A power control method for non-cellular massive MIMO system

Abstract

The present invention discloses a power control method for a non-cellular large-scale MIMO system, comprising the following steps: S1. performing AP selection based on pilot allocation; S2. obtaining the minimum spectrum efficiency (SE) of all terminals in the network by using an improved particle swarm algorithm to achieve power control; the present invention selects access points based on pilot allocation, can select appropriate access points, reduce the complexity of calculation, and uses the improved particle swarm algorithm to perform joint optimization to solve the maximum and minimum fair power control optimization problem of the non-cellular large-scale MIMO system, which can improve the total spectrum efficiency of users in the system while providing users with relatively consistent services.

Description

Technical Field

[0001] The present invention relates to the field of mobile communication technology, and more particularly to a power control method for a non-cellular large-scale MIMO system. Background Art

[0002] In recent years, mobile communications have developed rapidly. In order to meet the growing number of users, higher data transmission rates and more stringent service quality requirements, non-cellular massive MIMO has emerged as a new solution. User-centric non-cellular massive MIMO eliminates the concept of traditional cells and cell edges, and deploys access points far more than the number of users in the area to jointly serve users, which can reduce interference between users while providing users with more consistent and efficient services.

[0003] However, due to the large number of access points, when multiple access points collaborate and a single access point serves multiple users, it is particularly important to select the appropriate service access point and reasonably control the transmission power of each access point to reduce co-channel interference and improve the overall performance of the system.

[0004] Therefore, how to propose a power control method for a non-cellular massive MIMO system is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0005] In view of this, the present invention provides a power control method for a non-cellular massive MIMO system, and the present invention adopts the following technical solutions:

[0006] A power control method for a non-cellular massive MIMO system is studied under a user-centric non-cellular massive MIMO system model. Due to the large number of access points, how to select a suitable service access point and reasonably control the transmission power of each access point to reduce co-channel interference and improve the overall performance of the system. First, an access point selection algorithm is needed to make the selection. A user-centric method is used to form AP clusters serving each UE. The access point selection algorithm based on pilot allocation is analyzed to reduce the complexity of the calculation. Then, an optimization equation for the power control problem is established and the specific formulas in the uplink and downlink are derived. The power control algorithm uses an improved particle swarm algorithm to solve the maximum and minimum fair power control problem of the system and improve the overall transmission rate.

[0007] In order to achieve the above purpose, the following steps are included:

[0008] S1. Perform AP selection based on pilot allocation: respectively P The user UEs are assigned from the first to the τth P mutually orthogonal pilots, and then to the τth P+1 to K-th user UEs are allocated pilot signals one by one to complete AP selection;

[0009] S2. By using the improved particle swarm algorithm to maximize the minimum spectrum efficiency SE of all terminals in the network to achieve power control:

[0010] S21. Use vector p = [p 1 ,...,p K ] T Denotes all uplink powers. The uplink SE of UEk is determined by the effective SINR associated with p. The effective SINR of UEk in the uplink is expressed as SINR k (p); By maximizing SINR k (p) minimum value to maximize the minimum spectrum efficiency SE of all terminals in the network; by obtaining The maximum value of is equivalent to seeking the maximum SINR k (p)Minimum SINR is the signal to interference and noise ratio;

[0011] S22. Settings The initial lower and upper limits of ; the lower limit is set to a value that tends to zero, and the upper limit is set to Initialize the solution variables to zero: p opt =0 k＝K ;

[0012] S23. Initialize the particle swarm algorithm parameters and use the lower limit as the initial local optimal solution P id0 Set to zero and use the upper limit as the initial global optimal solution P gd0 Set as Calculate P according to the fitness function id0 and P gd0 The corresponding initial fitness; the fitness function is:

[0013]

[0014] S24. Calculation by particle swarm algorithm The corresponding power solution p is obtained;

[0015] S25. Update the weight inertia factor ω:

[0016]

[0017] Among them, f is the current fitness, f min is the minimum fitness, f vag is the average fitness, ω max With ω min are the preset maximum and minimum inertia weight factors respectively;

[0018] S26. Update the particle's velocity v according to the updated weight inertia factor ω id and position x id , and update the output of the current local optimal solution p id and the global optimal solution p gd ; Assign the value of the currently obtained power solution p to the optimal power solution p opt ;

[0019] S27. When the global optimal solution p gd With the local optimal solution p id When the difference is greater than the accuracy of the solution ε, the contents of S24-S26 are executed repeatedly until p gd With p id The difference is less than ε, where ε>0, and the last updated p is output opt .

[0020] Preferably, S1 is directed to the τth P The pilots are allocated one by one from the first to the Kth UEs. The specific contents of completing AP selection include:

[0021] S11. Determine the optimal and suboptimal access points corresponding to the channel state of the kth user UEk, denoted as AP1 and AP1', respectively, and save the suboptimal access point AP1' into the set N; UEk uses pilot k, where k = 1, ..., τp;

[0022] S12. Assign the pilot with the least pilot contamination at AP1 to UEk; and obtain the pilot with the second smallest interference at AP1 τ', and assign τ' to the remaining UEs;

[0023] S13. Set parameter r to indicate the upper limit of pilot interference for the AP in the system to normally serve the UE:

[0024]

[0025] If the pilot interference of the AP with the strongest current channel gain cannot meet the constraint condition of r, then the AP is abandoned as the main service AP; the suboptimal AP is selected from the set N as the main AP and the calculation is performed again until the constraint condition of r is met;

[0026] S14. When all UEs have been allocated pilots, cluster creation is completed and AP selection is performed: each AP identifies the UE with the maximum channel gain using each pilot as the AP service object.

