A low pilot overhead hybrid precoding method for a super large antenna array

Abstract

The application discloses a kind of multi-user low pilot overhead hybrid precoding methods of super large antenna array, including steps: 1) user sends pilot signal simultaneously, base station adopts far-field wide beam reception, and input GNN beam estimation model, obtains the candidate codeword of user according to the output of model;2) user sends pilot signal again, base station adopts candidate codeword reception, and allocates optimal near-field codeword for user;Based on optimal near-field codeword, base station designs analog precoder;3) user sends pilot signal for the third time, base station adopts optimal near-field codeword reception, and estimates effective channel matrix, designs digital precoder;4) base station carries out the transmission of downlink signal by deploying analog precoder and digital precoder.The application utilizes graph neural network to fully explore and utilize the correlation between users in the same system, reduces the pilot overhead required in beam training and channel estimation steps in hybrid precoding scheme under near-field environment, and improves the estimation accuracy of beam training.

Description

Technical Field

[0001] This invention relates to a low pilot overhead hybrid precoding method for very large-scale antenna arrays, belonging to the field of wireless communication physical layer technology, and involving beam training technology, beam allocation technology and precoder design technology. Background Technology

[0002] In fifth-generation (5G) wireless communication networks, massive MIMO (Multiple-Input Multiple-Output) technology has effectively improved the spectral efficiency of communication systems by deploying large-scale antenna arrays at base stations, and has therefore received widespread attention. However, the spectral efficiency improvements achieved through the application of MIMO are insufficient to meet the rapidly increasing spectral efficiency requirements of sixth-generation (6G) wireless communication networks. To meet the spectral efficiency demands of 6G networks, very large-scale antenna array (VMA) technology has been proposed. By deploying antennas at base stations that far exceed those in MIMO, VMA technology can achieve considerable beam gain, thereby further improving the spectral efficiency of communication systems to meet the needs of 6G wireless networks. In VMA technology, achieving considerable beam gain depends on the design of reliable analog and digital precoders at the base station. However, the design of VMA precoders faces challenges. On the one hand, the base station needs to perform codebook-based beam training to obtain the optimal beam for each user, thereby constructing the analog precoder. The massive antenna aperture of VMI (Very Large Scale Infrared) antennas necessitates consideration of near-field communication and near-field codebooks. However, the near-field codebook, containing a large number of codewords, drastically increases the pilot overhead required for beam training, posing a challenge to analog precoder design. On the other hand, obtaining a digital precoder depends on reliable channel estimation. However, in VMIs, the increased dimension of the channel matrix leads to a sharp rise in the complexity of channel estimation and pilot overhead, further posing challenges to digital precoder design. In summary, the design of low-pilot-overhead hybrid precoders for VMI technology is of significant research importance and also presents technical challenges. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a low-pilot-overhead hybrid precoding method for very large-scale antenna arrays. This method aims to fully explore and utilize the correlations between users within the same system by leveraging graph neural networks, thereby reducing the pilot overhead required for beam training and channel estimation steps in hybrid precoding schemes in near-field environments and improving the estimation accuracy of beam training. This invention also reduces beam collisions and interference between users by designing a reasonable beam allocation scheme and a digital precoder, effectively improving the downlink information transmission rate. Specifically, based on a trained GNN beam estimation model, this invention allows the base station to estimate the optimal near-field codeword for each user by testing only the far-field wide-beam codeword, thus significantly reducing pilot overhead.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A low-pilot-overhead hybrid precoding method for ultra-large-scale antenna arrays includes the following steps:

[0006] Step S1: The base station constructs and trains a GNN beam estimation model based on a graph neural network, and obtains and fixes the model parameters of the graph neural network.

[0007] Step S2: The base station is configured with N BS Root antenna and N RF There are K radio frequency links, each with a single antenna. All users simultaneously send mutually orthogonal pilot signals to the base station. The base station sequentially uses each codeword in a predefined far-field wide-beam codebook to receive all user pilot signals. The base station obtains the corresponding received signals and constructs them into vector form.

[0008] Step S3: The base station inputs the received signal in vector form into the GNN beam estimation model. The GNN beam estimation model outputs K probability vectors, where the i-th element of the k-th probability vector represents the estimated probability of the i-th near-field codeword in the near-field codebook for user k.

[0009] Step S4: Based on the probability vector of each user, the base station obtains the candidate codewords for each user;

[0010] Step S5: Each user sequentially sends time-orthogonal pilot signals to the base station, and the base station sequentially uses the candidate codewords of the corresponding user to receive the pilot signal of that user.

