PMI global optimization method based on genetic algorithm

Abstract

The invention relates to a PMI global optimization method based on a genetic algorithm, and belongs to the technical field of communication signal processing. According to the method, the law of article competition and survival of the fittest in the nature is used for survival of the victims and the inferior. And the algorithm can converge more quickly. In addition, according to the algorithm, gene mutation, crossover, forced mutation and other algorithms are carried out at proper positions, so that the algorithm has the gene evolution capacity, and the algorithm can jump out of the local optimal solution more quickly.

Description

Technical Field

[0001] The invention belongs to the technical field of communication signal processing, and in particular relates to a PMI global optimization method based on a genetic algorithm. Background Art

[0002] In order to solve the problem of large average calculation amount of the traditional calculation method of precoding matrix indication (PMI) in 5G system, a low-complexity PMI calculation algorithm is proposed. This algorithm combines genetic algorithm and uses the law of "survival of the fittest" in nature to carry out "survival of the fittest". It can make the algorithm converge more quickly.

[0003] All 5G downlink physical channel demodulation depends on the DM-RS corresponding to the channel. In addition, the terminal needs to assume that the network has precoded the data part and DM-RS in the same way. Therefore, any downlink multi-antenna precoding used by the network is transparent to the terminal, and the network can freely decide the downlink precoding.

[0004] The impact of downlink multi-antenna precoding on the protocol is mainly focused on how the terminal performs measurements and reports in order to support the network side in selecting the precoding matrix for downlink PDSCH transmission. These precoding-related measurements and reporting mechanisms are part of the CSI reporting framework.

[0005] Therefore, the PMI reported by the terminal represents the optimal precoding matrix that the terminal considers to be available for downlink transmission. Each PMI corresponds to a precoding matrix. The precoding matrices corresponding to all possible PMIs are collectively called a precoding codebook, from which the terminal selects the optimal PMI. How to quickly and accurately search for an effective PMI requires the personnel in this field to focus on. Summary of the invention

[0006] The present invention provides a PMI global optimal search solution in 5G based on a genetic algorithm with evolutionary capabilities, which has the ability of genetic evolution and can enable the solution to jump out of the local optimal solution more quickly.

[0007] The present invention is achieved through the following technical solutions:

[0008] A PMI global optimal method based on genetic algorithm, the specific process is as follows:

[0009] Step 1: Establish a 5G codebook, where the 5G codebook adopts a two-level codebook structure W=W1W2; W1 describes the long-term broadband characteristics of the channel, and W2 describes the short-term subband characteristics of the channel, which are used to perform column selection and phase adjustment on the beams in W;

[0010] The first-level codebook is based on a block diagonal structure, where each diagonal block represents a beam group in a polarization direction, which is obtained by calculating the Kronecker product of the beam grouping in the first dimension and the beam grouping in the second dimension.

[0011]

[0012] Among them, i 11 represents the index of the first dimension beam group X1, i 12 represents the index of the second-dimensional beam group X2; X1 is an L1XN1-dimensional matrix consisting of L1 DFT vectors of length N1; each vector is oversampled by O1 times and is expressed as:

[0013]

[0014] X2 is an N 2X The L2-dimensional matrix is ​​composed of L2 DFT vectors of length N2; each vector is oversampled by O2 times and is expressed as:

[0015]

[0016] The expression is as follows:

[0017]

[0018]

[0019] Among them, (s1,s 2) Defined as the beam group spacing in the first and second dimensions, it represents the difference between the indices of the first beams in two adjacent beam groups. (p1,p 2) Defined as the beam spacing within a beam group, it represents the spacing between adjacent beams in X1, X2;

[0020] W2 represents the beam selection and the phase between the two polarization directions, Y∈{e i ,i=1,2,3,4,e i , is a column vector of length 4, the i-th element is 1, and the rest are 0};

[0021] Step 2: Perform a quick PMI search in the 5G codebook using a genetic algorithm.

