Extensible communication method based on RIS assisted rate division multiple access

Abstract

The invention discloses an extensible communication method based on RIS-assisted rate division multiple access, which comprises the following steps of: establishing an RIS-assisted multi-device downlink communication system model based on rate division multiple access, transmitting data to a user by a base station through rate division multiple access, constructing a public data stream and a private data stream, and transmitting the public data stream and the private data stream to the base station through the RIS-assisted multi-device downlink communication system model. A base station uses a pre-coding matrix to transmit a signal, a user receives the signal after RIS reflection, a neural network RISNet is trained to learn mapping of available CSI to a reflection phase shift matrix of the RIS, a target function is constructed to maximize the weighting and rate of a model, a public data stream and a private data stream are independently decoded at a receiving end, and the decoding efficiency is improved. And after decoding, combining the information symbols of the public data stream and the private data stream to reconstruct the original message. According to the invention, the signal transmission range is expanded while the communication quality is guaranteed, and the problem of high-dimensional expandability caused by a large number of RIS reflection units is solved.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to an extensible communication method based on RIS-assisted rate division multiple access. Background Art

[0002] Reconfigurable Intelligent Surface (RIS) has attracted much attention in recent years. It is a two-dimensional thin-layer artificial electromagnetic surface structure with programmable electromagnetic properties. It is composed of many passive, reconfigurable units. It can reflect incident electromagnetic waves with complex moving phases by adjusting the electromagnetic parameters of electromagnetic waves (such as phase and amplitude), thereby enhancing or attenuating the reflected signal, optimizing channel characteristics, improving communication quality and extending signal coverage. In most typical scenarios, in order to establish sufficient link budget, it is usually necessary to deploy large RIS with hundreds to thousands of reflection units. The deployment of a large number of RIS reflection units leads to cost issues and also makes the acquisition of channel state information (CSI) a difficult problem.

[0003] The traditional method of obtaining CSI is mainly based on the pilot method. Even if the channel state is only changing slowly, the user CSI obtained by the transmitter may change dramatically due to the user moving too fast or occasionally blocked by large obstacles. Therefore, obtaining complete channel information requires high pilot overhead and computational complexity, resulting in the obtained CSI often being outdated or incomplete.

[0004] Many researchers have used machine learning, deep learning, meta-learning and other methods to train models to predict channels in order to optimize the configuration of RIS, but the complexity of the network model will also increase with the increase in the number of RIS reflection units, and scalability remains an unresolved issue.

[0005] Existing technologies related to rate splitting multiple access communication assisted by reconfigurable smart surfaces include:

[0006] For example, the optimization method of wireless energy rate splitting communication network assisted by dual IRS with patent publication number CN115103375A takes into account maximizing the communication rate of the minimum user, thereby maximizing the user communication rate and taking fairness into account under the user energy collection requirements. In addition, due to the deployment strategy of dual RIS, a higher communication rate can be achieved under limited available resources, improving the utilization rate of communication resources and communication coverage;

[0007] For example, patent publication number CN116419245A discloses a method for optimizing the energy efficiency of a multi-cell communication system with rate-splitting multiple access based on intelligent reflective surface assistance. This method proposes an iterative solution to the beamforming, rate-splitting and phase-shift optimization sub-problems for a non-convex optimization problem with the goal of maximizing the system energy efficiency.

[0008] However, the above-mentioned disclosed solutions all have the problem that all the reflective elements of RIS set in the system model can obtain the ideal CSI, or the number of RIS reflective units set is far from the required number of RIS to achieve the necessary link budget in many scenarios. In practice, the incompleteness of CSI makes the beamformer designed based on CSI inaccurate and unable to match the actual channel, resulting in a decrease in the data transmission performance and security performance of the system. The prior art has not yet considered the expansion of RIS reflective units and the scalable communication technology based on rate splitting with only limited CSI available. Summary of the invention

[0009] In order to overcome the defects and shortcomings of the prior art, the present invention provides a scalable communication method based on RIS-assisted rate split multiple access, which is used to solve the high-dimensional scalability problem caused by a large number of RIS reflection units, and optimize the configuration of CSI through a neural network architecture to significantly expand the coverage of the network when limited CSI is obtained in an actual system; at the same time, considering the robustness of rate split multiple access technology to non-ideal CSI, RSMA is introduced into the RIS-assisted multi-device downlink communication system to further reduce the difficulty of channel estimation, thereby effectively improving the reliability of communication and the system and rate.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions:

