A S-CSI estimation model construction method and application for MIMO multi-user wireless communications

Abstract

The present invention discloses a method and application for constructing an S-CSI estimation model for MIMO multi-user wireless communications, which belongs to the field of wireless communications and includes: building an S-CSI estimation model and training the S-CSI estimation model using a noise-free training sample set, where the training sample is a base station's received signal and the label is the S-CSI corresponding to the base station's received signal; the S-CSI estimation model is a neural network; a denoising layer is provided between the input layer and the hidden layer of the neural network; the denoising layer is used to learn the process of calculating high-order cumulative amounts of training samples, and the order of the high-order cumulative amount is J ≥ 3; and the neural network is subjected to loss training to obtain a trained S-CSI estimation model. The method of the present invention can improve the generalization ability of the S-CSI estimation model and the accuracy of S-CSI estimation, and can achieve accurate estimation of the S-CSI of a time-nonstationary channel, and can also achieve a high estimation accuracy rate under low received signal-to-noise ratio.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communications, and more specifically, relates to a method for constructing an S-CSI estimation model for MIMO multi-user wireless communications and its application. Background Art

[0002] Massive Multiple-Input Multiple-Output (MIMO) technology is one of the key technologies for 5G. Massive MIMO systems typically deploy antenna arrays consisting of multiple antennas at the base station. More antennas provide greater spatial freedom, further improving transmission rates and communication reliability. Random matrix theory demonstrates that when the number of antennas in a base station approaches infinity, the transmit power of a single antenna can be reduced to zero through simple linear signal detection and appropriate transceiver design, thus avoiding multi-user interference and significantly increasing system capacity.

[0003] Accurately acquiring statistical channel state information (S-CSI) is crucial to realizing the many advantages of massive MIMO technology. From an information theory perspective, it can be shown that, to achieve the same performance, when the base station fully knows the channel state information (CSI), the transmit power of each single-antenna user is inversely proportional to the number of antennas at the base station. When CSI is incompletely known, the user's transmit power is inversely proportional to the square root of the number of base station antennas. Therefore, obtaining accurate S-CSI can significantly improve the system's channel capacity and energy efficiency.

[0004] In existing technologies, S-CSI estimation is primarily performed under the condition of a fixed noise variance. That is, S-CSI estimation is performed under the condition of unchanged statistical characteristics of the noise, resulting in poor generalization capability. In reality, the noise statistical characteristics of the signal to be estimated may vary significantly in different scenarios. If existing methods are still used for S-CSI estimation, the accuracy of the obtained S-CSI information will be low.

[0005] Furthermore, most existing research on S-CSI estimation is based on the assumption that the massive MIMO system channel exhibits wide-sense stationary properties, meaning that S-CSI remains stable over a long period of time. However, S-CSI in real-world scenarios does not meet this assumption. The relative motion of users and surrounding scatterers causes channel statistics to change. These factors cause the S-CSI of all users to vary over time, indicating that the massive MIMO channel exhibits temporal non-stationary characteristics. In this case, algorithms based on the wide-sense stationary assumption still perform poorly in S-CSI estimation. Summary of the Invention

[0006] In response to the defects of the existing technology and the need for improvement, the present invention provides a method and application for constructing an S-CSI estimation model for MIMO multi-user wireless communications, which aims to improve the generalization ability of the S-CSI estimation model and the accuracy of S-CSI estimation.

[0007] To achieve the above objectives, according to a first aspect of the present invention, a method for constructing an S-CSI estimation model for MIMO multi-user wireless communications is provided, comprising:

[0008] Building an S-CSI estimation model and training the S-CSI estimation model using a noise-free training sample set, where the training samples are received signals from a base station and their labels are the S-CSI corresponding to the received signals from the base station; wherein the base station is equipped with a MIMO antenna array;

[0009] The S-CSI estimation model is a neural network; a denoising layer is provided between the input layer and the hidden layer of the neural network; the denoising layer is used to learn the process of calculating the high-order cumulants of the training samples to remove additive Gaussian noise in the base station received signal; and the order J of the high-order cumulants is ≥3;

[0010] With the goal of minimizing the feature loss of the neural network, the S-CSI estimation model is trained until the loss converges or a preset number of training rounds is reached, thereby obtaining a trained S-CSI estimation model. The feature loss is the sum of the loss between the output of the denoising layer node and the high-order cumulant and the loss between the user S-CSI predicted by the neural network and the label.

