High energy efficiency dynamic adaptive hybrid precoding method, module and system in cell-free massive MIMO system

Abstract

The application discloses a high-energy-efficiency dynamic adaptive hybrid precoding method, module and system in a cell-free massive MIMO system, comprising: constructing an antenna grouping dynamic adaptive connection hybrid precoding architecture on each access point; constructing a Markov game model, all access point agents are participants, the state of each access point agent includes a channel estimation value, a user side reachable rate and an access point power consumption, the action is a precoding control scheme of the hybrid precoding architecture, and the reward is a weighted sum of the proportionality of the user side reachable rate and the inverse proportionality of the access point power consumption; each access point implements precoding according to the precoding control scheme and issues to a user; the user calculates the reachable rate according to the received signal and feeds back to the access point agent; the access point calculates the access point power consumption and feeds back to each access point agent, driving the access point agent to generate a precoding control scheme at the next moment. The application has higher energy efficiency.

Description

Technical Field

[0001] This invention relates to wireless communication technology, and more particularly to a high-efficiency dynamic adaptive hybrid precoding method, module, and system for non-cellular massive MIMO systems. Background Technology

[0002] The demands of 6G networks for ultra-high data rates, extremely low latency, and ubiquitous coverage are driving the development of wireless communication systems towards millimeter-wave bands and ultra-dense architectures. As a key method for achieving 6G goals, cellular-free massive MIMO systems overcome the limitations of cellular architectures by deploying distributed access points, theoretically achieving ultra-high spectral efficiency and seamless coverage. However, the proliferation of access points (APs) and antennas significantly increases hardware power consumption and coordination complexity, constituting a critical energy efficiency bottleneck. Employing efficient precoding techniques can yield considerable array gain and interference suppression benefits, providing strong support for efficient transmission. However, the energy efficiency of traditional precoding techniques requires further optimization. Summary of the Invention

[0003] To address the problems existing in the prior art, the purpose of this invention is to provide a high-energy-efficiency dynamic adaptive hybrid precoding method, module, and system for non-cellular massive MIMO systems with higher energy efficiency.

[0004] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0005] A high-energy-efficiency dynamic adaptive hybrid precoding method for cellular-free massive MIMO systems includes the following steps:

[0006] (1) Construct a hybrid precoding architecture that supports dynamic adaptive connection of antenna groups at each access point, wherein each phase shifter in the analog precoder of the hybrid precoding architecture is connected to multiple antennas through several switches;

[0007] (2) Construct a Markov game model with the goal of maximizing energy efficiency. In the Markov game model, all agents at the access points are game participants. The state of each agent at the access point includes the channel estimate, the reachable rate on the user side and the power consumption of the access point. The action is a precoding control scheme of a hybrid precoding architecture. The reward is the difference between the reachable rate on the user side and the weighted value of the power consumption of the access point.

[0008] (3) Each access point performs precoding according to the precoding control scheme output by the access point agent at the current time, and transmits the precoded data to the user through the antenna;

[0009] (4) The user calculates its achievable rate based on the signal it receives and feeds it back to the access point agent;

[0010] (5) Each access point calculates the power consumption at each access point, drives the access point agent to update the policy, and generates the precoded control scheme for the next moment.

[0011] Furthermore, the hybrid precoding hardware architecture specifically includes: a digital precoder connected in sequence, A radio frequency chain and an analog pre-encoder; the analog pre-encoder includes a phase shifter network and a switching network, the phase shifter network including... Each phase shifter, each RF chain connected to A phase shifter; the switching network includes A switch, For the number of access point antennas, The antennas are divided into g groups. Each phase shifter passes through Each switch is connected to Each antenna, linked by each RF chain Each phase shifter is linked to a different antenna and is activated when the RF chain is linked to at least one antenna.

