A RIS-assisted massive MIMO deep learning approach

Abstract

The present invention discloses a RIS-assisted large-scale MIMO deep learning method. First, RIS is introduced for uplink unauthorized transmission, thereby cascading to form a communication channel link between a user and RIS, and between RIS and a base station, thereby effectively controlling the quality of wireless signal transmission; then, a deep learning system under the Turbo message transmission idea is established, and RIS-assisted cascade channel state information is learned and estimated through a large amount of training data; finally, the result of RIS-assisted large-scale machine-type communication channel estimation is statistically analyzed, and the parsing performance of each layer of the message transmission network is analyzed through state evolution, and the convergence result of the RIS-assisted large-scale machine-type uplink communication system is obtained. The method of the present invention has strong flexibility, excellent performance and low complexity, and can achieve better channel estimation performance with fewer learning layers.

Description

Technical Field

[0001] The present invention relates to the field of mobile communication technology, and in particular to a RIS-assisted large-scale MIMO deep learning method. Background Art

[0002] MIMO (Multiple-input Multiple-output) technology has been successfully applied to the 5G standard. This technology has significant performance gains in both spectrum and energy efficiency, ensuring that base stations equipped with a large number of antennas can serve a large number of devices in the system and bring rich spatial freedom.

[0003] For 6G mobile communication systems, the types of services and application scope supported will be more extensive. At the same time, there is a huge amount of data to be processed. The industry has begun to study and deploy new large-dimensional antenna array architectures. This large-dimensional antenna array can generally be deployed together with large infrastructure, such as building walls in airports, high-speed rail stations or stadiums, and electromagnetic surfaces in large smart surfaces. By controlling the phase shift unit of the reconfigurable intelligent metasurface (RIS) and performing beamforming, the signal propagation strength can be enhanced, interference signals can be eliminated, and green large-scale coverage enhancement can be achieved. In the prior art, for large-scale IoT (Internet of Things) systems, especially for machine-type communication systems equipped with large-scale MIMO, how to control and improve the harsh channel propagation environment through RIS to improve system performance is still an open problem.

[0004] On this basis, the performance of IoT systems can be improved by combining deep learning and RIS technology. This method uses the idea of ​​compressed sensing to apply the deep learning framework to large-scale IoT systems, enhance channel estimation, and improve data recovery performance. However, since the base station is equipped with a large-scale antenna array and the number of online devices in the entire system is huge, how to reduce the computational burden of the system is also an important aspect that needs to be studied. Summary of the invention

[0005] The problem to be solved by the present invention is to provide a RIS-assisted large-scale MIMO deep learning method, by establishing a RIS-assisted large-scale IoT uplink transmission system, Gaussian approximation of the RIS cascade link, and according to the model-driven idea, a deep-learnable message passing network architecture is proposed, and a generative adversarial network and a denoising convolutional neural network are integrated into the designed message passing network to perform channel estimation for large-scale machine-type communications.

[0006] The present invention adopts the following technical solution: a RIS-assisted large-scale MIMO deep learning method, comprising the following steps:

[0007] Step S1, establishing a RIS-assisted large-scale machine type uplink communication system, and approximating a RIS-assisted cascade channel through a Gaussian random distribution;

[0008] Step S2: Based on the Turbo message passing method, a RIS-assisted large-scale machine type uplink communication system deep learning network is constructed and layered, a Bayesian multi-dimensional measurement vector is estimated for each layer, and the loop is iterated until the final convergence to estimate the channel posterior information on the base station antenna;

[0009] Step S3, converting the channel posterior information estimation module in S2 into a fusion module of a joint generative adversarial network and a denoising convolutional neural network, learning the RIS-assisted cascade channel state information, and obtaining the cascade channel estimation result after denoising;

[0010] Step S4: Analyze the parsing performance of each layer of the deep learning network of the RIS-assisted large-scale machine type uplink communication system through state evolution, and obtain the convergence result of the RIS-assisted large-scale machine type uplink communication system.

[0011] Specifically, in the RIS-assisted large-scale machine type uplink communication system, there are channel links between the base station and the device, the channel link between the device and the RIS, and the channel link between the RIS and the base station. A base station configured with M antennas serves N single-antenna online devices, and the communication between the base station and the device is assisted by a RIS configured with K reflection elements.