[0027] Preferably, the specific content of S21 is:

[0028] The numerator of SINR depends on the power p of the desired signal. k, the interference term in the denominator depends on all power coefficients of p, and the effective SINR of UEk in the uplink is

[0029]

[0030] in represents the average channel gain of the desired signal, c k =[c k1 ...c kK ] T represents the average channel gain vector of each interference signal, represents the effective noise variance, and

[0031]

[0032]

[0033] In the considered uplink scenario, there are K individual transmit power constraints, so by maximizing SINR k (p) minimizes the minimum value to maximize the minimum spectrum efficiency SE of all terminals in the network:

[0034]

[0035] Preferably, the maximum SINR is sought in S21 k (p)Minimum The specific contents include:

[0036] Introducing an auxiliary variable t, t represents the lowest SINR among all UEs, then we can obtain The problem is transformed into obtaining the maximum value of t Get the maximum value of t according to the constraints The constraints are:

[0037] SINR k (p)≥t,k=1,...,K

[0038]

[0039] t opt Represents the optimal objective value of the problem;

[0040] Will get The problem is transformed into obtaining the maximum total power question.

[0041] Preferably, the specific contents of the initialization algorithm parameters in S25 include: the number of particles n, the number of iterations m, the learning factor c 1 ,c 2 , upper and lower limits of weighted inertia factor ω max,ω min , local optimal solution p id , the global optimal solution p gd , the accuracy of the solution ε>0, update the initial velocity v of the particle id and the initial position x id , the initial fitness f of each particle.

[0042] Preferably, in S27, the particle velocity v is updated according to the updated weighted inertia factor ω. id and position x id The specific contents include:

[0043] The specific position of the i-th particle in the search space is represented by a D-dimensional vector

[0044] x i =(x i1 ,...,x iD ),i=1,...,N

[0045] The velocity of the i-th particle is also expressed as a D-dimensional vector:

[0046] v i =(v i1 ,...,v iD ),i=1,...,N

[0047] After each iteration, the local optimal solution found for each individual is saved and recorded as p id The global optimal solution found by the entire group is denoted as p gd ;

[0048] According to the local optimal solution p id and the global optimal solution is denoted by p gd The velocity v of the i-th particle id To update:

[0049] v id =ωv id +c 1 r 1 (p id -x id )+c 2 r 2 (p gd -x id )

[0050] Among them, c 1 ,c 2 is the learning factor, which is determined according to the value range of the independent variable; r 1 ,r 2 is a random number in the range [0,1]; the position x of the i-th particleid renew:

[0051] x id =x id +v id .

[0052] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a power control method for a non-cellular large-scale MIMO system, in which AP selection and an improved particle swarm joint optimization algorithm are used to jointly optimize and solve the maximum and minimum power control problem of the non-cellular large-scale MIMO system. First, the AP selection algorithm based on pilot allocation selects part of the AP to work, which can achieve performance close to full AP transmission and lower calculation complexity, and can also avoid unnecessary power waste. Then, the improved particle swarm algorithm is used to solve the maximum and minimum fair power control problem. The inertia factor of the particle swarm optimization algorithm is optimized and updated through the fitness of each iteration to find the global optimal solution faster, which can improve the convergence speed and the average and maximum spectrum efficiency of users. The present invention can protect the fairness of users and improve the overall service quality while solving the maximum and minimum power control problem of the non-cellular large-scale MIMO system, while satisfying the premise of providing users with relatively consistent service performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0054] Figure 1 A schematic diagram of a complex scalar sent by an AP provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0055] Figure 2 A schematic diagram of a complex multiplication quantity sent by an AP provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0056] Figure 3 A CDF schematic diagram of SE provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0057] Figure 4 A CDF schematic diagram of an uplink SE provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0058] Figure 5A CDF schematic diagram of the minimum SE of the uplink provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0059] Figure 6 A schematic diagram of the fitness of a particle swarm algorithm provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0060] Figure 7 It is a schematic diagram of SE comparison of three algorithms provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0061] Figure 8 A schematic diagram of a CDF diagram of a downlink SE provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0062] Fig. 9 A schematic diagram of a CDF diagram of a downlink minimum SE provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0063] Fig.10 A fitness curve of a particle swarm algorithm provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention;

[0064] Fig.11 The figure shows a SE comparison of three algorithms provided in an embodiment of a power control method for a non-cellular massive MIMO system of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] The embodiment of the present invention discloses a power control method for a non-cellular massive MIMO system, including the following contents:

[0067] A power control method for a non-cellular massive MIMO system comprises the following steps:

[0068] S1. Perform AP selection based on pilot allocation: respectively P The user UEs are assigned from the first to the τth P mutually orthogonal pilots, and then to the τth P +1 to K-th user UEs are allocated pilot signals one by one to complete AP selection;

[0069] S2. By using the improved particle swarm algorithm to maximize the minimum spectrum efficiency SE of all terminals in the network to achieve power control:

[0070] S21. Use vector p = [p 1 ,...,p K ] T Denotes all uplink powers. The uplink SE of UEk is determined by the effective SINR associated with p. The effective SINR of UEk in the uplink is expressed as SINR k (p); By maximizing SINR k (p) minimum value to maximize the minimum spectrum efficiency SE of all terminals in the network; by obtaining The maximum value is equivalent to seeking the maximum SINR k (p)Minimum SINR is the signal to interference and noise ratio;

[0071] S22. Settings The initial lower and upper limits of ; the lower limit is set to a value that tends to zero, and the upper limit is set to Initialize the solution variables to zero: p opt =0 k＝K ;

[0072] S23. Initialize the particle swarm algorithm parameters and use the lower limit as the initial local optimal solution P id0 Set to zero and use the upper limit as the initial global optimal solution P gd0 Set as Calculate P according to the fitness function id0 and P gd0 The corresponding initial fitness; the fitness function is:

[0073]

[0074] S24. Calculation by particle swarm algorithm The corresponding power solution p is obtained;

[0075] S25. Update the weight inertia factor ω:

[0076]

[0077] Among them, f is the current fitness, f min is the minimum fitness, f vag is the average fitness, ω max With ω min are the preset maximum and minimum inertia weight factors respectively;

[0078] S26. Update the particle's velocity v according to the updated weight inertia factor ωid and position x id , and update the output of the current local optimal solution p id and the global optimal solution p gd ; Assign the value of the currently obtained power solution p to the optimal power solution p opt ;

[0079] S27. When the global optimal solution p gd With the local optimal solution p id When the difference is greater than the accuracy of the solution ε, the contents of S24-S26 are executed repeatedly until p gd With p id The difference is less than ε, where ε>0, and the last updated p is output opt .

