Polar code construction method based on genetic algorithm acceleration convergence

Abstract

The application discloses a polar code construction method based on genetic algorithm acceleration convergence, which comprises the following steps: step 1, setting parameters; step 2, initial genetic algorithm population; step 3, selecting two parents from the population by using a roulette algorithm; step 4, crossing the two parent information bits to generate offspring; step 5, mutating the unlocked information bits of the offspring by using the non-locked frozen bits; step 6, adding the mutated offspring to the population, selecting the optimal one according to the fitness, updating the population and recording the population information; step 7, repeating steps 3 to 6 until the number of genetic iterations reaches a number of spans, judging whether the optimal fitness decreases to a limit or not, if yes, reducing the number of locked bits, temporarily releasing the channel, repeating steps 3 to 6 until the number of genetic iterations reaches a minimum number of spans, if the optimal fitness decreases and the temporarily released channel changes, updating the channel locking condition according to the temporarily released channel, if no, directly repeating steps 3 to 6; and step 8, ending when the genetic iteration stopping condition is met. The application can solve the problem of slow convergence speed by dynamically locking the channel.

Description

Technical Field

[0001] This invention belongs to the field of channel coding, specifically relating to a method for constructing polar codes based on a genetic algorithm to accelerate convergence. Background Technology

[0002] Polar code is Erdal In 2009, a class of linear block codes based on the channel polarization phenomenon was proposed. It is the first channel code that can be theoretically proven to achieve the capacity of arbitrary binary-input discrete memoryless channels (BI-DMC). It also has low encoding and decoding complexity and deterministic construction, and is currently a hot topic in academic and industrial research on channel coding theory and applications.

[0003] Currently, the construction of Polar codes is usually achieved using genetic algorithms. However, the traditional genetic algorithm searches for the entire code length in the process of constructing Polar codes, resulting in a large search space, slow convergence speed, and a large number of generations required for training to reach the convergence state. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention provides a polar code construction method based on a genetic algorithm to accelerate convergence. The technical problem to be solved by this invention is achieved through the following technical solution:

[0005] A method for constructing polar codes based on a genetic algorithm to accelerate convergence includes:

[0006] Step 1: Set parameters, including polar code-related parameters and genetic algorithm-related parameters;

[0007] Step 2: Initialize the genetic algorithm population based on the set parameters; evaluate the fitness of individuals in the population; sort individuals according to their fitness to obtain the initial population; wherein, the individuals in the population represent the Polar code channel construction, which is divided into information channel and frozen channel, and the frozen bits and information bits in the individuals are locked according to the maximum number of locks corresponding to the relevant parameters of the genetic algorithm.

[0008] Step 3: Use the roulette wheel algorithm to select two parent individuals from the current population;

[0009] Step 4: Cross-process the information bits in the two parent individuals to generate a child;

[0010] Step 5: Using the unlocked frozen bits of the offspring, perform mutation processing on the unlocked information bits to obtain the mutated offspring;

[0011] Step 6: After evaluating the fitness of the mutated offspring, add them to the current population, sort and select individuals according to their fitness, realize the population update after one genetic step, and record the current population information, including channel construction, optimal fitness, and channel locking status.

[0012] Step 7: Repeat the genetic process corresponding to steps 3 to 6 until the number of genetic iterations reaches the genetic iteration span in the relevant parameters of the genetic algorithm. Determine whether the decrease in optimal fitness has reached its limit. If so, reduce the number of locked bits to temporarily open the corresponding channel. Repeat steps 3 to 6 until at least one genetic iteration span is reached. If the optimal fitness decreases and the temporarily opened channel changes, then the temporarily opened channel is designated as the definitely opened channel, and the channel locking status in the current population information is updated accordingly. If not, keep the number of locked bits unchanged and repeat steps 3 to 6.

[0013] Step 8: For each genetic iteration, the iteration ends when the genetic iteration stopping condition is met.

[0014] In one embodiment of the present invention, the polar code related parameters include: code length N, information bit length K, frozen bit length NK, code rate R = K / N, channel conditions, modulation and demodulation method, and decoding method;

[0015] The parameters related to the genetic algorithm include: population size M, upper limit of the number of generations (Ineration). max The number of generations T after fitness convergence, the concentration of fitness samples α in the population, the mutation probability β, the number of construct signal-to-noise ratios J used to calculate fitness, and the construct signal-to-noise ratio (SNR). j The genetic span S used to monitor population fitness decline, and the maximum number of locked information bits L. I,max Maximum number of frozen bits L F,max Where, j∈{1,2,...,J}, L I,max <K,L F,max <(NK), S <T;N、K、N-K、M、Ineration max J, S, T, L I,max L F,max All are integers greater than 0.

[0016] In one embodiment of the present invention, step 2, initializing the genetic algorithm population based on the set parameters, includes:

[0017] For the i-th individual in the population, the corresponding channel index is represented as Where i∈{1,2,...,M}; the information bits are represented as follows: The freeze bit is represented as

[0018] Index it as The channel is set as the information bit channel, and the index is... The channel is set to a frozen channel;

[0019] At index Randomly selected from the channels (KL) I,max ) channels are set as information bits, the rest (NKL) F,max Set ) channels as frozen bits to obtain the channel construction for the i-th individual;

[0020] The initialized genetic algorithm population is obtained by constructing channels from M individuals.

[0021] In one embodiment of the present invention, in step 2, the fitness of individuals in the population is evaluated using a formula including:

[0022]

[0023] Where i∈{1,2,...,M}; BLER represents the fitness of the i-th individual in the population. j This indicates that the i-th individual is constructed based on the signal-to-noise ratio (SNR). j The obtained block error rate.

[0024] In one embodiment of the present invention, step 3, which involves selecting two parent individuals from the current population using a roulette wheel algorithm, includes:

[0025] For the i-th individual in the current population, based on the concentration of fitness α in the population, calculate the proportion of the i-th individual that will be selected as the parent as e. -αi Where i∈{1,2,...,M};

[0026] The interval corresponding to the i-th individual is determined as follows: The lower limit of the interval corresponding to the first individual is 0;

[0027] Summing the proportions of all individuals in the current population selected as parents yields the following summation:

[0028] exist Take a random number within the range, and determine the corresponding individual as the selected parent individual based on the interval in which the random number falls. In this way, select two different parent individuals P1 and P2 from the current population.

[0029] In one embodiment of the present invention, step 4 involves cross-processing the information bits in the two parent individuals to generate a child, including:

[0030] Information bit P1 of the parent individual and the parent individual P2 information bit Take the intersection as

[0031] Sure right supplement as well as right supplement Among them, set The set represents the index in the channel of the different information bits between parent individual P1 and parent individual P2. The index in the channel represents the different information bits of parent individual P2 compared to parent individual P1;

[0032] Choose a random integer n from [1, (c-1)] as the number of crossovers. cross Corresponding exchange and The first to the nth cross 1 element, get Crossed set as well as Crossed set in,

[0033] Will respectively with Taking the union of the two sets, we obtain the set of the channel indices for the two exchanged information bits. as well as

[0034] Random selection and One of the index sets in the channel as child information bits based on One offspring was obtained from this genetic process.