[0011] The present invention provides an extensible communication method based on RIS-assisted rate division multiple access, comprising the following steps:

[0012] Establish a RIS-assisted multi-device downlink communication system model based on rate splitting multiple access;

[0013] The base station sends a message to the user, splits the message into a private part and a public part, encodes the public part into a public data stream, and encodes the private part into a private data stream;

[0014] The base station uses a precoding matrix to transmit the signal, and the user receives the signal after reflection from the RIS;

[0015] Construct a neural network RISnet, generate training sets and test sets with different random user positions based on ray tracing channel data, train the neural network RISnet to learn the mapping of available CSI to RIS reflection phase shift matrix, and construct an objective function to maximize the weighted sum rate of the model;

[0016] Calculate the public data stream precoding matrix and private data stream precoding matrix under the current RIS configuration, obtain the reflection phase shift matrix of each data sample, and calculate the mixing channel and objective function of each data sample. After iterative training, output the optimized public data stream precoding matrix, private data stream precoding matrix and reflection phase shift matrix;

[0017] The public data stream and the private data stream are decoded separately at the receiving end, and after decoding, the information symbols of the public data stream and the private data stream are combined to reconstruct the original message.

[0018] As a preferred technical solution, the multi-device downlink communication system model includes a base station equipped with M antennas, K single-antenna users, and a reconfigurable intelligent reflective surface RIS with N reflective units;

[0019] The base station transmits data to users through Rate Splitting Multiple Access (RSMA).

[0020] As a preferred technical solution, the base station uses a precoding matrix to transmit the signal, and the user receives the signal after being reflected by the RIS, which specifically includes:

[0021] The base station uses the precoding matrix V = [V c ,V1,V2,..,V K ]Transmission signal s = [s c ,s1,s2,..,s K ] T , where V c represents the common data stream precoding matrix, V k represents the private data stream precoding matrix, s c Indicates public data flow, s k represents the private data stream of the kth user, k∈K;

[0022] The signal received by the user after being reflected by RIS is:

[0023] y=(GΦL+D)Vs+n=HVs+n

[0024] Among them, G represents the communication channel from RIS to the user, L represents the communication channel from the base station to RIS, Φ represents the reflection phase shift matrix of RIS, D represents the communication channel from the base station to the user, n represents the noise interference matrix, and H=GΦL+D is the mixed channel.

[0025] As a preferred technical solution, the common data stream precoding matrix is ​​specifically expressed as:

[0026]

[0027] The private data stream precoding matrix is ​​specifically expressed as:

[0028]

[0029] Where K represents the number of users. P c =(1-t)P t , are the power allocated to public messages and private messages respectively, t∈(0,1] is the power allocation factor, P t Indicates the maximum transmit power, I N represents the identity matrix of size N×N, h k represents the mixed channel of the kth user, h i represents the mixed channel of the i-th user.

[0030] As a preferred technical solution, both the public data stream and the private data stream are of standardized power, and E stands for expectation.

[0031] As a preferred technical solution, the neural network RISnet is trained to learn the mapping of the available CSI to the reflection phase shift matrix of RIS, specifically including:

[0032] Taking the channel feature Γ as the equivalent CSI representation, the neural network RISnet maps the equivalent CSI to the reflection phase shift matrix of RIS;

[0033] Assume the characteristic between user u and RIS reflection unit n is γ un , complete channel characteristics It is a three-dimensional tensor, the corresponding second dimension index is user u, and the third dimension index is the element of reflection unit n is Υ un .