[0012] Furthermore, the settings for each access point agent in step (2) are as follows:

[0013] state: ;action: ;

[0014] award: ;

[0015] in, Indicates time slot The state of the access point agent m, including the channel estimation between access point m and user k in time slot t. Time slot The achievable rate of user k and power consumption at access point m All of these were obtained through local observation, where K represents the number of users. Constructing a time slot The status of all access point agents Indicates time slot The action performed by the access point agent m This indicates the connection status between access point m and user k; the value is 1 when a connection is established and 0 otherwise. This represents the phase matrix of the phase shifter at access point m. The element in the nth row and ith column This represents the phase code value of phase shifter n connected to the i-th RF chain. If the i-th RF chain is not connected to phase shifter n... If the i-th RF chain is connected to phase shifter n, then the phase of phase shifter n is encoded into a usable phase set. In A binary number, and assign it to , This represents the switching state matrix of the switching network. Represents the first [unit] at access point m. The switching state matrix of the group of switches represents the first group of switches. In the group root antenna and The switching states of the switches between the phase shifters On each antenna and each RF chain The switching states of all switches between phase shifters are encoded into a single... A binary number, where b represents the number of quantization bits in the phase shifter. Where g represents the number of RF chains, and g represents the number of antenna groups. Indicates time slot The reward for the access point agent m. This represents the achievable rate of user k in time slot t. This represents the power consumption at access point m in time slot t. This represents the penalty factor.

[0016] Furthermore, the actions of each access point agent are reduced in dimensionality as follows:

[0017] An action embedding method is used to introduce an action embedding space. Strategy function From motion embedding space Sampling an embedded action ,Right now Then, a preset action is used to embed the mapping function. Map the sampled embedded actions to action sets. One of the actions, namely This allows high-dimensional 0-1 discrete actions to be mapped to a low-dimensional continuous action space.

[0018] Furthermore, the achievable rate mentioned in step (4) is calculated using the following formula:

[0019]

[0020] In the formula, This represents the achievable rate of user k in time slot t. Indicates channel bandwidth. , These represent the connection status between access point m and users k and k', respectively. The value is 1 when a connection is established and 0 otherwise. This represents the channel estimation between access point m and user k in time slot t. and Let these represent the switch matrix of the switch network and the phase shifter matrix of the phase shifter network, respectively. and This represents the digital precoding matrix output by the digital precoder. Indicates the number of data streams transmitted in parallel. Indicates the transmission power. This indicates the channel noise power.

[0021] Furthermore, the power consumption at each access point in step (5) is calculated using the following formula:

[0022] =

[0023] Where K represents the number of users. Indicates access point Whether it serves users; if yes, the value is 1; otherwise, it is 0. Indicates the efficiency of the power amplifier. Indicates the transmission power. This indicates the power consumption of baseband signal processing. , , , , , These represent the number of active RF chains, RF chain power consumption, number of active phase shifters, phase shifter power consumption, number of switches, and switch power consumption, respectively. and They represent access points respectively. Fixed power consumption of the fronthaul link between the CPU and the network traffic-related power consumption factor.

[0024] The loss function used by the driving access point agent to update the policy in step (5) is:

[0025]

[0026] In the formula, Indicates loss, This represents the mean square error function. This represents the joint control action of all access point agents. This represents the joint embedding action of all access point agents. express Possible values, , Embed a mapping function for the preset action. Represents probability. , These represent the states of time slots t and t+1, respectively.

[0027] A high-efficiency dynamic adaptive hybrid precoding module for a cellular-free massive MIMO system, the module being disposed at each access point, comprising:

[0028] The hybrid precoding architecture supports dynamic adaptive connection of antenna groups, where each phase shifter in the analog precoder is connected to multiple antennas via several switches.

[0029] An access point agent has a state consisting of a channel estimate, a user-side reachable rate, and access point power consumption. Its actions are based on a precoding control scheme using a hybrid precoding architecture. Its reward is the weighted difference between the user-side reachable rate and the access point power consumption. It is used as a game participant with other access point agents to form a Markov game model with the optimization objective of maximizing energy efficiency. The reachable rate is calculated by the user based on the received signal and fed back to the agent.

[0030] The precoding controller is used to precode data according to the precoding control scheme output by the access point agent at the current time, and then send the precoded data to the user.

[0031] The power consumption calculation module is used to calculate the power consumption at the current access point, drive the access point agent to update the policy, and generate the precoded control scheme for the next moment.