[0011] Step S6: The base station sorts the candidate codewords of each user according to the energy of the pilot signal; the base station assigns the final optimal beam to each user according to the sorting result; based on the assigned optimal near-field codeword, the base station designs an analog precoder.

[0012] Step S7: All users simultaneously send orthogonal pilot signals to the base station. The base station uses the optimal near-field codeword of each user to receive the pilot signals of all users, and estimates the effective channel matrix of each user and designs a digital precoder based on the received pilot signals.

[0013] The specific steps for constructing and training the GNN beam estimation model based on a graph neural network in step S1 are as follows:

[0014] Step S101: The base station constructs a GNN beam estimation model; the GNN beam estimation model includes a feature update module and an output module; the feature update module consists of three layers of graph neural network, and the output module consists of two fully connected network layers and one Softmax layer;

[0015] Step S102: The base station collects multiple sets of user channel data and user coordinates from a multi-user communication system with a large scale of multiple inputs and multiple outputs; the base station determines the optimal near-field codeword for the user based on the user coordinates and uses it as the tag for the user's channel data;

[0016] Step S103: The base station uses the collected multi-user channel data and their corresponding labels as training samples for the graph neural network, and trains the GNN beam estimation model using the Adama optimizer and a learning rate decay strategy. During the training process, the model parameters are continuously updated to minimize the loss function. The expression for the loss function is as follows:

[0017]

[0018] Where S represents the number of codewords in the near-field codebook; p represents the actual probability that the optimal codeword for user k, as estimated by the GNN beam estimation model, is the k-th near-field codeword; k,s The actual estimated probability that user k's optimal codeword is the k-th near-field codeword: when user k's optimal codeword is the k-th near-field codeword, p k,s =1, otherwise p k,s =0;

[0019] When the loss function approaches convergence, save and fix the model parameters.

[0020] In step S2, the base station uses a far-field wide-beam codebook to receive pilot signals from all users, specifically including:

[0021] Step S201: All users simultaneously send orthogonal pilot signals consisting of K symbols to the base station; based on a predefined far-field wide-beam codebook, the base station sequentially selects N symbols. RF A far-field wide-beam codeword is used to form an analog precoder and a digital precoder is set as an identity matrix to receive pilot signals;

[0022] Step S202: The base station demodulates the pilot signals based on their orthogonality to obtain the pilot signal vector for each user. Each pilot signal vector contains N... RF Each of the N elements corresponds to the use of these N elements. RF The base station obtains the pilot signal of the user when receiving a codeword;

[0023] Step S203: Repeat steps S201 and S202 until all codewords in the far-field wide-beam codebook have been tested. The base station obtains the pilot signal of each user under all different far-field wide-beam codewords and constructs them into vector form respectively.

[0024] In step S4, the base station obtains the candidate codewords for each user, specifically including:

[0025] Step S401: Based on the probability vector of each user obtained in step S3, for the k-th probability vector, the base station selects the N with the largest value. RF Estimated probabilities and obtained these N RF N corresponding to each estimated probability RF One near-field codeword; the obtained N RF The near-field codewords are the candidate codewords for user k;

[0026] Step S402: Repeat step S401 until the base station obtains the candidate codewords of all users.

[0027] In step S5, the base station sequentially tests the candidate codewords of each user, specifically including:

[0028] Step S501: User k sends a pilot symbol to the base station; the base station selects N of user k. RF The candidate codewords are used to form an analog precoder, and the digital precoder is set as an identity matrix to receive the pilot signal; the base station obtains an N RF A received signal vector of dimension, wherein the i-th element represents the pilot signal of user k received by the base station when the i-th candidate codeword is received;

[0029] Step S502, repeat step S501: Users sequentially send pilot symbols to the base station, and the base station sequentially receives them using the corresponding candidate codewords, until all users have finished sending, and the base station obtains K N-codewords. RF A dimensional received signal vector.

