A pilot power control method and system based on multi-strategy differential evolution

Abstract

This invention relates to the field of wireless communication technology and discloses a pilot power control method and system based on multi-strategy differential evolution, comprising the following specific steps: S1, constructing an uplink training phase signal model; S2, initializing a multi-strategy differential evolution algorithm; S3, finding the initial optimal individual; S4, using an unbiased mutation evolution strategy to select the optimal individual and update the iteration count; S5, if the maximum number of iterations is reached, dividing the individual into three subpopulations; otherwise, repeating step S4; S6, based on the second-stage multi-strategy differential evolution algorithm, applying corresponding mutation strategies to the three subpopulations to update each individual and the iteration count; S7, if the maximum number of iterations is reached, selecting the current globally optimal individual; otherwise, repeating step S6; S8, outputting the pilot and its corresponding pilot power coefficient. This invention solves the problems of premature convergence and getting trapped in local optima that existing technologies easily encounter during the solution process, and also has the characteristic of effectively increasing system throughput.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically, to a pilot power control method and system based on multi-strategy differential evolution. Background Technology

[0002] Since its inception, cellular networks have gradually become the core architecture for commercial communications due to their advantages of simple implementation and high frequency band resource utilization. With the advent of 5G, to meet the demand for higher speeds, base station deployments in communication systems may become more dense, leading to more severe inter-cell interference. Edge effects may become a bottleneck limiting future network performance.

[0003] Based on this, a decellularized massive MIMO architecture was proposed. In this system, all terminals share time-frequency resources and transmit pilot signals simultaneously. These pilot signals can overlap and interfere with each other, leading to pilot pollution. Pilot pollution severely affects the accuracy of channel estimation, thereby impacting the reliability of data transmission. Although reasonable pilot allocation can improve this situation, as the number of user terminals continues to increase, the terminal distribution becomes increasingly dense, resulting in massive connection scenarios. At this point, terminals using the same pilot may still be close together, weakening the pilot allocation effect and causing severe pilot pollution. Therefore, introducing a pilot power control mechanism can further suppress pilot pollution by controlling the pilot transmit power of users reusing the same pilot, thereby reducing interference between them and improving system throughput. Differential evolutionary algorithms, as mature and classic evolutionary algorithms, have the advantages of few control parameters, simple implementation, and strong robustness, and are suitable for solving multidimensional problems. However, traditional differential evolutionary algorithms are prone to premature convergence and getting trapped in local optima. Therefore, they cannot directly solve the problem of controlling the pilot power coefficients of all terminals.

[0004] Existing technology offers a method for optimizing uplink spectral efficiency in decellularized massive MIMO systems. This method includes: establishing an uplink channel model for a decellularized massive MIMO system with a Rayleigh flat fading channel; introducing AQNM to model the low-resolution ADC quantization output during pilot training and data transmission phases; calculating the achievable spectral efficiency of the transmission mode (MT); and establishing a system-wide spectral efficiency optimization problem under constraints of the upper limit of the forward link capacity and the upper limit of the MT data transmission power, and allocating resources accordingly. Then, based on different optimization parameters, the non-convex problem is transformed into two sub-problems: ADC quantization bit design and optimal MT power allocation, which are solved to obtain a suboptimal combination of ADC quantization bits and power control factors, ultimately achieving the goal of optimizing the system's overall spectral efficiency.

[0005] However, existing technologies are prone to premature convergence and getting trapped in local optima during the solution process. Therefore, how to invent a pilot power control method based on multi-strategy differential evolution is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the problems of premature convergence and getting trapped in local optima in existing technologies, this invention provides a pilot power control method and system based on multi-strategy differential evolution, which can effectively increase system throughput.

[0007] To achieve the above-mentioned objectives of this invention, the technical solution adopted is as follows:

[0008] A pilot power control method based on multi-strategy differential evolution includes the following specific steps:

[0009] S1. Construct the uplink training phase signal model of the decellularized large-scale MIMO architecture;

[0010] S2. Initialize the multi-strategy differential evolution algorithm;

[0011] S3. Calculate the fitness value of each individual based on the initial population, and find the initial optimal individual;

[0012] S4. Based on the first-stage multi-strategy differential evolution algorithm, an unbiased mutation evolution strategy is adopted for all individuals in the population. Corresponding control parameters are set, and individual elements are updated through crossover and selection to select the optimal individual. Then, the number of iterations is updated.

[0013] S5. Determine whether the maximum number of iterations has been reached. If it has, divide the current population into three subpopulations based on the individual fitness values. Otherwise, repeat step S4.

[0014] S6. Based on the second-stage multi-strategy differential evolution algorithm, the corresponding mutation strategies are adopted for the three subpopulations respectively, the corresponding control parameters are set, and each individual is updated through crossover and selection, and then the iteration number is updated.

[0015] S7. Determine whether the maximum number of iterations has been reached. If it has, select the current global optimal individual as the coefficient combination of the user corresponding to this pilot. Otherwise, repeat step S6.

[0016] S8. After the pilot power coefficients of all users have been calculated, output the pilots used by the users and their corresponding pilot power coefficients.

[0017] Preferably, in step S1, the signal model during the uplink training phase is represented as follows:

[0018]

[0019] Where Yp,m g represents the pilot signal received by the m-th AP during the uplink training phase. mk This represents the channel vector between the k-th user terminal and the m-th AP. β mk It is a large-scale fading phenomenon, and is known a priori by the system; h mk This is a small-scale fading phenomenon, where the elements are independent and identically distributed random variables following the CN(0,1) distribution, and the channel coherence time is τ. c There exists a length τ used for uplink training, and τ satisfies τ < τ c , It is the pilot sequence of terminal k, where It is a normalized pilot sequence that satisfies k = 1, 2, ..., K; the pilot used by each terminal is one of the orthogonal pilot sets Φ; η k The pilot power coefficient satisfies 0 < η k ≤1; K is the number of single-antenna terminals, M is the number of APs, and each AP is equipped with L antennas. It is the normalized signal-to-noise ratio during the uplink training phase, p p For uplink pilot standard transmit power, Noise power; W p,m Let be an L×τ additive noise matrix, whose elements are independent and identically distributed random variables following CN(0,1).