[0034] As a preferred technical solution, the neural network RISnet includes multiple common layers and extended layers;

[0035] The output features of user u and RIS reflection unit n after being processed by the i-th common layer are expressed as:

[0036]

[0037] in, is the trainable bias of the cc-th class in the i-th layer, is the trainable weight of the cc-th class in the i-th layer, is the trainable bias of the ca-th class in the i-th layer, is the trainable weight of the ca-th class in the i-th layer, is the trainable bias of the oc-th class in the i-th layer, is the trainable weight of the oc-th class in the i-th layer, is the trainable bias of the oa-th class in the i-th layer, is the trainable weight of the oa-th class in the i-th layer, N represents the number of layers, U represents the total number of users, the cc class is the current user and the current RIS reflection unit, the ca class is the current user and all available RIS reflection units, the oc class is other users and the current RIS reflection unit, and the oa class is other users and all available RIS reflection units;

[0038] For the cc class in the i-th layer, after the local features of user u and RIS element n are extracted in the convolutional layer, the corresponding global output features are calculated through the fully connected layer, while the other classes take the weighted average of the results during information processing;

[0039] In the extended layer, multiple information processing units are used for feature processing for each category, and the RIS reflection unit with CSI is defined as the anchor element. By applying the same information processing unit to the adjacent RIS reflection units with the same relative position as the anchor element, the output of the RIS reflection unit n using the information processing unit j is calculated as follows:

[0040]

[0041] Among them, v(n,j) refers to the index of other RIS reflection units processed when the input RIS reflection unit is n. In the last layer, an information processing unit is used to process the information and attribute the characteristics of different users to phase shifts.

[0042] As a preferred technical solution, the objective function is expressed as:

[0043]

[0044] sttr(VV H )≤P t ,

[0045] |φ nn |=1,

[0046] |φ nn' | n≠n' =0.

[0047] Among them, a k represents the weight of user k, R ck represents the instantaneous achievable rate of the public data flow, R krepresents the instantaneous achievable rate of the private data flow, K represents the number of users, V c represents the common data stream precoding matrix, V k represents the private data stream precoding matrix, Φ represents the reflection phase shift matrix of RIS, V represents the precoding matrix, P t represents the maximum transmit power, tr represents the trace, |Φ nn | represents the diagonal element of the nth row and nth column of Φ, |Φ nn′ | n≠n′ represents off-diagonal elements.

[0048] As a preferred technical solution, the instantaneous achievable rate of the public data flow and the instantaneous achievable rate of the private data flow are calculated according to the Shannon formula, which is specifically expressed as:

[0049] R ck =log2(1+SINR c,k )

[0050] R k =log2(1+SINR p,k )

[0051]

[0052] Among them, SINR c,k Represents the signal-to-interference-to-noise ratio of the public data stream, SINR p,k represents the signal-to-interference-to-noise ratio of private data streams, P c =(1-t)P t , are the power allocated to public messages and private messages respectively, t∈(0,1] is the power allocation factor, P t Indicates the maximum transmit power, represents the noise interference to the kth user, K represents the number of users, and h k represents the hybrid channel of the kth user, H=GΦL+D is the hybrid channel, G represents the communication channel from RIS to the user, L represents the communication channel from the base station to RIS, Φ represents the reflection phase shift matrix of RIS, and D represents the communication channel from the base station to the user.

[0053] As a preferred technical solution, the public data stream and the private data stream are decoded separately at the receiving end, specifically including:

[0054] First, the public data stream is decoded, and all private data streams are regarded as interference or noise. The public data stream is eliminated through serial interference cancellation technology, and then the private data stream sent to each user is decoded, and the private data streams of other users are regarded as interference or noise.

[0055] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0056] (1) The scalable communication method based on RIS-assisted rate division multiple access proposed in the present invention can combine the robustness advantage of RSMA for partial CSI and RISnet for optimizing RIS configuration, optimize the phase shift of all RIS reflection units and optimize the precoding matrix on the base station side under the condition of limited CSI acquisition, while ensuring the communication quality, it also expands the signal transmission range, thereby achieving a good compromise between hardware complexity and performance.

[0057] (2) Compared with traditional SDMA, under different channel conditions where only partial CSI is known, the present invention introduces RSMA into the RIS-assisted multi-device downlink communication system to further reduce the difficulty of channel estimation. RSMA can effectively improve the weighted sum rate of the system.