[0030] In step S6, the base station sorts the candidate codewords for each user and assigns beams to each user, specifically including:

[0031] Step S601: Based on the K N values ​​obtained in step S5 RFGiven a dimensional received signal vector, for the k-th received signal vector, the base station performs a modulo operation on its elements and sorts them in descending order; the i-th received signal in the reordered received signal vector is denoted as c. k,i The corresponding candidate codeword represents the candidate codeword of user k whose priority is in the i-th position;

[0032] Step S602: Repeat step S601 until the base station obtains the priority of all users' candidate codewords;

[0033] Step S603: Based on the priority information of the candidate codewords of all users, the base station initializes the optimal codeword of each user as the candidate codeword with the highest priority of that user;

[0034] Step S604: The base station optimizes and adjusts the optimal near-field codeword for each user: When the optimal near-field codewords of user m and user n are the same, the base station compares c. m,1 and c n,1 The magnitude of the modulus, if c m,1 The modulus is greater than c n,1 If the modulus value is such that the optimal near-field codeword of user m remains unchanged, the base station will change c. n,1 Delete the nth received signal vector and select the ith received signal c from the current received signal vector. n,1 The corresponding candidate codeword is taken as the optimal near-field codeword for user n, and vice versa;

[0035] Step S605: Repeat step S604 until the optimal near-field codewords of any two users are different.

[0036] Step S606: The base station obtains the K optimal near-field codewords from the K users. The K near-field codewords form the analog precoder at the base station.

[0037] Step S7, which involves the base station estimating the user's effective channel matrix and designing the digital precoder, specifically includes:

[0038] Step S701: All users simultaneously send orthogonal pilot signals consisting of K symbols to the base station; the base station uses the analog precoder obtained in step S6 and sets the digital precoder as an identity matrix to receive the pilot signals from all users; the base station obtains K signal reception vectors, denoted as r1, r2, ..., r K ;

[0039] Step S702: The base station estimates the effective channel matrix based on the K signal reception vectors obtained in step S701. The expression for the effective channel matrix is:

[0040] Step S703: Based on the effective channel matrix obtained in step S702, the base station designs a digital precoder. The expression for the digital precoder is: Where P dl and These represent the transmission power and noise power of the base station's downlink information transmission, respectively.

[0041] The beneficial effects of this invention are:

[0042] Compared to existing hybrid precoding designs, this invention considers near-field communication caused by VMI arrays and replaces the traditional exhaustive search with a graph neural network-based (GNN) beam estimation model, significantly reducing the pilot overhead required for beam training and improving its accuracy. Furthermore, this invention reduces beam collisions and interference between users through a reasonable beam allocation method, and reduces the pilot overhead and complexity required for designing the digital precoder by estimating the effective channel. Attached Figure Description

[0043] Figure 1 This is a flowchart of a low pilot overhead hybrid precoding method for a very large-scale antenna array;

[0044] Figure 2 This is a schematic diagram of the hybrid precoder for a very large-scale antenna array.

[0045] Figure 3 This is a network architecture diagram of the GNN beam estimation model of the present invention;

[0046] Figure 4 The diagram shows the spectral efficiency results obtained from the embodiments of the present invention and the exhaustive search scheme. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example 1

[0049] See Figure 1 , Figure 2 and Figure 3 The specific implementation method of the present invention includes the following steps:

[0050] Step 1: In a wireless communication system with a very large-scale antenna array, the base station is equipped with N... BSRoot antenna and N RF There are 3 radio frequency links, and K users are each configured with a single antenna. First, the base station needs to build and train a GNN beam estimation model based on a graph neural network to obtain the model parameters.

[0051] a1) Base station construction as follows Figure 3 The diagram shows a graph neural network (GNN)-based beam estimation model, which includes a feature update module and an output module. The feature update module consists of a three-layer graph neural network, with 256 neurons in each layer, and each layer uses the ReLU activation function. The output module consists of two fully connected layers and one softmax layer. The first fully connected layer has 128 neurons, and the number of neurons in the second fully connected layer is equal to the number of codewords in the near-field codebook. Both fully connected layers use the ReLU activation function.

[0052] a2) The base station randomly initializes all model parameters of the GNN beam estimation model. The input is set to the received signal obtained by the base station, and the output is set to the estimated probability of all codewords in the near-field codebook.

[0053] a3) The base station collects user channel data and user coordinates from multiple sets of VMI arrays in a multi-user communication system. The base station determines the optimal near-field codeword for the user based on the user coordinates and uses it as the tag for the user's channel data.

[0054] a4) The base station uses the collected multi-user channel data and their corresponding labels as training samples for the graph neural network, and employs the Adama optimizer and a learning rate decay strategy to train the GNN beam estimation model. During the training process, the model parameters are continuously updated to minimize the loss function. The expression for the loss function is as follows:

[0055]

[0056] Where S represents the number of codewords in the near-field codebook; p represents the actual probability that the optimal codeword for user k, as estimated by the GNN beam estimation model, is the k-th near-field codeword; k,s The actual estimated probability that user k's optimal codeword is the k-th near-field codeword: when user k's optimal codeword is the k-th near-field codeword, p k,s =1, otherwise p k,s =0.