[0011] Furthermore, the process of constructing the training sample set includes:

[0012] Get the transmission signal vectors of N users that establish links with the base station Among them, P i represents the transmit power of the i-th user;

[0013] For the i-th user, the channel vector of the i-th user at time t is obtained by sampling the Rayleigh distribution with S-CSI as the variance. Repeat this process for the N users to obtain the uplink channel matrix from each user in the cell to the base station. Where M represents the number of antennas in the MIMO antenna array equipped with the base station;

[0014] Under each S-CSI, the channel model y t =H t x t Generate a receiving signal y of the base station tAs a training sample; the label is the received signal y t Corresponding S-CSI.

[0015] Furthermore, The Rayleigh distribution is:

[0016]

[0017] in, The elements in α and β all follow the standard normal distribution.

[0018] Furthermore, the number of nodes in the denoising layer is related to the base station receiving signal y t The total number of higher-order cumulants is consistent;

[0019] The base station receives signal y t The total number of high-order cumulants is K J ; Wherein, K represents the number of antennas contained in each subarray, and the subarray refers to the subarray that evenly divides the MIMO antenna array equipped with the base station; J represents the order of the high-order cumulant.

[0020] Furthermore, the high-order cumulant is a fourth-order cumulant.

[0021] Furthermore, the neural network is a fully connected neural network.

[0022] Furthermore, during the training process, the loss function of the S-CSI estimation model is a cross entropy loss function.

[0023] According to a second aspect of the present invention, a method for estimating S-CSI for MIMO multi-user wireless communication is provided, comprising:

[0024] The actual base station received signal y to be estimated t Input into the S-CSI estimation model constructed by the S-CSI estimation model construction method according to any one of the first aspects to obtain the base station received signal y t Corresponding user S-CSI.

[0025] According to a third aspect of the present invention, there is provided an S-CSI estimation system for MIMO multi-user wireless communications, comprising a computer-readable storage medium and a processor;

[0026] The computer-readable storage medium is used to store executable instructions;

[0027] The processor is configured to read the executable instructions stored in the computer-readable storage medium to execute the S-CSI estimation model construction method according to any one of the first aspects, or / and to execute the S-CSI estimation method according to the second aspect.

[0028] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the program implements the S-CSI estimation model construction method as described in any one of the first aspects, or / and implements the S-CSI estimation method as described in the second aspect.

[0029] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0030] (1) In order to realize the S-CSI estimation of base station received signals with different noise statistical characteristics, the present invention uses training samples without noise to construct a training sample set. However, the received signal of the actual base station contains noise whose statistical characteristics will change. At the same time, considering that the third-order and above cumulative amount of additive Gaussian noise is 0, the present invention sets a denoising between the input layer and the hidden layer of the neural network to simulate the high-order cumulative amount calculation process of the base station received signal, thereby removing the additive Gaussian noise in the base station received signal. That is, no matter whether the received signal-to-noise ratio of the signal input into the neural network is high or low, after the denoising process, the additive Gaussian noise contained in the signal is basically removed, so that the constructed S-CSI estimation model has good estimation ability for the base station received signal under any received signal-to-noise ratio, thereby improving the generalization ability of the S-CSI estimation model and the accuracy of S-CSI estimation.

[0031] (2) Furthermore, the dataset construction method of the present invention generates received signal training data by simulating a channel model without any noise. The trained neural network has good estimation performance for received signals under any signal-to-noise ratio. There is no need to train multiple networks for different received signal-to-noise ratios, thus saving training overhead. In addition, the present invention transforms the S-CSI estimation problem into a neural network classification problem. By generating a corresponding received signal under each S-CSI to construct a training sample set, S-CSI with multiple labels is generated. Through the learning of the neural network, the characteristics of different distributions of the received signals can be learned, so that the S-CSI estimation model can better cope with input signals with different distributions, and is suitable for statistical channel state information estimation of non-stationary channels.