[0080] In order to further implement the above technical solution, S1 P The pilots are allocated one by one from the first to the Kth UEs. The specific contents of completing AP selection include:

[0081] S11. Determine the optimal and suboptimal access points corresponding to the channel state of the kth user UEk, denoted as AP1 and AP1', respectively, and save the suboptimal access point AP1' into the set N; UEk uses pilot k, where k = 1, ..., τp;

[0082] S12. Assign the pilot with the least pilot contamination at AP1 to UEk; and obtain the pilot with the second smallest interference at AP1 τ', and assign τ' to the remaining UEs;

[0083] S13. Set parameter r to indicate the upper limit of pilot interference for the AP in the system to normally serve the UE:

[0084]

[0085] If the pilot interference of the AP with the strongest current channel gain cannot meet the constraint condition of r, then the AP is abandoned as the main service AP; the suboptimal AP is selected from the set N as the main AP and the calculation is performed again until the constraint condition of r is met;

[0086] S14. When all UEs have been allocated pilots, cluster creation is completed and AP selection is performed: each AP identifies the UE with the maximum channel gain using each pilot as the AP service object.

[0087] It needs further explanation:

[0088] Mutually orthogonal pilots τ p must be reused among UEs and allocated to them to limit pilot contamination. Pilot allocation is a combinatorial problem. pIn the set of equations for pilot frequencies, there is (τ p ) K possible values, so their computational complexity grows exponentially with the number of UEs. A suboptimal approach is feasible in practice, iteratively assigning pilots to UEs, always choosing the pilot that causes the least pilot pollution.

[0089] Clustering means that UE k is represented by only the set The AP services in the cluster are closely related to the pilot allocation. A basic AP clustering algorithm is to first allocate pilots to UEs and then let each AP serve a clear τ p UEs; then, for each pilot, the AP serves the subset of UEs that have been assigned the pilot with the strongest channel gain.

[0090] The pilot allocation algorithm consists of two steps.

[0091] First, for all τ P UEs are assigned from the first to the τth P mutually orthogonal pilots: UEk uses pilot k, where k = 1, ..., τ P . From the τth P UEs +1 to K are then randomly assigned pilots one by one. UEk first determines which AP has the best channel status with it and marks the performance index of the AP as

[0092]

[0093] Then select the AP with the suboptimal channel status and mark it as

[0094]

[0095] Save the suboptimal AP to the collection In this case, the AP is not selected to serve UEk temporarily.

[0096] Prioritizing the allocation of the pilot with the second least pilot contamination to UEk can eliminate the risk that the UE may not be served by any AP. For each pilot t, the AP can calculate the average channel gain β of the UE that has been allocated to the pilot il The AP selects the pilot with the smallest pilot interference.

[0097]

[0098] The AP then selects the pilot with the second smallest pilot interference.

[0099]

[0100] The suboptimal pilot is allocated to the UE, and then the pilot is allocated to the next UE.

[0101] Determine whether the pilot interference of the AP with the strongest current channel gain meets the constraint condition of r. If not, abandon the AP as the service main AP and select the second best AP from the set N as the main AP to calculate again until formula (5) is satisfied.

[0102] When all UEs are assigned pilots, clusters can be created. Each AP goes through each pilot and identifies the UE with the largest channel gain using that pilot, and the AP will serve this UE. Table 1 summarizes the process.

[0103] Table 1 AP selection algorithm based on pilot allocation

[0104]

[0105] The key idea of ​​the algorithm is:

[0106] In the first step, whenever a new UE is admitted to the network, it will first identify the AP with the strongest channel gain. The accessing UE calculates and designates AP1 as its primary AP according to formula (2). This selection is user-centric.

[0107] In the second step, the UE also uses the broadcast signal to synchronize with the AP. The master AP calculates the preferred pilot τ using formula (4) and notifies the surrounding APs of the presence of the new UE.

[0108] In the third step, the surrounding APs can determine whether they should change the service to the UE with pilot τ. This algorithm can also be applied when the UE is highly mobile and its channel statistics change.

[0109] In order to further implement the above technical solution, the specific content of S21 is:

[0110] The numerator of SINR depends on the power p of the desired signal. k , the interference term in the denominator depends on all power coefficients of p, and the effective SINR of UEk in the uplink is

[0111]

[0112] in represents the average channel gain of the desired signal, c k =[c k1 …c kK ] T represents the average channel gain vector of each interference signal, represents the effective noise variance, and

[0113]

[0114]

[0115] In the considered uplink scenario, there are K individual transmit power constraints, so by maximizing SINR k (p) minimizes the minimum value to maximize the minimum spectrum efficiency SE of all terminals in the network:

[0116]

[0117] It needs further explanation:

[0118] The power control problem in the non-cellular massive MIMO system is the focus of research on the non-cellular massive MIMO system. By adjusting the transmit power on the user side or the access point side, the power control that maximizes the minimum user transmission rate can reduce the co-channel interference and improve the overall performance of the de-cellular massive MIMO system. The goal of maximum-minimum fairness is to maximize the minimum SE among all terminals in the network.