[0020] Furthermore, in step S2, the multi-strategy differential evolution algorithm is initialized, specifically through the following steps:

[0021] Initialize the first generation population For individuals Let t = 0 represent the initial population, i = 1, 2, ..., B; B be the number of individuals in the population; D = T be the individual dimension, i.e., the dimension of the objective optimization problem; the upper limit of the number of iterations is N; and each individual in the population represents the population. All are combinations of pilot power coefficients for reusing the same pilot users, where D = T = K / τ represents the number of times each orthogonal pilot is reused; the elements of each individual in the population are randomly initialized on (0,1] to ensure population diversity.

[0022] Furthermore, in step S3, the fitness value of each individual is calculated, specifically as follows:

[0023]

[0024] subject to 0<η k ≤1,k=1,…,T

[0025] Where A kIt is the set of APs that serve the k-th user, and is usually selected based on prior knowledge of large-scale fading; This represents minimizing the channel estimation error for all users, and is typically used to improve overall system performance. This represents minimizing the maximum user channel estimation error.

[0026] Furthermore, in step S4, based on the first-stage multi-strategy differential evolution algorithm, an unbiased mutation evolution strategy is adopted for all individuals in the population. Corresponding control parameters are set, and individual elements are updated through crossover and selection to select the optimal individual. The specific steps are as follows:

[0027] An unbiased mutation evolution strategy, DE / rand / 2, is adopted. This strategy leverages the unbiasedness of the mutation strategy to improve the instability of random initialization, while also providing direction for subsequent optimization, balancing global search and local exploitation. Individual elements are then updated through crossover and selection. The DE / rand / 2 strategy is represented as follows: For the mutated individual, It is a unique individual randomly selected from the t-th generation population that is related to the target individual. Individuals that are not the same; F is the scaling factor, used to control the scaling ratio of the difference values, and its magnitude affects the convergence speed of the population; crossover is performed on the target individuals. According to certain criteria and the mutated individuals Perform element swaps to obtain the crossover individuals. The specific cross-reference criteria are as follows:

[0028]

[0029] Where d = 1, 2, ..., D, CR is the crossover rate, r is a random number in (0, 1), and d rand Let V be a random integer in [1, D], used to guarantee that after crossover, individual V... i t At least one element in the sample comes from a mutated individual. Individual V after crossover i t For candidate individuals, the entire population undergoes unbiased mutation evolution, with a large F0 and a small CR0 set, which helps to roughly determine the optimization direction globally; after crossover, the candidate individuals V are first... i t Boundary checks are performed on the elements.

[0030] Furthermore, after the crossover, the candidate individuals V are first... i t When performing boundary checks on elements, the specific boundary check operation is boundary absorption, which means taking the boundary value for coefficients that exceed the boundary, and then considering the candidate individual V.i t With the original target individual By comparing fitness values, the next generation of target individuals is selected based on their optimal performance.

[0031] Furthermore, in step S5, the current population is divided into three subpopulations based on individual fitness values. Specifically, according to the Pareto criterion and individual fitness values, the top 20% of individuals in the entire population are first divided into one subpopulation, denoted as X1, and then the bottom 80% of individuals are equally divided into two subpopulations, denoted as X2 and X3. By dividing the population into subpopulations, the top 20% of individuals with better fitness values ​​can perform a refined search to obtain a better solution, while the bottom 80% of individuals can perform a large-scale search using different control parameters, thus ensuring population diversity while avoiding getting trapped in local optima.

[0032] Furthermore, in step S6, based on the second-stage multi-strategy differential evolution algorithm, corresponding mutation strategies are applied to the three subpopulations, corresponding control parameters are set, and each individual is updated through crossover and selection. Specifically, corresponding mutation strategies are applied to the three subpopulations, corresponding control parameters are set, and individuals are updated through crossover and selection. For the first subpopulation X1, the DE / best / 2 mutation strategy is used to improve its mining capacity and further search for the optimal solution. For subpopulations X2 and X3, the DE / rand / 1 strategy is used for further unbiased global search, and the search and mining capacity of the two subpopulations are balanced by setting different control parameters. The DE / best / 2 strategy is expressed as follows: Both DE / rand / 1 and DE / rand / 2 belong to unbiased mutation strategies. The DE / rand / 1 strategy is expressed as follows: The crossover and selection operations are the same as step S4.

[0033] Furthermore, in step S6, the current population is divided into subpopulations again before each iteration to ensure population diversity and avoid getting trapped in local optima.

[0034] A pilot power control system based on multi-strategy differential evolution includes a model building module, an initialization module, a fitness value calculation module, a multi-strategy differential evolution module, and a data output module.

[0035] The model building module is used to build the uplink training phase signal model for a decellularized massive MIMO architecture;

[0036] The initialization module is used to initialize the multi-strategy differential evolution algorithm;

[0037] The fitness value calculation module is used to calculate the fitness value of each individual based on the initial population and find the initial optimal individual;

[0038] The multi-strategy differential evolution module is used based on the first-stage multi-strategy differential evolution algorithm. It employs an unbiased mutation evolution strategy for all individuals in the population, sets corresponding control parameters, updates individual elements through crossover and selection, selects the optimal individual, and then updates the iteration count. It then checks if the maximum iteration count has been reached; if so, it divides the current population into three subpopulations based on the individual's fitness value; otherwise, it repeats the first-stage multi-strategy differential evolution algorithm. Based on the second-stage multi-strategy differential evolution algorithm, it employs corresponding mutation strategies for each of the three subpopulations, sets corresponding control parameters, updates each individual through crossover and selection, and then updates the iteration count. It then checks if the maximum iteration count has been reached; if so, it selects the current globally optimal individual as the coefficient combination for the user corresponding to this pilot; otherwise, it repeats the second-stage multi-strategy differential evolution algorithm.