[0057] When the loss function approaches convergence, save and fix the model parameters.

[0058] Step 2: All users simultaneously send mutually orthogonal pilot signals to the base station. The base station sequentially uses each codeword from a predefined far-field wide-beam codebook to receive all user pilot signals. The base station obtains the corresponding received signals and constructs them into vector form.

[0059] b1) All users simultaneously transmit orthogonal pilot signals consisting of K symbols to the base station. Based on a predefined far-field wide-beam codebook, the base station sequentially selects N symbols... RF An analog precoder is composed of far-field wide-beam codewords, and a digital precoder is set up as an identity matrix to receive pilot signals.

[0060] b2) The base station demodulates the pilot signals based on their orthogonality to obtain the pilot signal vector for each user. Each vector contains N... RF Each of the N elements corresponds to the use of these N elements. RF The base station obtains the pilot signal of the user when receiving a codeword.

[0061] b3) Repeat steps b1) and b2) until all codewords in the far-field wide-beam codebook have been tested. The base station obtains the pilot signal of each user under all different far-field wide-beam codewords and constructs them into vector form respectively.

[0062] Step 3: The base station inputs the received signal in vector form into the GNN beam estimation model. The model outputs K probability vectors, where the i-th element of the k-th probability vector represents the estimated probability that user k's optimal near-field codeword is the i-th near-field codeword in the near-field codebook. The higher the estimated probability, the greater the probability that user k's optimal near-field codeword is the i-th near-field codeword.

[0063] Step 4: Based on the probability vector of each user, the base station obtains the candidate codewords for each user.

[0064] c1) Based on the probability vector of each user obtained in step 3, for the k-th probability vector, the base station selects the N with the largest value. RF Estimated probabilities and obtained these N RF N corresponding to each estimated probability RF N near-field codewords. The obtained N RF The near-field codewords are the candidate codewords for user k.

[0065] c2) Repeat step c1) until the base station obtains the candidate codewords for all users.

[0066] Step 5: Each user sends time-orthogonal pilot signals to the base station in sequence, and the base station uses the candidate codewords of the corresponding user to receive the pilot signals of that user in sequence.

[0067] d1) User k sends a pilot symbol to the base station. The base station selects N for user k.RF The base station obtains an N candidate codeword to form an analog precoder and sets the digital precoder as an identity matrix to receive the pilot signal. RF A received signal vector of dimension , where the i-th element represents the pilot signal of user k received by the base station when the i-th candidate codeword is received.

[0068] d2) Repeat step d1): Users 1 to k sequentially send pilot symbols to the base station, and the base station receives them sequentially using the corresponding candidate codewords. This continues until all users have sent their symbols, at which point the base station obtains K N-codewords. RF A dimensional received signal vector.

[0069] Step 6: The base station sorts the candidate codewords for each user in descending order based on the energy of the pilot signal. The higher the ranking, the higher the priority of the candidate codeword corresponding to the received signal. The base station assigns the final optimal beam to each user based on the ranking result. Based on the assigned optimal near-field codeword, the base station designs an analog pre-encoder.

[0070] e1) Based on the K N values ​​obtained in step 5 RF Given a dimensional received signal vector, for the k-th received signal vector, the base station performs a modulo operation on its elements and sorts them in descending order. The i-th received signal in the reordered vector is denoted as c. k,i The corresponding candidate codeword represents the candidate codeword of user k with priority in the i-th position.

[0071] e2) Repeat step e1) until the base station obtains the priority of all users' candidate codewords.

[0072] e3) Based on the priority information of the candidate codewords of all users, the base station initializes the optimal codeword of each user as the candidate codeword with the highest priority of that user.

[0073] e4) To avoid beam collisions, i.e., different users cannot have the same optimal near-field codeword, the base station optimizes and adjusts the optimal near-field codeword for each user: when the optimal near-field codewords of user m and user n are the same, the base station compares c. m,1 and c n,1 The magnitude of the modulus, if c m,1 The modulus is greater than c n,1 If the modulus value is such that the optimal near-field codeword of user m remains unchanged, the base station will change c. n,1 Delete the nth received signal vector and select the i-th received signal c from the current vector. n,1 (i.e., c in the original vector) n,2 The candidate codeword corresponding to ) is taken as the optimal near-field codeword for user n. Conversely, the candidate codeword is taken as the optimal near-field codeword for user n.