[0119] Due to SE k (p) = log 2 (1+SINR k In (p)), the SE of UEk is the effective SINR, which is the SINR k (p), so maximizing the minimum SE is the same as maximizing the minimum effective SINR of all UEs. Assume that the transmit power vector has R linear constraints:

[0120]

[0121] The fixed vector is the weight coefficient of the power coefficient of each UE, p max is the maximum allowed power of all UEs, the maximum and minimum fairness problem can be expressed as

[0122]

[0123] In the uplink, the vector p = [p 1 ,…,p K ] T represents all uplink powers, and they affect all UEs. The uplink SE of UEk is determined by the effective SINR associated with p. The numerator of SINR depends on the power of its desired signal, p. k , the interference term in the denominator depends on all power coefficients of p. The effective SINR of UEk in the uplink can be expressed as

[0124]

[0125] in represents the average channel gain of the desired signal, ck =[c k1 …c kK ] T represents the average channel gain vector of each interference signal, represents the effective noise variance, and

[0126]

[0127] Regarding the maximum and minimum fairness problem, in the uplink scenario considered, there are K separate transmit power constraints, so the maximum and minimum fairness optimization problem can be concretized as

[0128]

[0129] The maximum and minimum fairness problem focuses on an extreme type of fairness, which only helps the UE with the worst channel condition in the non-cellular network system, but sacrifices the interests of all other UEs. This paper proposes to use an improved particle swarm optimization algorithm to solve the maximum and minimum power control problem of non-cellular massive MIMO systems, and to improve the SE of all UEs as much as possible while ensuring a relatively uniform quality of service for users.

[0130] In order to further implement the above technical solution, the maximum SINR is obtained in S21. k (p)Minimum The specific contents include:

[0131] Introducing an auxiliary variable t, t represents the lowest SINR among all UEs, then we can obtain The problem is transformed into obtaining the maximum value of t Get the maximum value of t according to the constraints The constraints are:

[0132] SINR k (p)≥t,k=1,…,K

[0133]

[0134] t opt Represents the optimal objective value of the problem;

[0135] The goal of the power control feasibility problem is to find an arbitrary solution p that satisfies the constraints; then we will use a solution with the same constraints but a total power of The biggest question replaces the feasibility question; it will obtain The problem is transformed into obtaining the maximum total power Problem: Get according to constraints The maximum value of , where the constraints are:

[0136] SINRk (p)≥p gd ,k=1,...,K

[0137]

[0138] p≥0 K

[0139] in are R linear constraints on the transmit power vector, and the fixed vector is the weight coefficient of the power coefficient of each UE, p max is the maximum allowed power of all UEs.

[0140] In order to further implement the above technical solution, the specific contents of the algorithm parameters initialized in S25 include: the number of particles n, the number of iterations m, the learning factor c 1 ,c 2 , upper and lower limits of weighted inertia factor ω max ,ω min , local optimal solution p id , the global optimal solution p gd , the accuracy of the solution ε>0, update the initial velocity v of the particle id and the initial position x id , the initial fitness f of each particle.

[0141] In order to further implement the above technical solution, in S27, the particle velocity v is updated according to the updated weight inertia factor ω. id and position x id The specific contents include:

[0142] The specific position of the i-th particle in the search space is represented by a D-dimensional vector

[0143] x i =(x i1 ,...,x iD ),i=1,...,N

[0144] The velocity of the i-th particle is also expressed as a D-dimensional vector:

[0145] v i =(v i1 ,…,v iD ),i=1,…,N

[0146] After each iteration, the local optimal solution found for each individual is saved and recorded as p id The global optimal solution found by the entire group is denoted as p gd ;

[0147] According to the local optimal solution pid and the global optimal solution is denoted by p gd The velocity v of the i-th particle id To update:

[0148] v id =ωv id +c 1 r 1 (p id -x id )+c 2 r 2 (p gd -x id )

[0149] Among them, c 1 ,c 2 is the learning factor, which is determined according to the value range of the independent variable; r 1 ,r 2 is a random number in the range [0,1]; the position x of the i-th particle id renew:

[0150] x id =x id +v id .

[0151] Particle swarm optimization (PSO) is a heuristic algorithm. Assume there is a D-dimensional search space, in which there is a group of N particles. The goal of the group is to find an optimal position in the search space through continuous search. The specific position of the i-th particle in the group in the search space is represented by a D-dimensional vector

[0152] x i =(x i1 ,...,x iD ),i=1,...,N (11)

[0153] The velocity of the i-th particle can also be expressed as a D-dimensional vector

[0154] v i =(v i1 ,...,v iD ),i=1,...,N (12)

[0155] After each iteration, the local optimal solution found for each individual is saved and recorded as p id The global optimal solution found by the entire group is denoted as p gd .

[0156] Then, according to the optimal solution obtained previously, the velocity v of the i-th particle isid To update, the formula is as follows

[0157] v id =ωv id +c 1 r 1 (p id -x id )+c 2 r 2 (p gd -x id ) (13)

[0158] Among them, ω is the inertia weight, which reflects the influence of the local optimal solution selected by the individual in the past on the present; c 1 ,c 2 is the learning factor (or acceleration constant); r 1 ,r 2 is a random number in the range [0,1].

[0159] The i-th particle updates its position x according to the following formula id :

[0160] x id =x id +v id (14)

[0161] It needs further explanation:

[0162] There are two ways to terminate the particle swarm algorithm: one is when the preset maximum number of iterations is reached during the running process; the other is when the difference between the global optimal solutions obtained two or more times in a row meets the preset minimum allowable value.

[0163] In the early stage of particle swarm algorithm operation, a larger inertia weight is conducive to faster convergence of particles and approaching the global optimal solution. However, due to the large inertia weight, the position jump of each iteration update is too large, and the global optimal solution is missed. In the later stage of operation, a smaller weight fine-tunes the speed and position of the particles, which is conducive to local search, but increases the risk of falling into the local optimum, and may not be able to find the best global optimal solution.