[0032] Furthermore, the hybrid precoding hardware architecture specifically includes: a digital precoder connected in sequence, A radio frequency chain and an analog pre-encoder; the analog pre-encoder includes a phase shifter network and a switching network, the phase shifter network including... Each phase shifter, each RF chain connected to A phase shifter; the switching network includes A switch, For the number of access point antennas, The antennas are divided into g groups. Each phase shifter passes through Each switch is connected to Each antenna, linked by each RF chain Each phase shifter is linked to a different antenna and is activated when the RF chain is linked to at least one antenna.

[0033] A high-energy-efficiency dynamic adaptive hybrid precoding system for cellular-free massive MIMO systems includes:

[0034] The access point is equipped with the high-efficiency dynamic adaptive hybrid precoding module of the aforementioned non-cellular massive MIMO system.

[0035] The user is used to calculate its achievable rate based on the signals it receives and to feed back to the access point agent.

[0036] Compared with existing technologies, the beneficial effects of this invention are as follows: Through dynamic design of adaptive precoding, this invention achieves three core values: First, by utilizing energy-efficient hardware structures, it reduces hardware energy consumption and improves energy efficiency with acceptable transmission rate loss; second, the array gain focuses the transmit power on the target user, reducing interference and minimizing energy waste in non-target directions; third, it adapts to the time-varying characteristics of the channel, rapidly adjusting the beam direction to maintain link stability in dynamic scenarios such as "rain attenuation" and "blockage" in high-frequency bands like millimeter waves, avoiding energy redundancy caused by signal retransmission. Therefore, this invention provides strong support for efficient transmission in cellular-free massive MIMO systems. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the high-efficiency dynamic adaptive hybrid precoding method in a cellular-free massive MIMO system provided by the present invention.

[0038] Figure 2 This is a schematic diagram of the hybrid precoding architecture provided by the present invention;

[0039] Figure 3 The simulation results of global energy efficiency for different precoding algorithms with different APs are shown in the figure.

[0040] Figure 4 The figure shows the simulation results of the global energy efficiency of the present invention under different signal-to-noise ratios under different precoding structures. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0042] Example 1

[0043] This invention provides a high-energy-efficiency dynamic adaptive hybrid precoding method for cellular-free massive MIMO systems, such as... Figure 1 As shown, it includes the following steps:

[0044] (1) Construct a hybrid precoding architecture that supports dynamic adaptive connection of antenna groups at each access point.

[0045] In this embodiment of the invention, a cellular-free massive MIMO system has Each configuration has The connection point of the root antenna and Each configuration has For users of the root antenna, these distributed APs provide services to users on the same time-frequency resources.

[0046] In this hybrid precoding architecture, each phase shifter in the analog precoder is connected to multiple antennas via several switches. Specifically, as shown... Figure 2 As shown, the hybrid precoding architecture includes: a digital precoder connected in sequence, A radio frequency chain and an analog pre-encoder; the analog pre-encoder includes a phase shifter network and a switching network, the phase shifter network including... A phase shifter, the phase of the phase shifter adopts... Bit quantization, each RF chain is connected to A phase shifter; the switching network includes A switch, The antennas are divided into g groups. Each phase shifter passes through Each switch is connected to Each antenna has a switch that controls the on / off state of its corresponding link. Each RF link connects to... Each phase shifter is linked to a different antenna to prevent reducing the degree of freedom of phase control, and is activated when the RF chain is linked to at least one antenna.

[0047] Each AP has RF chain and One antenna, parallel transmission The data streams, in parallel, are first processed by a digital precoder, then by the RF chain and phase shifters in the analog precoder, and finally mapped to... The antenna transmits data.

[0048] (2) Construct a Markov game model with the goal of maximizing energy efficiency.

[0049] In the Markov game model, all access point agents are participants. Each agent's state includes a channel estimate, user-side reachable rate, and access point power consumption. Actions are precoding control schemes using a hybrid precoding architecture, and rewards are the weighted difference between the user-side reachable rate and the access point power consumption. All access point agents are in a game-like relationship, each aiming to maximize their own user-side reachable rate and minimize energy efficiency. Deep reinforcement learning is employed by each agent to reach an equilibrium, achieving an optimal balance that simultaneously considers global energy efficiency and reachable rate.