[0074] e5) Repeat step e4) until the optimal near-field codewords of any two users are different.

[0075] e6) The base station obtains the K optimal near-field codewords from K users. The K near-field codewords form the analog precoder at the base station.

[0076] Step 7: All users simultaneously send orthogonal pilot signals to the base station. The base station uses the optimal near-field codeword of each user to receive the pilot signals of all users, and estimates the effective channel matrix of each user and designs a digital precoder based on the received pilot signals.

[0077] f1) All users simultaneously transmit orthogonal pilot signals consisting of K symbols to the base station. The base station uses the analog precoder obtained in step S6 and sets the digital precoder as an identity matrix to receive the pilot signals from all users. The base station obtains K signal reception vectors, denoted as r1, r2, ..., r K .

[0078] f2) Based on the K signal reception vectors obtained in step f1), the base station estimates the effective channel matrix. The expression for the effective channel matrix is:

[0079] f3) Based on the effective channel matrix obtained in step f2), the base station designs a digital precoder. The expression for the digital precoder is: Where P dl and These represent the transmission power and noise power of the base station's downlink information transmission, respectively.

[0080] The results of the above specific embodiments are as follows: Figure 4 As shown in the figure, the results illustrate the performance of the proposed scheme, the exhaustive search scheme, the traditional neural network scheme, and the traditional channel estimation scheme in terms of sum and rate. This example demonstrates that the present invention can approach the exhaustive search scheme in terms of sum and rate performance, and outperforms the other existing schemes.

Claims

1. A low-pilot-overhead hybrid precoding method for ultra-large-scale antenna arrays, characterized in that, Includes the following steps: Step S1: The base station constructs and trains a GNN beam estimation model based on a graph neural network, and obtains and fixes the model parameters of the graph neural network. Step S2: The base station is configured with N BS Root antenna and N RF There are 1 radio frequency link, and K users are each configured with a single antenna. All users simultaneously send mutually orthogonal pilot signals to the base station. The base station sequentially uses each codeword in the predefined far-field wide beam codebook to receive the pilot signals of all users. The base station receives the corresponding received signal and constructs it into a vector form; Step S3: The base station inputs the received signal in vector form into the GNN beam estimation model. The GNN beam estimation model outputs K probability vectors, where the i-th element of the k-th probability vector represents the estimated probability of the i-th near-field codeword in the near-field codebook for user k. Step S4: Based on the probability vector of each user, the base station obtains the candidate codewords for each user; Step S5: Each user sequentially sends time-orthogonal pilot signals to the base station, and the base station sequentially uses the candidate codewords of the corresponding user to receive the pilot signal of that user. Step S6: The base station sorts the candidate codewords of each user according to the energy of the pilot signal; the base station assigns the final optimal beam to each user according to the sorting result; based on the assigned optimal near-field codeword, the base station designs an analog precoder. Step S7: All users simultaneously send orthogonal pilot signals to the base station. The base station uses the optimal near-field codeword of each user to receive the pilot signals of all users, and estimates the effective channel matrix of each user and designs a digital precoder based on the received pilot signals. In step S5, the base station sequentially tests the candidate codewords of each user, specifically including: Step S501: User k sends a pilot symbol to the base station; the base station selects N of user k. RF The candidate codewords are used to form an analog precoder, and the digital precoder is set as an identity matrix to receive the pilot signal; the base station obtains an N RF A received signal vector of dimension, wherein the i-th element represents the pilot signal of user k received by the base station when the i-th candidate codeword is received; Step S502, repeat step S501: Users sequentially send pilot symbols to the base station, and the base station sequentially receives them using the corresponding candidate codewords, until all users have finished sending, and the base station obtains K N-codewords. RF A dimensional received signal vector; In step S6, the base station sorts the candidate codewords for each user and assigns beams to each user, specifically including: Step S601: Based on the K N values ​​obtained in step S5 RF Given a dimensional received signal vector, for the k-th received signal vector, the base station performs a modulo operation on its elements and sorts them in descending order; the i-th received signal in the reordered received signal vector is denoted as c. k,i The corresponding candidate codeword represents the candidate codeword of user k whose priority is in the i-th position; Step S602: Repeat step S601 until the base station obtains the priority of all users' candidate codewords; Step S603: Based on the priority information of the candidate codewords of all users, the base station initializes the optimal codeword of each user as the candidate codeword with the highest priority of that user; Step S604: The base station optimizes and adjusts the optimal near-field codeword for each user: When the optimal near-field codewords of user m and user n are the same, the base station compares c. m,1 and c n,1 The magnitude of the modulus, if c m,1 The modulus is greater than c n,1 If the modulus value is such that the optimal near-field codeword of user m remains unchanged, the base station will change c. n,1 Delete the nth received signal vector and select the ith received signal c from the current received signal vector. n,1 The corresponding candidate codeword is taken as the optimal near-field codeword for user n, and vice versa; Step S605: Repeat step S604 until the optimal near-field codewords of any two users are different. Step S606: The base station obtains the K optimal near-field codewords of K users, and the K near-field codewords form the analog precoder at the base station. Step S7, which involves the base station estimating the user's effective channel matrix and designing the digital precoder, specifically includes: Step S701: All users simultaneously send orthogonal pilot signals consisting of K symbols to the base station; the base station uses the analog precoder obtained in step S6 and sets the digital precoder as an identity matrix to receive the pilot signals from all users; the base station obtains K signal reception vectors, denoted as r1, r2, ..., r K ; Step S702: The base station estimates the effective channel matrix based on the K signal reception vectors obtained in step S701. The expression for the effective channel matrix is: Step S703: Based on the effective channel matrix obtained in step S702, the base station designs a digital precoder. The expression for the digital precoder is: Where P dl and These represent the transmission power and noise power of the base station's downlink information transmission, respectively.

