Beam forming and unit position joint optimization method of mobile RIS enhanced wireless communication system

Abstract

The invention provides a beam forming and unit position joint optimization method for a movable RIS enhanced wireless communication system, and aims to improve the system performance by optimizing a base station end active beam forming vector, an RIS end passive beam forming vector and a unit position. The method comprises the following steps: constructing an optimization problem model with system sum rate maximization as a target; constructing constraint conditions of base station transmitting power, RIS phase shift constraint, RIS unit position and distance between RIS units; the deep reinforcement learning algorithm is adopted to solve the optimization problem, the optimal strategy can be learned in a self-adaptive mode, and the method can well adapt to the complex and changeable wireless communication environment.

Description

Technical Field

[0001] The invention belongs to the technical field of wireless communication, and in particular relates to a beamforming and unit position joint optimization method of a mobile RIS enhanced wireless communication system. Background Art

[0002] Reconfigurable Intelligent Surface (RIS) is an emerging technology in the field of communications. The RIS surface is composed of a large number of small, low-cost, programmable components. Its core is to improve the quality of signal transmission by dynamically adjusting the signal properties of each component on its surface, such as phase and amplitude. Unlike traditional relay stations, the characteristics of mobile RIS are that its components can be flexibly moved and reconfigured, which means that they can change their arrangement and state as needed to optimize the wireless propagation environment. This flexibility enables mobile RIS to respond quickly in various application scenarios. For example, in large-scale events or emergency rescue scenarios, it can dynamically enhance the signal strength in local areas, improve communication quality and reliability, and optimize wireless link performance in real time according to environmental changes, which is crucial to achieving the high performance goals of 5G and future communication systems.

[0003] When a reconfigurable smart surface (RIS) has a large number of transmitting units, each system parameter update will trigger a large amount of computing requirements, which will lead to extremely high computational complexity. Although non-convex optimization can solve certain optimization problems, the optimization variables are often high-dimensional reflection beamforming vectors, which makes the processing complexity extremely high. In contrast, deep reinforcement learning has shown superior performance in processing massive data, nonlinear non-convex problems, and high-computational intensity tasks. Since the intelligent agent of deep reinforcement learning can learn and make optimal decisions through interaction with the environment, including transmission power allocation and beamforming design, it is widely used by researchers to solve resource optimization problems in communication systems. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a beamforming and unit position joint optimization method for a mobile RIS enhanced wireless communication system.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a beamforming and unit position joint optimization method for a mobile RIS enhanced wireless communication system, comprising:

[0006] Step S1, constructing an optimization problem model with system and rate maximization as the optimization goal;

[0007] Step S2, setting four constraints for the optimization problem model constructed in step S1; the constraints include: base station transmission power constraint, RIS unit phase shift constraint, RIS unit position constraint, and distance constraint between two RIS units;

[0008] Step S3, solving the optimization problem after setting the constraint conditions, and obtaining an optimization solution that maximizes the system energy efficiency.

[0009] The present invention further includes the following preferred embodiments:

[0010] The mobile RIS is composed of M mobile units, where M = M row ×M col , M row Indicates the number of rows of RIS units, M col represents the number of columns of RIS units. The phase shift and position of each RIS unit are controlled by the coefficient adjustment module and the component positioning module of the unit respectively. The central controller sends instructions according to the current communication environment and optimization goals. The phase shift of each RIS unit is controlled by the coefficient adjustment module, and the position is adjusted by the component positioning module to control the horizontal and vertical slide rails, so that each RIS unit can move within a certain range. Without loss of generality, assume that the mobile RIS enhanced wireless communication system includes a t A base station with 1 antenna, a mobile RIS and K single-antenna users, such as Figure 1 As shown. is the passive beamforming vector at the RIS end, is the phase offset matrix of RIS, θ m is the phase offset of the mth reflection unit of RIS, diag(·) represents the conversion of the vector into a diagonal matrix, e j· Represents the exponential form of a complex number, m=1, 2, …M. is the position of the mth RIS unit, and The horizontal and vertical coordinate values ​​of the Cartesian coordinate system, m=1, 2,…M.

[0011] The optimization problem model is as follows:

[0012]

[0013] Among them, W k represents the beamforming matrix of user k, is the position vector of the movable unit of RIS, is the achievable data rate of user k, E k (p ris )=diag(g k (p ris )G(p ris)) is the cascade channel from the base station to user k. In addition, G(p ris ) is the channel between the slave base station and RIS, g k (p ris ) is the channel between RIS and user k, σ 2 is Gaussian white noise.