[0035] In one embodiment of the present invention, step 5, utilizing the unlocked frozen bits of the offspring, performs mutation processing on the unlocked information bits to obtain mutated offspring, including:

[0036] Based on the indices of the locked and frozen information bits, and the set of indices of the unlocked information bits of the offspring in the channel. Obtain the set of indices of the unlocked frozen bits of the offspring in the channel. in, This represents the complete channel index of this offspring. L I L represents the number of information bits currently locked. FIndicates the number of currently frozen bits locked;

[0037] Based on the mutation probability β and the code length N, in Based on the principle of equal probability selection, a random integer is determined to obtain the number of mutations m;

[0038] Will Replace m random elements with Given m random elements, mutated offspring are obtained based on the replacement results.

[0039] In one embodiment of the present invention, step 7, determining whether the decrease in optimal fitness has reached a limit state, includes:

[0040] When the current number of inheritance iterations has accumulated to at least two inheritance intervals, determine whether the optimal fitness satisfies either the first or second locking number adjustment condition. If so, it is determined that the decrease in optimal fitness has reached its limit; otherwise, it is determined that the decrease in optimal fitness has not reached its limit.

[0041] The first locking number adjustment condition includes: the optimal fitness of the population does not decrease after the genetic iteration corresponding to the current genetic number span, and the optimal fitness of the population decreases within the genetic iteration of the previous genetic number span.

[0042] The second locking number adjustment condition includes: the optimal fitness of the population does not decrease after two consecutive genetic iterations corresponding to two genetic spans.

[0043] In one embodiment of the present invention, step 7, reducing the number of locking bits to temporarily open the corresponding channel, includes:

[0044] When the currently locked information bit is greater than 0, reduce the locked information bit by one, thereby temporarily releasing the channel corresponding to the reduced locked information bit; where the released information bit is the first information bit in the index of all the previously locked information bits;

[0045] and / or

[0046] When the currently locked frozen bit is greater than 0, reduce the locked frozen bit by one, thereby temporarily releasing the channel corresponding to the reduced locked frozen bit; where the released frozen bit is the last frozen bit indexed among all the previously locked frozen bits.

[0047] In one embodiment of the present invention, step 8, the genetic iteration stopping condition includes:

[0048] Ineration maxEither the optimal fitness of the population converges to the upper limit of the cutoff genetic generations T, or the optimal fitness in the current population information reaches the target value.

[0049] For the problem of genetic algorithm code construction, especially Polar code construction, existing schemes, while producing high-performance constructs, suffer from slow convergence speed and high computational cost. This invention employs a dynamic channel-locking scheme, which, compared to existing schemes, reduces the search space of the genetic algorithm while maintaining construction performance, thereby accelerating population fitness convergence, reducing the number of generations required for genetic engineering, and improving training speed. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating a polar code construction method based on a genetic algorithm to accelerate convergence, provided in an embodiment of the present invention.

[0051] Figure 2(a) shows the performance convergence curve of the Polar code construction using the genetic algorithm in Example 2;

[0052] Figure 2(b) shows the performance curve of the genetic algorithm for constructing Polar codes in Example 2;

[0053] Figure 3(a) shows the performance convergence curve of the Polar code construction using the genetic algorithm in Example 3;

[0054] Figure 3(b) shows the performance curve of the genetic algorithm for constructing Polar codes in Example 3;

[0055] Figure 4(a) shows the performance convergence curve of the Polar code construction using the genetic algorithm in Example 4;

[0056] Figure 4(b) shows the performance curve of the genetic algorithm for constructing Polar codes in Example 4. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Example 1

[0059] Existing genetic algorithms suffer from large search spaces, slow fitness convergence during training, and require numerous generations to reach convergence. This invention accelerates training by dynamically changing the search space, providing a method for constructing polar codes based on a genetic algorithm to accelerate convergence. Figure 1 As shown, the method may include the following steps:

[0060] Step 1: Set parameters, including polar code-related parameters and genetic algorithm-related parameters;

[0061] In one alternative implementation,

[0062] The polar code related parameters include: code length N, information bit length K, frozen bit length NK, code rate R = K / N, channel conditions, modulation and demodulation methods, and decoding methods.

[0063] The parameters related to the genetic algorithm include: population size M, upper limit of the number of generations (Ineration). max The number of generations T after fitness convergence, the concentration of fitness samples α in the population, the mutation probability β, the number of construct signal-to-noise ratios J used to calculate fitness, and the construct signal-to-noise ratio (SNR). j The genetic span S used to monitor population fitness decline, and the maximum number of locked information bits L. I,max Maximum number of frozen bits L F,max Where, j∈{1,2,...,J}, L I,max <K,L F,max <(NK), S <T;N、K、N-K、M、Ineration max J, S, T, L I,max L F,max All are integers greater than 0.

[0064] Channel conditions refer to the type of channel used, such as AWGN channel or independent Rayleigh fading channel. Modulation and demodulation methods include QPSK modulation and demodulation. Decoding methods include SCL-Genie and SCL decoding.

[0065] Step 2: Initialize the genetic algorithm population based on the set parameters; evaluate the fitness of individuals in the population; sort the individuals according to their fitness to obtain the initial population;

[0066] In this context, the individual population represents the Polar code channel construction, which is divided into an information channel and a frozen channel. The frozen bits and information bits in the individual are locked according to the maximum number of locks corresponding to the relevant parameters of the genetic algorithm.

[0067] Specifically, in this embodiment of the invention, each individual in the population represents a Polar code channel construction, and each individual includes N bits, i.e., the code length is N. Each individual corresponds to N channels, divided into information channels and frozen channels; each bit of the individual is either an information bit or a frozen bit, and the information bits and frozen bits are distributed in the channels according to a certain arrangement. The information bits and frozen bits refer to the noiseless useful channels and noisy useless channels, respectively, which are divided by channel polarization. The noiseless useful channels corresponding to the information bits are used to directly transmit information; the noisy useless channels corresponding to the frozen bits are not used to transmit data, and fixedly place the bit information known to both the sender and receiver. The locking of information bits and frozen bits is first assumed to be a more reliable noiseless channel and a more unreliable noisy channel, respectively. Then, through the evolution of a genetic algorithm, the number of locked channels is gradually reduced and the real noiseless channel and noisy channel are determined.

[0068] During the execution of the method of this invention, some of the information bits and frozen bits of an individual are locked, and the unlocked parts correspond to the search space of the genetic algorithm. Specifically, in step 2, the frozen bits and information bits of an individual are first locked according to the maximum number of locked bits corresponding to the relevant parameters of the genetic algorithm. In subsequent processing, when the optimal fitness of the population has decreased to the limit, the number of locked bits is reduced to gradually open the search space. In this way, by gradually opening the search space by limiting it, the search is not performed across the entire code length at the beginning, thus improving the convergence speed and reducing the number of generations required for training to reach the convergence state.

