Interference suppression method of multi-antenna 5G AeroMACS to FSS system

Abstract

The present invention relates to a method for suppressing interference of a multi-antenna 5G AeroMACS to an FSS system, and belongs to the field of wireless communication technology. The interference suppression method in the prior art has a large pilot overhead, a complex interference suppression process, and does not fully consider the interference suppression problem in dynamic scenes. The interference suppression method of the present invention takes the ground-to-air channel matrix as input, adaptively maps it to the HBF weight matrix and outputs it. At the same time, a data-driven generative adversarial network-gated cyclic unit ground-to-air channel matrix prediction method is designed to obtain accurate prediction accuracy. The proposed AO HBF interference suppression method realizes a lower mean square error MSE model and realizes accurate beamforming.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication, and specifically relates to a method for suppressing interference of a multi-antenna 5G AeroMACS system with an FSS system, aiming at the high timeliness and interference suppression requirements of communications under fast ground-to-space time variations. Background Art

[0002] In the existing technology, the 5091-5150 MHz frequency band is used for the airport ground 5G Aeronautical Mobile Airport Communications System (AeroMACS) network to provide efficient wireless communication rates. However, this frequency band is also allocated to the uplink of the Fixed Satellite Service (FSS) system of non-geostationary satellites. That is, when the AeroMACS network coexists with the FSS system, there is FSS radio frequency interference. In order to achieve the goal of airport surface spectrum compatibility, the interference threshold must not exceed 2% of the equivalent noise value of the FSS satellite receiver. In order to solve the problem of interference caused by Ground to Air (G2A) transmission to the satellite FSS system, the Multiple-Input Multiple-Output (MIMO) technology currently uses directional beams to adaptively protect the communication system from interference. However, under the conditions of rapidly changing airport surface communication channels, MIMO interference suppression is still a challenging problem: First, the frequent changes in the characteristics of the airport surface channel require continuous transmission of pilot signals for real-time channel estimation, which will generate unbearable pilot overhead and significantly reduce communication efficiency. Secondly, the beamforming matrix should be updated as the channel changes determined by the coherence time, which requires the implementation of interference suppression beamforming technology with relatively low computational complexity. Therefore, it is crucial to design a time-efficient MIMO interference suppression scheme to improve transmission reliability in environments with fast channel changes.

[0003] In MIMO interference suppression technology, channel estimation and beamforming are two key aspects. Existing pilot-based channel estimation algorithms, such as Least Squares (LS) and Minimum Mean Square Error (MMSE), can provide accurate channel estimation, but will result in large pilot overhead. In addition, although blind channel estimation algorithms do not require pilots, they need to estimate the channel by decomposing the signal subspace and noise subspace, which often leads to high computational complexity. Based on channel estimation, MIMO Hybrid Beamforming (HBF) technology can be used to suppress interference and enhance the signal strength required by the receiving end. HBF combines the advantages of analog beamforming and digital beamforming, and can provide performance close to that of full digital beamforming while reducing the number of RF links. However, most existing studies focus on applications in static scenarios, but do not fully consider interference suppression in dynamic scenarios.

[0004] In view of the above problems, the present invention proposes a method for suppressing interference of a multi-antenna 5G AeroMACS to an FSS system. An interference suppression model of an iterative optimization (AO) HBF is designed, and the ground-to-air channel matrix is ​​used as input, adaptively mapped to the HBF weight matrix and used as output. The data-driven Generative Adversarial Networks-Gated Recurrent Unit (GAN-GRU) ground-to-air channel matrix prediction algorithm proposed in the present invention obtains accurate prediction accuracy, and the proposed iterative optimization hybrid beamforming (AO HBF) interference suppression method achieves a lower mean squared error (MSE) and realizes accurate beamforming. Summary of the invention

[0005] In view of the above problems, the present invention provides a method for suppressing interference of a multi-antenna 5G AeroMACS to an FSS system. Firstly, the ground-to-air channel matrix is ​​tracked by a data-driven GAN-GRU method. Secondly, the interference of AeroMACS to the FSS system is reduced by an AO HBF interference suppression method. Finally, the scheme proposed in the present invention is verified in an airport fast channel scenario. When the number of historical ground-to-air channel matrix samples is insufficient and the accuracy is low, accurate prediction results are obtained. Under the premise of satisfying the maximum interference power constraint on the satellite receiver, accurate beamforming and spectrum compatibility effects are achieved.