[0014] The system communication channel model further comprises:

[0015] Step S11: Channel model between the base station and the RIS;

[0016] Step S12: Channel model between RIS and user.

[0017] The constraints of the optimization problem include:

[0018] Constraint 1:

[0019] Constraint 2: θ m ∈[0,2π)

[0020] Constraint 3:

[0021] Constraint 4:

[0022] in, m=1,2,…M。P m is the movable area of ​​the mth RIS unit, d min is the minimum distance that needs to be satisfied between RIS units, constraint 1 is the maximum transmission power constraint of the base station, constraint 2 is the phase shift constraint of the RIS unit, constraint 3 is the position constraint of the RIS unit, and constraint 4 is the distance constraint between RIS units.

[0023] The method for solving the optimization problem model after setting the constraint conditions includes:

[0024] Step S31: Initialize the training actor network, critic network, and target actor and critic network parameters;

[0025] Step S32: Obtaining the initial state S (0) , action and next moment state;

[0026] Step S33: (t) , a (t) , r (t) , S (t+1) ) is stored in the experience pool;

[0027] Step S34: Construct the loss function for training the critic network

[0028] Step S35: Randomly sample n samples from the experience pool and calculate the critic network Q function

[0029] Step S36: Perform stochastic gradient descent in the training critic network, the target critic network, and the training actor network to obtain and

[0030] Step S37: Update the critic network and actor network of the training and target respectively;

[0031] Step S38: taking the new state as the input of DNN;

[0032] Step S39: Repeat steps S32 to S38 until the specified number of rounds is completed.

[0033] The step S37 further comprises:

[0034] The update formula for training the critic network is:

[0035]

[0036] The update formula for training the actor network is

[0037]

[0038] The update formula of the target critic network and the target actor network is:

[0039] BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a system architecture diagram of the beamforming and unit position joint optimization method of the mobile RIS enhanced wireless communication system in this specification.

[0041] Figure 2 This is an illustration of deep reinforcement learning. DETAILED DESCRIPTION

[0042] In order to overcome the shortcomings of the prior art, the present invention provides a joint optimization method for beamforming and unit position of a mobile RIS enhanced wireless communication system. The present invention utilizes a mobile RIS to enhance user communication performance, so that users distributed around the RIS can utilize the RIS to enhance the received signal, and improves the system and rate by optimizing the base station active beamforming vector and RIS phase shift.

[0043] The present invention provides a beamforming and unit position joint optimization method for a mobile RIS enhanced wireless communication system. The specific implementation is as follows:

[0044] Step S1, constructing an optimization problem model with system and rate maximization as the optimization goal;

[0045] The mobile RIS consists of M mobile units, where M = M row ×M col , M row Indicates the number of rows of RIS units, M col represents the number of columns of RIS units. The phase shift and position of each RIS unit are controlled by the coefficient adjustment module and the component positioning module of the unit respectively. The central controller sends instructions according to the current communication environment and optimization goals. The phase shift of each RIS unit is controlled by the coefficient adjustment module, and the position is adjusted by the component positioning module to control the horizontal and vertical slide rails, so that each RIS unit can move within a certain range. Without loss of generality, assume that the mobile RIS enhanced wireless communication system includes a t A base station with 1 antenna, a mobile RIS and K single-antenna users, such as Figure 1 As shown. is the passive beamforming vector at the RIS end, is the phase offset matrix of RIS, θ m is the phase offset of the mth reflection unit of RIS, dag(·) represents the conversion of the vector into a diagonal matrix, e j· Represents the exponential form of a complex number, m=1, 2,…M. is the position of the mth RIS unit, and The horizontal and vertical coordinate values ​​of the Cartesian coordinate system, m=1, 2,…M.

[0046] The communication channel model described in step S1 further includes:

[0047] Step S11: Channel model between the base station and the RIS;

[0048] Step S12: Channel model between RIS and user.

[0049] S11: The channel model between the base station and the RIS further includes:

[0050] and denote the total number of transmit and receive channel paths from BS to RIS. At the base station, the elevation angle and azimuth angle of the i-th transmit path are respectively and indicates that, and Then the field response vector at the base station is in, Base station antenna location To the origin O bs The signal propagation distance of the i-th transmission channel path between. Therefore, N t The field response matrix corresponding to the root transmitting antenna is in is the position of the nth antenna, n = 1, 2, ... N t .