[0012] Furthermore, the pilot signal in the RIS-assisted large-scale machine type uplink communication system is W, which is expressed as follows:

[0013] ;

[0014] in, represents the pilot sequence set of N single-antenna online devices with a length of B, , indicating the The pilot sequence of length B for each device, .

[0015] The received signal Y at the base station is expressed as follows:

[0016] ;

[0017] in, , , , Used to indicate the The received signal on the root antenna; Indicates A direct link between a device and a base station. The superscript T indicates transposition. ; It is the channel link between the device and RIS. ; It is the channel link between RIS and the base station. ; represent the complex domain signal sets of B, K and M dimensions respectively, is the phase shift matrix of RIS;

[0018] use Representing the cascade channel, the expression of the received signal is rewritten as:

[0019] ;

[0020] in, is complex Gaussian white noise with a mean of , the variance is .

[0021] Further, in step S1, based on the central limit theorem, the cascaded channel The approximation is a Bernoulli Gaussian random distribution, and the corresponding probability distribution P is expressed as follows:

[0022] ;

[0023] in, is the active probability of the device, is the Dirac function, Indicates Channel status information of each device, , are path loss parameters, corresponding to the channel link between the device and RIS, the channel link between RIS and the base station, and the direct channel link between the device and the base station. The function represents the complex Gaussian probability density function and I represents the identity matrix.

[0024] Specifically, in step S2, a message passing network is constructed to divide the network of the RIS-assisted large-scale machine type uplink communication system into Layer, for the Layer, the input is the received signal at the base station , the pilot signal is , the output is the estimated value of the channel .

[0025] Furthermore, based on the multidimensional measurement vector rule, the generative adversarial network and denoising convolutional neural network methods are integrated into the message passing network, and each iteration of channel estimation is used as each layer in the network. layer, the initial input is The output of the layer is iterated and the base station is The external information on the root antenna is estimated.

[0026] Specifically, in step S3, the estimation module of the channel posterior information is transformed into a fusion module of a joint generative adversarial network and a denoising convolutional neural network, the generative adversarial network is used to generate a generated sample closest to the true value, and the additive white Gaussian noise training model is used as the input of the denoising convolutional neural network;

[0027] The denoising convolutional neural network consists of three layers. The first layer consists of 64 The second layer consists of 64 filters and rectified linear units. The third layer consists of a single The filter is used to reconstruct the signal;

[0028] The output result of the fusion module is the noise value contained in the channel. The noise value is subtracted from the input signal to obtain the channel estimation result after noise reduction processing.

[0029] Specifically, in step S4, the state evolution is performed on each layer of the RIS-assisted large-scale machine type uplink communication system network by introducing the measurement coefficient and mean square error function of the pilot matrix; using and Indicates The two transfer functions of the layer are used for state evolution. layer, through and The continuous updates between them obtain the convergence results of RIS-assisted large-scale machine-type uplink communication systems.

[0030] The technical solution of the present invention also provides: an electronic device, comprising:

[0031] one or more processors;

[0032] a storage device having one or more programs stored thereon;

[0033] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the above-mentioned RIS-assisted massive MIMO deep learning methods.

[0034] The technical solution of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in any of the above-mentioned RIS-assisted large-scale MIMO deep learning methods are implemented.

[0035] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0036] 1. The RIS-assisted massive MIMO deep learning method of the present invention is highly flexible and can achieve satisfactory performance with a smaller number of learning layers, and has satisfactory practical value.