[0069] In one optional implementation, step 2, initializing the genetic algorithm population based on the set parameters, includes:

[0070] 1) For the i-th individual in the population, the corresponding channel index is represented as:

[0071] Where i∈{1,2,...,M}; This represents the complete channel indexes constructed for the i-th individual. It can be understood that the index represents the channel number; for the i-th individual, the information bits are represented as... The freeze bit is represented as

[0072] 2) Index it as The channel is set as the information bit channel, and the index is... The channel is set to a frozen channel;

[0073] As can be seen, in this step, the locked freeze bits are at the beginning of the channel, and the locked information bits are at the end of the channel. The number of locked freeze bits is the maximum number L of frozen bits that can be locked. F,max The number of locked information bits is the maximum number of information bits locked, L.I,max However, during subsequent processing, the number of bits L locked in the freeze state... F And the number of bits L of the locked information bits I It will be dynamically adjusted based on the decline of the population's optimal fitness, which will be explained in detail later.

[0074] 3) At index Randomly selected from the channels (KL) I,max ) channels are set as information bits, the rest (NKL) F,max Set ) channels as frozen bits to obtain the channel construction for the i-th individual;

[0075] In this step, among all channels corresponding to an individual, except for the locked frozen bits and locked information bits, the frozen bits and information bits of all the middle bits are set randomly so that the information bits and frozen bits in the individual meet their respective quantities. That is, under the premise that the code length is N, the information bit length reaches K and the frozen bit length reaches NK.

[0076] In this embodiment of the invention, as an optional program implementation, 0 and 1 can be used to represent the freeze bit and the information bit, respectively; thus, it can be understood that each individual can be a string of length N represented by 0 and 1.

[0077] Here is an example to aid understanding: If N = 20, K = 10, L I,max =L F,max =5, using 0 to represent the freeze bit and 1 to represent the information bit, the generated individual is constructed in the form of 00000**********11111. The middle * part is randomly distributed with 5 freeze bits and 5 information bits, thus finally obtaining an individual.

[0078] 4) The initialized genetic algorithm population is obtained by constructing the channels of M individuals.

[0079] For each individual, the processing method of steps 1) to 3) above is used to obtain the channel construction of M individuals, thereby obtaining the initialized genetic algorithm population.

[0080] After obtaining the initialized genetic algorithm population, it is necessary to evaluate the fitness of each individual.

[0081] In one optional implementation, in step 2, the fitness of individuals in the population is evaluated using a formula including:

[0082]

[0083] In this embodiment of the invention, the Block Error Rate (BLER) is used as the evaluation metric, and the fitness of an individual in the population is defined as the performance at several signal-to-noise ratios (SNR). Specifically, the following methods are employed: The fitness of the i-th construction is defined as J distinct SNRs. j The product of BLER.

[0084] Where i∈{1,2,...,M}; BLER represents the fitness of the i-th individual in the population. j This indicates that the i-th individual is constructed based on the signal-to-noise ratio (SNR). j The obtained block error rate. Wherein, based on the constructed signal-to-noise ratio (SNR)... j The obtained block error rate BLER j Please refer to the relevant technical explanations for the process details, which will not be elaborated upon here. However, it is understandable that each individual receives... The values ​​may not be the same.

[0085] Furthermore, in this embodiment of the invention, the fitness of each individual obtained through genetics needs to be evaluated, and the method used is consistent and will not be repeated in the following text.

[0086] After assessing the fitness of individuals in the obtained population, the individuals are sorted in ascending order of fitness to obtain the initial population. Those skilled in the art will understand that a lower fitness indicates better individual performance.

[0087] Step 3: Use the roulette wheel algorithm to select two parent individuals from the current population;

[0088] In one optional implementation, step 3 involves selecting two parent individuals from the current population using a roulette wheel algorithm, including:

[0089] Step 31: For the i-th individual in the current population, calculate the proportion e of the i-th individual being selected as the parent based on the sample fitness concentration α in the population. -αi ;

[0090] Where i∈{1,2,...,M}; therefore, it can be understood that each individual receives a corresponding proportion of the selection as the parent.

[0091] Step 32, determine the interval corresponding to the i-th individual.

[0092] Specifically, each individual can calculate its own interval based on the proportion of itself selected as a parent and the proportion of previous individuals selected as parents, using the formula above; where the lower limit of the interval corresponding to the first individual is 0, i.e., when i=1.

[0093] Step 33: Sum the proportions of all individuals in the current population that were selected as parents, and obtain the summation result as follows:

[0094] Step 34, in Take a random number within the range, and determine the corresponding individual as the selected parent individual based on the interval in which the random number falls. In this way, select two different parent individuals P1 and P2 from the current population.

[0095] Specifically, if the parent individuals selected by two random numbers are the same, the selection is performed again until the two parent individuals selected from the current population are different.

[0096] Here is an example to aid understanding: If, according to step 31, the proportions of individuals 1 to 5 selected as parents in the current population are calculated to be 10, 7, 5, 3, and 1 respectively; according to step 32, the intervals corresponding to each individual are calculated to be [0,10), [10,17), [17,22), [22,25), and [25,26); according to step 33, the sum of the proportions of these five individuals selected as parents is 10+7+5+3+1=26; according to step 34, random numbers 7.125 and 13.754 are drawn from the random numbers [0,26). The two random numbers fall within the intervals [0,10) and [10,17) respectively, therefore individuals 1 and 2 are selected as parents P1 and P2.

[0097] Compared to directly selecting the two individuals with the best fitness from the population, using the roulette wheel algorithm to select parents has the following characteristics and advantages: it ensures that all individuals in the population have a probability of being selected as parents, it can prevent dominant individuals in the population from entering a self-fertilization cycle, and thus can effectively prevent genetic evolution from getting stuck in a local optimum.

[0098] Step 4: Cross-process the information bits in the two parent individuals to generate a child.

[0099] In one optional implementation, step 4 involves cross-processing the information bits in the two parent individuals to generate a child, including:

[0100] Step 41, process the information bit of the parent individual P1. and the parent individual P2 information bit Take the intersection as

[0101] Step 42, confirm right supplement as well as right supplement

[0102] Among them, set The set represents the index in the channel of the different information bits between parent individual P1 and parent individual P2. This represents the index in the channel of the different information bits between parent individual P2 and parent individual P1.

[0103] Step 43, select a random integer n from [1, (c-1)] as the number of crossovers. cross Corresponding exchange and The first to the nth cross 1 element, get Crossed set as well as Crossed set

[0104] in, Corresponding swaps refer to swaps between the first and nth digits. cross One, exchange and The first element, the second element, ..., up to the nth element. cross Each element.

[0105] Step 44, will respectively with Taking the union of the two sets, we obtain the set of the channel indices for the two exchanged information bits. as well as

[0106] Step 45, randomly select and One of the index sets in the channel as child information bits based on One offspring was obtained from this genetic process.

