Terminal direction finding method and apparatus

Abstract

The application provides a terminal direction finding method and device, the method comprises the following steps: preprocessing a reference signal sent by a terminal to obtain a virtual spatial spectrum; reconstructing the virtual spatial spectrum through a neural network and a conjugate gradient algorithm to obtain a reconstructed spatial spectrum; and performing a spectrum peak search on the reconstructed spatial spectrum, and taking an angle corresponding to the spectrum peak as an angle of arrival. The terminal direction finding method and device provided by the application can perform AoA estimation in the presence of hardware damage by using a neural network spatial spectrum correction and a conjugate gradient algorithm to iteratively reconstruct a spatial spectrum, and a large amount of training data is not necessary, thereby improving the accuracy of terminal direction finding.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a terminal direction finding method and apparatus. Background Technology

[0002] Hardware impairments often affect the accuracy of terminal positioning. Traditional model-driven Angle of Arrival (AoA) estimation algorithms are susceptible to hardware impairments, which may lead to errors in the array response used during the estimation process. To address this issue, purely data-driven methods, such as direction-finding techniques based on deep neural networks (DNNs), have been extensively studied. However, the accuracy of these methods largely depends on training with large amounts of data, resulting in lower feasibility in practical applications.

[0003] In summary, the main drawbacks of existing technologies are that purely model-driven methods cannot overcome the impact of hardware impairments and cannot perform high-precision AoA estimation; purely data-driven methods rely on a large amount of training data, which makes them difficult to apply widely, and insufficient training data ultimately leads to low AoA estimation accuracy. Summary of the Invention

[0004] This application provides a terminal direction finding method and apparatus to solve the technical problem of low terminal direction finding accuracy caused by hardware damage in related technologies.

[0005] In a first aspect, embodiments of this application provide a terminal direction finding method, including:

[0006] The reference signal sent by the terminal is preprocessed to obtain the virtual spatial spectrum;

[0007] The virtual spatial spectrum is reconstructed using a neural network and the conjugate gradient algorithm to obtain the reconstructed spatial spectrum;

[0008] A peak search is performed on the reconstructed spatial spectrum, and the angle corresponding to the peak is taken as the angle of arrival.

[0009] In some embodiments, the reference signal transmitted by the terminal is preprocessed to obtain a virtual spatial spectrum, including:

[0010] Determine the covariance matrix of the reference signal transmitted by the terminal;

[0011] Vectorize the covariance matrix to obtain the array signal corresponding to the reference signal;

[0012] The array signal is estimated using array digital beamforming to obtain the virtual spatial spectrum.

[0013] In some embodiments, the virtual spatial spectrum is reconstructed using a neural network and a conjugate gradient algorithm to obtain a reconstructed spatial spectrum, including:

[0014] The virtual spatial spectrum is input into a neural network, and the neural network is used to calibrate the spatial spectrum.

[0015] The conjugate gradient algorithm is used to solve the calibrated spatial spectrum to obtain the reconstructed spatial spectrum of one iteration;

[0016] After multiple iterations of calibration and solution, the final reconstructed spatial spectrum is obtained.

[0017] In some embodiments, the weights of the neural network are shared in each iteration.

[0018] In some embodiments, it also includes:

[0019] The neural network is trained using a training dataset, with the goal of minimizing the mean square error between the array signal and the actual value of the reconstructed spatial spectrum.

[0020] In some embodiments, the neural network is a one-dimensional convolutional neural network.

[0021] Secondly, embodiments of this application provide a network device, including a memory, a transceiver, and a processor;

[0022] A memory for storing computer programs; a transceiver for sending and receiving data under the control of the processor; and a processor for reading the computer programs from the memory and executing the terminal direction finding method as described in the first aspect above.

[0023] Thirdly, embodiments of this application provide a terminal direction-finding device, comprising:

[0024] The preprocessing module is used to preprocess the reference signal sent by the terminal to obtain the virtual spatial spectrum;

[0025] The reconstruction module is used to reconstruct the virtual spatial spectrum using a neural network and a conjugate gradient algorithm to obtain the reconstructed spatial spectrum.

[0026] The search module is used to search for spectral peaks in the reconstructed spatial spectrum and take the angle corresponding to the spectral peak as the angle of arrival.

[0027] In some embodiments, the reconstruction module includes:

[0028] The calibration unit is used to input the virtual spatial spectrum into the neural network and use the neural network to calibrate the spatial spectrum.