[0037] 2. The RIS-assisted massive MIMO deep learning method of the present invention has excellent performance and low complexity, and the analytical results can efficiently predict the accuracy of the simulation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flowchart of the steps of the RIS-assisted massive MIMO deep learning method of the present invention;

[0039] Figure 2 Schematic diagram of a massive MIMO system assisted by RIS of the present invention;

[0040] Figure 3 This is a diagram of the deep learning architecture for channel estimation of the present invention;

[0041] Figure 4 A diagram of the module of the present invention for integrating the generative adversarial network and the denoising convolutional neural network;

[0042] Figure 5 is a performance diagram between the signal-to-noise ratio and the mean square error of the deep learning method of the present invention;

[0043] Figure 6 This is a performance diagram between the number of RIS units and the mean square error of the deep learning method of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the application is further elaborated in detail below in conjunction with the accompanying drawings. The described embodiments are only a part of the embodiments involved in the present invention. All non-innovative embodiments of other researchers in the field on this embodiment belong to the protection scope of the present invention. At the same time, for the step numbering in the embodiment of the present invention, it is only set for the convenience of explanation, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0045] The present invention provides a RIS-assisted massive MIMO deep learning method, which uses a DFT matrix that does not require storage as a measurement matrix, estimates channel state information through Bayesian posterior estimation and external soft message transmission, and obtains satisfactory channel estimation performance, such as Figure 1 As shown, the following steps are included:

[0046] Step S1, establishing a RIS-assisted large-scale machine type uplink communication system, and approximating a RIS-assisted cascade channel through a Gaussian random distribution;

[0047] Step S2: Based on the Turbo message passing method, a RIS-assisted large-scale machine type uplink communication system deep learning network is constructed and layered, a Bayesian multi-dimensional measurement vector is estimated for each layer, and the loop is iterated until the final convergence to estimate the channel posterior information on the base station antenna;

[0048] Step S3, converting the channel posterior information estimation module in S2 into a fusion module of a joint generative adversarial network and a denoising convolutional neural network, learning the RIS-assisted cascade channel state information, and obtaining the cascade channel estimation result after denoising;

[0049] Step S4: Analyze the parsing performance of each layer of the deep learning network of the RIS-assisted large-scale machine type uplink communication system through state evolution, and obtain the convergence result of the RIS-assisted large-scale machine type uplink communication system.

[0050] It should be noted that, under the background of deep learning, the present invention takes into account the intelligent implementation and hardware overhead of large-scale arrays of next-generation mobile communication systems. On the basis of RIS assistance, channel estimation is performed by integrating message passing, generative adversarial networks and denoising convolutional neural networks. Compared with some other current methods, satisfactory channel estimation performance can be obtained, which is suitable for RIS-assisted large-scale MIMO machine-type communication systems.

[0051] In one embodiment of the present invention, a RIS-assisted massive MIMO deep learning method is as follows:

[0052] The first step is to establish a RIS-assisted large-scale machine-type uplink communication system model, such as Figure 2 As shown, a base station configured with M antennas serves N single-antenna online devices in the system, and the communication between the two is assisted by a RIS configured with K reflection elements.

[0053] In this system, there are channel links between the base station and the device, between the device and the RIS, and between the RIS and the base station. The pilot signal is:

[0054] ;

[0055] in, represents the pilot sequence set of N single-antenna online devices with a length of B, , indicating the The pilot sequence of length B for each device, ;

[0056] The received signal at the base station is:

[0057] ;

[0058] in, , , , Used to indicate the The received signal on the root antenna; Indicates A direct link between a device and a base station. The superscript T indicates transposition. ; It is the channel link between the device and RIS. ; It is the channel link between RIS and the base station. ; Respectively represent the complex domain signal sets of B dimension, K dimension and M dimension;

[0059] The phase shift matrix of RIS is expressed as:

[0060] ;

[0061] in, It is the first The magnitude of the elements, For the Phase shift, ; j represents the imaginary part of the complex domain signal; Functions represent natural constants Bottom , The function is used to construct a diagonal matrix.

[0062] Furthermore, the expression of the signal received by the base station can be rewritten as:

[0063] ;

[0064] in, is complex Gaussian white noise with a mean of , the variance is .

[0065] According to the central limit theorem, the cascade channel It can be approximated as a Bernoulli Gaussian random distribution, and the corresponding probability distribution expression is:

[0066] ;

[0067] in, is the active probability of any device, is the Dirac function, , , and are path loss parameters, corresponding to the channel link between the device and RIS, the channel link between RIS and BS, and the direct channel link between the device and BS. The function represents the complex Gaussian probability distribution function, and I represents the identity matrix.

[0068] The second step is to build a RIS-assisted large-scale machine-type uplink communication system deep learning network based on the Turbo message passing method, and divide the network into Layer, such as Figure 3 shown.

