Cellular-free large-scale MIMO network deployment method with maximum benefit-cost ratio

Abstract

The invention relates to a cellular-free large-scale MIMO (Multiple Input Multiple Output) network deployment method with a maximum benefit-cost ratio. Comprising the following steps: dividing a coverage area of the cellular-free large-scale MIMO network into N coverage units according to a wireless access point coverage distance, and calculating the transmission capacity of the nth coverage unit; wherein n belongs to (0, N); calculating the service capability of the nth coverage unit according to the transmission capacity of the nth coverage unit and the average transmission capacity demand in the nth coverage unit; calculating a benefit-cost ratio according to the service capabilities of all the coverage units and the average deployment cost of the coverage area; based on the benefit-cost ratio, establishing a cellular-free large-scale MIMO network model; and carrying out optimization solution on the non-cellular large-scale MIMO network model to obtain an optimal wireless access point deployment decision. According to the method and the device, the non-cellular large-scale MIMO network model is optimized and solved to obtain a network deployment scheme which is constructed at the minimum cost and has the most benefit.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a method for deploying a cellular-free large-scale MIMO network with a maximized benefit-cost ratio. Background Art

[0002] As a key technology of 6G mobile networks, non-cellular massive multiple input multiple output (MIMO) networks can provide users with collaborative services, eliminate cell boundaries, and achieve unified high-speed network coverage by deploying a large number of wireless access points (APs). In the 6G era, the demand for mobile traffic varies significantly due to factors such as user equipment type (UE), connection density, and service type. Fortunately, the performance of non-cellular massive MIMO networks can be more accurately shaped by optimizing the number and location of deployed APs to match different traffic requirements.

[0003] In the related technologies, in the non-cellular massive MIMO system, although the AP deployment guided by traffic demand can significantly improve the energy efficiency of the network, due to the rapid expansion of the network, the mobile network has borne extremely high deployment costs, and the network operation and construction are facing huge economic pressure. It is crucial for the non-cellular massive MIMO network to achieve green economy and coordinated and efficient development.

[0004] Therefore, it is necessary to provide a new technical solution to improve one or more problems existing in the above solutions.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present application, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of the present application is to provide a method for deploying a cellular-free massive MIMO network with a maximized benefit-cost ratio, thereby overcoming one or more problems caused by limitations and defects of related technologies at least to a certain extent.

[0007] According to an embodiment of the present application, a method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio is provided, the method comprising:

[0008] The coverage area of ​​the non-cellular massive MIMO network is divided into N coverage units according to the coverage distance of the wireless access point, and the transmission capacity of the nth coverage unit is calculated; wherein n∈(0,N];

[0009] Calculating the service capacity of the nth coverage unit according to the transmission capacity of the nth coverage unit and the average transmission capacity requirement in the nth coverage unit;

[0010] Calculating a benefit-cost ratio according to the service capabilities of all the coverage units and the average deployment cost of the coverage area;

[0011] Based on the benefit-cost ratio, a cell-free massive MIMO network model is established;

[0012] The non-cellular massive MIMO network model is optimized and solved to obtain the optimal wireless access point deployment decision.

[0013] In an embodiment of the present application, the coverage area of ​​the non-cellular massive MIMO network is divided into N coverage units according to the coverage distance of the wireless access point, and the transmission capacity of the nth coverage unit is calculated, including:

[0014] In the time division duplex mode, each wireless access point is configured with N T antennas, the wireless access point is at L candidate locations, x=(x1,…,x l ,…,x L ) is the deployment decision vector of the wireless access point, where x l is equal to 0 or 1, x l =1 indicates that the wireless access point is deployed at the first candidate location, and the signal to interference noise ratio of the kth user is expressed as:

[0015]

[0016] Where ρ is the uplink transmission power, τ is the length of the pilot sequence, and ξ lk is the mean square value of the estimated channel between the lth wireless access point and the kth user, β lk is the large-scale fading coefficient between the lth wireless access point and the kth user, L represents the number of candidate locations, β li represents the large-scale fading coefficient between the lth wireless access point and the i-th user;

[0017] Calculating the transmission capacity of the nth coverage unit according to the signal to interference plus noise ratio of the kth user includes:

[0018]

[0019] In the formula, represents the coverage area of ​​the nth coverage unit, It indicates that the kth user equipment type is located in the nth coverage unit.