[0164] In order to solve the defects of the particle swarm algorithm, the strategy of adaptively updating the inertia weight factor is used to improve the particle swarm algorithm. When the particles in the particle swarm tend to the same target value or fall into the local optimum, the inertia weight value is increased. When the target values ​​of the particles are inconsistent, the inertia weight value is reduced to suppress the dispersion trend of the particles.

[0165] Improved content of the improved particle swarm algorithm: pre-set the maximum and minimum weighted inertia factors. Before the i-th particle updates its velocity, the inertia weight factor ω is updated according to the current fitness of the particle swarm algorithm, which can be expressed as

[0166]

[0167] Among them, f is the current fitness of the particle swarm optimization algorithm, f min is the minimum fitness, f vag is the average fitness, ω max With ω min are the maximum and minimum inertia weight factors that are preset respectively. In the specific algorithm, the various parameters and fitness values ​​of the algorithm can be adjusted according to different user needs and environments to obtain the global optimal solution that meets the specific requirements.

[0168] According to the maximum and minimum fair power control problem of non-cellular massive MIMO system, the fitness function of the improved particle swarm algorithm used to solve this problem is set as

[0169]

[0170] In each iteration of the algorithm, the fitness value of the particle is updated through the fitness function formula, and then the inertia weight factor is updated according to the fitness, and then the particle's own position and speed are updated to find the local optimal solution and then the global optimal solution, and the local optimal solution and the global optimal solution are updated according to the fitness value.

[0171] The improved particle swarm optimization algorithm process is shown in Table 2.

[0172] Table 2 Improved particle swarm algorithm

[0173]

[0174]

[0175] The present invention will be further described below through simulation:

[0176] Define a non-cellular massive MIMO system. Table 3 gives the key parameters of the simulation. Assume that the APs are uniformly and randomly distributed in the coverage area. In the simulation, it is assumed that each UE transmits at full power during the pilot and data transmission phases. In each coherent block, the pilot sequence length is τ. p =10, then the sequence length for uplink data transmission is τ u =τ c -τ p =190.

[0177] Table 3 System simulation parameter settings

[0178]

[0179]

[0180] Simulation 1:

[0181] The simulation using the AP selection algorithm based on pilot allocation is compared with the simulation using all APs. As the number of users increases, the average number of complex scalars (data plus pilots) sent from the AP to the CPU in each coherent block is compared. The number of Monte Carlo simulation experiments is 100.

[0182] like Figure 1 As shown in the figure, using this algorithm (each AP serves τ UEs), the complex scalars transmitted will basically not increase as the number of UEs increases. When all APs serve all terminals, it will grow with the number of UEs. Since the right AP is selected to serve the user, the meaningless data transmitted by the AP with poor service performance is reduced, thereby reducing the number of complex scalars sent to the CPU. When the number of UEs is large, the advantage of this algorithm that requires less signal transmission is reflected.

[0183] Simulation 2:

[0184] When using MMSE estimation, the average number of complex multiplications required to calculate the local combination vector in each coherent block when using the AP selection algorithm based on pilot allocation and using all APs, as the number of users increases. The number of Monte Carlo simulation experiments is 100.

[0185] from Figure 2 It can be seen that the average number of complex multiplications required to calculate the local combination vector is the largest when K = 20. Because the matrix that needs to be inverted is the same when all APs serve all UEs, it is calculated once for all UEs. In the algorithm, each UE performs matrix inversion separately, resulting in the highest complexity when K = 20 compared to the MMSE estimation combination when all APs serve all UEs. However, as the number of UEs increases, the computational complexity of the algorithm decreases, while when all APs serve all UEs, the computational complexity increases because calculations need to be performed for each UE. Therefore, when the number of UEs is large, using this algorithm can effectively reduce the computational complexity.

[0186] Simulation 3:

[0187] When using L-MMSE and MR estimation, the CDF distribution of the SE of the pilot allocation-based AP selection algorithm and full AP selection is used, that is, the probability distribution function. The number of Monte Carlo simulation experiments is 100.

[0188] from Figure 3 It can be seen that the SE provided by the AP selection algorithm based on pilot allocation is not as large as that provided by using all AP services, but the difference is not much. In particular, the SE estimated by MR is basically the same. That is to say, the AP selection algorithm using pilot allocation can select the appropriate AP to serve the user without basically affecting the user's SE, which can reduce the energy consumption of the system. And combined with the previous analysis, the use of this algorithm can also reduce the complex scalar and complex multiplication during calculation, which can reduce the calculation complexity and reduce the calculation pressure on the AP side and the central processor side. This shows the superiority of using this algorithm for AP selection.

[0189] Table 4 gives the parameter settings of the improved PSO algorithm in this simulation.

[0190] Table 4 Improved particle swarm algorithm parameter settings

[0191]

[0192] Simulation 4:

[0193] In the uplink of the non-cellular massive MIMO system, under the premise of using MMSE estimation to process the signal, the particle swarm algorithm and the improved particle swarm algorithm are used to solve the maximum and minimum fair power control problem. The number of Monte Carlo simulation experiments is 100. The simulation results are as follows:

[0194] from Figure 4 It can be seen that in the uplink, both the particle swarm algorithm and the improved particle swarm algorithm can provide higher SE than the basic maximum-minimum fairness algorithm, and the distribution of SE is very uniform, which is in line with the idea of ​​maximum-minimum fairness to provide UEs with a more consistent service experience. Using the full power allocation and combined rate maximization algorithm, assuming that all UEs transmit at maximum power, it provides the theoretical upper limit of SE that all UEs can achieve under completely ideal conditions. However, in actual situations, due to hardware loss or severe interference, UEs with poor channel conditions may be allocated 0 power.