[0006] The present invention provides a method for suppressing interference of a multi-antenna 5G AeroMACS to an FSS system, and the specific steps are as follows:

[0007] Step 1. Construct the minimum mean square error problem at the receiving end under the maximum interference power constraint;

[0008] Step 2. Construct a ground-to-air channel matrix prediction model based on GAN-GRU; use the ground-to-air channel matrix prediction model based on GAN-GRU to obtain the predicted ground-to-air channel matrix;

[0009] Step 3. Based on the predicted ground-to-air channel matrix, an interference suppression model based on AO HBF is constructed;

[0010] Step 4. Use the AO HBF-based interference suppression model obtained in step 3 to obtain anti-interference hybrid beamforming, and use the anti-interference hybrid beamforming to reduce interference to the FSS system.

[0011] Optionally, the specific steps of constructing the receiving end minimum mean square error problem under the maximum interference power constraint in step 1 are:

[0012] Constructing a receiving signal model of the AC end on the subcarrier in the data transmission stage, the receiving signal model includes a ground-to-air GS-AC channel matrix of the subcarrier, a baseband precoding matrix of the GS terminal carrier, a radio frequency precoding matrix of the GS end, a baseband combination matrix of the subcarrier at the AC end, and a radio frequency combination matrix at the AC end;

[0013] A mean square error (MSE) model is constructed based on the received signal model at the AC end of the subcarrier in the data transmission stage.

[0014] Optionally, the GAN-GRU based ground-to-air channel matrix prediction model includes a GAN based ground-to-air channel data enhancement module and a GRU based channel prediction module.

[0015] Optionally, the GAN-based ground-to-air channel data enhancement module includes a GAN input layer, a generation model, a discrimination model, a training model and a model evaluation module.

[0016] Optionally, the GRU-based channel prediction module includes a GRU input layer, a GRU unit and an output layer.

[0017] Optionally, by minimizing the loss function of the GAN-GRU based ground-to-air channel matrix prediction model Train the ground-to-air channel matrix prediction model based on GAN-GRU, the expression is:

[0018]

[0019] in, is the ground-to-air channel matrix prediction model based on GAN-GRU The ground-to-air channel matrix predicted by each frame, For the Frame simulation ground-to-air channel matrix; represents the mean symbol; L represents the total number of frames.

[0020] Optionally, the AO-based HBF model includes a radio frequency precoder, a baseband precoder, a radio frequency combiner and a baseband combiner.

[0021] Compared with the prior art, the present invention has at least the following beneficial effects:

[0022] (1) The GAN-GRU ground-to-air channel matrix prediction method proposed by the method of the present invention can perform ground-to-air channel data enhancement to improve the prediction accuracy of the ground-to-air channel matrix for multiple future frame lengths when the number of historical ground-to-air channel matrix samples is insufficient and the accuracy is low;

[0023] (2) The spectrum compatibility scheme obtained by the invented method satisfies the constraint that the interference to the satellite does not exceed the maximum power, and uses the AO algorithm to implement the combined hybrid beamforming of transmission and reception to reduce the interference to the FSS system;

[0024] (3) The interference suppression method of AO HBF proposed by the invented method adopts the manifold optimization (MO) algorithm to solve the constant modulus constraint problem of the RF precoder and the RF combiner. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings are only for the purpose of illustrating particular embodiments and are not to be construed as limiting the invention.

[0026] Figure 1 Schematic diagram of GAN-GRU ground-to-air channel matrix prediction for the interference suppression method of the present invention.

[0027] Figure 2 It is a structural schematic diagram of the AO HBF interference suppression model of the interference suppression method of the present invention.

[0028] Figure 3 The present invention is a flowchart of the GAN-GRU ground-to-air channel matrix prediction model of the interference suppression method.

[0029] Figure 4 Flow chart of the AO HBF interference suppression model of the interference suppression method of the present invention.

[0030] Figure 5 Flow chart of the interference suppression method of the present invention. DETAILED DESCRIPTION

[0031] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. In addition, the present invention can also be implemented in other ways different from those described herein, and therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0032] A specific embodiment of the present invention, as Figure 1-Figure 5 , discloses a method for suppressing interference of a multi-antenna 5G AeroMACS to an FSS system, the specific steps of which are as follows:

[0033] Step 1: Construct the minimum mean square error problem at the receiving end under the maximum interference power constraint;