[0020] In an embodiment of the present application, the service capability of the nth coverage unit includes:

[0021]

[0022] In the formula, C n (x) represents the transmission capacity of the nth coverage unit, represents the average transmission capacity requirement within the nth coverage unit.

[0023] In an embodiment of the present application, the benefit-cost ratio includes:

[0024]

[0025] In the formula, represents the average deployment cost of the coverage area, A represents the total coverage area of ​​the non-cellular massive MIMO network, Indicates the total number of wireless access points in the coverage area, c AP Represents the normalized cost of deploying a single wireless access point.

[0026] In an embodiment of the present application, the non-cellular massive MIMO network model includes:

[0027]

[0028] In an embodiment of the present application, the step of optimizing and solving the non-cellular massive MIMO network model to obtain an optimal wireless access point deployment decision includes:

[0029] The spatiotemporal graph neural network model is used to optimize and solve the non-cellular massive MIMO network model to obtain the optimal wireless access point deployment decision; wherein the spatiotemporal graph neural network model includes a spatial learning network and a temporal learning network;

[0030] or,

[0031] The agent-based deep deterministic policy gradient algorithm is used to optimize and solve the non-cellular large-scale MIMO network model to obtain the optimal wireless access point deployment decision.

[0032] In an embodiment of the present application, the method of optimizing and solving the non-cellular massive MIMO network model using a spatiotemporal graph neural network model to obtain the optimal wireless access point deployment decision includes:

[0033] The non-cellular massive MIMO network model is converted into a space-time graph data, wherein the space-time graph data includes: in, represents the node feature set at time t, x trepresents the characteristics of the wireless access point at time t, R t represents the characteristics of the user at time t, C t Represents the characteristics of the covering unit at time t; represents the edge feature set at time t, t=1,...,T, represents the edge feature from the user to the wireless access point at time t, represents the edge feature from the user to the coverage unit at time t, represents the edge feature from the coverage unit to the wireless access point at time t;

[0034] The spatiotemporal graph data is used as input of the graph attention network, the graph attention network is used to learn the spatial learning network, node features and edge features are weighted according to the traffic demand at time t, and spatial state features of wireless access point deployment are extracted;

[0035] The output of the graph attention network learning at time T is used as the input of the gated recurrent unit, and the gated recurrent unit is used to learn the time learning network, learn the time characteristics and time evolution law of the wireless access point deployment decision at time T, and optimize the wireless access point deployment decision that meets the traffic demand at time T; wherein the wireless access point deployment decision that meets the traffic demand at time T includes:

[0036] x T+1 =σ2(W hx H T ) (6);

[0037] Where W hx represents the hidden state calculated by the gated recurrent unit, H T is the hidden state updated by the gated recurrent unit at time T, σ2 represents the output layer activation function, which is a sigmoid function here;

[0038] A loss function is defined according to the non-cellular massive MIMO network model, and the spatiotemporal graph neural network model is trained by minimizing the loss function, and the trained and optimized spatiotemporal graph neural network model outputs the optimal wireless access point deployment decision; wherein the loss function is defined as the negative benefit-cost ratio with constraint penalty at the average T moment, and the negative benefit-cost ratio is expressed as:

[0039]

[0040] In the formula, λ represents the penalty coefficient, λ>0, represents the average transmission capacity requirement within the nth coverage unit at time t, B t (x T+1 ) represents the benefit-cost ratio at time t obtained by taking into account the optimal wireless access point deployment decision, C n,t(x T+1 ) represents the achievable throughput of the nth coverage unit at time t calculated by the optimal wireless access point deployment decision.

[0041] In an embodiment of the present application, the agent-based deep deterministic policy gradient algorithm is used to optimize and solve the non-cellular massive MIMO network model to obtain an optimal wireless access point deployment decision, including:

[0042] Defining an optimization model of the multi-agent deep deterministic policy gradient algorithm, the optimization model comprising: state, action and reward;

[0043] The non-cellular massive MIMO network model is optimized and solved by the optimization model to obtain an optimal wireless access point deployment decision;

[0044] Among them, state: the i-th agent obtains its local information from the network environment at time t, and the local information includes the candidate location A of the wireless access point p (t), user location distribution U p (t) and mobile traffic demand The state of the ith agent at time t is defined as Then the joint state of all the agents at time t is defined as I represents the number of agents;

[0045] Action: Based on the current state and network environment at time t, the action space of the i-th agent at time t represents the wireless access point deployment strategy at the current time t, and the joint action of all the agents at time t is

[0046] Reward: Define the rewards of all the agents according to the cellular-free massive MIMO network model and construct a reward function.