[0195] The fractional power control method is also only applicable to the situation where all APs serve UEs at the same time. Although the allocation scheme can be adjusted through the fractional power control coefficient to provide UEs with more average services or to pursue the maximum service quality of a single UE. However, the coefficient of fractional power control cannot be changed according to the actual situation and can only be set in advance. When the coefficient of the fractional power control method is v = -0.5, a relatively similar SE is provided for UEs with poor channel conditions, which can achieve the purpose of maximum and minimum fair power control to a certain extent, but at the same time sacrifice the SE of other UEs with better channel conditions.

[0196] from Figure 4From the CDF curve of the improved particle swarm algorithm in , it can be seen that the curve is close to the overall right shift of the particle swarm algorithm, that is, the algorithm can provide a higher SE than the particle swarm algorithm, and can improve the SE upper limit of users with better performance without affecting the SE of UEs with poor performance. The upper and lower ends of the curve are close to horizontal, indicating that it can provide users with a nearly continuous SE distribution, which is in line with the maximum and minimum fair power control idea, can well protect the fairness of users, and will not sacrifice the service quality of a single UE.

[0197] Figure 5 This is the CDF distribution diagram of the minimum SE obtained in each round of uplink simulation. It can be seen from the figure that the distribution of the minimum SE provided by the improved particle swarm algorithm is close to that of the basic maximum-minimum fairness algorithm, and is better than the particle swarm algorithm, and is significantly stronger than the full power allocation and combined rate maximization algorithm. It shows that the power allocation scheme using the improved particle swarm algorithm can guarantee the minimum SE provided for users with poor channel conditions, will not sacrifice its performance to improve the overall service quality, can improve fairness between users, and conforms to the maximum-minimum fairness power control idea. The full power allocation and combined rate maximization algorithm cannot guarantee fairness between users, and the minimum SE that can be provided is lower. Although the fractional power control method can provide performance similar to that of the proposed algorithm when the coefficient is v = -0.5, it has limitations and cannot be adjusted according to actual conditions. When v = 0.5, the minimum SE provided by the algorithm is very low because the forced allocation coefficient cannot take care of users with poor channel conditions. This is the limitation of the fractional power control method.

[0198] Figure 6 This is the fitness curve of the particle swarm algorithm and the improved particle swarm algorithm in this simulation. The improved particle swarm algorithm can preferentially achieve high fitness when the number of iterations is low, that is, the algorithm converges faster. This shows that the improved particle swarm algorithm can save computing time, reduce the occupation of system computing resources, and improve the operating efficiency of the system while providing better performance. This proves the good performance of the improved particle swarm algorithm in solving this problem.

[0199] Figure 7 It is a SE comparison of the basic algorithm, particle swarm algorithm and improved particle swarm algorithm for solving the maximum and minimum fair power control problem of uplink in non-cellular massive MIMO systems, including the minimum, maximum, average and standard deviation of SE obtained by each algorithm.

[0200] First, compare the minimum SE, from Figure 7 It can be seen that the minimum SE values ​​provided by the three algorithms are close, that is, the improved particle swarm algorithm can maintain the lowest SE of the UE with the worst channel status and will not sacrifice the user's SE in order to improve the overall performance of the system. Figure 7From the average SE of , we can see that the particle swarm algorithm and the improved particle swarm algorithm can improve the average SE of UE, and the improved particle swarm algorithm has better performance. Figure 7 It can be seen from the maximum SE that both the particle swarm algorithm and the improved particle swarm algorithm can provide a maximum SE that is much higher than the basic maximum-minimum fairness algorithm. However, the maximum SE provided by the improved particle swarm algorithm is smaller than that of the particle swarm algorithm, which can improve the SE upper limit of some UEs without sacrificing the SE of any UE, reflecting the advantage of using the particle swarm algorithm to solve the maximum-minimum fairness optimization problem.

[0201] The standard deviation of SE of all UEs by the three algorithms shows that the improved particle swarm algorithm has the smallest standard deviation, indicating that the SE distribution of all UEs obtained by this algorithm is more concentrated and less uncertain. This means that this algorithm can provide users with a more consistent user experience than the basic maximum-minimum fairness algorithm, protecting user fairness, which is also one of the advantages of this algorithm.

[0202] Simulation 5:

[0203] In the downlink, when using P-MMSE estimation to process the signal, the particle swarm algorithm and the improved particle swarm algorithm are used to solve the maximum and minimum fair power control problem. The number of Monte Carlo simulation experiments is 100. The simulation results are shown in Figure 2. Figure 8 shown.

[0204] The simulation is similar to the uplink power control. Figure 8 It can be seen that in the downlink, the particle swarm algorithm and the improved particle swarm algorithm can also provide higher SE than the basic maximum-minimum fairness algorithm, but the distribution of SE is not as uniform as in the uplink. When using equal power allocation, assuming that all APs transmit with equal power, the CDF curve of equal power allocation within the coverage of the system network is the leftmost curve at the bottom and the rightmost curve at the top of the tail among all curves, and it is very different from other curves. The reason is similar to the full power allocation in the uplink. If each AP is allocated with equal power, it means that UEs with good channel conditions can get better services, while UEs with worse channel conditions receive worse services. It is also impossible to provide the minimum performance guarantee for UEs and cannot guarantee user fairness. The situation reflected by this curve is different from that of the uplink. The combined rate maximization algorithm performs better than the equal power allocation within the network, and there will be no UEs with poor service quality.

[0205] from Figure 8It can be seen that the particle swarm algorithm and the improved particle swarm algorithm can provide the UE with an SE close to that of the fractional power control method, especially when the SE is less than 6bit / s / Hz. However, since the fractional power control method assumes that all APs serve the UE, it is not in line with reality. The curve of fractional power control can be used as a reference upper limit for maximum and minimum fair power control. From the CDF curve of the improved particle swarm algorithm, it can be seen that when the SE is less than 5.5bit / s / Hz, the curve is on the left of the particle swarm algorithm curve, and when the SE is greater than 7bit / s / Hz, the curve is on the right of the particle swarm algorithm curve. That is, in the downlink power control of the improved particle swarm algorithm, the SE distribution of the UE is not as concentrated as that of the particle swarm algorithm. When the SE is large or small, the SE distribution has a certain deviation, that is, the fairness of the user will be relatively poor. Therefore, in the downlink, the particle swarm algorithm better conforms to the maximum and minimum fairness idea than the improved particle swarm algorithm, providing users with more consistent services.