[0069] For Layer, the input is the received signal at the base station , the pilot sequence is , the output is the estimated value of the channel Based on the multi-dimensional measurement vector rule, each iteration of channel estimation can be regarded as each layer in the entire network.

[0070] Specifically, for the The initialization input is the The output of the layer is:

[0071] ;

[0072] Wherein, the subscripts A and B represent modules A and B, wherein module A performs a posteriori estimation on the vector and transmits the external expectation and the external variance, and module B performs a posteriori estimation on the signal and estimates the channel information of each submodule; the superscript pri represents the prior value, and the superscript ext represents the external value. represents the variance, represents the expected function, Respectively represent Layer A module On the antenna The prior expectation and prior variance of Respectively represent Layer B module On the antenna The external expectation and external variance of .

[0073] The loop iteration process is:

[0074] S2.1. Order , Respectively represent The product signal of the layer channel and the pilot signal and the channel state information are used, and the superscript post represents the posterior value to estimate the base station end On the antenna The posterior information and External information, for each antenna In terms of calculation and , respectively representing the Tier In the A module of the antenna and The posterior expectation and posterior variance of The posterior expectation and posterior variance of ;

[0075] S2.2. Joint channel estimation is performed for all online devices in the RIS-assisted large-scale machine type uplink communication system. When The posterior expectation and posterior variance are and , the formula is:

[0076] ;

[0077] ;

[0078] in, represents the expression to be solved, represents the expression for solving the variance, represents diagonalization; Indicates Tier The a priori expectation of the channel of each device in module B; Indicates Tier The channel prior variance of each device in module B;

[0079] S2.3, for the base station The external information on the root antenna is estimated. When , we can calculate and , respectively representing the Tier In the B module of the antenna The posterior expectation and posterior variance of Tier In the B module of the antenna The prior expectation and prior variance of .

[0080] The third step is to transform the estimation module of the channel posterior information into a fusion module of the joint generative adversarial network and the denoising convolutional neural network, such as Figure 4 shown.

[0081] First, a generative adversarial network is used to generate samples that are closest to the true value. , and the new additive white Gaussian noise training model. For the simplicity of formula expression, the superscript is omitted. , the generated sample is expressed as:

[0082] ;

[0083] in, Represent the channel and noise closest to the real generated sample respectively. Then, the generated sample As the input of the denoising convolutional neural network, the denoising convolutional neural network consists of three layers. The first one is 64 The first type consists of 64 filters and rectified linear units (ReLU). The filter, batch normalization and ReLU are composed of a single The filter is used to reconstruct the signal.

[0084] Finally, the final output result of the denoising convolutional neural network is the noise contained in the channel. By subtracting the noise value from the input signal, the channel estimation result after denoising can be obtained.

[0085] Step 4: For each layer of the RIS-assisted large-scale machine-type uplink communication system deep learning network established in this solution, the state evolution formula is:

[0086] ;

[0087] ;

[0088] ;

[0089] in, The initialization can be directly based on the cascade channel Perform calculations, are the measurement coefficients of the pilot matrix, For the The mean square error (MSE) function of the channel estimation of each device.

[0090] and Indicates The layer is used for the transfer function of state evolution. layer, through and They are continuously updated with each other until the final convergence result is obtained.

[0091] The method of this embodiment is used to perform simulation tests on different signal-to-noise ratios, and the MSE results under different signal-to-noise ratios are compared. Figure 5 As shown, direct path is used to represent direct link, and RIS path is used to represent RIS-assisted link, specifically including: approximate message passing scheme of direct path without deep hierarchical learning, approximate message passing network scheme of direct path with deep hierarchical learning, Turbo message passing scheme of direct path without deep hierarchical learning, Turbo message passing network scheme of direct path with deep hierarchical learning, RIS-assisted Turbo message passing scheme of direct path+RIS path without deep hierarchical learning, RIS-assisted Turbo message passing network scheme of direct path+RIS path with deep hierarchical learning, RIS-assisted Turbo message passing scheme of RIS path without deep hierarchical learning, and RIS-assisted Turbo message passing network scheme of RIS path with deep hierarchical learning.