[0107] It should be noted that the index set of the offspring information bits in the channel is obtained. Then, based on the index of the locked frozen bits, the remaining indexes can be determined as the indexes of the unlocked frozen bits, thus determining the complete sub-channel index.

[0108] The following example is provided for easier understanding: If parent 1's channel is set to 000010101111, and parent 2's channel is set to 000101010111, but c = 3, cross_num = 1, that is, n cross =1, after swapping fragments Received Random selection and One of them as according to You can obtain the offspring from this inheritance.

[0109] Step 5: Using the unlocked frozen bits of the offspring, perform mutation processing on the unlocked information bits to obtain the mutated offspring;

[0110] In one optional implementation, step 5 involves using the unlocked frozen bits of one offspring to mutate the unlocked information bits, resulting in a mutated offspring, including:

[0111] Step 51: Based on the indices of the locked information bits and frozen bits, and the set of indices of the unlocked information bits of the offspring in the channel. Obtain the set of indices of the unlocked frozen bits of the offspring in the channel.

[0112] in, This represents the complete channel index of this offspring. L I L represents the number of information bits currently locked. F Indicates the number of currently frozen bits locked;

[0113] Step 52, based on the mutation probability β and the code length N, in Based on the principle of equal probability selection, a random integer is determined to obtain the number of mutations m;

[0114] Step 53, will Replace m random elements with Given m random elements, mutated offspring are obtained based on the replacement results.

[0115] As can be seen from the above process, the locked area does not participate in the mutation process. That is, when setting the mutation range, the locked frozen bits and information bits must be excluded. Instead, the unlocked information bits are replaced with unlocked frozen bits to achieve mutation.

[0116] Similar to step 4, Replace m random elements with After obtaining m random elements, the mutated offspring can be obtained based on the indices of the locked information bits and the frozen bits.

[0117] Here is an example to help you understand: If L I =3,L F =3. Then the locked information bits Unlocked information bits Locked freeze position Unlocked freeze position If m=1, after a random one-bit mutation It becomes {4,7,8,10,11,12}.

[0118] Step 6: After evaluating the fitness of the mutated offspring, add them to the current population, sort and select individuals according to their fitness, realize the population update after one genetic step, and record the current population information, including channel construction, optimal fitness, and channel locking status.

[0119] The method for evaluating the fitness of the mutant offspring is described in Step 2 above, using the formula and related descriptions. It will not be repeated here. After evaluating the fitness of the mutant offspring and adding it to the current population, the population size becomes M+1. These M+1 individuals are then sorted by fitness from smallest to largest, and the individual with the worst fitness is eliminated to maintain the updated population size.

[0120] The channel construction in the current population information includes the channel construction represented by each individual; the optimal fitness is the optimal fitness in the current population; the channel locking status includes the number of locked information bits and frozen bits, as well as the corresponding channel index.

[0121] Step 7: Repeat the genetic process corresponding to steps 3 to 6 until the number of genetic iterations reaches the genetic iteration span in the relevant parameters of the genetic algorithm. Determine whether the decrease in optimal fitness has reached its limit. If so, reduce the number of locked bits to temporarily open the corresponding channel. Repeat steps 3 to 6 until at least one genetic iteration span is reached. If the optimal fitness decreases and the temporarily opened channel changes, then the temporarily opened channel is designated as the definitely opened channel, and the channel locking status in the current population information is updated accordingly. If not, keep the number of locked bits unchanged and repeat steps 3 to 6.

[0122] In this embodiment of the invention, steps 3 to 6 represent one inheritance cycle. When the number of inheritance cycles reaches the inheritance cycle span S, it indicates that one inheritance cycle has been completed. S can be set as needed, such as 50 cycles, etc.

[0123] In this embodiment of the invention, the number of locked information bits can be L. I This indicates that the number of frozen bits can be locked using L. F Therefore, in step 2, L I =L I,max LF =L F,max .

[0124] In this embodiment of the invention, the state where the optimal fitness has reached its limit means that the optimal fitness no longer decreases, indicating that no better individual can be found in the current search space, and therefore the search space needs to be adjusted.

[0125] In one optional implementation, step 7, determining whether the decrease in optimal fitness has reached a limit, includes:

[0126] When the current number of inheritances has accumulated to at least two inheritance spans, determine whether the optimal fitness meets the first or second locking number adjustment condition. If so, determine that the decrease in optimal fitness has reached the limit state; otherwise, determine that the decrease in optimal fitness has not reached the limit state.

[0127] Specifically, in this embodiment of the invention, after completing the first genetic span, it does not determine whether the decline in optimal fitness has reached its limit. Instead, starting from the completion of the second genetic span, for each completed genetic span, it monitors the decline in optimal fitness in the population. If the optimal fitness satisfies either the first locking number adjustment condition or the second locking number adjustment condition, it is determined that the decline in optimal fitness has reached its limit, and the search space needs to be adjusted. If neither the first nor the second locking number adjustment condition is satisfied, it is determined that the decline in optimal fitness has not reached its limit, and the search space does not need to be adjusted.

[0128] in,

[0129] The first locking number adjustment condition includes: the optimal fitness of the population does not decrease after the genetic iteration corresponding to the current genetic number span, and the optimal fitness of the population decreases within the genetic iteration of the previous genetic number span.

[0130] The second locking number adjustment condition includes: the optimal fitness of the population does not decrease after two consecutive genetic iterations corresponding to two genetic spans.

[0131] It is understood that the first lock number adjustment condition and the second lock number adjustment condition will not be satisfied simultaneously.

[0132] The significance of setting the first and second locking number adjustment conditions in this embodiment of the invention is that the fitness should gradually decrease during the genetic evolution of the population until the maximum number of inheritances is reached or the fitness reaches the target index, at which point it should be actively stopped. Since the search space of the genetic algorithm is artificially limited, the fitness of the population may reach the limit under the current limitations. At this point, it is necessary to try to reduce the limitations and expand the search space, thereby further decreasing the fitness of the population.

[0133] The first condition for adjusting the number of locked individuals is for a population whose fitness has decreased to a limit within a certain number of inheritance cycles during the process of genetic evolution.

[0134] The second condition for adjusting the number of locked individuals is for a population whose fitness remains at a limit within a certain number of inheritance cycles during its genetic evolution and no longer decreases.

[0135] In step 7, if it is determined that the decrease in optimal fitness has reached the limit, the number of locked bits is reduced to temporarily open the corresponding channel. Steps 3 to 6 are repeated until at least one genetic span is reached. If the optimal fitness decreases and the temporarily opened channel changes, the temporarily opened channel is designated as the definitely opened channel, and the channel locking status in the current population information is updated accordingly.

[0136] In one optional implementation, reducing the number of locking bits to temporarily release the corresponding channel includes:

[0137] When the currently locked information bit is greater than 0, reduce the locked information bit by one, thereby temporarily releasing the channel corresponding to the reduced locked information bit; where the released information bit is the first information bit in the index of all the previously locked information bits;

[0138] and / or

[0139] When the currently locked frozen bit is greater than 0, reduce the locked frozen bit by one, thereby temporarily releasing the channel corresponding to the reduced locked frozen bit; where the released frozen bit is the last frozen bit indexed among all the previously locked frozen bits.