[0039] The data output module is used to output the pilot signal used by the user and its corresponding pilot power coefficient.

[0040] The beneficial effects of this invention are as follows:

[0041] This invention constructs an uplink training phase signal model for a decellularized massive MIMO architecture and employs a multi-strategy differential evolution algorithm. Based on population fitness calculation, iterative optimization is performed to obtain the optimal pilot power coefficients for all users. This ensures population diversity while avoiding getting trapped in local optima, enabling direct control of the pilot power coefficients of all terminals. Furthermore, by improving the differential evolution algorithm, the system throughput is effectively increased. Attached Figure Description

[0042] Figure 1 This is a flowchart of the pilot power control method based on multi-strategy differential evolution of the present invention.

[0043] Figure 2 This is a flowchart of the algorithm in an embodiment of the present invention.

[0044] Figure 3 This is a CDF comparison chart of system throughput in an embodiment of the present invention.

[0045] Figure 4 This is a graph showing the number of failures in an embodiment of the Graph algorithm.

[0046] Figure 5 This is a comparison chart of system throughput CDF between the embodiments of the present invention and existing pilot power control algorithms.

[0047] Figure 6 This is a power consumption comparison chart of an example of the present invention.

[0048] Figure 7 This is a complexity comparison chart of examples of the present invention. Detailed Implementation

[0049] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0050] Example 1

[0051] like Figure 1 As shown, a pilot power control method based on multi-strategy differential evolution includes the following specific steps:

[0052] S1. Construct the uplink training phase signal model of the decellularized large-scale MIMO architecture;

[0053] S2. Initialize the multi-strategy differential evolution algorithm;

[0054] S3. Calculate the fitness value of each individual based on the initial population, and find the initial optimal individual;

[0055] S4. Based on the first-stage multi-strategy differential evolution algorithm, an unbiased mutation evolution strategy is adopted for all individuals in the population. Corresponding control parameters are set, and individual elements are updated through crossover and selection to select the optimal individual. Then, the number of iterations is updated.

[0056] S5. Determine whether the maximum number of iterations has been reached. If it has, divide the current population into three subpopulations based on the individual fitness values. Otherwise, repeat step S4.

[0057] S6. Based on the second-stage multi-strategy differential evolution algorithm, the corresponding mutation strategies are adopted for the three subpopulations respectively, the corresponding control parameters are set, and each individual is updated through crossover and selection, and then the iteration number is updated.

[0058] S7. Determine whether the maximum number of iterations has been reached. If it has, select the current global optimal individual as the coefficient combination of the user corresponding to this pilot. Otherwise, repeat step S6.

[0059] S8. After the pilot power coefficients of all users have been calculated, output the pilots used by the users and their corresponding pilot power coefficients.

[0060] In this embodiment, the distribution of APs and user terminals in a decellularized massive MIMO system is first simulated on a two-dimensional plane of size D*D. Their positions are randomly generated, and an infinitely large network is simulated by surrounding the APs, i.e., eight neighbors surround the square area boundary of each AP. No cells or cell boundaries are defined in the entire system, so handover does not occur when a terminal moves within the area. Assume that for user k, only the APs with the smallest large-scale fading are selected to serve it, thus avoiding resource waste, and denoted as A. kIn this example, the location information of the AP and user terminals is used to determine the AP service objects, and users are divided into τ groups by random user grouping. Each group of users is assigned an orthogonal pilot, and different groups of users reuse the pilot. Each pilot is reused T = K / τ times. If the number of users K is not an integer multiple of τ, the remaining users are recorded as an additional group. Users using the same pilot have their pilot power coefficient combinations as the solution to be solved. Each individual in the proposed algorithm can be regarded as a feasible solution. This invention mainly calculates the τ T-dimensional pilot coefficient combinations separately. The channel model mainly consists of large-scale fading and small-scale fading. Large-scale fading is modeled by path loss and uncorrelated log-normal shadowing fading. The elements of small-scale fading are independent and identically distributed random variables following CN(0,1). The path loss uses the Hata-COST231 three-slope model, and the path loss PL is expressed in dB. mk It is given by the following formula:

[0061]

[0062] L = 46.3 + 33.9lg(f) - 13.82lg(h) AP )-(1.1lg(f)-0.7)h u +(1.56lg(f)-0.8)

[0063] Where, d mk d represents the distance (m) between the m-th AP and the k-th terminal; d0 and d1 are the distance thresholds (m) that distinguish different loss types; f is the carrier power (MHz); h AP h is the height of the AP antenna (m). u PL is the height of the terminal antenna (m); mk It is d mk A continuous function, and when d mk There is no shadow fading when d1 is less than or equal to d1. To more intuitively describe the signal transmission model, this method uses a normalized signal-to-noise ratio to express the signal power. This paper uses ρ... p The normalized signal-to-noise ratio (SNR) represents the signal-to-noise ratio (SNR) when the UE transmits pilot signals during the uplink pilot training phase, denoted by ρ. u This represents the normalized signal-to-noise ratio (SNR) when the UE transmits signals during the uplink data transmission phase. The specific expression is as follows:

[0064]

[0065] Where, p p and p u These represent the uplink pilot transmission power and the uplink data transmission power (W), respectively. The noise power (W) is calculated using the following expression:

[0066]

[0067] Among them, B d For bandwidth (MHz), k B =1.381×10 -23 The Boltzmann constant (Joule / Kelvin) is given, T0 = 290 is the noise temperature (Kelvin), and N is the noise level. f This represents the noise figure (dB).