[0029] The solver unit is used to solve the calibrated spatial spectrum using the conjugate gradient algorithm to obtain the reconstructed spatial spectrum of one iteration;

[0030] An iterative unit is used to iterate through multiple calibrations and solutions to obtain the final reconstructed spatial spectrum. Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program for causing a computer to execute the terminal direction finding method described in the first aspect above.

[0031] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the terminal direction finding method described in the first aspect above.

[0032] The terminal direction finding method and apparatus provided in this application employ iterative reconstruction of the spatial spectrum using neural network spatial spectrum correction and conjugate gradient algorithm. This allows for AoA estimation even in the presence of hardware impairments, eliminating the need for a large amount of training data and improving the accuracy of terminal direction finding. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic flowchart of the terminal direction finding method provided in the embodiments of this application;

[0035] Figure 2 This is a schematic diagram of a model-driven deep neural network with spatial spectrum correction capability provided in an embodiment of this application;

[0036] Figure 3 This is a schematic diagram of the logic flow of a model-driven deep neural network algorithm with spatial spectrum correction capability provided in an embodiment of this application;

[0037] Figure 4 This is a schematic diagram of the overall iterative structure of a model-driven deep neural network with spatial spectrum correction capability provided in an embodiment of this application;

[0038] Figure 5 This is a schematic diagram of the structure of the convolutional neural network provided in the embodiments of this application;

[0039] Figure 6 This is a schematic diagram of a single iteration structure of a model-driven deep neural network with spatial spectrum correction capability provided in an embodiment of this application;

[0040] Figure 7 This is a comparison chart of spatial spectrum estimation results provided in the embodiments of this application;

[0041] Figure 8 This is a comparison chart of RMSE of algorithms under different SNRs provided in the embodiments of this application;

[0042] Figure 9 This is a comparison chart of RMSE of the algorithm under different damage coefficients provided in the embodiments of this application;

[0043] Figure 10 This is a comparison chart of the standard deviation of the algorithm provided in this application as a function of epochs;

[0044] Figure 11 This is a schematic diagram of the structure of a network device provided in an embodiment of this application;

[0045] Figure 12 This is a schematic diagram of the structure of a terminal direction finding device provided in an embodiment of this application. Detailed Implementation

[0046] In the future development of wireless communication, cutting-edge technological paradigms need to be continuously explored and integrated to generate diverse new applications. As a value-added service, high-precision positioning plays an increasingly crucial role in ensuring a perfect user communication experience and realizing various applications (such as emergency response, asset tracking, and location-based services). Although positioning in wireless communication systems is still in its early stages, the increasing number of 5G mobile base stations (i.e., 5G gNBs) and their large-scale deployment indoors and outdoors provide a platform for the continuous development of communication positioning services. Therefore, from the physical layer to the application layer of communication, the ability to perceive the surrounding environment and provide location-aware services has gained new driving forces. Version 16 of the 3GPP protocol, released in 2021, introduced several 5G positioning schemes, including downlink time difference of arrival (DL-TDoA), uplink time difference of arrival (UL-TDoA), downlink angle of departure (DL-AoD), uplink angle of arrival (UL-AoA), and multiple round-trip time (Multi-RTT) positioning. Among these schemes, UL-AoA is considered to be able to overcome multipath effects and improve positioning performance more effectively because it conforms to the trend of increasing antenna array size and bandwidth.

[0047] Hardware impairments often affect the accuracy of 5G positioning. Traditional purely model-driven AoA estimation algorithms are affected by hardware impairments, which may lead to errors in the array responses used in the estimation process. To solve this problem, purely data-driven methods, for example, direction finding technologies based on Deep Neural Network (DNN), have been extensively studied. Nevertheless, the accuracy of these methods relies heavily on training with a large amount of data, resulting in low feasibility in practical applications. In summary, the main shortcomings of the prior art are that the purely model-driven methods cannot overcome the influence of hardware impairments and cannot perform high-precision AoA estimation; the purely data-driven methods rely on a large amount of training data, cannot be widely applied, and insufficient training data ultimately also leads to low AoA estimation accuracy.