[0140] Specifically, under the current condition that L is satisfied I ,L F When the value is greater than 0, the search space can be expanded by temporarily releasing one information bit and / or freezing the channel.

[0141] Let's take the example given earlier as an illustration: For instance, if the structure of an individual is 00000**********11111, the 0 at the beginning indicates that there are 5 locked freeze bits, and the 1 at the end indicates that there are 5 locked information bits. If one freeze bit and one information bit are released, then the first and last bits of the ********** part are released, and the structure becomes 0000************1111.

[0142] After the channel is temporarily released, the process returns to the execution of the first genetic iteration corresponding to steps 3 to 6 until the number of genetic iterations reaches at least one genetic iteration span S. This can be set as needed, such as reaching one genetic iteration span S, two genetic iteration spans S, three genetic iteration spans S, etc. After reaching this point, if the optimal fitness decreases and the temporarily released channel changes (i.e., the information bit becomes the frozen bit or the frozen bit becomes the information bit), then the temporarily released channel is designated as the confirmed released channel, and the channel locking status in the current population information determined in step 6 is updated according to the confirmed released channel.

[0143] As can be understood from the above process, in this embodiment of the invention, after reaching the second genetic number span, if the optimal fitness has reached its limit when a genetic number span S is reached, the number of locked channels is dynamically adjusted to open the channels, thereby gradually expanding the range of the genetic algorithm's search space. Therefore, it can solve the problem that in the process of constructing Polar codes, the traditional genetic algorithm has a search space of the entire code length, resulting in a large search space, slow convergence speed, and a large number of genetic generations required for training to reach the convergence state.

[0144] In step 7, if it is determined that the decrease in optimal fitness has not reached the limit, then the number of locked positions remains unchanged, and steps 3 to 6 are repeated. It can be understood that this process continues until another genetic span S is reached, at which point it is again determined whether the decrease in optimal fitness has reached the limit.

[0145] Step 8: For each genetic iteration, the iteration ends when the genetic iteration stopping condition is met.

[0146] In step 8, the genetic iteration stopping condition includes:

[0147] Ineration max Either the optimal fitness of the population converges to the upper limit of the cutoff genetic generations T, or the optimal fitness in the current population information reaches the target value.

[0148] Among these, it is understandable that the upper limit of the number of generations in inheritance is... max The number of generations T is much greater than the number of generations S represented by the number of generations. After the optimal fitness of the population converges, the upper limit of the truncation generation T is greater than the number of generations S represented by the number of generations, which can be set according to the specific code information. The target value can be set as needed and is not restricted here. After the iteration ends, the population information of the current population can be output.

[0149] It should be noted that, in this embodiment of the invention, step 8 is not limited to being executed sequentially after step 7. Rather, for each genetic iteration, a parallel processing approach can be adopted to monitor in real time whether the genetic iteration stops. For example, the detection can be performed after steps 6 and 7. Figure 1 In the diagram, the relevant connections are represented by dashed lines to indicate one possible example connection, but are not intended to limit the embodiments of the present invention.

[0150] For the problem of genetic algorithm code construction, especially Polar code construction, existing schemes, while producing high-performance constructs, suffer from slow convergence speed and high computational cost. This invention employs a dynamic channel-locking scheme, which, compared to existing schemes, reduces the search space of the genetic algorithm while maintaining construction performance, thereby accelerating population fitness convergence, reducing the number of generations required for genetic engineering, and improving training speed.

[0151] To facilitate understanding of the embodiments of the present invention, three specific embodiments are given below for illustration.

[0152] Example 2: Taking the construction of Polar codes under a 128-code-length AWGN channel as an example, the specific steps of implementing the method of this embodiment are as follows:

[0153] For step 1, the parameters are set as follows: Polar code related parameters: code length N = 128; information bit length K = 64; frozen bit length NK = 128 - 64 = 64; code rate R = K / N = 1 / 2; channel condition is AWGN channel; modulation and demodulation method is QPSK modulation and demodulation; decoding method is SCL-Genie, List = 8 decoding. Genetic algorithm related parameters: population size M = 400, upper limit of the number of generations Ineration max =100000, cutoff generation number after fitness convergence T=300, sample fitness concentration in the population α=0.03, mutation probability β=0.01, number of construct signal-to-noise ratios used to calculate fitness J=2; construct signal-to-noise ratio SNR j =1.74, 2.76 dB; The genetic number span S = 50 used to monitor population fitness decline, and the maximum number of locked information bits and frozen bits L are set respectively. I,max =40, L F,max =40.

[0154] For step 2, the construction corresponding to the i-th individual in the population i∈{1,2,...,400} is indexed as C. 89 C 90 ,…,C 128 The channel is set as the information bit channel, with indices C1, C2, ..., C 40 The channel is set to a frozen bit channel. The index is C.41 C 42 ,...,C 87 C 88 The channels are randomly initialized, with 24 channels randomly selected as information bits and the remaining 24 channels set as freeze bits. After generating the above individuals, the fitness of each individual in the population is evaluated. The individuals in the population are sorted from smallest to largest fitness to obtain the initial population.

[0155] For step 3, the roulette wheel algorithm is used to select two parent individuals from the current population. Specifically, for the i-th individual in i∈{1,2,...,400}, the proportion of individuals selected as parents is e. -0.03i Determine the interval corresponding to the i-th individual as follows: The lower limit of the interval corresponding to the first individual is 0; the summation result obtained by accumulating the proportions of individuals in the population is: exist Take a random number within the range, select the individual whose random number falls within the interval, and obtain two different parent individuals P1 and P2.

[0156] For step 4, take the intersection of the information bits of the two parent individuals P1 and P2. right The complement is right The complement is Choose a random integer n from [1, (c-1)] as the number of crossovers. cross Corresponding exchange and The first to the nth cross 1 element, get and respectively with Take the union of the results, randomly select one of them, and retain it as the index set of the offspring information bits in the channel for this genetic process. based on One offspring was obtained from this genetic process.

[0157] Regarding step 5, based on Obtain the set of index values ​​of the unlocked information bits of the offspring in the channel. The set of indices of the unlocked frozen bits of this offspring in the channel. Used for offspring mutation. Based on the mutation probability β = 0.01 and code length N = 128, in Determine the number of mutations, m, where m is a random integer with equal probability selected from the range [0,1]. Select m random elements and replace them with m... Random elements are selected from the offspring, and the offspring undergo mutation.

[0158] For step 6, the fitness of the offspring is calculated, and the offspring are sorted in the current population according to their fitness. The individual with the worst fitness in the population is eliminated, and the population is updated after one genetic event. The current population information, including channel construction, optimal fitness, and channel locking status, is recorded.