[0034] The present invention considers the forward link transmission of the ground-to-air broadband MIMO system as the communication with a rapidly changing channel, and the number of antennas is The number of ground stations (GS) and an antenna is In order to save costs, both the ground station GS and the aircraft AC use antenna units formed by uniform planar arrays (UPA) and HBF architecture. The ground station GS uses an orthogonal frequency division multiplexing (OFDM) communication solution. The subcarriers transmit independent data streams. The signal vector transmitted on the subcarrier , ,satisfy The number of data streams transmitted on each subcarrier To ensure the effectiveness of communication, ,in, and The present invention considers that ground-to-air transmission (i.e., G2A) will affect the radio frequency chain number at the ground station GS and the aircraft AC. The low-orbit satellite FSS system with the root antenna generates interference. Therefore, real-time tracking of the ground-to-air channel is realized in a fast time-varying environment. At the same time, a hybrid beamforming scheme that meets the above conditions is designed to ensure that the interference power does not exceed the specified maximum interference power threshold.

[0035] Specifically, the receiving signal model of the AC end on the subcarrier in the data transmission stage is constructed, and the expression is:

[0036] (1)

[0037] in, Representative Frame No. The received signal at the AC end on the subcarrier; Indicates the AC end Frame No. The baseband combination matrix of subcarriers; Indicates the AC end RF combination matrix of the frame; Indicates Frame No. The ground-to-air GS-AC channel matrix of subcarriers; Indicates GS end RF precoding matrix of the frame; Indicates GS end Frame No. The baseband precoding matrix of subcarriers is ; , represents a complex value, is the number of antennas at the ground station, Number of RF chains on the aircraft GS side; On behalf of GS Frame No. The transmitted signal vector of subcarriers; , , , Indicates the number of antennas at the AC end. Indicates the number of RF chains at the AC end; Representative Frame No. The subcarriers have zero mean and variance of Gaussian white noise vector; , ; Indicates the number of data streams transmitted on each subcarrier, represents the variance of the noise, express The identity matrix of H represents the conjugate transpose.

[0038] Furthermore, the ground-to-air communication system (multi-antenna communication system) divides time resources into multiple frames, each frame includes two stages, namely, a channel estimation stage and a data transmission stage.

[0039] Furthermore, considering the propagation environment of the G2A scenario, a method with 1 LoS path (i.e., direct path) and The geometric Saleh-Valenzuela channel model of the NLoS path (i.e., non-direct path) is used to represent the Frame No. The ground-to-air GS-AC channel matrix of subcarriers is expressed as:

[0040] (2)

[0041] in, and The steering vectors represent the departure angle at the GS end and the arrival angle at the AC end, represents the conjugate transpose, , ; , , , , , , and The azimuth angle of the LoS path departure angle at the GS end, the elevation angle of the LoS path departure angle at the GS end, and the The azimuth of the NLoS path departure angle, the LGS end The elevation angle of the NLoS path departure angle, the azimuth angle of the AC end LoS path arrival angle, the elevation angle of the AC end LoS path arrival angle, the The azimuth of the NLoS path departure angle and the AC end The elevation angle of the NLoS path departure angle; and Respectively represent Frame No. The LoS path of the subcarrier and the The complex gain of each NLoS path; is the Ricean factor, i.e., the power ratio between the LoS path and the NLoS path; Indicates the total number of NLoS paths.

[0042] Furthermore, the complex gain of the NLoS path The expression is:

[0043] (3)

[0044] in, and Respectively represent NLoS in Large-scale fading gain and delay of frames; represents an index; represents an imaginary unit; Represents the OFDM symbol period; Indicates the total number of subcarriers.

[0045] Furthermore, the complex gain of the LoS path The expression is:

[0046] (4)

[0047] in, and They represent the Los path in the Large-scale fading gain and delay of frames.

[0048] Furthermore, based on Frame No. Subcarrier ground-to-air GS-AC channel matrix , GS terminal Frame No. The transmitted signal vector of subcarriers , GS terminal Frame No. The baseband precoding matrix of subcarriers , GS terminal RF precoding matrix of the frame 、AC end Frame No. The baseband combination matrix of subcarriers , build the omitted frame index The mean square error MSE model is expressed as:

[0049] (5)

[0050] in, Indicates The mean square error of subcarriers; represents the mean square value; On behalf of GS The transmitted signal vector of subcarriers; Indicates The received signal vector of the AC end on the subcarrier; Indicates Scaling factor for a given transmit power and noise power on a subcarrier; It means finding the trace of the matrix; Indicates The ground-to-air channel matrix of subcarriers; and Respectively The precoding matrix and combination matrix of the GS side and AC side on the subcarriers, , , represents the RF precoding matrix at the GS end, Indicates the GS end The baseband precoding matrix of subcarriers is Represents the RF combination matrix at the AC end, Indicates the AC end The baseband combination matrix of subcarriers; represents the variance of the noise, express The identity matrix of .