[0058] (3) The present invention is applicable to the collection requirements of any number of RIS reflection units and any number of users, has practical applicability and flexibility, and is applicable to a variety of communication application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic diagram of the implementation architecture of the scalable communication method based on RIS-assisted rate division multiple access of the present invention;

[0060] Figure 2 This is a schematic diagram of the neural network RISnet architecture of the present invention;

[0061] Figure 3 It is a flow chart of RISnet training of the present invention;

[0062] Figure 4 A schematic diagram of the weighted sum rate in the solution proposed by the present invention under different channel conditions;

[0063] Figure 5 The figure is a schematic diagram of weighted sum rate comparison between the RSMA-based solution and the SDMA-based solution proposed in the present invention under different channel conditions. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0065] This embodiment provides a scalable communication method based on RIS-assisted rate division multiple access. With the assistance of RIS, some users that do not have a direct communication link with a base station or whose direct communication link is unreliable can receive signals or obtain enhanced signals, thereby expanding the coverage and reliability range of communication and realizing large-scale RIS deployment with low complexity. The method specifically includes the following steps:

[0066] S1: Establish a multi-device downlink communication system model assisted by RIS based on rate splitting multiple access, including a base station, RIS and several wireless users. There is no direct line-of-sight link between the base station and the users. RIS is used to assist in improving the signal transmission between the base station and the users.

[0067] like Figure 1 As shown in the figure, the multi-device downlink communication system includes a base station equipped with M antennas, K single-antenna users, and a reconfigurable intelligent reflective surface RIS with N reflective units. The base station transmits data to the user through rate splitting multiple access RSMA, but there is no direct communication link between the base station and the user. Each RIS reflective unit can receive the signal from the base station and reflect the signal to the user end through phase adjustment without changing the signal amplitude to assist communication, thereby expanding the reach of the signal. They are the communication channels from base station to user, base station to RIS and RIS to user respectively. It is the phase shift matrix of RIS, while only a few RIS reflection units are equipped with the RF chain for channel estimation.

[0068] S2: Based on the received signal of the downlink user, combined with the statistical characteristics of the channel, the phase shift matrix of the RIS and the precoding matrix on the base station side, the data transmission rate of the public data stream and the data transmission rate of all user private data streams are derived, and the objective function is constructed to maximize the weighted sum rate of the system model;

[0069] Determine the optimization variables as the phase shift matrix of the RIS and the precoding matrix on the base station side;

[0070] Forming constraint conditions: including RIS phase shift unit mode constraint, base station maximum transmit power constraint and public signal decoding constraint;

[0071] S3: Generate training sets and test sets with different random positions and build the neural network RISnet architecture;

[0072] In this embodiment, the input of RISnet in step S3 is channel characteristics and part of CSI information;

[0073] Since the original partial CSI cannot be directly used as the input of RISnet, it is necessary to define the channel feature Γ as an equivalent CSI representation. Specifically, a feature γ is defined for user u and RIS element n. un There are three channel matrices in the RIS-assisted communication system considered in this embodiment: Since the positions of BS and RIS are fixed, L can be assumed to be a constant, and D and G are N θ (RISnet) and changes with the user’s location. Since D cannot be directly mapped to an element of RIS, define J = DL + Therefore, the characteristics between user u and RIS reflection unit n can be defined as:

[0074]

[0075] Among them, |a| and arg(a) are the amplitude and phase of the complex number a, and the complete channel characteristics It is a three-dimensional tensor, the corresponding second dimension index is u, the third dimension index is n, and the element is γ un ;

[0076] S4: Train RISnet to learn the mapping between available CSI, optimal precoding, and phase adjustment, alternately optimize the precoding matrix and the phase shift matrix, and test the performance based on the test set after each training iteration;

[0077] In the offline training phase, we first define a neural network N with θ as a parameter. θ (Actually, it is RISnet) and maps the available CSI to the reflection phase shift matrix Φ of RIS, that is, Φ = N θ (Γ), where the channel characteristic Γ is an equivalent CSI representation. At this time, the objective function is determined by the channel characteristic Γ and the reflection phase shift matrix Φ, which can be written as f(Γ, Φ) = f(Γ, N θ (Γ); θ).

[0078] Preferably, parallel computing is used in step S4 to accelerate training and testing.