[0159] For step 7, repeat the genetic process corresponding to steps 3 to 6. Starting from the second 50 times of genetic inheritance, after each 50 times of genetic inheritance, determine whether the optimal fitness meets the first locking number adjustment condition or the second locking number adjustment condition. The first locking number adjustment condition is: the optimal fitness of the population does not decrease after every 50 times of genetic inheritance, and the optimal fitness of the population decreased within the previous 50 times of genetic inheritance. The second locking number adjustment condition is: the optimal fitness of the population does not decrease after every 100 consecutive times of genetic inheritance.

[0160] If so, reduce the number of locked information bits and / or frozen bits to temporarily open the corresponding channel, and repeat steps 3 to 6 until at least one 50-time inheritance is reached. If the optimal fitness decreases and the temporarily opened channel changes, the temporarily opened channel is designated as the definitely opened channel, and the channel locking status in the current population information is updated accordingly. If not, keep the number of locked bits unchanged and repeat steps 3 to 6.

[0161] For step 8, for each genetic iteration, the iteration ends when the genetic iteration stopping condition is met.

[0162] Figure 2(a) shows the performance convergence curve of the genetic algorithm Polar code construction in Example 2, and Figure 2(b) shows the performance curve of the genetic algorithm Polar code construction in Example 2. Figure 2(a) shows the optimal construction of the population at E... s The BLER convergence curves at / N0 = 2.76dB are shown in the figure. Different curves correspond to the cases of fixing a certain number of information bits and freezing bits, respectively. For example, lock5 is a case where the population is fixedly locked with 5 information bits and 5 frozen bits, and the remaining 118 channels are randomly initialized, crossovered, and mutated; lock0 is a case where all channels are randomly initialized, crossovered, and mutated, which is the performance convergence curve of the existing genetic algorithm code construction scheme; lock-learn is the performance convergence curve of the dynamic channel locking genetic algorithm proposed in this invention. E s It is the average energy of each transmitted symbol in the discrete-time channel, N0 is the one-sided noise power spectral density, and E s / N0 is one way to define signal-to-noise ratio (SNR).

[0163] Figure 2(b) shows the performance simulation curves corresponding to different constructions. The different curves in the figure correspond to the cases of fixing and locking a certain number of information bits and freezing bits, respectively. The curve corresponding to Gaussian approximation is the simulation performance curve of the Gaussian approximation construction, with a signal-to-noise ratio of -1.59 dB. The dynamic channel locking genetic algorithm of this embodiment is used... Figure 2a The convergence speed and convergence performance of the lock-learn curve are significantly better than those of a randomly initialized population. Figure 2a (The curve corresponding to lock0 in the middle). Because this invention can dynamically adjust the number of locked channels, compared to the case of fewer locked channels (such as... Figure 2a The learned channel locking (corresponding to the lock 20 curve) allows the population to converge faster; compared to cases with more locked channels (such as... Figure 2a Lock 30, Figure 2b (The curve corresponding to lock40 in the middle) shows that the learned channel locking can improve the performance of population construction; at the same time, according to Figure 2b The curves corresponding to lock-learn and lock0 show that the construction of the dynamic channel-locking genetic algorithm can achieve a speedup of 10 compared to the construction of a randomly initialized population. -5 The hierarchical performance improved by 0.2 dB (a common metric used to describe differences in simulation performance).

[0164] Example 3, taking the construction of Polar codes under a 256-code-length AWGN channel as an example, the specific steps of implementing the method of this embodiment are as follows:

[0165] For step 1, the parameters are set as follows: Polar code related parameters: code length N = 256; information bit length K = 128; frozen bit length NK = 256 - 128 = 128; code rate R = K / N = 1 / 2; channel condition is AWGN channel; modulation and demodulation method is QPSK modulation and demodulation; decoding method is SCL-Genie List = 8 decoding. Genetic algorithm related parameters: population size M = 400, upper limit of the number of generations Ineration max =100000, cutoff generation number after fitness convergence T=400, sample fitness concentration in the population α=0.03, mutation probability β=0.01, number of construct signal-to-noise ratios used to calculate fitness J=2; construct signal-to-noise ratio SNR j =1.17, 1.88 dB; The genetic number span S used to monitor population fitness decline is 50, and the maximum number of locked information bits and frozen bits L are set respectively. I,max =80, L F,max =80.

[0166] For step 2, the construction corresponding to the i-th individual in the population i∈{1,2,...,400} is indexed as C.177 C 178 ,…,C 256 The channel is set as the information bit channel, with indices C1, C2, ..., C 80 The channel is set to a frozen bit channel. The index is C. 81 C 82 ,...,C 175 C 176 The channels are randomly initialized, with 48 channels randomly selected as information bits and the remaining 48 channels set as freeze bits. After generating the above individuals, the fitness of each individual in the population is evaluated. The individuals in the population are sorted from smallest to largest fitness to obtain the initial population.

[0167] For step 3, the roulette wheel algorithm is used to select two parent individuals from the current population. Specifically, for the i-th individual in i∈{1,2,...,400}, the proportion of individuals selected as parents is e. -0.03i Determine the interval corresponding to the i-th individual as follows: The lower limit of the interval corresponding to the first individual is 0; the summation result obtained by accumulating the proportions of individuals in the population is: exist Take a random number within the range, select the individual whose random number falls within the interval, and obtain two different parent individuals P1 and P2.

[0168] For step 4, take the intersection of the information bits of the two parent individuals P1 and P2. right The complement is right The complement is Choose a random integer n from [1, (c-1)] as the number of crossovers. cross Corresponding exchange and The first to the nth cross 1 element, get and respectively with Take the union of the results, randomly select one of them, and retain it as the index set of the offspring information bits in the channel for this genetic process. based on One offspring was obtained from this genetic process.

[0169] Regarding step 5, based on Obtain the set of index values ​​of the unlocked information bits of the offspring in the channel. The set of indices of the unlocked frozen bits of this offspring in the channel. Used for offspring mutation. Based on the mutation probability β = 0.01 and code length N = 256, in Determine the number of mutations, m, where m is a random integer chosen with equal probability from the range [0, 1, 2]. Select m random elements and replace them with m... Random elements are selected from the offspring, and the offspring undergo mutation.

[0170] For step 6, the fitness of the offspring is calculated, and the offspring are sorted in the current population according to their fitness. The individual with the worst fitness in the population is eliminated, and the population is updated after one genetic event. The current population information, including channel construction, optimal fitness, and channel locking status, is recorded.

[0171] For step 7, repeat the genetic process corresponding to steps 3 to 6. Starting from the second 50 times of genetic inheritance, after each 50 times of genetic inheritance, determine whether the optimal fitness meets the first locking number adjustment condition or the second locking number adjustment condition. The first locking number adjustment condition is: the optimal fitness of the population does not decrease after every 50 times of genetic inheritance, and the optimal fitness of the population decreased within the previous 50 times of genetic inheritance. The second locking number adjustment condition is: the optimal fitness of the population does not decrease after every 100 consecutive times of genetic inheritance.