[0051] Similarly, at the RIS, the elevation angle and azimuth angle of the jth receiving path are expressed as and indicates that, and Then the field response vector at RIS is in, RIS unit location To the origin O ris The signal propagation distance of the jth receiving channel path between . Therefore, the corresponding field response matrix of the M mobile units is

[0052] Furthermore, the path response matrix Indicates from O bs to ris The response between all transmit and receive channel paths. In particular, The jth row and ith column of represents the response coefficient between the i-th transmission path and the j-th receiving path. Finally, the channel between BS and RIS can be expressed as:

[0053] The step S12: channel model between RIS and user k further comprises:

[0054] The total number of transmit and receive channel paths from RIS to user k is denoted by and For the i-th transmission path, its elevation angle and azimuth angle are expressed as and RIS unit location To the origin O ris The signal propagation distance of the jth receiving channel path between Therefore, the path response matrix corresponding to the RIS reflection side is obtained as follows: The path response matrix at RIS is

[0055] also, Indicates RIS origin O ris To the location of user k The path response matrix between. Among them, [γk ] j，i is the channel response coefficient between the i-th reflection path and the j-th reception path. Therefore, the channel between RIS and user k is used as the RIS element positioning vector p ris It can be expressed as Among them, 1 is the dimension A column vector of all ones.

[0056] Finally, with the help of RIS, the equivalent channel from BS to user k can be expressed as

[0057]

[0058] According to the above analysis, the signal received by user k is:

[0059]

[0060] Among them, w k represents the beamforming matrix of user k, is the position vector of the movable unit of RIS, S k is the signal of user k, n k represents the additive white Gaussian noise at user k.

[0061] The achievable data rate of user k can be expressed as:

[0062]

[0063] Among them, E k (p ris )=diag(g k (p ris )G(p ris )) is the cascade channel from the base station to user k. In addition, G(p ris ) is the channel between the slave base station and RIS, g k (p ris ) is the channel between RIS and user k, σ 2 is Gaussian white noise.

[0064] The optimization problem model described in step S1 is as follows:

[0065]

[0066] Step S2, setting four constraints for the optimization problem model constructed in step S1; the constraints include: base station transmission power constraint, RIS unit phase shift constraint, RIS unit position constraint, and distance constraint between two RIS units;

[0067] The constraints of the optimization problem include:

[0068] Constraint 1:

[0069] Constraint 2: θ m ∈[0,2π)

[0070] Constraint 3:

[0071] Constraint 4:

[0072] in, m=1,2,…M。P m is the movable area of ​​the mth RIS unit, d min is the minimum distance that needs to be satisfied between RIS units, constraint 1 is the maximum transmission power constraint of the base station, constraint 2 is the phase shift constraint of the RIS unit, constraint 3 is the position constraint of the RIS unit, and constraint 4 is the distance constraint between RIS units.

[0073] Step S3, solving the optimization problem after setting the constraint conditions, and obtaining an optimization solution that maximizes the system energy efficiency.

[0074] The deep Q network in the embodiment of the present invention is described below.

[0075] An illustration of deep reinforcement learning Figure 2 As shown, it mainly includes the experience pool, the training actor network, the target actor network, the training critic network, the target critic network, and the intelligent agents that act for different states. In this embodiment, the state State(t) is defined as the transmission power at the tth step, the receiving power of the user at the tth step, the action taken at the t-1th step, and the channels from the BS to the RIS and from the RIS to the user. The action action is defined as the beamforming matrix, the phase shift matrix and the position of the RIS unit; the sum rate is defined as the reward Reward. Constraints 1 and 2 are processed by the normalization method, and constraints 3 and 4 are hard constraints, so that they must satisfy constraints 3 and 4.

[0076] Step S31: Initialize the training actor network, critic network, and target actor and critic network parameters;

[0077] Set up an experience pool to store information such as states, actions, rewards, etc. in the experience; initialize the trained actor network, critic network, and target actor and critic network, set the parameters of the training and target actor networks to be the same, and set the parameters of their critic networks to be the same. Initialize the beamforming matrix and phase shift matrix to the identity matrix. Specify the number of rounds and the number of steps in each round, and set the noise process to explore the action space.

[0078] Step S32: Obtain the initial state, action and next moment state;

[0079] Collect and preprocess the channels from BS to RIS and from RIS to user. The state corresponding to time step t = 0 is denoted as S (0) , and get action a from the actor network (t) , under a given action, the new state reached by the system is denoted as S (t+1) , observe that after taking this action, the immediate reward obtained is recorded as r (t+1) .