[0079] like Figure 2 As shown, considering the trade-off between performance and complexity, the neural network RISnet architecture has 8 layers, including common layers and extended layers. Of course, the number of network layers and extended layers can also be adjusted according to specific application requirements;

[0080] Because the phase shift of each RIS reflector unit depends on all users and all RIS reflector elements, all users and available RIS reflector units are divided into the following four categories:

[0081] (1) Class cc: current user and current RIS reflection unit;

[0082] (2) ca class: current user and all available RIS reflection units;

[0083] (3) oc class: other users and current RIS reflection unit;

[0084] (4) Category oa: other users and all available RIS reflection units;

[0085] Use multiple information processing units to process local and global information and stack the outputs as input to the next layer, thus creating an appropriate information flow for complex decision making on Φ;

[0086] In the common layer, one information processing unit is used for each category, so the output features of user u and RIS reflection unit n after being processed by the i-th common layer can be calculated by the following formula:

[0087]

[0088] in, is the trainable bias of the cc-th class in the i-th layer, is the trainable weight of the cc-th class in the i-th layer, P i represents the input feature dimension of class cc, Q i Represents the output feature dimension. Similar definitions and the same dimensions apply to the ca, oc, and oa classes. For the cc class in the i-th layer, after the convolutional layer extracts local features of the user u and the RIS element n, a traditional fully connected layer is applied to calculate the corresponding global output features, while other classes take a weighted average of the results during information processing to enhance the nonlinear expression ability of the network.

[0089] Unlike the common layer, in the extended layer, 9 information processing units are used for feature processing for each category. Specifically, the RIS reflection units with CSI (that is, available) are defined as anchor elements, and each anchor element in a layer is expanded to 9 anchor elements by applying the same information processing unit to the adjacent RIS reflection units with the same relative position as the anchor element, such as Figure 2 As shown in the figure, the characteristics of the same RIS reflection unit and the adjacent 8 reflection units are output. By defining two such extension layers, the number of anchor elements in rows and columns is increased to 9 times the original. If only 16 RIS reflection units have CSI at the beginning, the phase shift of 1296 RIS reflection units can be obtained after two such extension layers.

[0090] The output of RIS reflection unit n using information processing unit j is calculated as follows:

[0091]

[0092] Among them, v(n,j) refers to the index of other RIS reflection units processed when the input RIS reflection unit is n. In the last layer, only one information processing unit is used to process the information. Because all users share the same phase shift, the characteristics of different users are attributed to the phase shift.

[0093] S5: Save model parameters, optimal precoding, and some information of training and test sets at certain iteration times;

[0094] like Figure 3 As shown in Figure 1, in the data preparation stage, the open source DeepMIMO ray tracing dataset is used to obtain real ray tracing channel data, and then the training set and test set with different random user positions are generated. In the training stage, RISnet is first randomly initialized, a batch of data samples are randomly selected, and then the precoding matrix V of the batch of samples under the current RIS configuration is calculated according to the precoding calculation formula. c 、V k , get the reflection phase shift matrix Φ of each data sample, calculate the mixing channel H of each data sample, and calculate the objective function of each data sample By calculating the objective function Regarding the gradient of the neural network, a stochastic gradient ascent step is performed to update the relevant parameters of the network. After each training round, the performance of the model is evaluated using the test set to adjust the learning rate. Finally, the saved model parameters can be used to obtain the objective function Maximize the optimal phase shift matrix Φ and precoding matrix V c 、V k .

[0095] As shown in Table 1 below, the relevant system and model parameters are summarized, where ADAM (Adaptive Moment Estimation) is an optimization algorithm commonly used in deep learning, which can adjust the learning rate of each parameter by calculating the first-order moment estimate and second-order moment estimate of the gradient.

[0096] Table 1 Related system and model parameters

[0097]

[0098] S6: Based on the optimal model parameters and precoding method obtained in the above steps, the weighted sum rate of the system is maximized while ensuring the expansion of the communication range.

[0099] The present invention combines the robustness of RSMA for partial CSI and the interference management capability with RISnet for optimizing RIS configuration to configure the phase shift matrix of RIS and the precoding matrix on the base station side when CSI acquisition is limited, thereby making the system performance stronger.