[0172] If so, reduce the number of locked information bits and / or frozen bits to temporarily open the corresponding channel, and repeat steps 3 to 6 until at least one 50-time inheritance is reached. If the optimal fitness decreases and the temporarily opened channel changes, the temporarily opened channel is designated as the definitely opened channel, and the channel locking status in the current population information is updated accordingly. If not, keep the number of locked bits unchanged and repeat steps 3 to 6.

[0173] For step 8, for each genetic iteration, the iteration ends when the genetic iteration stopping condition is met.

[0174] Figure 3(a) shows the performance convergence curve of the genetic algorithm Polar code construction in Example 3, and Figure 3(b) shows the performance curve of the genetic algorithm Polar code construction in Example 3. Figure 3(a) shows the optimal construction of the population at E... sThe BLER convergence curves at / N0 = 1.88dB are shown in the figure. Different curves correspond to the cases of fixing a certain number of information bits and freezing bits, respectively. For example, lock20 represents a population with 20 information bits and 20 frozen bits, with the remaining 216 channels undergoing random initialization, crossover, and mutation. lock0 represents the performance convergence curve of existing genetic algorithm code construction schemes, where all channels undergo random initialization, crossover, and mutation. lock-learn represents the performance convergence curve of the dynamic channel-locking genetic algorithm proposed in this invention. The comparison shows that the convergence speed of the dynamic channel-locking genetic algorithm in this embodiment is significantly better than that of the randomly initialized population.

[0175] Example 4, taking the construction of Polar codes under a 128-code-length independent Rayleigh fading channel as an example, the specific steps of implementing the method of this embodiment are as follows:

[0176] For step 1, the parameters are set as follows: Polar code related parameters: code length N = 128; information bit length K = 64; frozen bit length NK = 128 - 64 = 64; code rate R = K / N = 1 / 2; channel condition is independent Rayleigh fading channel; modulation and demodulation method is QPSK modulation and demodulation; decoding method is SCL-Genie List = 8 decoding. Genetic algorithm related parameters: population size M = 400, upper limit of the number of generations Ineration max =100000, cutoff generation number after fitness convergence T=300, sample fitness concentration in the population α=0.03, mutation probability β=0.01, number of construct signal-to-noise ratios used to calculate fitness J=2; construct signal-to-noise ratio SNR j =4.1, 5.4 dB; The genetic number span S used to monitor population fitness decline is 50, and the maximum number of locked information bits and frozen bits L are set respectively. I,max =40, L F,max =40.

[0177] For step 2, the construction corresponding to the i-th individual in the population i∈{1,2,...,400} is indexed as C. 89 C 90 ,...,C 128 The channel is set as the information bit channel, with indices C1, C2, ..., C 40 The channel is set to a frozen bit channel. The index is C. 41 C 42 ,...,C 87 C 88The channels are randomly initialized, with 24 channels randomly selected as information bits and the remaining 24 channels set as freeze bits. After generating the above individuals, the fitness of each individual in the population is evaluated. The individuals in the population are sorted from smallest to largest fitness to obtain the initial population.

[0178] For step 3, the roulette wheel algorithm is used to select two parent individuals from the current population. Specifically, for the i-th individual in i∈{1,2,...,400}, the proportion of individuals selected as parents is e. -0.03i Determine the interval corresponding to the i-th individual as follows: The lower limit of the interval corresponding to the first individual is 0; the summation result obtained by accumulating the proportions of individuals in the population is: exist Take a random number within the range, select the individual whose random number falls within the interval, and obtain two different parent individuals P1 and P2.

[0179] For step 4, take the intersection of the information bits of the two parent individuals P1 and P2. right The complement is right The complement is Choose a random integer n from [1, (c-1)] as the number of crossovers. cross Corresponding exchange and The first to the nth cross 1 element, get and respectively with Take the union of the results, randomly select one of them, and retain it as the index set of the offspring information bits in the channel for this genetic process. based on One offspring was obtained from this genetic process.

[0180] Regarding step 5, based on Obtain the set of index values ​​of the unlocked information bits of the offspring in the channel. The set of indices of the unlocked frozen bits of this offspring in the channel. Used for offspring mutation. Based on the mutation probability β = 0.01 and code length N = 128, in Determine the number of mutations, m, where m is a random integer with equal probability selected from the range [0,1]. Select m random elements and replace them with m... Random elements are selected from the offspring, and the offspring undergo mutation.

[0181] For step 6, the fitness of the offspring is calculated, and the offspring are sorted in the current population according to their fitness. The individual with the worst fitness in the population is eliminated, and the population is updated after one genetic event. The current population information, including channel construction, optimal fitness, and channel locking status, is recorded.

[0182] For step 7, repeat the genetic process corresponding to steps 3 to 6. Starting from the second 50 times of genetic inheritance, after each 50 times of genetic inheritance, determine whether the optimal fitness meets the first locking number adjustment condition or the second locking number adjustment condition. The first locking number adjustment condition is: the optimal fitness of the population does not decrease after every 50 times of genetic inheritance, and the optimal fitness of the population decreased within the previous 50 times of genetic inheritance. The second locking number adjustment condition is: the optimal fitness of the population does not decrease after every 100 consecutive times of genetic inheritance.

[0183] If so, reduce the number of locked information bits and / or frozen bits to temporarily open the corresponding channel, and repeat steps 3 to 6 until at least one 50-time inheritance is reached. If the optimal fitness decreases and the temporarily opened channel changes, the temporarily opened channel is designated as the definitely opened channel, and the channel locking status in the current population information is updated accordingly. If not, keep the number of locked bits unchanged and repeat steps 3 to 6.

[0184] For step 8, for each genetic iteration, the iteration ends when the genetic iteration stopping condition is met.

[0185] Figure 4(a) shows the performance convergence curve of the genetic algorithm for Polar code construction in Example 4, and Figure 4(b) shows the performance curve of the genetic algorithm for Polar code construction in Example 4. Figure 4(a) shows the optimal construction of the population at E... s The BLER curves at / N0 = 5.4dB are shown in the figure. Different curves correspond to the cases of fixing a certain number of information bits and freezing bits, respectively. For example, lock10 is the population that is constantly locked with 10 information bits and 10 frozen bits, and the remaining 108 channels are randomly initialized, crossover and mutated; lock0 is the performance convergence curve of the existing genetic algorithm code construction scheme, where all channels are randomly initialized, crossover and mutated; lock-learn is the performance convergence curve of the dynamic channel locking genetic algorithm proposed in this invention.

[0186] Figure 4(b) shows the performance simulation curves corresponding to different configurations. In the figure, lock0 corresponds to the genetic algorithm configuration with a randomly initialized population, lock-learn corresponds to the dynamic channel locking genetic algorithm configuration of this invention, and the curve corresponding to Gauss is the simulation performance curve of the Gaussian approximation configuration, with a signal-to-noise ratio of -1.59dB. It can be seen that the convergence speed of the dynamic channel locking genetic algorithm of this invention is significantly better than that of the randomly initialized population.