[0047] In an embodiment of the present application, the reward function is:

[0048]

[0049] In the formula, c t represents the penalty term when the constraint conditions are not met, B t (x) represents the benefit-cost ratio at time t.

[0050] In an embodiment of the present application, optimizing and solving the non-cellular massive MIMO network model by using the optimization model to obtain an optimal wireless access point deployment decision includes:

[0051] The i-th agent observes the current state at time t from the network environment And input it into the actor network to obtain the initial strategy for wireless access point deployment;

[0052] Each of the agents performs the action at the current time t and obtains the reward value r t , and update the joint state s of all agents at time t+1 t+1 ;

[0053] All of the agents will be four-tuple {s t ,a t ,r t ,s t+1} is stored in the experience replay pool, and the process is repeated until the experience replay pool is full, and F training samples are randomly selected from the experience replay pool to train the actor network and the centralized critic network to eliminate the time correlation between the training samples, and the joint state and joint action of all the agents are obtained through the centralized critic network.

[0054] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0055] In one embodiment of the present application, the coverage area is divided into N coverage units through the above method, and the benefit-cost ratio is defined according to the service capabilities of all coverage units and the average deployment cost of the coverage area. Based on the benefit-cost ratio, the wireless access point deployment problem is established as a cellular-free massive MIMO network model, and the cellular-free massive MIMO network model is optimized and solved to obtain the most effective network deployment plan built at the lowest cost.

[0056] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0058] Figure 1 A flowchart schematically illustrates a method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio in an exemplary embodiment of the present application;

[0059] Figure 2 The schematic diagram of the optimization principle of a non-cellular massive MIMO network based on a spatiotemporal graph neural network in an exemplary embodiment of the present application is shown;

[0060] Figure 3 The architecture diagram of the spatiotemporal graph neural network model in the exemplary embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0061] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0062] In addition, the accompanying drawings are only schematic illustrations of the present application and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0063] In this exemplary embodiment, a method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio is first provided. Figure 1 As shown in , the method includes:

[0064] Step S101: Divide the coverage area of ​​the non-cellular massive MIMO network into N coverage units according to the coverage distance of the wireless access point, and calculate the transmission capacity of the nth coverage unit; wherein n∈(0,N].

[0065] Step S102: Calculate the service capacity of the nth coverage unit according to the transmission capacity of the nth coverage unit and the average transmission capacity requirement in the nth coverage unit.

[0066] Step S103: Calculate the benefit-cost ratio according to the service capabilities of all coverage units and the average deployment cost of the coverage area.

[0067] Step S104: Based on the benefit-cost ratio, a non-cellular massive MIMO network model is established.

[0068] Step S105: Optimize and solve the non-cellular massive MIMO network model to obtain an optimal wireless access point deployment decision.

[0069] In one embodiment of the present application, the coverage area is divided into N coverage units through the above method, and the benefit-cost ratio is defined according to the service capabilities of all coverage units and the average deployment cost of the coverage area. Based on the benefit-cost ratio, the wireless access point deployment problem is established as a cellular-free massive MIMO network model, and the cellular-free massive MIMO network model is optimized and solved to obtain the most effective network deployment plan built at the lowest cost.

[0070] Next, we will refer to Figures 2 to 3 Each step of the above method in this example implementation is described in more detail.

[0071] In step S101, a cell-free massive MIMO network is considered, where M APs are distributed within the coverage area and connected to a central processing unit (CPU) via a backhaul link with unlimited capacity.

[0072] Furthermore, in time division duplex mode, each wireless access point is configured with N T antennas, wireless access points are located in L candidate locations, and are consistent with the real environment L≥M. x=(x1,…,x l ,…,x L ) is the deployment decision vector of the wireless access point, where x l is equal to 0 or 1, x l =1 means that the wireless access point is deployed at the first candidate location, and the signal to interference noise ratio of the kth user is expressed as:

[0073]

[0074] Where σ is the uplink transmit power, τ is the length of the pilot sequence, and ξ lk is the mean square value of the estimated channel between the lth wireless access point and the kth user, β lk is the large-scale fading coefficient between the lth wireless access point and the kth user, L represents the number of candidate locations, β li represents the large-scale fading coefficient between the lth wireless access point and the i-th user.