[0051] Based on the mean square error (MSE) model, the optimization problem of the minimum mean square error problem at the receiving end under the maximum interference power constraint is constructed, and the expression is:

[0052] (6a)

[0053] (6b)

[0054] (6c)

[0055] (6d)

[0056] in, represents the interference threshold; For the first The channel matrix from the ground station to the satellite on subcarriers, ; Represents the GS end radio frequency precoding matrix The constant modulus constraint, Represent the GS end radio frequency precoding matrix No. Line The index of the column element; Represents the AC end RF precoding matrix The constant modulus constraint, Respectively represent the AC end RF combination matrix No. Row, No. The index of the column element; .

[0057] Furthermore, the interference threshold The expression is:

[0058] (7)

[0059] in, is the Boltzmann constant; is the satellite receiver bandwidth; is the receiver noise temperature; Interference criterion.

[0060] The present invention can minimize the minimum mean square error (MMSE) between the AC-side receiving signal and the GS-side transmitting signal to characterize the reliability of transmission under the fast time-varying channel of the airport surface.

[0061] Step 2: Construct a ground-to-air channel matrix prediction model based on GAN-GRU;

[0062] Specifically, Figure 1 As shown, the GAN-GRU based ground-to-air channel matrix prediction model includes a GAN based ground-to-air channel data enhancement module and a GRU based channel prediction module.

[0063] Specifically, the GAN-based ground-to-air channel data enhancement module is used to enhance the quantity and diversity of the ground-to-air channel data set, which is used to subsequently enhance the training effect of the GRU channel prediction model. The GAN-based ground-to-air channel data enhancement module includes a GAN input layer, a generation model (GM), a discriminant model (DM), a training model, and a model evaluation module.

[0064] Specifically, the historical actual ground-to-space channel matrix is ​​sampled; the GAN input layer is used to sample each historical actual ground-to-space channel matrix Extract the real and imaginary parts, vectorize and connect them into each historical actual ground-to-air channel matrix vector ,in, Indicated in The historical actual ground-to-air channel matrix on all subcarriers in the frame; represents the real part set; represents the imaginary part set; Represents a vector set; represents the set of real numbers; Indicates the index of the current frame; Indicates the total number of historical frames.

[0065] Specifically, the generative model is used to take random noise that obeys Gaussian distribution as the input of the generative model, and to generate the same The actual ground-to-air channel matrix of the frame is similar to Frame Enhancement Channel Matrix .

[0066] Furthermore, the generative model includes four convolutional layers, a normalization (Batch Normalization, BN) layer and a parameter rectified linear unit (Rectified Linear Unit, ReLU) which are arranged in sequence; the four convolutional layers have good recognition performance for the spatiotemporal characteristics of the historical actual ground-space channel matrix; the normalization layer normalizes the output of each node in each layer to enhance the generalization and robustness of the model; the activation function of the parameter rectified linear unit (Rectified Linear Unit, ReLU) is used for nonlinear processing to improve the accuracy and efficiency of training.

[0067] Furthermore, the discriminant model generates the enhanced channel matrix of each frame generated by the generation model and the historical actual ground-to-air channel matrix The discriminant model includes 5 convolutional layers and 1 fully connected (FC) layer, a BN layer, a ReLU layer, and a random inactivation Dropout layer, which are used to extract the enhanced channel matrix of each frame output by the generative model. By combining these multi-dimensional features with the historical actual ground-to-air channel matrix vector corresponding to each frame, Compare and guide the training direction of the generative model to generate a more realistic enhanced channel matrix .

[0068] It can be understood that the binary classification result is whether the historical actual ground-to-air channel matrix is ​​real or synthetic.

[0069] The discriminant model of the present invention adds a BN layer, a ReLU layer and a Dropout layer, which improves the anti-overfitting ability of the neural network and enhances its versatility. In addition, the training model of the present invention includes a generative model and a discriminative model; the generative model is responsible for generating synthetic data, and the goal is to make the synthetic data as close as possible to the real data in terms of channel spatiotemporal characteristics; and the discriminative model (DM) learns how to distinguish between real data and generated synthetic data during the game process.

[0070] Furthermore, the goal of training the GAN-based ground-to-air channel data enhancement module is to minimize the loss function of the discriminant model, which is expressed as:

[0071] (8)

[0072] in, is the number of batches for training; is the discriminant model for The first training Enhanced channel matrix of the frame judgment, is the discriminant model for The first training The generative model of the frame is noisy in the input The judgment of the similarity between the enhanced channel matrix generated under the above conditions and the historical actual ground-to-air channel matrix; Indicates The noise of the generative model GM is input during the training.