2. The low pilot overhead hybrid precoding method for a very large-scale antenna array according to claim 1, characterized in that, The specific steps for constructing and training the GNN beam estimation model based on a graph neural network in step S1 are as follows: Step S101: The base station constructs a GNN beam estimation model; the GNN beam estimation model includes a feature update module and an output module; The feature update module consists of three graph neural network layers, and the output module consists of two fully connected network layers and one softmax layer. Step S102: The base station collects multiple sets of user channel data and user coordinates from the ultra-large-scale multiple-input multiple-output multi-user communication system; The base station determines the optimal near-field codeword for the user based on the user's coordinates and uses it as the tag for the user's channel data; Step S103: The base station uses the collected multi-user channel data and their corresponding labels as training samples for the graph neural network, and trains the GNN beam estimation model using the Adama optimizer and a learning rate decay strategy. During the training process, the model parameters are continuously updated to minimize the loss function. The expression for the loss function is as follows: Where S represents the number of codewords in the near-field codebook; p represents the actual probability that the optimal codeword for user k, as estimated by the GNN beam estimation model, is the k-th near-field codeword; k,s The actual estimated probability that user k's optimal codeword is the k-th near-field codeword: when user k's optimal codeword is the k-th near-field codeword, p k,s =1, otherwise p k,s =0; When the loss function approaches convergence, save and fix the model parameters.

3. The low pilot overhead hybrid precoding method for a very large-scale antenna array according to claim 1, characterized in that, In step S2, the base station uses a far-field wide-beam codebook to receive pilot signals from all users, specifically including: Step S201: All users simultaneously send orthogonal pilot signals consisting of K symbols to the base station; based on a predefined far-field wide-beam codebook, the base station sequentially selects N symbols. RF A far-field wide-beam codeword is used to form an analog precoder and a digital precoder is set as an identity matrix to receive pilot signals; Step S202: The base station demodulates the pilot signals based on their orthogonality to obtain the pilot signal vector for each user. Each pilot signal vector contains N... RF Each of the N elements corresponds to the use of these N elements. RF The base station obtains the pilot signal of the user when receiving a codeword; Step S203: Repeat steps S201 and S202 until all codewords in the far-field wide-beam codebook have been tested. The base station obtains the pilot signal of each user under all different far-field wide-beam codewords and constructs them into vector form respectively.

4. The low pilot overhead hybrid precoding method for a very large-scale antenna array according to claim 1, characterized in that, In step S4, the base station obtains the candidate codewords for each user, specifically including: Step S401: Based on the probability vector of each user obtained in step S3, for the k-th probability vector, the base station selects the N with the largest value. RF Estimated probabilities and obtained these N RF N corresponding to each estimated probability RF One near-field codeword; the obtained N RF The near-field codewords are the candidate codewords for user k; Step S402: Repeat step S401 until the base station obtains the candidate codewords of all users.

Images