[0068] The main purpose of the uplink pilot training phase is to estimate the channel. Based on channel reciprocity, the estimated channel can be used for decoding uplink data transmission. Therefore, the algorithm performance can be reflected by the user's uplink throughput. The closed-form expression for the user throughput during the uplink data transmission phase is:

[0069]

[0070] Where ρ u γ is the normalized signal-to-noise ratio when the UE transmits signals during the uplink data transmission phase. mk The mean square value of the channel estimated by the m-th AP for the k-th user can be expressed as:

[0071]

[0072] In one specific embodiment, in step S1, the signal model during the uplink training phase is represented as follows:

[0073]

[0074] Where Y p,m g represents the pilot signal received by the m-th AP during the uplink training phase. mk This represents the channel vector between the k-th user terminal and the m-th AP. β mk It is a large-scale fading phenomenon, and is known a priori by the system; h mk This is a small-scale fading phenomenon, where the elements are independent and identically distributed random variables following the CN(0,1) distribution, and the channel coherence time is τ. c There exists a length τ used for uplink training, and τ satisfies τ < τ c , It is the pilot sequence of terminal k, where It is a normalized pilot sequence that satisfies Each terminal uses one of the orthogonal pilot signals Φ; η k The pilot power coefficient satisfies 0 < η k ≤1; K is the number of single-antenna terminals, M is the number of APs, and each AP is equipped with L antennas. It is the normalized signal-to-noise ratio during the uplink training phase, pp For uplink pilot standard transmit power, Noise power; W p,m Let be an L×τ additive noise matrix, whose elements are independent and identically distributed random variables following CN(0,1).

[0075] In this embodiment, the specific system simulation parameters are shown in Table 1:

[0076] Table 1 System Simulation Parameters

[0077]

[0078] In one specific embodiment, step S2 involves initializing the multi-strategy differential evolution algorithm, specifically through the following steps:

[0079] Initialize the first generation population For individuals Let t = 0 represent the initial population, i = 1, 2, ..., B; B be the number of individuals in the population; D = T be the individual dimension, i.e., the dimension of the objective optimization problem; the upper limit of the number of iterations is N; and each individual in the population represents the population. All are combinations of pilot power coefficients for reusing the same pilot users, where D = T = K / τ represents the number of times each orthogonal pilot is reused; the elements of each individual in the population are randomly initialized on (0,1] to ensure population diversity.

[0080] In one specific embodiment, step S3 involves calculating the fitness value for each individual, specifically as follows:

[0081]

[0082] subject to 0<η k ≤1,k=1,…,T

[0083] Where A k It is the set of APs that serve the k-th user, and is usually selected based on prior knowledge of large-scale fading; This represents minimizing the channel estimation error for all users, and is typically used to improve overall system performance. Minimizing the maximum user channel estimation error is typically used to ensure system fairness. Simultaneously optimizing both factors constitutes a multi-objective, high-dimensional optimization problem, which is generally difficult and hard to solve. The multi-policy differential evolution algorithm proposed in this method can determine the iteration direction by setting a joint fitness function, achieving simultaneous optimization of both factors and thus improving overall system performance while ensuring fairness. The optimal individual in the initial population can then be determined based on this fitness function.

[0084] In one specific embodiment, in step S4, based on the first-stage multi-strategy differential evolution algorithm, an unbiased mutation evolution strategy is adopted for all individuals in the population, corresponding control parameters are set, and individual elements are updated through crossover and selection to select the optimal individual. The specific steps are as follows:

[0085] An unbiased mutation evolution strategy, DE / rand / 2, is adopted. This strategy leverages the unbiasedness of the mutation strategy to improve the instability of random initialization and provides direction for subsequent optimization. Corresponding control parameters F0 = 0.8 and CR0 = 0.3 are set to balance global search and local exploitation. Individual elements are then updated through crossover and selection. The DE / rand / 2 strategy is represented as follows: For the mutated individual, It is a unique individual randomly selected from the t-th generation population that is related to the target individual. The individuals are also different because this strategy does not use an optimal individual as a guide but randomly selects individuals for mutation, so the mutation is unbiased; F is the scaling factor, used to control the scaling ratio of the difference value, and its size affects the convergence speed of the population, usually taken from a value in [0,1]; crossover is performed on the target individual According to certain criteria and the mutated individuals Perform element swaps to obtain the crossover individuals. The specific cross-reference criteria are as follows:

[0086]

[0087] Where d = 1, 2, ..., D, CR is the crossover rate, r is a random number in (0, 1), and d rand Let V be a random integer in [1, D], used to guarantee that after crossover, individual V... i t At least one element in the sample comes from a mutated individual. Individual V after crossover i t As candidate individuals, in the first stage, the entire population undergoes unbiased mutation evolution, with a large F0 and a small CR0 set, which helps to roughly determine the optimization direction globally; after crossover, the candidate individuals V are first... i t The elements are subjected to boundary checks. This method uses boundary absorption, meaning that coefficients exceeding the boundary are taken at the boundary value. Then, the candidate individuals V are... i t With the original target individual By comparing fitness values, the next generation of target individuals is selected based on their optimal performance.

[0088] In one specific implementation, step S5 involves dividing the current population into three subpopulations based on individual fitness values. Specifically, according to the Pareto criterion and individual fitness values, the top 20% of individuals in the entire population are first grouped into one subpopulation, denoted as X1. Then, the remaining 80% of individuals are equally divided into two subpopulations, denoted as X2 and X3. The Pareto criterion states that for any population, the top 20% of individuals are the most important, while the remaining 80% are secondary. By dividing the population into subpopulations, the top 20% of individuals with better fitness values ​​can perform a refined search to find a better solution, while the remaining 80% of individuals can perform a large-scale search using different control parameters, ensuring population diversity while avoiding getting trapped in local optima.