[0073] Furthermore, the model evaluation module is used to test the channel spatiotemporal features generated by the generative model and the discriminative model (enhanced channel matrix ) is similar to the actual channel to evaluate the performance of the generative model and the discriminative model.

[0074] Specifically, the model evaluation module uses adversarial loss To measure whether the spatiotemporal characteristics of the channels generated by the generative model and the discriminative model are similar to the simulated ground-to-space channel matrix.

[0075] The GAN input layer is used to calculate the actual ground-to-space channel matrix for each historical Extract the real and imaginary parts, vectorize and connect them into each historical actual ground-to-air channel matrix vector

[0076] Specifically, against loss The expression is:

[0077] (9)

[0078] in, Indicates Frame simulation ground-to-air channel matrix vector Data distribution; Indicates Enhanced Channel Matrix in Frame data distribution.

[0079] It is understandable that when Frame Enhanced Channel Matrix in Frame Data distribution Close to Frame history actual ground-to-air channel matrix vector Data distribution When fighting against loss Close to 0.5, in this case, the discriminant model cannot distinguish the actually observed ground-to-air channel data vector and the enhanced channel matrix. That is, the generated data can be regarded as the real ground-to-air channel data.

[0080] Specifically, the GRU-based channel prediction module is used to extract the ground-to-air channel matrix (enhanced channel matrix) generated by the GAN-based ground-to-air channel data enhancement module. )’s temporal and spatial characteristics to predict the future channel state. The GRU-based channel prediction module includes a GRU input layer, a GRU unit, and an output layer.

[0081] Furthermore, the input layer is used to generate the first Frame Enhancement Channel Matrix and Frame Enhancement Channel Matrix Enter the GRU-based channel prediction module.

[0082] Furthermore, the GRU-based channel prediction module is used to update the gate and reset the gate control information flow to capture the first Frame Enhancement Channel Matrix With Frame Enhancement Channel Matrix The dependency of the channel matrix on multiple historical frames is enhanced.

[0083] Further, Frame Update Gate Control Frame Enhancement Channel Matrix The information is passed to the Frame Enhancement Channel Matrix The number of Frame update gate The expression is:

[0084] (10)

[0085] in, Indicates Frame update gate; is the Sigmoid activation function, is the weight matrix of the update gate; Indicates The hidden state of the frame.

[0086] Further, Frame reset gate Control Frame Enhancement Channel Matrix The number of forgettings. Allow the model to ignore unnecessary historical information and focus on the important features of the current frame enhanced channel matrix, expressed as:

[0087] (11)

[0088] in, Indicates Frame reset gate; is the weight matrix of the reset gate.

[0089] Further, Frame reset gate Store the past information about the ground-to-air channel matrix and update the new memory content as the candidate hidden state , No. Frame candidate hidden state The expression is:

[0090] (12)

[0091] in, is the activation function, is the weight matrix of the candidate hidden state.

[0092] Furthermore, The hidden state of the frame is Frame candidate hidden state Fusion, get the GRU unit Hidden state of the frame , the expression is:

[0093] (13)

[0094] Specifically, the output layer is used to output the predicted channel matrix of each frame , the predicted channel matrix of each frame Combining the real and imaginary parts, we get Frame prediction ground-to-air channel matrix set ,in, , and Respectively represent the predicted ground-to-air channel matrix of the first frame and the Frame prediction ground-to-air channel matrix.

[0095] The GRU-based channel prediction module of the present invention can effectively extract and utilize the temporal and spatial characteristics of the ground-to-air channel matrix generated by the GAN-based ground-to-air channel data enhancement module, thereby improving the accuracy of the prediction of the future channel state (i.e., the future ground-to-air channel matrix), reducing the pilot overhead, and improving the system performance.

[0096] Furthermore, the training is performed by minimizing the mean square error between the predicted ground-to-air channel matrix set output by the ground-to-air channel matrix prediction model based on GAN-GRU and the simulated ground-to-air channel matrix, thereby minimizing the loss function of the ground-to-air channel matrix prediction model based on GAN-GRU. The expression is:

[0097] (14)

[0098] in, is the ground-to-air channel matrix prediction model based on GAN-GRU The ground-to-air channel matrix is ​​predicted for each frame. For the Frame simulation ground-to-air channel matrix; represents the mean symbol; L represents the total number of predicted frames.