[0100] In this embodiment, the specific working process of the adopted rate split multiple access RSMA is as follows:

[0101] (1) The base station sends K independent messages to K users. The message sent to user k is expressed as W. k express, According to the design principle of rate splitting, each user message W k Split into private partsW pk and the public part W ck , using the codebook shared by all users to convert the public part {W c1 ,W c2 ,...,W cK} Jointly encoded into a common data stream s c , while the private part of user k is independently encoded into a private data stream s k , all data streams are power normalized, and E stands for expectation;

[0102] (2) At the maximum transmit power P t Under the constraint of , the base station uses the precoding matrix V = [V c ,V1,V2,..,V K ]Transmission signal s = [s c ,s1,s2,..,s K ] T , where V c represents the common data stream precoding matrix, V k represents the private data stream precoding matrix, K represents the total number of users;

[0103] (3) After being reflected by RIS, the signal received by the user end is y = (GΦL + D)Vs + n = HVs + n, where: is the noise interference matrix, n obeys the variance σ 2 , a distribution with a mean of 0, that is, n~CN(0,σ 2 ), H = GΦL + D is a mixed channel;

[0104] In this embodiment, the public stream adopts Maximum Ratio Transmission (MRT) precoding, and the private stream adopts Minimum Mean Square Error (MMSE) precoding, that is, in And P c =(1-t)P t , are the power allocated to public messages and private messages respectively, t∈(0,1] is the power allocation factor, I N Represents an N×N unit matrix. The diagonal elements of the unit matrix are all 1, and the elements in the other positions are 0. N represents N reflection units. h k represents the mixed channel of the kth user;

[0105] (4) At the receiving end, each user first sends a public data stream s c It decodes and treats all private data streams as interference or noise, then eliminates the public data stream through the Serial Interference Cancellation (SIC) technology, and then decodes the private data stream sent to each user. k , and regard the private data streams of other users as interference or noise, so as to achieve the purpose of rate splitting to eliminate some interference. The Signal to Interference plus Noise Ratio (SINR) of the public data stream at user k is: The SINR of the private data stream is According to Shannon's formula, the corresponding instantaneous achievable rates of the public data flow and the private data flow are R ck =log2(1+SINR c,k ), R k =log2(1+SINR p,k ), represents the noise interference to the kth user, which is the σ in the previous text 2 The user k part of

[0106] (5) After successfully decoding the public data stream and the private data stream, user k then merges the private and public information symbols to reconstruct the original message.

[0107] Given the system model considered above, the main goal set in this embodiment is to jointly optimize the phase shift matrix Φ of RIS and the precoding matrix V on the base station side. c 、V k , the common rate set C of all users is used to maximize the weighted sum rate (WSR) of all users. The optimization problem constructed is as follows: As shown:

[0108]

[0109] tr(VV H )≤P t ,

[0110] |φ nn |=1,

[0111] |φ nn' | n≠n' =0.

[0112] Among them, a k represents the weight of user k, C k is the user k portion of the public rate, R c Constraint 1 is the public rate constraint, which ensures that the public message can be successfully decoded by each user; Constraint 2 indicates that the public rate assigned to user k should be non-negative; Constraint 3 stipulates that the total transmission power cannot exceed the maximum transmission power P t ; Constraints 4 and 5 ensure that RIS does not amplify the received signal, |Φ nn | represents the diagonal element of the nth row and nth column of Φ, |Φ nn′ | n≠n′ represents the off-diagonal elements, because the column is not equal to the row, and tr represents the trace;

[0113] By inference The optimization problem can be Further simplified to At this time, the optimization variable C is no longer considered.

[0114]

[0115] sttr(VV H )≤P t ,

[0116] |φ nn |=1,

[0117] |φ nn' | n≠n' =0.

[0118] Solve the original equivalent optimization problem The purpose is to jointly optimize the precoding matrix V c 、V k and the phase shift matrix Φ of RIS to maximize WSR, but the high-dimensional complexity of the problem makes it difficult to find the optimal solution. Therefore, this implementation trains a neural network RISnet to learn the mapping between available CSI, optimal precoding and phase shift matrix.

[0119] In order to verify the effectiveness of the scheme of the present invention, three different channels are considered in the simulation process, namely, deterministic ray tracing channel, independent and identically distributed complex Gaussian gain channel and deterministic ray tracing channel plus complex Gaussian gain, which are respectively denoted as channel A, channel B and channel C. Under the simulation condition of channel A, since all RIS reflection units share the same propagation path, the CSI of all reflection units can be inferred through a small number of RIS reflection units; under the simulation condition of channel B, since the channel gain is an independent and identically distributed complex Gaussian random variable in the RIS reflection unit, the partial CSI cannot provide enough information about the complete channel, so it is difficult to infer the complete CSI; channel C is between the two, which is jointly determined by the strong deterministic ray tracing channel and the complex Gaussian gain caused by weak scattering. It is hoped that the advantage of RSMA compared to SDMA, "robust performance in the case of partial CSI", is introduced. Through simulation, it is found that under the conditions of channels B and C, the system based on the RSMA transmission scheme can also obtain good performance.