[0032] (3) As a preferred method, the received signal y t The fourth-order cumulative amount can reduce the computational complexity of the model while better eliminating y t The medium additive Gaussian noise makes the trained neural network have good estimation ability for the received signal under any signal-to-noise ratio.

[0033] (4) As a preference, the S-CSI estimation model adopts a fully connected neural network, so that the constructed model structure is simple and the complexity is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a fully connected neural network assisted by a high-order cumulant according to an embodiment of the present invention.

[0035] Figure 2 A flow chart of an S-CSI estimation method for massive MIMO multi-user wireless communications provided by an embodiment of the present invention.

[0036] Figure 3 Schematic diagram of a massive MIMO time-nonstationary channel provided by an embodiment of the present invention.

[0037] Figure 4 This article describes the impact of the received signal-to-noise ratio on the mean square error of S-CSI estimation under different estimation methods provided in the embodiments of the present invention.

[0038] Figure 5 This article describes the impact of the number of base station antennas on the mean square error of S-CSI estimation under different estimation methods provided in the embodiments of the present invention. DETAILED DESCRIPTION

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0040] In the present invention, the terms "first", "second", etc. in the present invention and the accompanying drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0041] like Figure 1 、 Figure 2 As shown, the S-CSI estimation model construction method for MIMO multi-user wireless communication of the present invention includes:

[0042] Build an S-CSI estimation model and use a training sample set without noise to train the S-CSI estimation model. The training sample is the received signal y of the base station. t , whose label is the base station received signal y t Corresponding S-CSI; wherein, the base station is equipped with a massive MIMO antenna array with M antennas.

[0043] The S-CSI estimation model is a neural network with a denoising layer between the input layer and the hidden layer. The denoising layer is used to learn the process of calculating the high-order cumulants of the training samples to remove the additive Gaussian noise in the input signal. The order of the high-order cumulants is J ≥ 3.

[0044] With the goal of minimizing the feature loss between the S-CSI of users in the cell at that moment predicted by the neural network and the label, the S-CSI estimation model is trained until the loss converges or a preset number of training rounds is reached, thereby obtaining a trained S-CSI estimation model. The feature loss is the sum of the loss between the output of the denoising layer node and the high-order cumulant and the loss between the S-CSI of users in the cell at that moment predicted by the neural network and the label. In this embodiment of the present invention, the loss between the output of the denoising layer node and the high-order cumulant and the loss between the S-CSI of users in the cell at that moment predicted by the neural network and the label both adopt cross-entropy loss.

[0045] Specifically, in the embodiment of the present invention, the process of constructing a data set includes:

[0046] Get the transmission signal vectors of N users that establish links with the base station Among them, P i represents the transmit power of the i-th user;

[0047] For the i-th user in the user group, the channel vector of the i-th user at time t is obtained by sampling the Rayleigh distribution with S-CSI as the variance. That is, It obeys the Rayleigh distribution with a mean of 0 and a variance of S-CSI. j represents the imaginary unit, where The elements in α and β all obey the standard normal distribution, M represents the number of antennas in the massive MIMO antenna array equipped by the base station; and repeat this process for N users in the user group to obtain the uplink channel matrix from each user in the cell to the base station

[0048] Under each S-CSI, the channel model y t =H t x t Generate a receiving signal y of the base station t As a training sample, its label is the received signal y t Its corresponding S-CSI.

[0049] In the embodiment of the present invention, 10,000 received signals corresponding to S-CSI=q are respectively taken as data samples, where q=1, 2, ..., 10; 80% are randomly selected as the selected sample set, and the remaining 20% ​​are used as the validation set. Preferably, the S-CSI estimation model in the embodiment of the present invention is a fully connected neural network, including a serial input layer, a denoising layer, a hidden layer, and an output layer;

[0050] Specifically, the number of nodes in the input layer is M, which is consistent with the number of antennas equipped with a massive MIMO antenna array in the base station.

[0051] The number of nodes in the denoising layer and the base station received signal y t The total number of high-order cumulants is consistent, which is K J ; Where K represents the number of antennas contained in each subarray of the massive MIMO antenna array equipped with the base station, which is evenly divided into subarrays; J represents the order of the high-order cumulant;

[0052] In the embodiment of the present invention, the base station receives the signal y t The fourth-order cumulant, y t The fourth-order cumulant cum can be defined as:

[0053]

[0054] in, Indicates taking an index m from the received signal of the nth subarray k In the embodiment of the present invention, k is 1, 2, 3 or 4; express The conjugate of ; R represents the number of sub-arrays into which the massive MIMO antenna array equipped by the base station is evenly divided, R = M / K.