[0099] Specifically, the specific steps for training the ground-to-air channel matrix prediction model based on GAN-GRU are:

[0100] S21. Given a historical actual ground-to-air channel matrix training set, set the number of training batches to , the learning rate is .

[0101] S22. Extract the real and imaginary parts of the historical actual ground-to-air channel matrix training set to obtain the simulated ground-to-air channel data vector ;

[0102] S23. Generate enhanced channel matrix using generative model , the discriminant model uses formula (8) to enhance the channel matrix Perform evaluation, and if converged, proceed to the next step; otherwise, update the weight matrix of the GAN-based ground-to-air channel data enhancement module, and repeat S22-S23;

[0103] S24. Update the gate to retain the characteristics of the historical enhanced channel matrix to obtain the first Frame Update Gate

[0104] S25. Reset the gate to focus on the characteristics of the current enhanced channel matrix to obtain the first Frame reset gate ;

[0105] S26. Output enhanced ground-to-air channel matrix;

[0106] S27. Combine the real part and the imaginary part of the enhanced channel matrix to obtain the predicted ground-to-space channel matrix; if converged, proceed to the next step, otherwise update the weight matrix of the GRU-based channel prediction module and repeat S24-S27.

[0107] When using the trained GAN-GRU-based ground-to-air channel matrix prediction model, the actual historical ground-to-air channel matrix is ​​input into GAN-GRU, and the predicted ground-to-air channel matrix is ​​output.

[0108] From the MSE derivation given in the optimization problem (5), we can see that only when and The present invention assumes that the satellite-to-ground channel The matrix is ​​known, and we only focus on tracking the ground-to-air channel. Under fast time-varying channels, observe the ground-to-air channel matrix Obviously, it will take up a lot of pilot overhead, thus reducing communication efficiency. In addition, since each and The optimization problem (5) must be solved whenever there are changes, so the proposed solution method should work under relatively low computational complexity. However, the optimization problem (5) is difficult to handle due to the highly coupled optimization variables and constant mode constraints. In order to solve the above two challenges, the present invention proposes a ground-to-air channel matrix prediction strategy in the second step, which uses the past observation set to predict the ground-to-air channel matrix of multiple frames in the future, with low pilot overhead and high robustness. Then in the third step, based on , execute the interference suppression algorithm of AO HBF, and alternately iterate at the transmitting and receiving ends to achieve accurate beamforming and spectrum compatibility effects.

[0109] The present invention adopts interpolation or compressed sensing methods to restore full-space observation, thereby overcoming the problem of interference projection being lacking rank caused by the projection subspace being smaller than the total subcarrier space.

[0110] The set of ground-to-air channel matrices based on historical multi-frame observations is: . Since the channel prediction using machine learning relies heavily on the amount of training data and data accuracy, the current historical channel matrix samples are small in number, and it is impossible to obtain high-precision channel data with sufficient spatiotemporal characteristics. The present invention proposes a GAN-GRU model to enhance the ground-to-air channel training data, thereby improving the accuracy of prediction, thereby reducing the pilot overhead caused by frequent ground-to-air channel matrix estimation. The model consists of a GAN cascade GRU network. GAN generates synthetic data similar to real data through adversarial learning of two sub-models, Generative Model (GM) and Discriminative Model (DM), to obtain a sufficient number of data sets; then the ground-to-air channel data set with high precision and spatiotemporal characteristics output by GAN is input into the GRU network to obtain a predicted ground-to-air channel matrix.

[0111] The GAN-GRU-based joint prediction channel model can not only enhance the ground-to-space channel data through GAN to generate the corresponding channel data set with spatiotemporal characteristics, but also extract sequence data features through GRU to perform joint prediction in the spatiotemporal domain.

[0112] Step 3: Based on the predicted ground-to-air channel matrix, an interference suppression model based on iterative optimization hybrid beamforming (AO HBF) is constructed.

[0113] In order to solve the minimum mean square error problem at the receiving end under the maximum interference power constraint, the optimization problem (6) can be solved based on the AO idea on the basis of the predicted ground-to-air channel matrix obtained in step 2. , , , Decouple and iterate to solve. Through the hierarchical iterative processing of external and internal loops, the external loop iterates between the precoder and the combiner, and the internal loop iterates between the baseband and RF processing. The present invention focuses on the algorithm design of the GS-side precoder for spectrum compatibility, and the AC-side combiner design problem can be solved in the same way.