[0050] Specifically, Markov game models include: state space Action space , These represent the action sets of access point agents 1, ..., M; and the transition function. ,in Represents the distribution of states; rewards for the access point agent. Each access point agent has an independent observation set. , as state space During execution, based on strategy Guided Actions. Specifically, the access point agent determines the action based on the state using a policy function, then determines the state at the next moment based on the executed action, calculates the reward, and updates the policy function based on the reward.

[0051] The settings for each access point agent are as follows:

[0052] state: ;

[0053] action: ;

[0054] award: ;

[0055] in, Indicates time slot The state of the access point agent m, including the channel estimation between access point m and user k in time slot t. Time slot The achievable rate of user k and power consumption at access point m All of these were obtained through local observation, where K represents the number of users. Constructing a time slot The status of all access point agents Indicates time slot The action performed by the access point agent m This indicates the connection status between access point m and user k; the value is 1 when a connection is established and 0 otherwise. This represents the phase matrix of the phase shifter at access point m. The element in the nth row and ith column This represents the phase code value of phase shifter n connected to the i-th RF chain. If the i-th RF chain is not connected to phase shifter n... If the i-th RF chain is connected to phase shifter n, then the phase of phase shifter n is encoded into a usable phase set. In A binary number, and assign it to , specifically middle Indicates connection to the i-th radio frequency chain Each phase shifter has a non-zero value, while the rest are 0. This represents the switching state matrix of the switching network. Represents the first [unit] at access point m. The switching state matrix of the group of switches represents the first group of switches. In the group root antenna and The switching states of the switches between the phase shifters On each antenna and each RF chain The switching states of all switches between phase shifters are encoded into a single... This is a binary number indicating whether there is a connection between the corresponding phase shifter and the antenna; 'b' represents the number of quantization bits of the phase shifter. Where g represents the number of RF chains, and g represents the number of antenna groups. Indicates time slot The reward for the access point agent m. This represents the achievable rate of user k in time slot t. This represents the power consumption at access point m in time slot t. This represents the penalty factor.

[0056] In addition, the actions of each access point agent are reduced in dimensionality as follows: an action embedding space is introduced using an action embedding method. Consider an action embedding mapping function and the internal strategy of the AP agent Parameterization strategy , can make , Specifically, the policy function Used from action embedding space Sampling an embedded action Then, a preset action is used to embed the mapping function. Map the sampled embedded actions to action sets. One of the actions, namely This allows high-dimensional 0-1 discrete actions to be mapped to a low-dimensional (d-dimensional) continuous action space, where d is the dimension of the action embedding space, thus achieving dimensionality reduction of the action space.

[0057] (3) Each access point performs precoding according to the precoding control scheme output by the access point agent at the current time, and sends the precoded data to the user through the antenna.

[0058] When the access point agent obtains the action to be performed based on its state, this action becomes the pre-coded control scheme output by the access point agent. This pre-coded control scheme sets the link status between the access point and the user, the on / off status of each switch in the switching network, and the setting status of each phase shifter in the phase shifter network. By setting this pre-coded control scheme, the pre-coding of the data stream can be achieved.

[0059] (4) The user calculates its achievable rate based on the signal it receives and feeds it back to the access point agent.

[0060] The signal received by the user can be represented as:

[0061] ,

[0062] in Indicates user The received signal at the location, Indicates sending to the user Data symbols, This represents the channel estimation between access point m and user k in time slot t. and Let these represent the switch matrix of the switch network and the phase shifter matrix of the phase shifter network, respectively. and This represents the digital precoding matrix output by the digital precoder. This represents additive white noise that follows a complex Gaussian distribution.

[0063] According to Shannon's theorem, the reachable rate on the user side is:

[0064]

[0065] In the formula, Indicates channel bandwidth. Indicates the transmission power. This indicates the channel noise power.

[0066] (5) Each access point calculates the power consumption at each access point, drives the access point agent to update the policy, and generates the precoded control scheme for the next moment.