[0089] In a specific implementation, in step S6, based on the second-stage multi-strategy differential evolution algorithm, corresponding mutation strategies are adopted for the three subpopulations, corresponding control parameters are set, and each individual is updated through crossover and selection. Specifically, corresponding mutation strategies are adopted for the three subpopulations, corresponding control parameters are set, and individuals are updated through crossover and selection. For the first subpopulation X1, its overall channel error performance is good and contains the current globally optimal coefficient combination. The DE / best / 2 mutation strategy is used to improve its mining capability and further search for the optimal solution. For subpopulations X2 and X3, although their channel estimation errors are large, they may still find the optimal solution due to the multiple extreme values ​​of the multi-peak problem. Therefore, the DE / rand / 1 strategy is used for further unbiased global search, and the search and mining capabilities of the two subpopulations are balanced by setting different control parameters. The DE / best / 2 strategy is expressed as follows: Both DE / rand / 1 and DE / rand / 2 belong to unbiased mutation strategies. The DE / rand / 1 strategy is expressed as follows: For control parameters, a larger F value means a greater influence of the coefficient difference, resulting in a stronger search capability of the algorithm; a smaller F value means less disturbance to the coefficient difference, resulting in a stronger mining capability of the algorithm. A larger CR value means that during crossover operations, the candidate individual V... i t Inherited variant individuals The more elements there are, the stronger the population diversity. A smaller CR value can improve the convergence speed. For subpopulation X1, in order to enable it to further mine the optimal solution and achieve fine-grained search while avoiding overly detailed mining that would lead to slow convergence, this paper sets F1 to 0.7 and CR1 to a smaller value of 0.2. For subpopulation X2, in order to enable it to have good mining ability while conducting unbiased search over a large range, this paper sets F2 to a larger value of 0.8 and CR2 to a smaller value of 0.2. For subpopulation X3, since the total channel estimation error corresponding to its coefficient combination individuals is relatively large, F3 is set to a larger value of 0.8 and CR3 to 0.9 to enhance its search ability and expand the search range of the optimal solution. The crossover and selection operations are the same as in step S4. In addition, before the start of each iteration, the current population is divided into subpopulations again to ensure population diversity and avoid getting trapped in local optima.

[0090] Example 2

[0091] In this embodiment, as Figure 2 As shown, the specific process of the pilot power control method based on multi-strategy differential evolution in a decellularized massive MIMO system of the present invention includes the following steps:

[0092] S1. Construct the uplink training phase signal model;

[0093] S12. The signal model for the uplink training phase is represented as follows:

[0094]

[0095] Where Y p,m g represents the pilot signal received by the m-th AP during the uplink training phase. mk This represents the channel vector between the k-th user terminal and the m-th AP. β mk It is a large-scale fading phenomenon, and is known a priori by the system; h mk This is a small-scale fading phenomenon, where the elements are independent and identically distributed random variables following the CN(0,1) rule. The channel coherence time is τ. c (symbol), where a length τ is used for uplink training, and τ satisfies τ<τ c . It is the pilot sequence of terminal k, where It is a normalized pilot sequence that satisfies Each terminal uses one of the orthogonal pilot signals Φ; η k The pilot power coefficient satisfies 0 < η k ≤1. K is the number of single-antenna terminals, M is the number of APs, and each AP is equipped with L antennas. It is the normalized signal-to-noise ratio during the uplink training phase, p pFor uplink pilot standard transmit power, Noise power; W p,m Let be an L×τ additive noise matrix, whose elements are independent and identically distributed random variables following CN(0,1).

[0096] S2. Initialize the parameters related to the multi-strategy differential evolution algorithm;

[0097] S21. The pilot power control method based on multi-strategy differential evolution in a decellularized massive MIMO system according to claim 1, wherein step S2, the initialization of the multi-strategy differential evolution algorithm includes initializing the initial population. For individuals Let t = 0 represent the initial population, i = 1, 2, ..., B; B be the number of individuals in the population; D = T be the individual dimension, i.e., the dimension of the objective optimization problem; N be the upper limit of the number of iterations; and each individual in the population be represented by a variable. All are combinations of pilot power coefficients for reusing the same pilot users, where D = T = K / τ represents the number of times each orthogonal pilot is reused; the elements of each individual in the population are randomly initialized on (0,1] to ensure population diversity.

[0098] The initialization parameter settings for the multi-strategy differential evolution algorithm are shown in Table 2:

[0099] Table 2 Initialization parameters related to the multi-strategy differential evolution algorithm

[0100]

[0101] S3. Calculate the fitness value of each individual based on the initial population, and find the initial optimal individual;

[0102] S31. The pilot power control method based on multi-strategy differential evolution in a decellularized massive MIMO system according to claim 1, in step S3, the individual fitness value is expressed as:

[0103]

[0104] subject to 0<η k ≤1,k=1,…,T

[0105] Where A k It is the set of APs that serve the k-th user, and is usually selected based on prior knowledge of large-scale fading; This represents minimizing the channel estimation error for all users, and is typically used to improve overall system performance. Minimizing the maximum user channel estimation error is typically used to ensure system fairness. Simultaneously optimizing both factors constitutes a multi-objective, high-dimensional optimization problem, which is generally difficult and hard to solve. The multi-policy differential evolution algorithm proposed in this method can determine the iteration direction by setting a joint fitness function, achieving simultaneous optimization of both factors and thus improving overall system performance while ensuring fairness. The optimal individual in the initial population can then be determined based on this fitness function.

[0106] S4. In the first stage of the multi-strategy differential evolution algorithm, an unbiased mutation evolution strategy is adopted for all individuals in the population. Corresponding control parameters are set, and individual elements are updated through crossover and selection to select the optimal individual. Then the number of iterations is updated.