[0120] like Figure 4 As shown, under the above simulation conditions, the weighted sum rate of the scheme proposed in the present invention in channel A is obtained. It can be seen that the scalable communication scheme based on rate splitting multiple access of the present invention has achieved better system weighted sum rate results, and the limited CSI has achieved performance similar to that of the full CSI.

[0121] like Figure 5 As shown, the comparison results of the scheme proposed in the present invention and SDMA in terms of weighted sum rate under three channel simulation conditions are obtained. The RISnet models used are all Partial CSI. It can be seen that: when the Partial CSI is known, RSMA is significantly better than SDMA, indicating that compared with traditional SDMA, under different channel conditions where only partial CSI is known, RSMA can achieve an effective improvement in the weighted sum rate of the system.

[0122] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A scalable communication method based on RIS-assisted rate division multiple access, characterized in that: The steps include: Establish a RIS-assisted multi-device downlink communication system model based on rate splitting multiple access; The base station sends a message to the user, splits the message into a private part and a public part, encodes the public part into a public data stream, and encodes the private part into a private data stream; The base station uses a precoding matrix to transmit the signal, and the user receives the signal after reflection from the RIS; Construct a neural network RISnet, generate training sets and test sets with different random user positions based on ray tracing channel data, train the neural network RISnet to learn the mapping of available CSI to RIS reflection phase shift matrix, and construct an objective function to maximize the weighted sum rate of the model; Calculate the public data stream precoding matrix and private data stream precoding matrix under the current RIS configuration, obtain the reflection phase shift matrix of each data sample, and calculate the mixing channel and objective function of each data sample. After iterative training, output the optimized public data stream precoding matrix, private data stream precoding matrix and reflection phase shift matrix; The public data stream and the private data stream are decoded separately at the receiving end, and after decoding, the information symbols of the public data stream and the private data stream are combined to reconstruct the original message.

2. The scalable communication method based on RIS-assisted rate division multiple access according to claim 1, characterized in that: The multi-device downlink communication system model includes a base station equipped with M antennas, K single-antenna users, and a reconfigurable intelligent reflective surface RIS with N reflective units; The base station transmits data to users through Rate Splitting Multiple Access (RSMA).

3. The scalable communication method based on RIS-assisted rate division multiple access according to claim 1, characterized in that: The base station uses the precoding matrix to transmit the signal, and the user receives the signal after reflection by the RIS. Specifically, it includes: The base station uses the precoding matrix V = [V c ,V1,V2,..,V K ]Transmission signal s = [s c ,s1,s2,..,s K ] T , where V c represents the common data stream precoding matrix, V k represents the private data stream precoding matrix, s c Indicates public data flow, s k represents the private data stream of the kth user, k∈K; The signal received by the user after being reflected by RIS is: y=(GΦL+D)Vs+n=HVs+n Among them, G represents the communication channel from RIS to the user, L represents the communication channel from the base station to RIS, Φ represents the reflection phase shift matrix of RIS, D represents the communication channel from the base station to the user, n represents the noise interference matrix, and H=GΦL+D is the mixed channel.

4. The scalable communication method based on RIS-assisted rate division multiple access according to claim 3, characterized in that: The common data stream precoding matrix is ​​specifically expressed as: The private data stream precoding matrix is ​​specifically expressed as: Where K represents the number of users. P c =(1-t)P t , are the power allocated to public messages and private messages respectively, t∈(0,1] is the power allocation factor, P t Indicates the maximum transmit power, I N represents the identity matrix of size N×N, h k represents the mixed channel of the kth user, h i represents the mixed channel of the i-th user.

5. The scalable communication method based on RIS-assisted rate division multiple access according to claim 3, characterized in that: Both public and private data streams are of normalized power, and E stands for expectation.

