large fitness values have a high probability of being selected.
6. The channel coding method according to claim 2, It is characterized in that The population member exchange sub-part is specifically, the crossover probability p c =0.8, take a random number a∈[0,1), compare it with the crossover probability, and when a is satisfied <p c , then the cross point is determined as cross_point=round(a*codelength), where codelength is the sequence length and round() is a rounding function.
7. The channel coding method according to claim 2, It is characterized in that The random bits of the population members are mutated specifically with the mutation probability p m =0.1, take a random number a∈[0,1), if a <p m , then the mutation condition is met, and a new variance of σ is generated at the mutation point mutation_point = round (a*codelength) p 2 The noise value.
8. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 7 when executed.
9. An electronic device comprising a memory and a processor, It is characterized in that A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 7.