[0067] The power consumption at each access point is calculated using the following formula:

[0068] =

[0069] in, Indicates access point Whether it serves users; if yes, the value is 1; otherwise, it is 0. Indicates the efficiency of the power amplifier. Indicates the transmission power. This indicates the power consumption of baseband signal processing. , , , , , These represent the number of active RF chains, RF chain power consumption, number of active phase shifters, phase shifter power consumption, number of switches, and switch power consumption, respectively. and They represent access points respectively. Fixed power consumption of the fronthaul link between the CPU and the network traffic-related power consumption factor.

[0070] The global energy efficiency can be calculated from the power consumption and achievable speed of all access points. :

[0071] ,

[0072] It can be seen that the reward function in this application is a reflection of global energy efficiency, and the optimal global energy efficiency can be achieved by setting the reward function.

[0073] The specific steps for training and policy updating of the access point agent are as follows:

[0074] Step 1: Initialize the parameters of the Actor network, target Actor network, Critic network, target Critic network, and action embedding network for each access point agent;

[0075] Step 2: Initialize the experience replay pool Batch sampling sample size Reward Discount Factor Learning rate and the soft update factor of the target network. ;

[0076] Step 3: Each access point agent acquires local observations. , as its state And input it into the Actor network to obtain the embedded action. ;

[0077] Step 4: The action embedding space maps the embedded actions to local pre-coded control actions. The access point agent performs this action;

[0078] Step 5: The CPU feeds back the global energy efficiency to the access point agent; the access point agent receives the next state. And calculate the reward State transition and will Store to middle;

[0079] Step 6, from Random sampling Each tuple updates the parameters of the Actor and Critic networks based on the policy gradient and loss function.

[0080] Step 7: Softly update the target network parameters based on the updated Actor and Critic network parameters;

[0081] Step 8: Based on the loss function Update the embedded network parameters.

[0082] In the formula, Indicates loss, This represents the mean square error function. This represents the joint control action of all access point agents. This represents the joint embedding action of all access point agents. express Possible values, , Embed a mapping function for the preset action. It represents probability.

[0083] Appendix Figure 3 In the simulation results, MAPPO (Multi-Agent Proximal Policy Optimization) is a classic distributed multi-agent deep reinforcement learning method, SAC (Soft Actor-Critic) is a centralized deep reinforcement learning method where the CPU is regarded as the sole agent, and AE-MASAC is the method of this invention. It is evident that this invention can maintain high energy efficiency even on larger network scales.

[0084] Appendix Figure 4 The simulation results show that fully connected hybrid precoding, where all RF chains are connected to all antennas, and subconnected hybrid precoding, where each RF chain is connected to a subset of antennas, are two classic hybrid precoding hardware architecture designs. The dynamic adaptive connection structure is the design of this patent. It is evident that this invention achieves the optimal balance between transmission performance and hardware power consumption through a grouping architecture and dynamic connection design.

[0085] Example 2

[0086] This invention provides a high-efficiency dynamic adaptive hybrid precoding module for a cellular-free massive MIMO system. The module is configured at each access point and includes:

[0087] The hybrid precoding architecture supports dynamic adaptive connection of antenna groups, where each phase shifter in the analog precoder is connected to multiple antennas via several switches.

[0088] An access point agent has a state consisting of a channel estimate, a user-side reachable rate, and access point power consumption. Its actions are based on a precoding control scheme using a hybrid precoding architecture. Its reward is the weighted difference between the user-side reachable rate and the access point power consumption. It is used as a game participant with other access point agents to form a Markov game model with the optimization objective of maximizing energy efficiency. The reachable rate is calculated by the user based on the received signal and fed back to the agent.

[0089] The precoding controller is used to precode data according to the precoding control scheme output by the access point agent at the current time, and then send the precoded data to the user.

[0090] The power consumption calculation module is used to calculate the power consumption at the current access point, drive the access point agent to update the policy, and generate the precoded control scheme for the next moment.

[0091] The modules provided in this embodiment of the invention can be used to execute the method provided in Embodiment 1 of the invention, and have the corresponding functions and beneficial effects of executing the method. For details not covered herein, please refer to Embodiment 1, which will not be repeated here.

[0092] It is worth noting that in the embodiments of the above modules, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of the present invention.