[0107] S41. The pilot power control method based on multi-strategy differential evolution in a decellularized massive MIMO system according to claim 1; in step S4, the first stage sets an unbiased mutation evolution strategy DE / rand / 2, which improves the instability of random initialization by utilizing the unbiasedness of the mutation strategy, and provides a general optimization direction for the second stage of the algorithm. Corresponding control parameters F0 = 0.8 and CR0 = 0.3 are set to balance global search and local exploitation, and then individual elements are updated through crossover and selection; the DE / rand / 2 strategy is expressed as... For the mutated individual, It is a unique individual randomly selected from the t-th generation population that is the same as the current target individual. The individuals are also different, t = 1, 2, ..., N. Because this strategy does not use an optimal individual as a guide but randomly selects individuals for mutation, the mutation is unbiased; F is the scaling factor, used to control the scaling ratio of the difference value, and its size affects the convergence speed of the population, usually taken from a value in [0,1]; crossover is performed on the target individual. According to certain criteria and the mutated individuals Perform element swaps to obtain the crossover individuals. The crossover criteria are as follows:

[0108]

[0109] Where d = 1, 2, ..., D, CR is the crossover rate, r is a random number in (0, 1), and d rand Let V be a random integer in [1, D], used to guarantee that after crossover, individual V... i t At least one element in the sample comes from a mutated individual. Individual V after crossover i tThese are the candidate individuals. In the first stage, the entire population undergoes unbiased mutation evolution, with a large F0 and a small CR0 set, which helps to roughly determine the optimization direction globally; after crossover, the candidate individuals V are first... i t The elements are subjected to boundary checks. This method uses boundary absorption, meaning that coefficients exceeding the boundary are taken at the boundary value. Then, the candidate individuals V are... i t With the original target individual By comparing fitness values, the next generation of target individuals is selected based on their optimal performance.

[0110] S5. Determine whether the maximum number of iterations has been reached. If it has, divide the current population into three subpopulations based on the individual fitness values. Otherwise, repeat step S4.

[0111] S51. The pilot power control method based on multi-strategy differential evolution in a decellularized massive MIMO system according to claim 1; in step S5, according to the Pareto criterion, combined with the individual fitness values, the first 20% of individuals in the entire population are first divided into a subpopulation, denoted as X1, and then the remaining 80% of individuals are equally divided into two subpopulations, denoted as X2 and X3. The Pareto criterion states that for any population, the first 20% of individuals are the most important, and the remaining 80% of individuals are secondary. By dividing into subpopulations, the first 20% of individuals with better fitness values ​​can perform a refined search to obtain a better solution, while the remaining 80% of individuals can perform a large-scale search by combining different control parameters, ensuring population diversity while avoiding getting trapped in local optima.

[0112] S6. In the second stage of the multi-strategy differential evolution algorithm, the corresponding mutation strategy is adopted for each of the three subpopulations, the corresponding control parameters are set, and each individual is updated through crossover and selection, and then the number of iterations is updated.

[0113] S61. The pilot power control method based on multi-strategy differential evolution in a decellularized massive MIMO system according to claim 1; in step S6, the second stage adopts corresponding mutation strategies and sets corresponding control parameters for the three subpopulations respectively, and updates individuals through crossover and selection; for the first subpopulation X1, its overall channel error performance is good, and it contains individuals with the current global optimal coefficient combination, so the DE / best / 2 mutation strategy is used to improve its mining ability, which is conducive to further searching for the optimal solution; for subpopulations X2 and X3, although their channel estimation errors are large, they may still find the optimal solution due to the multiple extreme values ​​of the multi-peak problem, so the DE / rand / 1 strategy is used for further unbiased global search, and the search and mining capabilities of the two subpopulations are balanced by setting different control parameters. The DE / best / 2 strategy is expressed as This strategy includes individuals with globally optimal coefficient combinations. Therefore, focusing on local mining causes the search direction to tend towards the current optimal solution. Although this results in a faster convergence speed, it is prone to premature convergence. DE / rand / 1 and DE / rand / 2 are both unbiased mutation strategies. Since all participating individuals are randomly selected, the search direction of this type of strategy is unbiased, which is beneficial for global search and suitable for solving multimodal functions. The DE / rand / 1 strategy is expressed as... For control parameters, a larger F value means a greater influence of the coefficient difference, resulting in a stronger search capability of the algorithm; a smaller F value means less disturbance to the coefficient difference, resulting in a stronger mining capability of the algorithm. A larger CR value means that during crossover operations, the candidate individual V... i t Inherited variant individuals The more elements there are, the stronger the population diversity. A smaller CR value can improve the convergence speed. For subpopulation X1, in order to enable it to further mine the optimal solution and achieve fine-grained search while avoiding overly detailed mining that would lead to slow convergence, this paper sets F1 to 0.7 and CR1 to a smaller value of 0.2. For subpopulation X2, in order to enable it to have good mining ability while conducting unbiased search over a large range, this paper sets F2 to a larger value of 0.8 and CR2 to a smaller value of 0.2. For subpopulation X3, since the total channel estimation error corresponding to its coefficient combination individuals is relatively large, F3 is set to a larger value of 0.8 and CR3 to 0.9 to enhance its search ability and expand the search range of the optimal solution. The crossover and selection operations are the same as in step S4. In addition, before the start of each iteration, the current population is divided into subpopulations again to ensure population diversity and avoid getting trapped in local optima.

[0114] S7. Determine whether the maximum number of iterations has been reached. If it has, select the current global best individual as the coefficient combination for the user corresponding to this pilot. Otherwise, repeat step S6.

[0115] S8. After all users' pilot power coefficients have been calculated, output the pilots used by each user and their corresponding pilot power coefficients.

[0116] Figure 3This is a comparison chart of the cumulative distribution function (CDF) of system throughput in embodiments of the present invention, illustrating the effect of the CDF on throughput in different algorithms. In a massive connection scenario (K=300, τ=10), the Graph scheme significantly improves the performance of terminals with low to medium throughput compared to the traditional GPA scheme, but reduces the performance of high-throughput terminals, failing to improve the overall system performance. The method of the present invention, while achieving a similar improvement in low to medium throughput terminals as the Graph scheme, has superior overall performance. This is mainly because the present invention optimizes both the channel estimation error of all users and the maximum user channel estimation error simultaneously by setting a joint optimization objective, thus avoiding the situation where the performance of high-throughput users is sacrificed for the performance of low-throughput users. Figure 4 It is known that in scenarios with a higher density of users (such as K=300, 400, 500, τ=10), the number of orthogonal pilots is relatively smaller, and the Graph scheme will fail, while the algorithm of this invention is not affected by this.