[0055] Therefore, the received signal y t The fourth-order cumulants are 4 4 =256; in the embodiment of the present invention, the number of nodes in the denoising layer is set to 256.

[0056] The hidden layer is used to perform nonlinear transformation on the output of the denoising layer, so that the model can learn more complex features.

[0057] In the embodiment of the present invention, three hidden layers are set, and the number of nodes is 512, 128, and 32 respectively; in other embodiments, other numbers of layers and nodes can also be selected according to actual needs.

[0058] Specifically, the number of nodes in the output layer is consistent with the type of label S-CSI; in the embodiment of the present invention, the number of nodes in the output layer is 10, corresponding to the statistical channel state information of the users in the system During the S-CSI estimation model training process, this embodiment of the present invention uses a cross-entropy loss function, a stochastic gradient descent optimizer, an adjusted learning rate of 0.01, and a momentum parameter of 0.5. The number of iterations is set to 2000. The estimation accuracy of the trained neural network is verified on a validation set. The network parameters and estimation accuracy are saved. This process is repeated, and the network parameters with the highest estimation accuracy on the validation set are selected as the final S-CSI estimation network.

[0059] Example 2

[0060] The present invention also provides an S-CSI estimation method for MIMO multi-user wireless communication, which is applied to a base station, comprising:

[0061] The actual received signal y to be estimated t Input into the S-CSI estimation model constructed by the S-CSI estimation model construction method for MIMO multi-user wireless communication in Example 1, and the output classification is y t Corresponding user group S-CSI scenario.

[0062] The following is Figure 3 The single-cell scenario shown, with a single base station and multiple single-antenna user terminals, further illustrates the method of the present invention. Because multiple users within a cell are closely distributed and have similar surrounding scattering environments, all users have the same S-CSI. However, due to simultaneous co-directional movement of users or the generation and disappearance of scatterers around them, the massive MIMO channel exhibits temporal non-stationarity, causing the user's S-CSI to vary over time.

[0063] The base station is equipped with a massive MIMO antenna array, and users in the cell send signals to the base station simultaneously through the uplink channel. t , the base station receives the signal y t Then the user's S-CSI is estimated.

[0064] In the embodiment of the present invention, the base station knows the number N of users in the cell and agrees on the transmit power of a single user. The S-CSI of the user is a positive integer not greater than 10.

[0065] Key steps: (1) Based on the number of base station antennas M, the number of users N and the user transmission power P i , generate training data from the channel model; (2) build a fully connected neural network with a denoising layer; (3) use the training data to train the neural network; (4) use the trained S-CSI estimation model to estimate the actual received signal y t Make S-CSI estimation.

[0066] In this embodiment, the estimated mean square error of the S-CSI is used as the performance measurement standard and is defined as:

[0067]

[0068] in, is the true value of the S-CSI corresponding to the k-th received signal in the test set, is its estimated value, ζ is the number of test data, and in this embodiment, ζ=3000.

[0069] Figure 4 This study demonstrates the impact of the presence or absence of a denoising layer on neural network estimation performance at different received signal-to-noise ratios (SNRs). The number of users (N) is set to 16, the number of base station antennas (M) is set to 128, and the user S-CSI sets are {2, 4, 6}. The comparative experimental settings used are: Setting 1 uses a fully connected neural network without a denoising layer, with a received SNR of 15dB for the training data; Setting 2 uses a fully connected neural network without a denoising layer, with a received SNR of -15dB for the training data; and Setting 3 uses the fully connected neural network of the present invention with a denoising layer, with noise-free training data. Figure 4 This demonstrates the excellent noise-resistance performance of the high-order cumulant-assisted S-CSI estimation model (Set 3). It can be seen that regardless of whether the received signal SNR is high or low, the estimated MSE of Set 3 does not change dramatically; Set 1 achieves the best estimation effect when the received signal-to-noise ratio is 15dB, and Set 2 achieves the best estimation effect at -15dB. In other words, Sets 1 and 2 can only achieve good estimation performance near SNRs similar to those of the training data. The high-order cumulant-assisted S-CSI estimation network of the present invention has better generalization ability, as shown in the following: in the high SNR range, compared with the performance of the network trained for high SNR data (Set 1), the estimation performance of Set 3 is only slightly lower than that of Set 1, and better than that of Set 2; in the low SNR range, compared with the performance of the network trained for low SNR data (Set 2), the estimation performance of Set 3 is close to that of the network in Set 2, and far better than that of Set 1. This shows that the high-order cumulant-assisted S-CSI estimation network of the present invention has good adaptability to received signals of different SNRs and has greater practical usability.