[0022] Furthermore, the process of genetic algorithm quickly searching in the 5G codebook is as follows:

[0023] Step 201, the genetic algorithm is defined as follows:

[0024] The population is Ω=[A1,A2,…,A J ], is the jth one and also the jth individual in the population;

[0025] Definition a. Let the number of individual gene positions be K, and individual A j The i gene loci in From the collection Any value in the set is the allele of the ith gene locus, where M is the total number of allele values;

[0026] Definition b. The matrix composed of the probabilities of the values ​​in the corresponding allele set that each gene position can take is called the probability matrix, which is expressed as P = [p mk ] MK ; Among them, p mk Indicates that at the kth gene position, the value is The probability of the mth element in the set, where m = 1, ..., M;

[0027] Definition c. Each column in the probability matrix P is called a probability vector, and the sum of all elements of each probability vector is 1;

[0028] Step 2: Initialize the probability matrix by using the allele equal probability method, and set each element of the initial probability matrix to

[0029] Define gene information entropy E, and its formula is as follows:

[0030]

[0031] The maximum gene information entropy value E max The evolution rate Ratio is expressed as follows:

[0032]

[0033] The evolution rate indicates the degree of civilization development, and its range is 0 to 1; the higher the evolution rate value, the higher the level of civilization and the higher the degree of evolution;

[0034] Step 3, using roulette wheel selection method to generate population individuals;

[0035] Step 4: randomly mutate a gene position of an individual in the population to its allele; after the mutation, force mutation on the same individuals in the population and completely randomly mutate the redundant individuals into another genotype;

[0036] Step 5, perform gene crossover on the mutated population individuals;

[0037] Step 6: UE first calculates the signal-to-noise ratio of each layer of each RB of each PMI under each RI represents possible RI; k∈{1,…N RB} represents the size of RB;,l∈{1,…r}

[0038] It indicates the signal-to-noise ratio of each layer at a certain RI and a certain PMI;

[0039]

[0040] in H CsiRs is the channel transmission matrix for each RB, W is the encoding under a certain RI, R uu Represents the covariance matrix of interference and noise, generally at the subband level;

[0041] W is strongly associated with chromosomes;

[0042] Objective function:

[0043]

[0044] [i11,i12,i13,i2]=arg max(value)

[0045] Step 7, Individual A j The i-th gene position in From the collection Any value in . is the allele of the ith gene locus, u j Indicates the competitive advantage of the jth genotype, and the default value is 0;

[0046] Select strategy:

[0047] First, if the jth genotype is the optimal genotype, then u j =u j +1;

[0048] Second, when the evolution degree Ratio≤0.5, if the jth genotype is a suboptimal genotype, then

[0049] Third, when the evolution degree Ratio≤0.25, if the jth genotype is the optimal genotype, then

[0050] Fourth, if the jth genotype is the worst genotype, then u j =u j -1;

[0051] Step 8, the probability vector is modified by the information of the best individual in each generation of the population, and the update rule of the probability vector is expressed as:

[0052] Pt+1 =(1-a)*P t +a*u

[0053] Among them, P t is the probability vector used by the algorithm in the tth iteration, a is the evolution rate in the competitive evolution mechanism, and u is the fitness of the genotype in natural selection;

[0054] when season is equal to the positive minimum;

[0055] The evolution rate a is expressed as:

[0056] After gene evolution, the probability distribution is normalized and the calculation formula is as follows:

[0057]

[0058] Step 9, through the modified probability matrix P t+1 Generate individuals in the next generation population and perform step 3 until the iteration termination condition is reached or the evolution rate Ratio reaches a preset threshold, and the optimal precoding matrix is ​​obtained.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The method provided by the present invention introduces a genetic algorithm to accelerate codebook search according to the law of "survival of the fittest" in nature. Simulation results show that the proposed search method significantly reduces the search complexity while ensuring system performance and will not fall into a local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a genetic diagram of NR codebook parameters according to an embodiment of the present invention.

[0062] Figure 2 It is a population individual formed by roulette in an embodiment of the present invention.

[0063] Figure 3 It is an allele map after gene mutation in an embodiment of the present invention.

[0064] Figure 4 It is a forced mutation gene map of an embodiment of the present invention.

[0065] Figure 5 It is a gene map after crossover in an embodiment of the present invention.