[0093] Example 3

[0094] This invention provides a high-efficiency dynamic adaptive hybrid precoding system for cellular-free massive MIMO systems, comprising:

[0095] The access point is equipped with a high-efficiency dynamic adaptive hybrid precoding module in the non-cellular massive MIMO system described in Embodiment 2;

[0096] The user is used to calculate its achievable rate based on the signals it receives and to feed back to the access point agent.

[0097] It should be understood that the embodiments and descriptions above are only the principles, main features and advantages of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the invention, and all such changes and modifications fall within the protection scope of the present invention.

Claims

1. A high-efficiency dynamic adaptive hybrid precoding method for cellular-free massive MIMO systems, characterized in that, Includes the following steps: (1) Construct a hybrid precoding architecture that supports dynamic adaptive connection of antenna groups at each access point, wherein each phase shifter in the analog precoder of the hybrid precoding architecture is connected to multiple antennas through several switches; (2) Construct a Markov game model with the goal of maximizing energy efficiency. In the Markov game model, all agents at the access points are game participants. The state of each agent at the access point includes the channel estimate, the reachable rate on the user side and the power consumption of the access point. The action is a precoding control scheme of a hybrid precoding architecture. The reward is the difference between the reachable rate on the user side and the weighted value of the power consumption of the access point. (3) Each access point performs precoding according to the precoding control scheme output by the access point agent at the current time, and transmits the precoded data to the user through the antenna; (4) The user calculates its achievable rate based on the signal it receives and feeds it back to the access point agent; (5) Each access point calculates the power consumption at each access point, drives the access point agent to update the policy, and generates the precoded control scheme for the next moment. The hybrid precoding architecture specifically includes: a digital precoder connected in sequence, A radio frequency chain and an analog pre-encoder; the analog pre-encoder includes a phase shifter network and a switching network, the phase shifter network including... Each phase shifter, each RF chain connected to A phase shifter; the switching network includes A switch, For the number of access point antennas, The antennas are divided into g groups. Each phase shifter passes through Each switch is connected to Each antenna, linked by each RF chain Each phase shifter is linked to a different antenna and is activated when the RF chain is linked to at least one antenna.

2. The high-efficiency dynamic adaptive hybrid precoding method in a cellular-free massive MIMO system according to claim 1, characterized in that, The settings for each access point agent in step (2) are as follows: state: ; action: ; award: ; in, Indicates time slot The state of the access point agent m, including the channel estimation between access point m and user k in time slot t. Time slot The achievable rate of user k and power consumption at access point m K represents the number of users. Constructing a time slot The states of all access point agents, where M is the number of access point agents. Indicates time slot The action performed by the access point agent m This indicates the connection status between access point m and user k; the value is 1 when a connection is established and 0 otherwise. This represents the phase matrix of the phase shifter at access point m. The element in the nth row and ith column This represents the phase code value of phase shifter n connected to the i-th RF chain. If the i-th RF chain is not connected to phase shifter n... If the i-th RF chain is connected to phase shifter n, then the phase of phase shifter n is encoded into a usable phase set. In A binary number, and assign it to , This represents the switching state matrix of the switching network. Represents the first [unit] at access point m. The switching state matrix of the group of switches represents the first group of switches. In the group root antenna and The switching states of the switches between the phase shifters On each antenna and each RF chain The switching states of all switches between phase shifters are encoded into a single... A binary number, where b represents the number of quantization bits in the phase shifter. Indicates time slot The reward for the access point agent m. This represents the achievable rate of user k in time slot t. This represents the power consumption at access point m in time slot t. Indicates the penalty factor. The number of access point antennas, For the number of phase shifters, For the number of radio frequency chains, The number of phase shifters connected to each RF chain.

3. The high-efficiency dynamic adaptive hybrid precoding method in a cellular-free massive MIMO system according to claim 2, characterized in that, The actions of each access point agent are reduced in dimensionality as follows: An action embedding method is used to introduce an action embedding space. Strategy function From motion embedding space Sampling an embedded action ,Right now Then, a preset action is used to embed the mapping function. Map the sampled embedded actions to action sets. One of the actions, namely This allows high-dimensional 0-1 discrete actions to be mapped to a low-dimensional continuous action space.