[0075] In order to achieve accurate design of service capacity, the transmission capacity of the nth coverage unit is calculated according to the signal-to-interference-to-noise ratio of the kth user, including:

[0076]

[0077] In the formula, represents the coverage area of ​​the nth coverage unit, It indicates that the kth user equipment type is located in the nth coverage unit.

[0078] In step S102, in order to evaluate the matching degree between the achievable capacity and the traffic demand, the service capacity is defined, and specifically, the service capacity of the nth coverage unit is calculated according to the transmission capacity of the nth coverage unit and the average transmission capacity demand in the nth coverage unit.

[0079] Furthermore, the service capabilities of the nth coverage unit include:

[0080]

[0081] In the formula, C n (x) represents the transmission capacity of the nth coverage unit, Represents the average transmission capacity requirement within the nth coverage unit.

[0082] In step S103, in order to measure the deployment cost and service capability, a benefit-cost ratio is defined, and specifically, the benefit-cost ratio is calculated based on the service capability of all coverage units and the average deployment cost of the coverage area.

[0083] Furthermore, the benefit-cost ratio includes:

[0084]

[0085] In the formula, represents the average deployment cost of the coverage area, A represents the total coverage area of ​​the non-cellular massive MIMO network, Indicates the total number of wireless access points in the coverage area, c AP Represents the normalized cost of deploying a single wireless access point.

[0086] In step S104, based on the benefit-cost ratio, the non-cellular massive MIMO AP deployment optimization problem with the maximum benefit-cost ratio is established as a non-cellular massive MIMO network model.

[0087] Further, the non-cellular massive MIMO network model includes:

[0088]

[0089] In step S105, the non-cellular massive MIMO network model is optimized and solved to obtain an optimal wireless access point deployment decision.

[0090] It should be understood that by optimizing and solving formula (5), the optimal AP deployment decision that meets the traffic demand with the minimum deployment cost can be obtained. To this end, the cell-free massive MIMO network will be accurately planned so that its service capacity can meet different traffic demands.

[0091] In this regard, the present application provides the first method or the second method for optimization and solution. The details are as follows:

[0092] The first method is to use the spatiotemporal graph neural network model to optimize and solve the non-cellular massive MIMO network model to obtain the optimal wireless access point deployment decision; Figure 3 As shown, the spatiotemporal graph neural network model includes a spatial learning network and a temporal learning network.

[0093] or,

[0094] The second method is to use the agent-based deep deterministic policy gradient algorithm to optimize and solve the non-cellular massive MIMO network model to obtain the optimal wireless access point deployment decision.

[0095] When the first method is used, considering that the traffic demand has the characteristics of temporal and spatial non-uniformity, the spatiotemporal graph neural network is used to solve formula (5). Figure 2 The optimization principle of non-cellular massive MIMO network based on the spatiotemporal graph neural network model architecture is shown in the figure. First, the non-cellular massive MIMO AP deployment optimization problem is transformed into a spatiotemporal graph data. The relationship between nodes and edges in the graph is used to represent the spatial topological relationship between APs, users and coverage units in the mobile communication system, and the logical association between known parameters is represented by node and edge features.

[0096] Secondly, the specific architecture of the spatiotemporal graph neural network model is as follows Figure 3 As shown in the figure, it consists of a spatial learning network and a temporal learning network. The above initialized spatiotemporal graph data is used as input. First, the Graph Attention Network (GAT) is used for the spatial learning network to weight the features of nodes and edges according to the traffic demand at time t, and extract the spatial state features of AP deployment. Then, the Gate Recurrent Unit (GRU) is used to learn the temporal learning network, learn the temporal features and temporal evolution laws, obtain dynamic information in the time dimension, and optimize AP deployment decisions.

[0097] Finally, the corresponding loss function is designed according to the cellular-free massive MIMO network model that maximizes the benefit-cost ratio. The cellular-free massive MIMO network model is trained by minimizing the loss function. The spatiotemporal neural network finally outputs the number and location of AP deployment, enabling the cellular-free massive MIMO network to achieve on-demand service capabilities with minimal deployment cost.