[0066] Figure 6 4 is a flow chart of a genetic algorithm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0068] The 5G codebook uses a parameterized codebook determined by a unified codebook framework combined with several codebook parameters, and adopts a two-level codebook structure W=W1W2. W1 describes the long-term broadband characteristics of the channel; W2 describes the short-term subband characteristics of the channel, which is used to select columns and adjust phases of beams in W.

[0069] The first-level codebook is based on a block diagonal structure, where each diagonal block represents a beam group in a polarization direction, which is obtained by performing a Kronecker product calculation on the beam grouping in the first dimension and the beam grouping in the second dimension.

[0070]

[0071] Among them, i 11 represents the index of the first dimension beam group X1, i 12 Represents the index of the second-dimensional beam group X2. X1 is an L1XN1-dimensional matrix consisting of L1 DFT vectors of length N1. Each vector is oversampled by O1 times and is expressed as:

[0072]

[0073] X2 is an N 2X The L2-dimensional matrix is ​​composed of L2 DFT vectors of length N2. Each vector is oversampled by O2 times and is expressed as

[0074]

[0075] The expression is as follows:

[0076]

[0077] Where (s1, s2) is defined as the beam group spacing of the first and second dimensions, indicating the difference between the indices of the first beams in two adjacent beam groups. (p1, p2) is defined as the beam spacing within a beam group, indicating the spacing between adjacent beams in X1, X2.

[0078] W2 represents the beam selection (when the beam is 4) and the phase between the two polarization directions. Y∈{e i ,i=1,2,3,4,e i is a column vector of length 4, with the i-th element being 1 and the rest being 0}.

[0079] This algorithm combines genetic algorithm to perform fast PMI search. First, the following definitions are given:

[0080] The population is Ω = [A1, A2, ..., A J ], is the jth one and also the jth individual in the population.

[0081] Definition 1: Let the number of individual gene positions be K, and individual A j The i gene loci in From the collection Any value in . is the allele of the ith gene locus, where M is the total number of allele values.

[0082] Definition 2. The matrix consisting of the probabilities of each gene position taking the value in the corresponding allele set is called the probability matrix, which can be expressed as P = [p mk ] MK Among them, p mk Indicates that at the kth gene position, the value is The probability of the mth element in the set, where m=1,...,M.

[0083] Definition 3. Each column in the probability matrix P is called a probability vector, and the sum of all elements of each probability vector is set to 1;

[0084] The codebook elements in NR are wideband codebook i1 and narrowband codebook i2, where i1 contains i 11 、i 12 、i 13 Three elements, we will NR codebook parameter i 11 、i 12 、i 13 The four elements i and i2 are defined as the four genes of an individual. Figure 1 shown.

[0085] In order to ensure that each allele can be selected fairly, we use the allele equal probability method to initialize the probability matrix and set each element of the initial probability matrix to

[0086] Define gene information entropy E, and its formula is as follows:

[0087]

[0088] In the early stages of evolution, the probability of each gene position taking its allele value is equal, and the gene information entropy value is the maximum E max The evolution rate Ratio is expressed as follows:

[0089]

[0090] The evolution rate indicates the degree of civilization development, and its range is 0 to 1. The higher the evolution rate value, the higher the civilization and the higher the degree of evolution. This value affects the evolution rate, natural selection, and algorithm convergence.

[0091] Once the gene probability matrix is ​​known, it is needed to guide the generation of individuals in the population. The Roulette Wheel Selection (RWS) method is used to generate individuals in the population. Figure 2 shown.

[0092] Randomly mutate a gene position in the population and change it into its allele. Its goal is to increase the genetic diversity of the population, and the probability of mutation is relatively small. Figure 3 shown.

[0093] In the later stages of population evolution, the probability of some genotypes is very high, so after roulette, identical individuals may appear in the population. At this time, forced mutation is needed to randomly mutate the redundant individuals into another genotype. The goal is to increase the genetic diversity of the population. Figure 4 shown.

[0094] Genetic crossover also occurs during the evolution process. Crossover is the crossing of gene fragments between two chromosomes. Where the crossover occurs is random, and whether the crossover occurs is also random. Figure 5 shown.