[0206] Fig. 9 This is the CDF distribution diagram of the minimum SE obtained in each round of downlink simulation. It can be seen from the figure that the distribution of the minimum SE that the improved particle swarm algorithm can provide is close to that of the basic maximum-minimum fairness algorithm, and is significantly stronger than the equal power allocation and combined rate maximization algorithm. This algorithm and the particle swarm algorithm have their own advantages and disadvantages, and are not significantly stronger than the latter. However, it is significantly lower than the basic maximum-minimum fairness algorithm in the minimum SE part, which shows that the power allocation scheme using the improved particle swarm algorithm can guarantee the performance of users with poor channel conditions to improve the overall service quality. The performance is lower than the level of the algorithm in the uplink. The performance of the equal power allocation scheme is poor, because this allocation scheme does not provide a certain minimum SE for users, but will sacrifice a small part to ensure the service quality for users with poor channel conditions. The SE of these users is often very low, and fairness to users cannot be guaranteed.

[0207] Fig.10 The fitness curves of the particle swarm algorithm and the improved particle swarm algorithm in this simulation are similar to the results in the uplink. The improved particle swarm algorithm can achieve a higher fitness when the number of iterations is low. However, in the later stage of iteration, the fitness of the particle swarm algorithm is higher than that of the improved particle swarm algorithm, which shows that the improved particle swarm algorithm does not have a great advantage over the particle swarm in solving the maximum and minimum fair power control problem of the downlink. Figure 7 This can also be seen from the CDF curves of the two algorithms.

[0208] Fig.11This paper compares the SE of the basic algorithm, particle swarm algorithm and improved particle swarm algorithm to solve the maximum and minimum fair power control problem of the downlink in user-centric cell-free massive MIMO systems, including the minimum, maximum, average and standard deviation of the SE obtained by each algorithm.

[0209] First, compare the minimum SE, from Fig.11 It can be seen that the minimum SE values ​​that the three algorithms can provide decrease in turn. The minimum SE provided by the improved particle swarm algorithm for UE is the smallest, even worse than the particle swarm algorithm. When solving the maximum and minimum fair power allocation problem, the improved particle swarm algorithm and the particle swarm algorithm sacrifice the SE of some UEs, that is, the improved particle swarm algorithm and the particle swarm algorithm cannot improve the overall SE under the premise of maintaining the minimum SE of several UEs with the worst channel status. Therefore, when the improved particle swarm algorithm and the particle swarm algorithm allocate power to users with relatively poor channel status in the downlink, they do not fully comply with the maximum and minimum fairness idea, and the performance is poor.

[0210] Compare Fig.11 The average SE of UE shows that the particle swarm algorithm and the improved particle swarm algorithm can improve the average SE of UE, and the performance of the two algorithms is close, and the improved particle swarm algorithm is slightly stronger. The improved particle swarm algorithm can improve the average SE of UE by 15.01% compared with the basic maximum-minimum fairness algorithm, which is higher than the percentage in the uplink. This shows that the improved particle swarm algorithm is better than the uplink performance in improving the average SE of downlink UE.

[0211] from Fig.11 It can be seen from the maximum SE that, similar to the simulation results of the uplink, both the particle swarm algorithm and the improved particle swarm algorithm can provide a maximum SE much higher than the basic maximum-minimum fair algorithm. Unlike the uplink, the maximum SE that the improved particle swarm algorithm can provide is larger. Combined with the previous analysis of the SE of the three algorithms, the particle swarm algorithm and the improved particle swarm algorithm still sacrifice the SE of a small part of UEs to improve the SE upper limit of some UEs. Although they can improve the overall SE of UEs in the system, they do not fully comply with the idea of ​​maximum-minimum fair power control and do not provide users with more consistent services.

[0212] The standard deviation of SE of all UEs by the three algorithms shows that the improved particle swarm algorithm has the largest standard deviation, which means that the SE distribution of all UEs obtained by this algorithm is more dispersed and has greater uncertainty. This is completely different from the situation in the uplink. This shows that the particle swarm algorithm and the improved particle swarm algorithm are weaker than the basic maximum-minimum fairness algorithm in providing users with more consistent services when solving the power control problem of the downlink.

[0213] This paper proposes to use AP selection and improved particle swarm optimization algorithm to jointly optimize and solve the maximum and minimum power control problem of non-cellular large-scale MIMO system. First, the AP selection algorithm based on pilot allocation is used to select AP, and then the improved particle swarm algorithm is used to optimize the maximum and minimum power control problem after AP selection. The AP selection algorithm based on pilot allocation is theoretically analyzed and simulated, which proves its superiority over full AP selection, that is, selecting some APs can achieve performance close to full AP transmission with lower computational complexity and avoid unnecessary power waste. On this basis, the maximum and minimum fair power control optimization problem is established. Then, the improved particle swarm algorithm is proposed to solve the maximum and minimum fair power control problem. The algorithm optimizes and updates the inertia factor of the particle swarm optimization algorithm through the fitness of each iteration to find the global optimal solution faster. Through simulation experiments and analysis, it is concluded that the use of the improved particle swarm algorithm can improve the convergence speed and improve the average and maximum spectrum efficiency of users. It is proved that the use of AP selection and improved particle swarm joint optimization algorithm can protect user fairness and improve the overall service quality while providing users with relatively consistent service performance in solving the maximum and minimum power control problem of non-cellular large-scale MIMO systems.