[0187] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A polar code construction method based on genetic algorithm to accelerate convergence, characterized in that, include: Step 1: Set parameters, including polar code-related parameters and genetic algorithm-related parameters; Step 2: Initialize the genetic algorithm population based on the set parameters; evaluate the fitness of individuals in the population; sort individuals according to their fitness to obtain the initial population; wherein, the individuals in the population represent the Polar code channel construction, which is divided into information channel and frozen channel, and the frozen bits and information bits in the individuals are locked according to the maximum number of locks corresponding to the relevant parameters of the genetic algorithm. Step 3: Use the roulette wheel algorithm to select two parent individuals from the current population; Step 4: Cross-process the information bits in the two parent individuals to generate a child; Step 5: Using the unlocked frozen bits of the offspring, perform mutation processing on the unlocked information bits to obtain the mutated offspring; Step 6: After evaluating the fitness of the mutated offspring, add them to the current population, sort and select individuals according to their fitness, realize the population update after one genetic step, and record the current population information, including channel construction, optimal fitness, and channel locking status. Step 7: Repeat the genetic process corresponding to steps 3 to 6 until the number of genetic iterations reaches the genetic iteration span in the relevant parameters of the genetic algorithm. Determine whether the decrease in optimal fitness has reached its limit. If so, reduce the number of locked bits to temporarily open the corresponding channel. Repeat steps 3 to 6 until at least one genetic iteration span is reached. If the optimal fitness decreases and the temporarily opened channel changes, then the temporarily opened channel is designated as the definitely opened channel, and the channel locking status in the current population information is updated accordingly. If not, keep the number of locked bits unchanged and repeat steps 3 to 6. Step 8: For each genetic iteration, the iteration ends when the genetic iteration stopping condition is met.

2. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 1, characterized in that, The polar code related parameters include: code length N, information bit length K, frozen bit length NK, code rate R = K / N, channel conditions, modulation and demodulation methods, and decoding methods. The parameters related to the genetic algorithm include: population size M, upper limit of the number of generations (Ineration). max The number of generations T after fitness convergence, the concentration of fitness samples α in the population, the mutation probability β, the number of construct signal-to-noise ratios J used to calculate fitness, and the construct signal-to-noise ratio (SNR). j The genetic span S used to monitor population fitness decline, and the maximum number of locked information bits L. I,max Maximum number of frozen bits L F,max Where, j∈{1,2,...,J}, L I,max <K,L F,max <(NK), S <T;N、K、N-K、M、Ineration max J, S, T, L I,max L F,max All are integers greater than 0.

3. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 2, characterized in that, Step 2, which initializes the genetic algorithm population based on the set parameters, includes: For the i-th individual in the population, the corresponding channel index is represented as Where i∈{1,2,...,M}; the information bits are represented as follows: The freeze bit is represented as Index it as The channel is set as the information bit channel, and the index is... The channel is set to a frozen channel; At index Randomly selected from the channels (KL) I,max ) channels are set as information bits, the rest (NKL) F,max Set ) channels as frozen bits to obtain the channel construction for the i-th individual; The initialized genetic algorithm population is obtained by constructing channels from M individuals.

4. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 2, characterized in that, In step 2, the fitness of individuals in the population is assessed using formulas including: Where i∈{1,2,...,M}; BLER represents the fitness of the i-th individual in the population. j This indicates that the i-th individual is constructed based on the signal-to-noise ratio (SNR). j The obtained block error rate.

5. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 2 or 3, characterized in that, In step 3, the roulette wheel algorithm is used to select two parent individuals from the current population, including: For the i-th individual in the current population, based on the concentration of fitness α in the population, calculate the proportion of the i-th individual that will be selected as the parent as e. -αi Where i∈{1,2,...,M}; The interval corresponding to the i-th individual is determined as follows: The lower limit of the interval corresponding to the first individual is 0; Summing the proportions of all individuals in the current population selected as parents yields the following summation: exist Take a random number within the range, and determine the corresponding individual as the selected parent individual based on the interval in which the random number falls. In this way, select two different parent individuals P1 and P2 from the current population.

6. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 5, characterized in that, In step 4, the information bits in the two parent individuals are cross-processed to generate a child, including: Information bit P1 of the parent individual and the parent individual P2 information bit Take the intersection as Sure right supplement as well as right supplement Among them, set The set represents the index in the channel of the different information bits between parent individual P1 and parent individual P2. The index in the channel represents the different information bits of parent individual P2 compared to parent individual P1; Choose a random integer n from [1, (c-1)] as the number of crossovers. cross Corresponding exchange and The first to the nth cross 1 element, get Crossed set as well as Crossed set in, Will respectively with Taking the union of the two sets, we obtain the set of the channel indices for the two exchanged information bits. as well as Random selection and One of the index sets in the channel as child information bits based on One offspring was obtained from this genetic process.

7. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 6, characterized in that, In step 5, the unlocked information bits of the offspring are mutated using its unlocked frozen bits to obtain mutated offspring, including: Based on the indices of the locked and frozen information bits, and the set of indices of the unlocked information bits of the offspring in the channel. Obtain the set of indices of the unlocked frozen bits of the offspring in the channel. in, This represents the complete channel index of this offspring. L I L represents the number of information bits currently locked. F Indicates the number of currently frozen bits locked; Based on the mutation probability β and the code length N, in Based on the principle of equal probability selection, a random integer is determined to obtain the number of mutations m; Will Replace m random elements with Given m random elements, mutated offspring are obtained based on the replacement results.

8. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 7, characterized in that, Step 7, determining whether the decrease in optimal fitness has reached a limit, includes: When the current number of inheritance iterations has accumulated to at least two inheritance intervals, determine whether the optimal fitness satisfies either the first or second locking number adjustment condition. If so, it is determined that the decrease in optimal fitness has reached its limit; otherwise, it is determined that the decrease in optimal fitness has not reached its limit. The first locking number adjustment condition includes: the optimal fitness of the population does not decrease after the genetic iteration corresponding to the current genetic number span, and the optimal fitness of the population decreases within the genetic iteration of the previous genetic number span. The second locking number adjustment condition includes: the optimal fitness of the population does not decrease after two consecutive genetic iterations corresponding to two genetic spans.

9. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 8, characterized in that, In step 7, reducing the number of locked bits to temporarily open the corresponding channel includes: When the currently locked information bit is greater than 0, reduce the locked information bit by one, thereby temporarily releasing the channel corresponding to the reduced locked information bit; where the released information bit is the first information bit in the index of all the previously locked information bits; and / or When the currently locked frozen bit is greater than 0, reduce the locked frozen bit by one, thereby temporarily releasing the channel corresponding to the reduced locked frozen bit; where the released frozen bit is the last frozen bit indexed among all the previously locked frozen bits.

10. The polar code construction method based on genetic algorithm to accelerate convergence according to claim 2, characterized in that, In step 8, the genetic iteration stopping conditions include: Ineration max Either the optimal fitness of the population converges to the upper limit of the cutoff genetic generations T, or the optimal fitness in the current population information reaches the target value.

Images