[0095] The UE first calculates the signal-to-noise ratio of each layer of each RB for each PMI under each RI represents possible RI; k∈{1,…N RB} represents the size of RB;,l∈{1,…r}

[0096] It indicates the signal-to-noise ratio of each layer at a certain RI and a certain PMI.

[0097]

[0098] in H CsiRs is the channel transmission matrix for each RB, W is the encoding under a certain RI, R uu Represents the covariance matrix of interference and noise, generally at the subband level.

[0099] W is affected by i11, i12, i13, and i2, that is, it is strongly related to chromosomes

[0100] Objective function:

[0101]

[0102] [i11,i12,i13,i2]=arg max(value)

[0103] The genotype of individuals after natural selection should account for a larger proportion in the next generation, so how to increase the probability of the genotype after natural selection requires certain strategies.

[0104] Individual A j The i-th gene position in From the collection Any value in . is the allele of the ith gene locus, u j Represents the competitive advantage of the j-th genotype, and the default value is 0.

[0105] Select strategy:

[0106] 1. If the jth genotype is the optimal genotype, then u j =u j +1

[0107] 2. When the evolution degree Ratio≤0.5, if the jth genotype is a suboptimal genotype, then

[0108] 3. When the evolution degree Ratio≤0.25, if the jth genotype is the optimal genotype, then

[0109] (In particular, condition 3 does not apply to i 13 , i2)

[0110] 4. If the jth genotype is the worst genotype, then u j =u j -1

[0111] In the process of population evolution, the information of the best individual in each generation of the population is used to correct the probability vector, and the update rule of the probability vector can be expressed as:

[0112] P t+1 =(1-a)*P t +a*u

[0113] Among them, P t is the probability vector used by the algorithm in the tth iteration, a is the evolution rate in the competitive evolution mechanism, and u is the fitness of the genotype in natural selection.

[0114] In particular, due to "survival of the inferior", It may be a negative number, which does not meet the definition of probability. season Equal to the positive minimum

[0115] Assume that the evolution rate is set to a fixed value. If the evolution rate is set too high, the population can quickly move toward the optimal solution in the solution space in the early stage of evolution, thereby improving the efficiency of the algorithm. However, in the later stage of evolution, the excessive evolution rate will lead to a large step size, reducing the accuracy of the algorithm. Similarly, if the evolution rate is set too low, although the accuracy of the algorithm in the later stage of evolution can be guaranteed, it will cause the population in the early stage of evolution to explore the optimal solution in the solution space slowly, reducing the efficiency of the algorithm.

[0116] The evolution rate in the algorithm can be expressed as:

[0117]

[0118] After genetic evolution, the probability distribution needs to be normalized to facilitate the evolution of the next generation of populations. The calculation formula is as follows:

[0119]

[0120] Using the modified probability matrix P t+1 Generate individuals in the next generation population;

[0121] Finally, the competition and population evolution process is repeated, gradually approaching the optimal solution until the iteration termination condition is reached. After the evolution is completed, each element of the probability vector will approach 0 or 1 respectively.

[0122] When the number of algorithm iterations reaches the upper limit, or the evolution rate Ratio reaches the preset threshold, it means that the algorithm has converged.

[0123] The overall process is as follows Figure 6 shown.

[0124] It should be noted that the above description is only a preferred application example of the present invention and is not intended to limit the protection scope of the present invention. Any technical solution using equivalent replacement or equivalent transformation is within the protection scope of the present invention.