[0117] Figure 5 This is a comparison of the cumulative distribution function (CDF) of system throughput between the present invention and existing pilot power control algorithms. It can be seen that in a massive connectivity scenario (K=300, τ=10), while existing PPC algorithms show a significant improvement, they do not guarantee an overall performance improvement; in fact, the performance of high-throughput terminals may even decrease. This is because the PPC algorithm only optimizes the maximum user channel estimation error, without considering the overall user channel estimation error. The present invention, by simultaneously optimizing the channel estimation errors of all users and the maximum user channel estimation error, ensures both system fairness and improves overall performance.

[0118] Figure 6 This is a comparison chart of the power consumption of this invention, existing pilot power control algorithms, and full-power transmission. It shows that in multi-user scenarios, compared to full-power pilot transmission, pilot power control can significantly reduce the terminal's transmission power and energy consumption. Combined with... Figure 5 and Figure 6 It can be seen that the present invention outperforms the PPC algorithm under similar power consumption conditions. This further verifies the effectiveness of the present invention in multi-user scenarios.

[0119] The time complexity of each iteration of the PPC algorithm is known to be O(3K+1). 3 / 2 (K+1) 2 The complexity of this invention is O(NB(K)). 2 +2Kτ+τ 2 The expression is defined as logB) / τ), where K is the total number of users, τ is the number of orthogonal pilots, N is the number of algorithm iterations, and B is the custom population size. In multi-user scenarios, although the number of iterations required by this invention is much greater than that of the PPC algorithm, the overall complexity is much lower. Figure 7It can be seen that when K=300, the complexity of this invention is 39.1% of that of the PPC algorithm; when K=500, the complexity of this invention is 29.6% of that of the PPC algorithm. Therefore, in scenarios with massive connections, this invention can achieve better performance with lower complexity under approximately equal power consumption conditions, and the more user terminals there are, the more obvious the advantage becomes.

[0120] Example 3

[0121] A pilot power control system based on multi-strategy differential evolution includes a model building module, an initialization module, a fitness value calculation module, a multi-strategy differential evolution module, and a data output module.

[0122] The model building module is used to build the uplink training phase signal model for a decellularized massive MIMO architecture;

[0123] The initialization module is used to initialize the multi-strategy differential evolution algorithm;

[0124] The fitness value calculation module is used to calculate the fitness value of each individual based on the initial population and find the initial optimal individual;

[0125] The multi-strategy differential evolution module is used based on the first-stage multi-strategy differential evolution algorithm. It employs an unbiased mutation evolution strategy for all individuals in the population, sets corresponding control parameters, updates individual elements through crossover and selection, selects the optimal individual, and then updates the iteration count. It then checks if the maximum iteration count has been reached; if so, it divides the current population into three subpopulations based on the individual's fitness value; otherwise, it repeats the first-stage multi-strategy differential evolution algorithm. Based on the second-stage multi-strategy differential evolution algorithm, it employs corresponding mutation strategies for each of the three subpopulations, sets corresponding control parameters, updates each individual through crossover and selection, and then updates the iteration count. It then checks if the maximum iteration count has been reached; if so, it selects the current globally optimal individual as the coefficient combination for the user corresponding to this pilot; otherwise, it repeats the second-stage multi-strategy differential evolution algorithm.

[0126] The data output module is used to output the pilot signal used by the user and its corresponding pilot power coefficient.

[0127] This invention constructs an uplink training phase signal model for a decellularized massive MIMO architecture and employs a multi-strategy differential evolution algorithm. Based on population fitness calculation, iterative optimization is performed to obtain the optimal pilot power coefficients for all users. This ensures population diversity while avoiding getting trapped in local optima, enabling direct control of the pilot power coefficients of all terminals. Furthermore, by improving the differential evolution algorithm, the system throughput is effectively increased.

[0128] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A pilot power control method based on multi-strategy differential evolution, characterized in that: The specific steps include the following: S1. Construct the uplink training phase signal model for the decellularized massive MIMO architecture; the uplink training phase signal model is represented as: in Indicates the first m Pilot signals received by each AP during the uplink training phase Indicates the first k The user terminal and the first m Channel vectors between APs , It is a large-scale fading phenomenon, and it is known a priori by the system. This is a small-scale fading phenomenon, where the elements are independent and identically distributed random variables that follow a set pattern. The channel coherence time is , which includes length Used for uplink training, and satisfy , It is a terminal k The pilot sequence, in which It is a normalized pilot sequence that satisfies , The pilot signals used by each terminal are a set of orthogonal pilot signals. one of the; For pilot power coefficient, satisfying ; K This refers to the number of terminals per antenna. M For the number of APs, each AP is equipped with L Root antenna, It is the normalized signal-to-noise ratio during the uplink training phase. For uplink pilot standard transmit power, Noise power; for Additive noise matrix, whose elements follow the rules of... Independent and identically distributed random variables; S2. Initialize the multi-strategy differential evolution algorithm; S3. Calculate the fitness value of each individual based on the initial population to find the initial optimal individual; The calculation of the fitness value of each individual is specifically as follows: in This is the first k The set of APs serving a user is typically selected based on prior knowledge of large-scale fading. This represents minimizing the channel estimation error for all users, and is typically used to improve overall system performance. This represents minimizing the maximum user channel estimation error; S4. Based on the first-stage multi-strategy differential evolution algorithm, an unbiased mutation evolution strategy is adopted for all individuals in the population. Corresponding control parameters are set, and individual elements are updated through crossover and selection to select the optimal individual. Then, the number of iterations is updated. S5. Determine whether the maximum number of iterations has been reached. If it has, divide the current population into three subpopulations based on the individual fitness values. Otherwise, repeat step S4. S6. Based on the second-stage multi-strategy differential evolution algorithm, the corresponding mutation strategies are adopted for the three subpopulations respectively, the corresponding control parameters are set, and each individual is updated through crossover and selection, and then the iteration number is updated. S7. Determine whether the maximum number of iterations has been reached. If it has, select the current global optimal individual as the coefficient combination of the user corresponding to this pilot. Otherwise, repeat step S6. S8. After the pilot power coefficients of all users have been calculated, output the pilots used by the users and their corresponding pilot power coefficients.