[0092] Figure 5 The simulation parameter settings are shown in Table 1 below:

[0093] Table 1

[0094]

[0095] It can be seen that when the signal-to-noise ratio value increases from 14dB to 19dB, the MSE results of the scheme in this embodiment and the MSE results of the comparative scheme both decrease with the increase of the signal-to-noise ratio value; when the deep learning strategy and RIS assistance are adopted, the MSE performance of the scheme in this embodiment is better than the MSE performance of the comparative scheme, and the gain effect is obvious when the signal-to-noise ratio value becomes larger.

[0096] The MSE performance of channel estimation under different RIS unit values ​​is as follows: Figure 6 shown.

[0097] Figure 6 The simulation parameter settings are shown in Table 2 below:

[0098] Table 2

[0099]

[0100] It can be seen that the MSE performance of the scheme of this embodiment and the MSE performance of the comparative scheme both increase with the increase of the number of RIS units; when the number of RIS units is fixed, the MSE values ​​of the scheme guided by deep learning are lower than the MSE of the scheme without deep learning guidance. It can be found that the method of this embodiment, with the assistance of deep learning and RIS, increases the number of RIS units and can compensate for the performance loss in the direct link case. Therefore, in an actual system, the method of this embodiment can reasonably set the number of RIS reflection units, thereby obtaining an excellent balance between performance and power consumption.

[0101] In summary, it can be seen that the RIS-assisted large-scale MIMO deep learning method of the present invention integrates RIS, generative adversarial networks and denoising convolutional neural networks, achieves excellent channel estimation performance and low computational complexity, has low training overhead and high practical value, and provides reference value for the development of future large-scale intelligent IoT systems.

[0102] In an embodiment of the present invention, an electronic device is also provided, including: one or more processors; a storage device on which one or more programs are stored; when the one or more programs are executed by the one or more processors, the one or more processors implement the RIS-assisted massive MIMO deep learning method described in any of the above embodiments.

[0103] In an embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the program is executed by a processor, the steps in any one of the RIS-assisted massive MIMO deep learning methods in the above embodiments are implemented.

[0104] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that for ordinary technicians in this technical field, any modifications, equivalent substitutions or improvements made without departing from the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A RIS-assisted massive MIMO deep learning method, characterized in that: The following steps are involved: Step S1, establishing a RIS-assisted large-scale machine type uplink communication system, and approximating a RIS-assisted cascade channel through a Gaussian random distribution; In the RIS-assisted large-scale machine type uplink communication system, there are channel links between the base station and the device, between the device and the RIS, and between the RIS and the base station. The base station configured with M antennas serves N single-antenna online devices, and the communication between the base station and the device is assisted by the RIS configured with K reflection elements. The pilot signal is , which is expressed as follows: ； in, represents the pilot sequence set of N single-antenna online devices with a length of B, , indicating the The pilot sequence of length B for each device, ; Received signal at the base station , which is expressed as follows: ； in, , , , Used to indicate the The received signal on the root antenna; Indicates A direct link between a device and a base station. The superscript T indicates transposition. ; It is the channel link between the device and RIS. ; It is the channel link between RIS and the base station. ; Respectively represent the complex domain signal sets of B dimension, K dimension and M dimension; Phase Shift Matrix of RIS , which is expressed as follows: ； in, It is the first The magnitude of the elements, For the Phase shift, ; j represents the imaginary part of the complex domain signal, The function represents an exponential function with the natural constant e as the base. Function is used to construct a diagonal matrix; use Represents the cascade channel, and rewrites the expression of the received signal at the base station as follows: ； in, is complex Gaussian white noise with a mean of , the variance is , Used to represent the product of the pilot signal and the concatenated channel; Based on the central limit theorem, the cascaded channel The approximation is a Bernoulli Gaussian random distribution, and the corresponding probability distribution P is expressed as follows: ； in, is the active probability of the device, is the Dirac function, Indicates Channel status information of each device, ; are path loss parameters, corresponding to the channel link between the device and RIS, the channel link between RIS and the base station, and the direct channel link between the device and the base station. Function represents the complex Gaussian probability distribution function, I represents the identity matrix; Step S2: Based on the Turbo message passing method, a RIS-assisted large-scale machine type uplink communication system deep learning network is constructed and layered, a Bayesian multi-dimensional measurement vector is estimated for each layer, and the loop is iterated until the final convergence to estimate the channel posterior information on the base station antenna; Construct a RIS-assisted large-scale machine-type uplink communication system deep learning network and divide it into Layer, for the Layer, the input is the received signal at the base station , the pilot signal is , the output is the estimated value of the channel ; Based on the multi-dimensional measurement vector rule, the generative adversarial network and denoising convolutional neural network methods are integrated into the message passing network. Each iteration of channel estimation is used as each layer in the network for cyclic iteration to The a posteriori information on the root antenna is estimated; For layer, the initial input is Output of the layer: ； Wherein, subscripts A and B represent module A and module B, wherein module A performs a posteriori estimation on the vector and transmits the external expectation and the external variance, and module B performs a posteriori estimation on the signal and estimates the channel information of each submodule; The superscript pri represents a priori value, and the superscript ext represents an external value. represents the variance, represents the expected function; , Respectively represent Layer A module On the antenna The prior expectation and prior variance of , Respectively represent Layer B module On the antenna The external expectation and external variance of To the base station The cyclic iteration of estimating the a posteriori information on the root antenna includes the following sub-steps: S2.