[0114] Specifically, the design precoding design sub-problem is expressed as:

[0115]

[0116] (15b)

[0117] (15c)

[0118] In the formula, Indicates The ground-to-air channel matrix on subcarriers Combined matrix with AC end The equivalent ground-to-space channel matrix is ​​multiplied by Indicates The negative first power of the scaling factor for the transmit power and noise power on the subcarrier and the baseband precoding matrix The equivalent baseband precoding matrix multiplied by .

[0119] Further, ,

[0120] Furthermore, the objective function of the precoding design subproblem (Eq. (15a)) is obtained relative to the Baseband combining matrix on subcarriers When the transmit power is at the maximum value, Scaling factor for a given transmit power and noise power on subcarriers The value of is expressed as:

[0121] (16)

[0122] In the formula, represents the RF precoding matrix, Indicates The equivalent baseband precoding matrix on subcarriers is: Indicates The conjugate transpose of the equivalent baseband precoding matrix on the subcarriers, represents the conjugate transpose of the RF precoding matrix.

[0123] Then, we obtain the first The closed-form solution of the baseband precoder on the subcarriers is expressed as:

[0124] (17)

[0125] In the formula, is the equivalent ground-to-space channel matrix Its conjugate transpose The product of Represents the conjugate transpose of the combined matrix Multiply the combined matrix Trace of.

[0126] Further, , .

[0127] The first The closed-form solution of the baseband precoder on the subcarriers and Scaling factor for a given transmit power and noise power on subcarriers Substituting into the objective function (15a) in the precoding design subproblem, we obtain the rewritten mean square error MSE model, which is expressed as:

[0128] (18)

[0129] In the formula, Represents the rewritten mean square error MSE model about the baseband combination matrix Function of Indicates Conjugate transpose of the channel matrix from the ground station to the FSS system on the subcarrier , the conjugate transposed matrix of the RF combiner , RF combination matrix , and the channel matrix from ground station to FSS system The product of the four.

[0130] Further, .

[0131] Furthermore, the mean square error MSE model (5) is simplified and transformed into the RF combination matrix The optimization problem is expressed as:

[0132] (19a)

[0133] (19b)

[0134] Furthermore, in view of the fact that the constant modulus constraint of the above RF encoder causes the optimization problem to be non-convex, the MO algorithm is used to search the step size of the gradient descent through the Armijo backtracking line, and the step size is used to update the RF precoding matrix to obtain the final RF precoding matrix. The specific steps are as follows:

[0135] 1) Calculate formula (18) to rewrite the mean square error MSE model with respect to the baseband combination matrix Function

[0136] Corresponding RF precoding matrix The Euclidean gradient of .

[0137] 2) Through the Euclidean gradient Orthogonal projection onto the tangent space computes the Riemann gradient.

[0138] 3) Calculate the conjugate gradient direction: Update the RF precoding matrix using the conjugate gradient direction and the gradient descent step size obtained by the Armijo backtracking line search .

[0139] 4) Update the RF precoding matrix Recover to the complex circle flow to get the final RF precoding matrix .

[0140] Accordingly, when solving the RF combination matrix Also perform the above operations.

[0141] Furthermore, an AO HBF model is constructed to solve the beamforming matrix, such as Figure 2 As shown in the figure, the AO HBF model includes four modules: RF precoder, baseband precoder, RF combiner, and baseband combiner. The specific description is as follows:

[0142] 1) RF precoder module: This module uses the combination matrix and the baseband precoding matrix as constants to obtain equation (18), and then executes the MO algorithm to solve the RF combination matrix Assume that the RF precoder is The step size of the gradient descent obtained by the Armijo backtracking line search is , the RF precoder is first Sub-Riemannian Gradient Descent Updated to:

[0143] (20)

[0144] In the formula, Indicates i External iterations j +1 update formula after gradient descent; Indicates External iterations The RF precoding matrix obtained by gradient descent is Indicates External iterations The step size obtained by gradient descent is Indicates External iterations The RF precoding matrix obtained by gradient descent The Riemann gradient of Represents the RF precoding matrix The Riemann gradient of is used to find the Frobenius norm.

[0145] Then, the i External iterations j +1 update after gradient descent Returning to the complex circular manifold, we obtain:

[0146] (twenty one)

[0147] In the formula, Indicates i External iterations j +1 gradient descent to obtain the update matrix of the RF precoding matrix No. m Line n Elements of a column; Indicates a retraction operation; Indicates update array No. m Line n The modulus of the elements of the column; Indicates i External iterations j +1 gradient descent to obtain the RF precoding matrix at the GS end Line Elements of a column;

[0148] Use the MO algorithm to iterate until The output after the external iteration converges is the final RF precoding matrix .