[0070] Figure 5 It shows that the application of large-scale MIMO has a performance gain for the S-CSI estimation network assisted by high-order cumulants. The number of users is set to N = 16, and the test data of SNR = 15dB is used in setting 1, and the test data of SNR = -15dB is used in setting 2, so that the neural network works at the optimal performance point. The test data of SNR = -15dB is used in setting 3. With the increase in the number of antennas, the estimation performance of the three networks is improved. Compared with settings 1 and 2, the performance gain of multiple antennas for the S-CSI estimation network assisted by high-order cumulants of the present invention is more obvious. This is because more antenna array elements make it possible to divide more sub-arrays in the received signal, and when calculating the high-order cumulants of the received signal, more data samples can be used for calculation to obtain a better approximation effect. In other words, the method of the present invention is more suitable for large-scale MIMO.

[0071] Example 3

[0072] An embodiment of the present invention provides an S-CSI estimation system for MIMO multi-user wireless communications, including a computer-readable storage medium and a processor;

[0073] Computer-readable storage media for storing executable instructions;

[0074] The processor is configured to read executable instructions stored in a computer-readable storage medium to execute the S-CSI estimation model construction method for MIMO multi-user wireless communication in the above-mentioned embodiment 1, or / and, to execute the S-CSI estimation method for MIMO multi-user wireless communication in embodiment 2.

[0075] Example 4

[0076] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for constructing an S-CSI estimation model for MIMO multi-user wireless communication in the above-mentioned embodiment 1 is implemented, or / and the method for constructing an S-CSI estimation model for MIMO multi-user wireless communication in the above-mentioned embodiment 2 is implemented.

[0077] The present invention discloses a method for constructing an S-CSI estimation model for MIMO multi-user wireless communications. In order to achieve S-CSI estimation of base station received signals with different noise statistical characteristics, the present invention uses noise-free training samples to construct a training sample set. However, the received signal of an actual base station contains noise. Considering that the third-order and above cumulative amount of additive Gaussian noise is 0, the present invention sets denoising between the input layer and the hidden layer of the neural network to simulate the high-order cumulative amount calculation process of the base station received signal, thereby removing the additive Gaussian noise in the base station received signal. That is, regardless of whether the received signal-to-noise ratio of the signal input into the neural network is high or low, after the denoising process, the additive Gaussian noise contained in the signal is basically removed, so that the constructed S-CSI estimation model has good estimation capability for the base station received signal under any received signal-to-noise ratio, thereby improving the generalization capability of the S-CSI estimation model and the accuracy of the S-CSI estimation.

[0078] (2) Furthermore, the dataset construction method of the present invention generates received signal training data by simulating a channel model without any noise, and the trained neural network has good estimation performance for received signals under any signal-to-noise ratio. There is no need to train multiple networks for different received signal-to-noise ratios, thus saving training overhead. In addition, the present invention transforms the S-CSI estimation problem into a neural network classification problem. By generating a training sample set corresponding to each S-CSI to generate S-CSI with multiple labels, the neural network can learn the characteristics of different distributions of received signals, so that the S-CSI estimation model can better cope with input signals with different distributions, and is suitable for statistical channel state information estimation of non-stationary channels, which is crucial for improving the performance of large-scale MIMO communication systems.

[0079] (3) As a preferred method, the received signal y t The fourth-order cumulative amount can better eliminate y t The additive Gaussian noise in the model enables the trained neural network to have good estimation ability for the received signal under any signal-to-noise ratio; at the same time, it can also reduce the computational complexity of the model.