Claims

1. A PMI global optimization method based on genetic algorithm, characterized in that: The specific process is as follows: Step 1: Establish a 5G codebook, where the 5G codebook adopts a two-level codebook structure W=W1W2; W1 describes the long-term broadband characteristics of the channel, and W2 describes the short-term subband characteristics of the channel, which are used to perform column selection and phase adjustment on the beams in W; The first-level codebook is based on a block diagonal structure, where each diagonal block represents a beam group in a polarization direction, which is obtained by calculating the Kronecker product of the beam grouping in the first dimension and the beam grouping in the second dimension. Among them, i 11 represents the index of the first dimension beam group X1, i 12 represents the index of the second-dimensional beam group X2; X1 is an L1XN1-dimensional matrix, consisting of L1 DFT vectors of length N1; each vector is oversampled by O1 times, expressed as: X2 is an N 2X The L2-dimensional matrix is ​​composed of L2 DFT vectors of length N2; each vector is oversampled by O2 times and is expressed as: The expression is as follows: Where (s1, s2) is defined as the beam group spacing of the first dimension and the second dimension, indicating the difference between the indices of the first beams in two adjacent beam groups. (p1, p2) is defined as the beam spacing within the beam group, indicating the spacing between adjacent beams in X1, X2; W2 represents the beam selection and the phase between the two polarization directions, is a column vector of length 4, the i-th element is 1 and the rest are 0}; Step 2: Perform a quick PMI search in the 5G codebook using a genetic algorithm.

2. The PMI global optimization method based on genetic algorithm according to claim 1, characterized in that: The process of genetic algorithm quickly searching in the 5G codebook is as follows: Step 201, the genetic algorithm is defined as follows: The population is Ω=[A1,A2,…,A J ], is the jth one and also the jth individual in the population; Definition a. Let the number of individual gene positions be K, and individual A j The i gene loci in From the collection Any value in the set is the allele of the ith gene locus, where M is the total number of allele values; Definition b. The matrix composed of the probabilities of the values ​​in the corresponding allele set that each gene position can take is called the probability matrix, which is expressed as P = [p mk ] MK ; Among them, p mk Indicates that at the kth gene position, the value is The probability of the mth element in the set, where m = 1, ..., M; Definition c. Each column in the probability matrix P is called a probability vector, and the sum of all elements of each probability vector is 1; Step 2: Initialize the probability matrix by using the allele equal probability method, and set each element of the initial probability matrix to Define gene information entropy E, and its formula is as follows: The maximum gene information entropy value E max The evolution rate Ratio is expressed as follows: The evolution rate indicates the degree of civilization development, and its range is 0 to 1; the higher the evolution rate value, the higher the level of civilization and the higher the degree of evolution; Step 3, using roulette wheel selection method to generate population individuals; Step 4: randomly mutate a gene position of an individual in the population to its allele; after the mutation, force mutation on the same individuals in the population and completely randomly mutate the redundant individuals into another genotype; Step 5, perform gene crossover on the mutated population individuals; Step 6: UE first calculates the signal-to-noise ratio of each layer of each RB of each PMI under each RI r∈{1,…UE RX } represents possible RI; k∈{1,…N RB } represents the size of RB;,l∈{1,…r} It indicates the signal-to-noise ratio of each layer at a certain RI and a certain PMI; in is the channel transmission matrix for each RB, W is the encoding under a certain RI, R uu Represents the covariance matrix of interference and noise, generally at the subband level; W is strongly associated with chromosomes; Objective function: [i11,i12,i13,i2]=arg max(value) Step 7, Individual A j The i-th gene position in From the collection Any value in is the allele of the ith gene locus, u j Indicates the competitive advantage of the jth genotype, and the default value is 0; Select strategy: First, if the jth genotype is the optimal genotype, then u j =u j +1; Second, when the evolution degree Ratio≤0.5, if the jth genotype is a suboptimal genotype, then Third, when the evolution degree Ratio≤0.25, if the jth genotype is the optimal genotype, then Fourth, if the jth genotype is the worst genotype, then u j =u j -1; Step 8, the probability vector is modified by the information of the best individual in each generation of the population, and the update rule of the probability vector is expressed as: P t+1 =(1-a)*P t +a*u Among them, P t is the probability vector used by the algorithm in the tth iteration, a is the evolution rate in the competitive evolution mechanism, and u is the fitness of the genotype in natural selection; when season is equal to the positive minimum; The evolution rate a is expressed as: After gene evolution, the probability distribution is normalized and the calculation formula is as follows: Step 9, through the modified probability matrix P t+1 Generate individuals in the next generation population and perform step 3 until the iteration termination condition is reached or the evolution rate Ratio reaches a preset threshold, and the optimal precoding matrix is ​​obtained.

Images