[0098] The specific process of using the first method is as follows:

[0099] In one embodiment, a spatiotemporal graph neural network model is used to optimize and solve a non-cellular massive MIMO network model to obtain an optimal wireless access point deployment decision, including:

[0100] First, the non-cellular massive MIMO network model is converted into a spatiotemporal graph data, which includes: in, represents the node feature set at time t, x t represents the characteristics of the wireless access point at time t, R t represents the characteristics of the user at time t, C t Represents the characteristics of the covering unit at time t; represents the edge feature set at time t, t=1,...,T, represents the edge feature from the user to the wireless access point at time t, represents the edge feature from the user to the coverage unit at time t, represents the edge feature from the coverage unit to the wireless access point at time t. Correspondingly, the edge feature is defined as is the large-scale fading coefficient from the kth user to the lth AP, Edge Features Indicates whether the k-th user is associated with the n-th coverage unit, Edge Features It is expressed as the impact of the traffic demand of the nth coverage unit on the deployment decision of the first AP. Due to the different mobile traffic demands, the values ​​of node and edge features change in different time and space.

[0101] Secondly, the spatiotemporal graph data is used as the input of the graph attention network, and the graph attention network is used to learn the spatial learning network. The node features and edge features are weighted according to the traffic demand at time t to extract the spatial state features of the wireless access point deployment.

[0102] In the spatial learning phase, GAT learns the spatial dependence of AP deployment according to the non-uniform traffic demand at time t. The achievable transmission rate R of each UE is calculated through AP-UE message passing. k,t Then, the transmission capacity C of the nth coverage unit is updated by the message transmission from the user to the coverage unit. n,t Finally, the node feature x of AP is optimized and updated through the attention mechanism t .

[0103] Thirdly, the output of graph attention network learning at time T is used as the input of the gated recurrent unit, and the gated recurrent unit is used to learn the time learning network to learn the time characteristics and time evolution laws of wireless access point deployment decisions at time T, and optimize the wireless access point deployment decisions that meet the traffic requirements at time T.

[0104] Specifically, GRU includes an update gate and a reset gate. The update gate determines the influence of the previous hidden state on the new hidden state, and the reset gate determines whether to forget the previous hidden state. The candidate hidden state W is calculated from the results of the update gate and the reset gate. hx and the final state H t Then, the AP deployment decision that meets the traffic demand at time T is output through the output layer.

[0105] The wireless access point deployment decisions that meet the traffic requirements at time T include:

[0106] x T+1 =σ2(W hx H T ) (6);

[0107] Where W hx represents the hidden state calculated by the gated recurrent unit, H T is the hidden state updated by the gated recurrent unit at time T, σ2 represents the output layer activation function, which is a sigmoid function here;

[0108] Finally, the loss function is defined according to the non-cellular massive MIMO network model, and the spatiotemporal graph neural network model is trained by minimizing the loss function. The trained and optimized spatiotemporal graph neural network model outputs the optimal wireless access point deployment decision. Among them, the loss function is defined as the negative benefit-cost ratio with constraint penalty in the average T time, and the negative benefit-cost ratio is expressed as:

[0109]

[0110] In the formula, λ represents the penalty coefficient, λ>0, represents the average transmission capacity requirement within the nth coverage unit at time t, B t (x T+1 ) represents the benefit-cost ratio at time t obtained by taking into account the optimal wireless access point deployment decision, C n,t (x T+1 ) represents the achievable throughput of the nth coverage unit at time t calculated by the optimal wireless access point deployment decision.

[0111] Formula (5) is a 0-1 nonlinear integer programming problem, which is an NP-hard problem. Therefore, the present application also provides a second method to optimize and solve Formula (5).

[0112] The specific process of optimizing and solving formula (5) by the second method is as follows:

[0113] In one embodiment, an agent-based deep deterministic policy gradient algorithm is used to optimize and solve a non-cellular massive MIMO network model to obtain an optimal wireless access point deployment decision, including:

[0114] Define the optimization model of the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, which includes: state, action, and reward;

[0115] The optimization model is used to optimize and solve the non-cellular massive MIMO network model to obtain the optimal wireless access point deployment decision;

[0116] Among them, state: the i-th agent obtains its local information from the network environment at time t, and the local information includes the candidate location A of the wireless access point p (t), user location distribution U p (t) and mobile traffic demand The state of the i-th agent at time t is defined as Then the joint state of all agents at time t is defined as I represents the number of agents;

[0117] Action: Based on the current state and network environment at time t, the action space of the i-th agent at time t represents the wireless access point deployment strategy at the current time t, and the joint action of all agents at time t is

[0118] Rewards: Define rewards for all agents based on the cell-free massive MIMO network model and construct a reward function.