4. The high-efficiency dynamic adaptive hybrid precoding method in a cellular-free massive MIMO system according to claim 1, characterized in that, The achievable rate mentioned in step (4) is calculated using the following formula: ， In the formula, This represents the achievable rate of user k in time slot t. Indicates channel bandwidth. , These represent the connection status between access point m and users k and k', respectively. The value is 1 when a connection is established and 0 otherwise. This represents the channel estimation between access point m and user k in time slot t. and Let these represent the switch state matrix of the switch network and the phase shifter matrix of the phase shifter network, respectively. When the phase shifter n connected to the i-th RF chain has an antenna connected to it via a switch, the phase shifter is in an active state. At this time, the phase shifter matrix... The element in the nth row and ith column When the phase shifter is inactive, this element is 0. The phase matrix of the phase shifter at access point m The element in the nth row and ith column, and This represents the digital precoding matrix output by the digital precoder. Indicates the number of data streams transmitted in parallel. Indicates the transmission power. This indicates the channel noise power.

5. The high-efficiency dynamic adaptive hybrid precoding method in a cellular-free massive MIMO system according to claim 1, characterized in that, The power consumption at each access point in step (5) is calculated using the following formula: = ， in, This represents the power consumption at access point m in time slot t. This represents the achievable rate for user k in time slot t, where K represents the number of users. Indicates access point Whether it serves users; if yes, the value is 1; otherwise, it is 0. Indicates the efficiency of the power amplifier. Indicates the transmission power. This indicates the power consumption of baseband signal processing. , , , , , These represent the number of active RF chains, RF chain power consumption, number of active phase shifters, phase shifter power consumption, number of switches, and switch power consumption, respectively. and These represent the fixed power consumption of the fronthaul link between access point m and the CPU, and the power consumption coefficient related to network traffic, respectively.

6. The high-efficiency dynamic adaptive hybrid precoding method in a cellular-free massive MIMO system according to claim 1, characterized in that, The loss function used by the driving access point agent to update the policy in step (5) is: ， In the formula, Indicates loss, This represents the mean square error function. This represents the joint control action of all access point agents. This represents the joint embedding action of all access point agents. express Possible values, , Embed a mapping function for the preset action. This represents the probability of a condition being met. , These represent the states of time slots t and t+1, respectively.

7. A high-efficiency dynamic adaptive hybrid precoding module for a cellular-free massive MIMO system, characterized in that, The module is installed at each access point and includes: The hybrid precoding architecture supports dynamic adaptive connection of antenna groups, where each phase shifter in the analog precoder is connected to multiple antennas via several switches. An access point agent has a state consisting of a channel estimate, a user-side reachable rate, and access point power consumption. Its actions are based on a precoding control scheme using a hybrid precoding architecture. Its reward is the weighted difference between the user-side reachable rate and the access point power consumption. It is used as a game participant with other access point agents to form a Markov game model with the optimization objective of maximizing energy efficiency. The reachable rate is calculated by the user based on the received signal and fed back to the agent. The precoding controller is used to precode data according to the precoding control scheme output by the access point agent at the current time, and then send the precoded data to the user. The power consumption calculation module is used to calculate the power consumption at the current access point, drive the access point agent to update the policy, and generate the precoded control scheme for the next moment. The hybrid precoding architecture specifically includes: a digital precoder connected in sequence, A radio frequency chain and an analog pre-encoder; the analog pre-encoder includes a phase shifter network and a switching network, the phase shifter network including... Each phase shifter, each RF chain connected to A phase shifter; the switching network includes A switch, For the number of access point antennas, The antennas are divided into g groups. Each phase shifter passes through Each switch is connected to Each antenna, linked by each RF chain Each phase shifter is linked to a different antenna and is activated when the RF chain is linked to at least one antenna.

8. A high-efficiency dynamic adaptive hybrid precoding system for a cellular-free massive MIMO system, characterized in that, include: The access point is equipped with a high-efficiency dynamic adaptive hybrid precoding module in the non-cellular massive MIMO system as described in claim 7; The user is used to calculate its achievable rate based on the signals it receives and to feed back to the access point agent.

Images