[0080] Step S33: (t) , a (t) , r (t) , S (t+1) ) is stored in the experience pool;

[0081] Step S34: constructing a loss function for training the critic network;

[0082] Step S35: randomly sample a small batch of samples with a capacity of nn from the experience pool for training; according to Calculate the critic network Q function;

[0083] Step S36: Perform stochastic gradient descent in the training critic network, the target critic network, and the training actor network to obtain and

[0084] Step S37: Update the critic network and actor network of the training and target respectively;

[0085] The update formula for training the critic network is:

[0086]

[0087] The update formula for training the actor network is

[0088]

[0089] The update formula of the target critic network and the target actor network is:

[0090]

[0091] Step S38: Set the new state S (t+1) As input to DNN;

[0092] Step S39: Repeat steps S32 to S38 until the specified number of rounds is completed.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A beamforming and unit position joint optimization method for a mobile RIS enhanced wireless communication system, characterized in that: The following steps are involved: Step S1, constructing an optimization problem model with system and rate maximization as the optimization goal; Step S2, setting four constraints for the optimization problem model constructed in step S1; the constraints include: base station transmission power constraint, RIS unit phase shift constraint, RIS unit position constraint, and distance constraint between two RIS units; Step S3, solving the optimization problem after setting the constraint conditions, and obtaining an optimization solution that maximizes the system energy efficiency.

2. The mobile RIS in the beamforming and unit position joint optimization method for a mobile RIS-enhanced wireless communication system according to claim 1, further comprising: The mobile RIS consists of M mobile units, where M = M row ×M col , M row Indicates the number of rows of RIS units, M col represents the number of columns of RIS units. The phase shift and position of each RIS unit are controlled by the coefficient adjustment module and the component positioning module of the unit respectively. The central controller sends instructions according to the current communication environment and optimization goals. The phase shift of each RIS unit is controlled by the coefficient adjustment module, and the position is adjusted by the component positioning module to control the horizontal and vertical slide rails, so that each RIS unit can move within a certain range. Without loss of generality, assume that the mobile RIS enhanced wireless communication system includes a t As shown in Figure 1, there are a base station with 1 antenna, a mobile RIS and K single-antenna users. is the passive beamforming vector at the RIS end, is the phase offset matrix of RIS, θ m is the phase offset of the mth reflection unit of RIS, dag(·) represents the conversion of the vector into a diagonal matrix, Represents the exponential form of a complex number, m=1, 2, …M. is the position of the mth RIS unit, and The horizontal and vertical coordinate values ​​of the Cartesian coordinate system, m=1, 2,…M.

3. The beamforming and unit position joint optimization method of a mobile RIS enhanced wireless communication system according to claim 1, characterized in that: The optimization problem model is as follows: Among them, W k represents the beamforming matrix of user k, is the position vector of the movable unit of RIS, is the achievable data rate of user k, E k (p ris )=diag(g k (p ris )G(p ris )) is the cascade channel from the base station to user k. In addition, G(p ris ) is the channel between the slave base station and RIS, g k (p ris ) is the channel between RIS and user k, σ 2 is Gaussian white noise.

4. The beamforming and unit position joint optimization method of a mobile RIS enhanced wireless communication system according to claim 1, characterized in that: The constraints of the optimization problem include: Constraint 1: Constraint 2: θ m ∈[0,2π) Constraint 3: Constraint 4: in, m=1,2,…M。P m is the movable area of ​​the mth RIS unit, d min is the minimum distance that needs to be satisfied between RIS units, constraint 1 is the maximum transmission power constraint of the base station, constraint 2 is the phase shift constraint of the RIS unit, constraint 3 is the position constraint of the RIS unit, and constraint 4 is the distance constraint between RIS units.

5. The beamforming and unit position joint optimization method of a mobile RIS enhanced wireless communication system according to claim 5, characterized in that: The optimization problem model after setting the constraint conditions is solved, and the method includes: Step S31: Initialize the training actor network, critic network, and target actor and critic network parameters; Step S32: Obtaining the initial state S (0) , action and next moment state; Step S33: (t) , a (t) , r (t) , S (t+1) ) is stored in the experience pool; Step S34: Construct the loss function for training the critic network Step S35: Randomly sample n samples from the experience pool and calculate the critic network Q function Step S36: Perform stochastic gradient descent in the training critic network, the target critic network, and the training actor network to obtain and Step S37: Update the critic network and actor network of the training and target respectively; Step S38: taking the new state as the input of DNN; Step S39: Repeat steps S32 to S38 until the specified number of rounds is completed.

6. The beamforming and unit position joint optimization method of a mobile RIS enhanced wireless communication system according to claim 5, characterized in that: The step S37 further comprises: The update formula for training the critic network is: The update formula for training the actor network is The update formula of the target critic network and the target actor network is:

Images