[0119] It should be understood that in the process of defining actions, the AP needs to select the action to be performed based on the current state of the non-cellular massive MIMO network system and the network environment. In the process of defining rewards, in the MADDGP algorithm, multiple agents cooperate with each other to maximize the total utility of the system. The deployment location and number of APs are optimized to maximize the system benefit-cost ratio. The rewards of all agents are defined according to formula (5).

[0120] Furthermore, the reward function is:

[0121] r t = r(s t ,a t )=B t (x)-c t (8);

[0122] In the formula, c trepresents the penalty term when the constraint conditions are not met, B t (x) represents the benefit-cost ratio at time t.

[0123] It should be understood that after defining the above states, actions, and rewards, the following specific MADDPG-based AP deployment optimization strategy is given.

[0124] In one embodiment, the non-cellular massive MIMO network model is optimized and solved by an optimization model to obtain an optimal wireless access point deployment decision, including:

[0125] The i-th agent observes the current state at time t from the network environment And input it into the actor network to obtain the initial strategy for wireless access point deployment;

[0126] Each agent performs the action at the current time t and obtains the reward value r t , and update the joint state s of all agents at time t+1 t+1 ;

[0127] All agents will be four-tuple {s t ,a t ,r t ,s t+1}Store them in the experience replay pool and repeat the process until the experience replay pool is full. Then, randomly select F training samples from the experience replay pool to train the actor network and the centralized critic network to eliminate the temporal correlation between the training samples, and obtain the joint state and joint action of all agents through the centralized critic network.

[0128] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary technical means in the art that are not disclosed in the present application.

Claims

1. A method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio, characterized in that: The method includes: The coverage area of ​​the non-cellular massive MIMO network is divided into N coverage units according to the coverage distance of the wireless access point, and the transmission capacity of the nth coverage unit is calculated; wherein n∈(0,N]; Calculating the service capacity of the nth coverage unit according to the transmission capacity of the nth coverage unit and the average transmission capacity requirement in the nth coverage unit; Calculating a benefit-cost ratio according to the service capabilities of all the coverage units and the average deployment cost of the coverage area; Based on the benefit-cost ratio, a cell-free massive MIMO network model is established; The non-cellular massive MIMO network model is optimized and solved to obtain the optimal wireless access point deployment decision.

2. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 1, characterized in that: The method of dividing the coverage area of ​​the non-cellular massive MIMO network into N coverage units according to the coverage distance of the wireless access point, and calculating the transmission capacity of the nth coverage unit, comprises: In the time division duplex mode, each wireless access point is configured with N T antennas, the wireless access point is at L candidate locations, x=(x1,…,x l ,…,x L ) is the deployment decision vector of the wireless access point, where x l is equal to 0 or 1, x l =1 indicates that the wireless access point is deployed at the first candidate location, and the signal to interference noise ratio of the kth user is expressed as: Where ρ is the uplink transmission power, τ is the length of the pilot sequence, and ξ lk is the mean square value of the estimated channel between the lth wireless access point and the kth user, β lk is the large-scale fading coefficient between the lth wireless access point and the kth user, L represents the number of candidate locations, β li represents the large-scale fading coefficient between the lth wireless access point and the i-th user; Calculating the transmission capacity of the nth coverage unit according to the signal to interference plus noise ratio of the kth user includes: In the formula, represents the coverage area of ​​the nth coverage unit, It indicates that the kth user equipment type is located in the nth coverage unit.

3. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 2, characterized in that: The service capability of the nth coverage unit includes: In the formula, C n (x) represents the transmission capacity of the nth coverage unit, C n represents the average transmission capacity requirement within the nth coverage unit.

4. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 3, characterized in that: The benefit-cost ratio includes: In the formula, represents the average deployment cost of the coverage area, A represents the total coverage area of ​​the non-cellular massive MIMO network, Indicates the total number of wireless access points in the coverage area, c AP Represents the normalized cost of deploying a single wireless access point.

5. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 4, characterized in that: The non-cellular massive MIMO network model includes:

6. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 5, characterized in that: The optimizing and solving the non-cellular massive MIMO network model to obtain an optimal wireless access point deployment decision includes: The spatiotemporal graph neural network model is used to optimize and solve the non-cellular massive MIMO network model to obtain the optimal wireless access point deployment decision; wherein the spatiotemporal graph neural network model includes a spatial learning network and a temporal learning network; or, The agent-based deep deterministic policy gradient algorithm is used to optimize and solve the non-cellular large-scale MIMO network model to obtain the optimal wireless access point deployment decision.

7. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 6, characterized in that: The adopting of a spatiotemporal graph neural network model to optimize and solve the non-cellular massive MIMO network model to obtain the optimal wireless access point deployment decision includes: The non-cellular massive MIMO network model is converted into a space-time graph data, wherein the space-time graph data includes: in, represents the node feature set at time t, x t represents the characteristics of the wireless access point at time t, R t represents the characteristics of the user at time t, C t Represents the characteristics of the covering unit at time t; represents the edge feature set at time t, t=1,...,T, represents the edge feature from the user to the wireless access point at time t, represents the edge feature from the user to the coverage unit at time t, represents the edge feature from the coverage unit to the wireless access point at time t; The spatiotemporal graph data is used as input of the graph attention network, the graph attention network is used to learn the spatial learning network, node features and edge features are weighted according to the traffic demand at time t, and spatial state features of wireless access point deployment are extracted; The output of the graph attention network learning at time T is used as the input of the gated recurrent unit, and the gated recurrent unit is used to learn the time learning network, learn the time characteristics and time evolution law of the wireless access point deployment decision at time T, and optimize the wireless access point deployment decision that meets the traffic demand at time T; wherein the wireless access point deployment decision that meets the traffic demand at time T includes: x T+1 =σ2(W hx H T ) (6); Where W hx represents the hidden state calculated by the gated recurrent unit, H T is the hidden state updated by the gated recurrent unit at time T, σ2 represents the output layer activation function, which is a sigmoid function here; A loss function is defined according to the non-cellular massive MIMO network model, and the spatiotemporal graph neural network model is trained by minimizing the loss function, and the trained and optimized spatiotemporal graph neural network model outputs the optimal wireless access point deployment decision; wherein the loss function is defined as the negative benefit-cost ratio with constraint penalty at the average T moment, and the negative benefit-cost ratio is expressed as: In the formula, λ represents the penalty coefficient, λ>0, represents the average transmission capacity requirement within the nth coverage unit at time t, B t (x T+1 ) represents the benefit-cost ratio at time t obtained by taking into account the optimal wireless access point deployment decision, C n,t (x T+1 ) represents the achievable throughput of the nth coverage unit at time t calculated by the optimal wireless access point deployment decision.

8. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 6, characterized in that: The method of optimizing and solving the non-cellular massive MIMO network model by using an agent-based deep deterministic policy gradient algorithm to obtain an optimal wireless access point deployment decision includes: Defining an optimization model of the multi-agent deep deterministic policy gradient algorithm, the optimization model comprising: state, action and reward; The non-cellular massive MIMO network model is optimized and solved by the optimization model to obtain an optimal wireless access point deployment decision; Among them, state: the i-th agent obtains its local information from the network environment at time t, and the local information includes the candidate location A of the wireless access point p (t), user location distribution U p (t) and mobile traffic demand The state of the ith agent at time t is defined as Then the joint state of all the agents at time t is defined as I represents the number of agents; Action: Based on the current state and network environment at time t, the action space of the i-th agent at time t represents the wireless access point deployment strategy at the current time t, and the joint action of all the agents at time t is Reward: Define the rewards of all the agents according to the cellular-free massive MIMO network model and construct a reward function.

9. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 8, characterized in that: The reward function is: In the formula, c t represents the penalty term when the constraint conditions are not met, B t (x) represents the benefit-cost ratio at time t.

10. The method for deploying a non-cellular massive MIMO network with a maximized benefit-cost ratio according to claim 9, characterized in that: The optimizing and solving the non-cellular massive MIMO network model by using the optimization model to obtain an optimal wireless access point deployment decision includes: The i-th agent observes the current state at time t from the network environment And input it into the actor network to obtain the initial strategy for wireless access point deployment; Each of the agents performs the action at the current time t and obtains the reward value r t , and update the joint state s of all agents at time t+1 t+1 ; All of the agents will be four-tuple {s t ,a t ,r t ,s t+1 } is stored in the experience replay pool, and the process is repeated until the experience replay pool is full, and F training samples are randomly selected from the experience replay pool to train the actor network and the centralized critic network to eliminate the time correlation between the training samples, and the joint state and joint action of all the agents are obtained through the centralized critic network.

Images