[0214] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0215] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A power control method for a non-cellular massive MIMO system, It is characterized in that The following steps are involved: S1. Perform AP selection based on pilot allocation: respectively P The user UEs are assigned from the first to the τth P mutually orthogonal pilots, and then to the τth P +1 to Kth user UEs are allocated pilots one by one to complete AP selection, τ P represents the length of the pilot sequence, that is, the number of orthogonal pilots that can be allocated in the system, and K represents the total number of all users in the system; S2. By using the improved particle swarm algorithm to maximize the minimum spectrum efficiency SE of all terminals in the network to achieve power control: S21. Use vector p = [p 1 ,…,p K ] T Denotes all uplink powers. The uplink SE of UEk is determined by the effective SINR associated with p. The effective SINR of UEk in the uplink is expressed as SINR k (p); By maximizing SINR k (p) minimum value to maximize the minimum spectrum efficiency SE of all terminals in the network; by obtaining The maximum value of is equivalent to seeking the maximum SINR k (p)Minimum SINR is the signal to interference and noise ratio; S22. Settings The initial lower and upper limits of ; the lower limit is set to a value that tends to zero, and the upper limit is set to Initialize the solution variables to zero: p opt =0 k＝K ; S23. Initialize the particle swarm algorithm parameters and use the lower limit as the initial local optimal solution P id0 Set to zero and use the upper limit as the initial global optimal solution P gd0 Set as Calculate P according to the fitness function id0 and P gd0 The corresponding current initial fitness; the fitness function is: Among them, τ c is the number of samples for each coherent block; S24. Calculation by particle swarm algorithm The corresponding power solution p is obtained; S25. Update the weight inertia factor ω: Among them, f is the current fitness, f min is the minimum fitness, f vag is the average fitness, ω max With ω min are the preset maximum and minimum inertia weight factors respectively; S26. Update the particle's velocity v according to the updated weight inertia factor ω id and position x id , and update the output of the current local optimal solution p id and the global optimal solution p gd ; Assign the value of the currently obtained power solution p to the optimal power solution p opt ; S27. When the global optimal solution p gd With the local optimal solution p id When the difference is greater than the accuracy of the solution ε, the contents of S23-S26 are executed repeatedly until p gd With p id The difference is less than ε, where ε>0, and the last updated p is output opt .

2. The power control method of a non-cellular massive MIMO system according to claim 1, It is characterized in that S1 to τ P The pilots are allocated one by one from the first to the Kth UEs. The specific contents of completing AP selection include: S11. Determine the optimal and suboptimal access points corresponding to the channel state of the kth user UEk, which are respectively denoted as and Suboptimal access point Save to set N; UEk uses pilot k, where k = 1, ..., τ P ; S12. Allocate UEk The pilot with the least pilot pollution is obtained; and The pilot signal τ' with the second smallest interference is selected and τ' is allocated to the remaining UEs; S13. Set parameter r to indicate the upper limit of pilot interference for the AP in the system to normally serve the UE: in, is the average channel gain; If the pilot interference of the AP with the strongest current channel gain cannot meet the constraint condition of r, then the AP is abandoned as the main service AP; the suboptimal AP is selected from the set N as the main AP and the calculation is performed again until the constraint condition of r is met; S14. When all UEs have been allocated pilots, cluster creation is completed and AP selection is performed: each AP identifies the UE with the maximum channel gain using each pilot as the AP service object.

3. The power control method of a non-cellular massive MIMO system according to claim 2, It is characterized in that The specific contents of S21 are as follows: The numerator of SINR depends on the power p of the desired signal. k , the interference term in the denominator depends on all power coefficients of p, and the effective SINR of UEk in the uplink is in represents the average channel gain of the desired signal, c k =[c k1 …c kK ] T represents the average channel gain vector of each interference signal, represents the effective noise variance, and In the considered uplink scenario, there are K individual transmit power constraints, so by maximizing SINR k (p) minimizes the minimum value to maximize the minimum spectrum efficiency SE of all terminals in the network:

4. The power control method of a non-cellular massive MIMO system according to claim 3, It is characterized in that S21 solves the problem of maximizing SINR k (p) Minimum value problem The specific contents include: Introducing an auxiliary variable t, t represents the lowest SINR among all UEs, then we can obtain The problem is transformed into obtaining the maximum value of t Get the maximum value of t according to the constraints The constraints are: SINR k (p)≥t,k=1,…,K t opt Represents the optimal objective value of the problem; Will get The problem is transformed into obtaining the maximum total power question.

5. The power control method of a non-cellular massive MIMO system according to claim 4, It is characterized in that The specific contents of the algorithm parameters initialized in S25 include: number of particles n, number of iterations m, learning factor c 1 ,c 2 , upper and lower limits of weighted inertia factor ω max ,ω min , local optimal solution p id , the global optimal solution p gd , the accuracy of the solution ε>0, update the initial velocity v of the particle id and the initial position x id , the initial fitness f of each particle.

6. The power control method of a non-cellular massive MIMO system according to claim 5, It is characterized in that In S27, the particle velocity v is updated according to the updated weighted inertia factor ω. id and position x id The specific contents include: The specific position of the i-th particle in the search space is represented by a D-dimensional vector x i =(x i1 ,…,x iD ),i=1,...,N The velocity of the i-th particle is also expressed as a D-dimensional vector: v i =(v i1 ,...,v iD ),i=1,...,N After each iteration, the local optimal solution found for each individual is saved and recorded as p id The global optimal solution found by the entire group is denoted as p gd ; According to the local optimal solution p id and the global optimal solution is denoted by p gd The velocity v of the i-th particle id To update: v id =ωv id +c 1 r 1 (p id -x id )+c 2 r 2 (p gd -x id ) Among them, c 1 ,c 2 is the learning factor, which is determined according to the value range of the independent variable; r 1 ,r 2 is a random number in the range [0,1]; The i-th particle moves to position x id renew: x id =x id +v id 。

Images