1. Order , it is estimated that the base station On the antenna The posterior information and External information, for each antenna In terms of calculation , , , and ; in, Respectively represent The product signal of the layer channel and the pilot and the channel state information. The superscript post represents the posterior value. , , , and Respectively represent Tier In the A module of the antenna and The posterior expectation and posterior variance of The posterior expectation and posterior variance of ; S2.2, performing joint channel estimation on all online devices of the RIS-assisted large-scale machine type uplink communication system, when hour, Indicates Tier The channel status information of each device is calculated The posterior expectation and posterior variance of are and , the formula is as follows: ； ； in, represents the expression to be solved, represents the expression for solving the variance, represents diagonalization; Indicates Tier The a priori expectation of the channel of each device in module B; Indicates Tier The channel prior variance of each device in module B; S2.3, for the base station The external information on the root antenna is estimated. When and , respectively representing the Tier In the B module of the antenna The posterior expectation and posterior variance of Tier In the B module of the antenna The prior expectation and prior variance of ; Step S3, converting the channel posterior information estimation module in S2 into a fusion module of a joint generative adversarial network and a denoising convolutional neural network, learning the RIS-assisted cascade channel state information, and obtaining the cascade channel estimation result after denoising; The estimation module of the channel posterior information is transformed into a fusion module of the joint generative adversarial network and the denoising convolutional neural network, and the generative adversarial network is used to generate the generated samples closest to the true value. , using the additive white Gaussian noise training model as the input of the denoising convolutional neural network, the generated samples are represented as follows: ； in, Respectively represent the channel and noise that are closest to the real generated samples; The denoising convolutional neural network consists of three layers. The first layer consists of 64 The second layer consists of 64 filters and rectified linear units. The third layer consists of a single The filter is used to reconstruct the signal; The fusion module outputs a noise value contained in the channel, and the noise value is subtracted from the input signal to obtain a channel estimation result after noise reduction processing; Step S4: Analyze the parsing performance of each layer of the deep learning network of the RIS-assisted large-scale machine type uplink communication system through state evolution, and obtain the convergence result of the RIS-assisted large-scale machine type uplink communication system; The state evolution of each layer of the RIS-assisted large-scale machine type uplink communication system deep learning network is performed as follows: ； ； ； in, Initialization of the cascade channel Calculation is performed to represent the element values ​​of the c-th layer channel variance matrix at the (1, 1), (2, 2) ..., (M, M) positions on the diagonal. are the measurement coefficients of the pilot matrix, For the The mean square error function of the channel estimation of each device; and Indicates The layer is used for the transfer function of state evolution. layer, through and The results are continuously updated until the convergence results of the RIS-assisted large-scale machine type uplink communication system are obtained.

2. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the RIS-assisted massive MIMO deep learning method as described in claim 1.

3. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the RIS-assisted large-scale MIMO deep learning method described in claim 1 is implemented.

Images