[0149] 2) Baseband precoder module:

[0150] This module directly applies the closed-form solution The final baseband precoding matrix after the outer iteration converges .

[0151] 3) RF combiner module:

[0152] The RF combination module is similar to the RF precoding module. Gradient descent steps update RF combination matrix , the MO algorithm iterates until After convergence, output RF combination matrix .

[0153] 4) Baseband combiner module:

[0154] Similar to the baseband precoder module, this module uses a closed-form solution for the The final baseband combination matrix after the outer iteration converges .

[0155] In order to improve the performance of the AO HBF-based interference suppression model, full digital interference suppression hybrid beamforming is used in the initial setting. The specific steps for training the AO HBF-based interference suppression model are as follows:

[0156] Step 31. Initialize the beamforming matrix of the transmitting and receiving ends and the ground-to-air channel matrix;

[0157] Step 32. Fixed receiving end combination matrix;

[0158] Step 33. Execute the MO algorithm to solve the transmitter RF precoding matrix ;

[0159] Step 34. Use closed form to solve the baseband precoding matrix ;

[0160] Step 35. Fix the transmitter precoding matrix;

[0161] Step 36. Execute the MO algorithm to solve the receiving end RF combination matrix ;

[0162] Step 37. Use closed form to solve the baseband combination matrix ;

[0163] Step 38: Determine whether the model converges. If so, end the training to obtain the interference suppression model based on AO HBF. Otherwise, repeat steps 32 to 37.

[0164] Step 4: Use the AO HBF-based interference suppression model obtained in step 3 to obtain anti-interference hybrid beamforming, and use the anti-interference hybrid beamforming to reduce interference to the FSS system.

[0165] Further, in the After the external iteration is completed, the final baseband precoding matrix is ​​obtained. , RF pre-coding matrix , baseband combination matrix and RF combination matrix .

[0166] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for suppressing interference of a multi-antenna 5G AeroMACS system to an FSS system, characterized in that: The specific steps are as follows: Step 1. Construct the minimum mean square error problem at the receiving end under the maximum interference power constraint; Step 2. Construct a ground-to-air channel matrix prediction model based on GAN-GRU; use the ground-to-air channel matrix prediction model based on GAN-GRU to obtain the predicted ground-to-air channel matrix; Step 3. Based on the predicted ground-to-air channel matrix, an interference suppression model based on AO HBF iterative optimization hybrid beamforming is constructed; Step 4. Use the interference suppression model based on AO HBF iterative optimization hybrid beamforming obtained in step 3 to obtain anti-interference hybrid beamforming, and use the anti-interference hybrid beamforming to reduce interference to the FSS system.

2. The interference suppression method according to claim 1, characterized in that: The specific steps of constructing the minimum mean square error problem at the receiving end under the maximum interference power constraint in step 1 are: Constructing a receiving signal model of the AC end on the subcarrier in the data transmission stage, the receiving signal model includes a ground-to-air GS-AC channel matrix of the subcarrier, a baseband precoding matrix of the GS terminal carrier, a radio frequency precoding matrix of the GS end, a baseband combination matrix of the subcarrier at the AC end, and a radio frequency combination matrix at the AC end; A mean square error (MSE) model is constructed based on the received signal model at the AC end of the subcarrier in the data transmission stage.

3. The interference suppression method according to claim 2, characterized in that: The GAN-GRU based ground-to-air channel matrix prediction model includes a GAN based ground-to-air channel data enhancement module and a GRU based channel prediction module.

4. The interference suppression method according to claim 3, characterized in that: The GAN-based ground-to-air channel data enhancement module includes a GAN input layer, a generation model, a discriminant model, a training model, and a model evaluation module.

5. The interference suppression method according to claim 4, characterized in that: The GRU-based channel prediction module includes a GRU input layer, a GRU unit, and an output layer.

6. The interference suppression method according to any one of claims 2 to 5, characterized in that: By minimizing the loss function of the GAN-GRU based ground-to-air channel matrix prediction model Train the ground-to-air channel matrix prediction model based on GAN-GRU, the expression is: in, is the ground-to-air channel matrix prediction model based on GAN-GRU The ground-to-air channel matrix predicted by each frame, For the Frame simulation ground-to-air channel matrix; represents the mean symbol; L represents the total number of frames.

7. The interference suppression method according to claim 1, characterized in that: The AO HBF-based model includes RF precoder, baseband precoder, RF combiner and baseband combiner.

Images