6. The scalable communication method based on RIS-assisted rate division multiple access according to claim 1, characterized in that: The neural network RISnet is trained to learn the mapping of available CSI to the reflected phase shift matrix of RIS, including: Taking the channel feature Γ as the equivalent CSI representation, the neural network RISnet maps the equivalent CSI to the reflection phase shift matrix of RIS; Assume the characteristic between user u and RIS reflection unit n is γ un , complete channel characteristics It is a three-dimensional tensor, the corresponding second dimension index is user u, and the third dimension index is the element of reflection unit n is γ un .

7. The scalable communication method based on RIS-assisted rate division multiple access according to claim 6, characterized in that: The neural network RISnet includes multiple common layers and extended layers; The output features of user u and RIS reflection unit n after being processed by the i-th common layer are expressed as: in, is the trainable bias of the cc-th class in the i-th layer, is the trainable weight of the cc-th class in the i-th layer, is the trainable bias of the ca-th class in the i-th layer, is the trainable weight of the ca-th class in the i-th layer, is the trainable bias of the oc-th class in the i-th layer, is the trainable weight of the oc-th class in the i-th layer, is the trainable bias of the oa-th class in the i-th layer, is the trainable weight of the oa-th class in the i-th layer, N represents the number of layers, U represents the total number of users, the cc class is the current user and the current RIS reflection unit, the ca class is the current user and all available RIS reflection units, the oc class is other users and the current RIS reflection unit, and the oa class is other users and all available RIS reflection units; For the cc class in the i-th layer, after the local features of user u and RIS element n are extracted in the convolutional layer, the corresponding global output features are calculated through the fully connected layer, while the other classes take the weighted average of the results during information processing; In the extended layer, multiple information processing units are used for feature processing for each category, and the RIS reflection unit with CSI is defined as the anchor element. By applying the same information processing unit to the adjacent RIS reflection units with the same relative position as the anchor element, the output of the RIS reflection unit n using the information processing unit j is calculated as follows: Among them, v(n,j) refers to the index of other RIS reflection units processed when the input RIS reflection unit is n. In the last layer, an information processing unit is used to process the information and attribute the characteristics of different users to phase shifts.

8. The scalable communication method based on RIS-assisted rate division multiple access according to claim 1, characterized in that: The objective function is expressed as: s.t.tr(VV H )≤P t , |f nn |=1, |f nn' | n≠n' =0. Among them, a k represents the weight of user k, R ck represents the instantaneous achievable rate of the public data flow, R k represents the instantaneous achievable rate of the private data flow, K represents the number of users, V c represents the common data stream precoding matrix, V k represents the private data stream precoding matrix, φ represents the reflected phase shift matrix of RIS, V represents the precoding matrix, P t represents the maximum transmit power, tr represents the trace, |Φ nn | represents the diagonal element of the nth row and nth column of Φ, |Φ nn′ | n≠n′ represents off-diagonal elements.

9. The scalable communication method based on RIS-assisted rate division multiple access according to claim 8, characterized in that: The instantaneous achievable rate of the public data flow and the instantaneous achievable rate of the private data flow are calculated according to the Shannon formula, which is specifically expressed as: R ck =log2(1+SINR c,k ) R k =log2(1+SINR p,k ) Among them, SINR c,k Represents the signal-to-interference-to-noise ratio of the public data stream, SINR p,k represents the signal-to-interference-to-noise ratio of private data streams, P c =(1-t)P t , are the power allocated to public messages and private messages respectively, t∈(0,1] is the power allocation factor, P t Indicates the maximum transmit power, represents the noise interference to the kth user, K represents the number of users, and h k represents the hybrid channel of the kth user, H=GΦL+D is the hybrid channel, G represents the communication channel from RIS to the user, L represents the communication channel from the base station to RIS, φ represents the reflection phase shift matrix of RIS, and D represents the communication channel from the base station to the user.

10. The scalable communication method based on RIS-assisted rate division multiple access according to claim 1, characterized in that: Decoding the public data stream and the private data stream separately at the receiving end specifically includes: First, the public data stream is decoded, and all private data streams are regarded as interference or noise. The public data stream is eliminated through serial interference cancellation technology, and then the private data stream sent to each user is decoded, and the private data streams of other users are regarded as interference or noise.

Images