2. The pilot power control method based on multi-strategy differential evolution according to claim 1, characterized in that: In step S2, the multi-strategy differential evolution algorithm is initialized, and the specific steps are as follows: Initialize the first generation population For individuals ,pass To represent the initial population, ; B The number of individuals in the population; For the individual dimension, i.e., the dimension of the target optimization problem; the upper limit of the number of iterations. N Each individual in the population , All of these are combinations of pilot power coefficients that reuse the same pilot users, among which This represents the number of times each orthogonal pilot is reused; for each individual element in the population... Random initialization is performed to ensure population diversity.

3. The pilot power control method based on multi-strategy differential evolution according to claim 2, characterized in that: In step S4, based on the first-stage multi-strategy differential evolution algorithm, an unbiased mutation evolution strategy is adopted for all individuals in the population. Corresponding control parameters are set, and individual elements are updated through crossover and selection to select the optimal individual. The specific steps are as follows: An unbiased mutation evolution strategy, DE / rand / 2, is adopted. This strategy leverages the unbiasedness of the mutation strategy to improve the instability of random initialization, while also providing direction for subsequent optimization, balancing global search and local exploitation. Individual elements are then updated through crossover and selection. The DE / rand / 2 strategy is represented as follows: , For the mutated individual, , , , , It is in the t Randomly selected distinct individuals from the population and related to the target individual They are also different individuals; F This is the scaling factor, used to control the scaling ratio of the difference values. Its magnitude affects the convergence speed of the population. Crossover is for target individuals According to certain criteria and the mutated individuals Perform element swaps to obtain the crossover individuals. The specific cross-reference criteria are as follows: in , CR Crossover rate, r for Random numbers in the data, for A random integer, used to guarantee the individuals after crossover. At least one element in the sample comes from a mutated individual. Individuals after crossover As candidate individuals, the entire population undergoes unbiased mutation evolution, setting... F 0 and CR 0 is helpful in roughly determining the optimization direction in the global context; after the crossover, the candidate individuals are first evaluated. Boundary checks are performed on the elements.

4. The pilot power control method based on multi-strategy differential evolution according to claim 3, characterized in that: After crossover, the candidate individuals are first... When performing boundary checks on elements, the specific boundary check operation is boundary absorption, which means taking the boundary value for coefficients that exceed the boundary, and then considering the candidate individuals. With the original target individual By comparing fitness values, the next generation of target individuals is selected based on their optimal performance. .

5. The pilot power control method based on multi-strategy differential evolution according to claim 4, characterized in that: In step S5, the current population is divided into three subpopulations based on individual fitness values. Specifically, according to the Pareto criterion and individual fitness values, the top 20% of individuals in the entire population are first divided into one subpopulation, denoted as X1. Then, the bottom 80% of individuals are equally divided into two subpopulations, denoted as X2 and X3. By dividing the population into subpopulations, the top 20% of individuals with better fitness values ​​can perform a refined search to obtain a better solution, while the bottom 80% of individuals can perform a large-scale search using different control parameters, thus ensuring population diversity while avoiding getting trapped in local optima.

6. The pilot power control method based on multi-strategy differential evolution according to claim 5, characterized in that: In step S6, based on the second-stage multi-strategy differential evolution algorithm, corresponding mutation strategies are applied to the three subpopulations, corresponding control parameters are set, and each individual is updated through crossover and selection. Specifically, corresponding mutation strategies are applied to the three subpopulations, corresponding control parameters are set, and individuals are updated through crossover and selection. For the first subpopulation X1, the DE / best / 2 mutation strategy is used to improve its mining capacity and further search for the optimal solution. For subpopulations X2 and X3, the DE / rand / 1 strategy is used for further unbiased global search, and the search and mining capacity of the two subpopulations are balanced by setting different control parameters. The DE / best / 2 strategy is expressed as follows: Both DE / rand / 1 and DE / rand / 2 belong to unbiased mutation strategies. The DE / rand / 1 strategy is represented as... The crossover and selection operations are the same as step S4.

7. The pilot power control method based on multi-strategy differential evolution according to claim 6, characterized in that: In step S6, the current population is divided into subpopulations again before each iteration to ensure population diversity and avoid getting trapped in local optima.

8. A pilot power control system based on multi-strategy differential evolution, characterized in that: The method for implementing the method as described in any one of claims 1 to 7 includes a model building module, an initialization module, a fitness value calculation module, a multi-strategy differential evolution module, and a data output module. The model building module is used to build the uplink training phase signal model for a decellularized massive MIMO architecture; The initialization module is used to initialize the multi-strategy differential evolution algorithm; The fitness value calculation module is used to calculate the fitness value of each individual based on the initial population and find the initial optimal individual; The multi-strategy differential evolution module is used based on the first-stage multi-strategy differential evolution algorithm. It employs an unbiased mutation evolution strategy for all individuals in the population, sets corresponding control parameters, updates individual elements through crossover and selection, selects the optimal individual, and then updates the iteration count. It then checks if the maximum iteration count has been reached; if so, it divides the current population into three subpopulations based on the individual's fitness value; otherwise, it repeats the first-stage multi-strategy differential evolution algorithm. Based on the second-stage multi-strategy differential evolution algorithm, it employs corresponding mutation strategies for each of the three subpopulations, sets corresponding control parameters, updates each individual through crossover and selection, and then updates the iteration count. It then checks if the maximum iteration count has been reached; if so, it selects the current globally optimal individual as the coefficient combination for the user corresponding to this pilot; otherwise, it repeats the second-stage multi-strategy differential evolution algorithm. The data output module is used to output the pilot signal used by the user and its corresponding pilot power coefficient.

Images