[0080] (4) As a preference, the S-CSI estimation model adopts a fully connected neural network, so that the constructed model structure is simple and the complexity is low.

[0081] In summary, the present invention learns the correspondence between the base station-side received signal and the S-CSI through a neural network assisted by a constructed high-order cumulant. The input of the network is the base station-side received signal at a certain moment, and the output of the network is the S-CSI of the users in the cell at that moment. In order to improve the noise resistance of the network, a denoising layer based on the high-order cumulant of the signal is added between the input layer and the hidden layer. During the training phase, a large number of received signals with known S-CSI generated by the channel model are used as training sets, and the network parameters are iteratively updated until the best estimation effect is achieved on the verification set. The estimation performance of the method of the present invention is better than the existing S-CSI estimation method based on the channel wide-sense stationary assumption, and a higher estimation accuracy can be achieved even under low received signal-to-noise ratio.

[0082] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing an S-CSI estimation model for MIMO multi-user wireless communication, characterized in that: include: Building an S-CSI estimation model and training the S-CSI estimation model using a noise-free training sample set, where the training samples are received signals from a base station and their labels are the S-CSI corresponding to the received signals from the base station; wherein the base station is equipped with a MIMO antenna array; The S-CSI estimation model is a neural network; a denoising layer is provided between the input layer and the hidden layer of the neural network; the denoising layer is used to learn the process of calculating the high-order cumulants of the training samples to remove additive Gaussian noise in the base station received signal; and the order J of the high-order cumulants is ≥3; With the goal of minimizing the feature loss of the neural network, the S-CSI estimation model is trained until the loss converges or a preset number of training rounds is reached, thereby obtaining a trained S-CSI estimation model. The feature loss is the sum of the loss between the output of the denoising layer node and the high-order cumulant and the loss between the user S-CSI predicted by the neural network and the label.

2. The S-CSI estimation model construction method according to claim 1, characterized in that: The process of constructing the training sample set includes: Get the transmission signal vectors of N users that establish links with the base station Among them, P i represents the transmit power of the i-th user; For the i-th user, the channel vector of the i-th user at time t is obtained by sampling the Rayleigh distribution with S-CSI as the variance. Repeat this process for the N users to obtain the uplink channel matrix from each user in the cell to the base station. Where M represents the number of antennas in the MIMO antenna array equipped with the base station; Under each S-CSI, the channel model y t =H t x t Generate a receiving signal y of the base station t As a training sample; the label is the received signal y t Corresponding S-CSI.

3. The S-CSI estimation model construction method according to claim 2, characterized in that: The Rayleigh distribution is: in, The elements in α and β all follow the standard normal distribution.

4. The S-CSI estimation model construction method according to claim 1, characterized in that: The number of nodes in the denoising layer is related to the base station receiving signal y t The total number of higher-order cumulants is consistent; The base station receives signal y t The total number of high-order cumulants is K J ; Wherein, K represents the number of antennas contained in each subarray, and the subarray refers to the subarray that evenly divides the MIMO antenna array equipped with the base station; J represents the order of the high-order cumulant.

5. The S-CSI estimation model construction method according to claim 4, characterized in that: The high-order cumulant is a fourth-order cumulant.

6. The S-CSI estimation model construction method according to any one of claims 1 to 5, characterized in that: The neural network is a fully connected neural network.

7. The S-CSI estimation model construction method according to any one of claims 1 to 5, characterized in that: During the training process, the loss function of the S-CSI estimation model is a cross entropy loss function.

8. A S-CSI estimation method for MIMO multi-user wireless communication, characterized in that: include: The actual base station received signal y to be estimated t The input is input into the S-CSI estimation model constructed by the S-CSI estimation model construction method according to any one of claims 1 to 7 to obtain the base station received signal y t Corresponding user S-CSI.

9. An S-CSI estimation system for MIMO multi-user wireless communication, characterized in that: comprising a computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read the executable instructions stored in the computer-readable storage medium to execute the S-CSI estimation model construction method according to any one of claims 1 to 7, or / and to execute the S-CSI estimation method according to claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, the S-CSI estimation model construction method according to any one of claims 1 to 7 is implemented, or / and the S-CSI estimation method according to claim 8 is implemented.

Images