[0048] First, the algorithm principle of the embodiments of the present application is introduced:

[0049] Considering the ideal signal output in the uplink positioning scenario, without loss of generality, the angle space of interest is quantized into a uniformly discrete AoA set θ=[θ1,θ2,…,θ l-1 , θ l , …, θ L T , 1≤l≤L, where L represents the number of quantized AoAs, and the angle interval is Δθ=θ l -θ l-1 ; For a 5G gNB, a uniform linear array with M elements is adopted for omnidirectional reception, and the array spacing is half a wavelength. Assuming that the subcarrier spacing Δf of the OFDM signal is much smaller than f0, where f0 represents the center carrier frequency of the OFDM signal, the expression of the received signal at the t-th time instant, on the m-th element, and for the k-th subcarrier is as follows:

[0050]

[0051] Wherein, s l,k (t) represents the complex waveform of the k-th subcarrier of the sounding reference signal with AoA of θ l , n k (t) represents additive white Gaussian noise (AWGN), s k (t)=[s 1,k (t),…,s l,k (t),…,s L,k (t)] T represents the complex waveform matrix of the sounding reference signal corresponding to the AoA set θ, a(θ l )=[1,…,exp{-jπ(M-1)sinθ l}]​T Let A(θ) represent the guiding vector, and A(θ) = [a(θ1), ..., a(θ)]. l ),…,a(θ L [)] represents the array manifold corresponding to the set θ of AoA. Since the ideal array manifold A in formula (1) is only related to the source angle and array settings, the channel state information (CSI) of the k-th subcarrier can be obtained by fast Fourier transform (FFT), as shown in the following expression:

[0052]

[0053] in, and These represent the signal and noise in the frequency domain, respectively, with h(k) equivalent to a single snapshot signal in the frequency domain. However, due to unavoidable non-ideal factors in the design and manufacturing of millimeter-wave arrays, namely hardware defects, the actual steering vector... The inconsistency between the model and the ideal steering vector a(θ) leads to a decrease in the estimation performance of pure model-driven positioning algorithms in practical 5G communication systems.

[0054] The effects of hardware defects broadly include element position errors, gain-phase inconsistencies, and mutual coupling. In most cases, hardware defects will cause antenna elements to have different phase errors for different AoA values. Considering the phase error in the signal model, the CSI of the received signal in a real system can be expressed by the following expression:

[0055]

[0056] Where ⊙ represents the Hadamard product operation. Indicate AoA as θ l The frequency domain form of the complex waveform of the probe reference signal, γ(θ) l )=[γ1(θ l ),…,γ m (θ l ),…,γ M (θ l )] T , 1≤m≤M, Γ=[γ(θ1),…,γ(θ l ),…,γ(θ L Coefficient γ m (θ l ) represents the m-th element with respect to angle θ l The phase error. Equation (3) shows that hardware defects lead to angle-independent and nonlinear phase errors in the actual system.

[0057] Furthermore, this invention proposes an iterative optimization scheme that utilizes the shared neural network weight coefficients between the neural network and the CG module to establish the overall MoD-DNN framework. Numerical results show that, compared with purely model-driven and purely data-driven methods, this invention achieves better AoA estimation performance and improves the accuracy of terminal direction finding.

[0058] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] Figure 1 This is a flowchart illustrating the terminal direction finding method provided in an embodiment of this application, as shown below. Figure 1 As shown, this application provides a terminal direction finding method, the executing entity of which can be a network device, such as a base station. The method includes:

[0060] Step 101: Preprocess the reference signal sent by the terminal to obtain the virtual spatial spectrum.

[0061] Specifically, the preprocessing of the reference signal sent by the terminal can be performed by first determining its covariance matrix based on the reference signal sent by the terminal, then vectorizing the covariance matrix to obtain the array signal corresponding to the reference signal, and finally using array digital beamforming to estimate the array signal to obtain the virtual spatial spectrum.

[0062] Alternatively, the virtual spatial spectrum can be obtained based on the reference signal sent by the terminal using related technologies, which will not be elaborated in this example.

[0063] Step 102: Reconstruct the virtual spatial spectrum using a neural network and the conjugate gradient algorithm to obtain the reconstructed spatial spectrum.

[0064] Specifically, in the embodiments of this application, the neural network may be a convolutional neural network or a recurrent neural network, etc.

[0065] For example, a one-dimensional convolutional neural network can be used to reduce computational load and improve detection efficiency.

[0066] The virtual spatial spectrum can be input into a neural network, the spatial spectrum can be calibrated using the neural network, and then the conjugate gradient algorithm can be used to solve the calibrated spatial spectrum to obtain the reconstructed spatial spectrum.

[0067] Alternatively, the virtual spatial spectrum can be input into a neural network, the spatial spectrum can be calibrated using the neural network, and then the conjugate gradient algorithm can be used to solve the calibrated spatial spectrum to obtain the reconstructed spatial spectrum obtained in one iteration. After multiple iterations of calibration and solving, the final reconstructed spatial spectrum can be obtained, which can make the obtained reconstructed spatial spectrum more accurate, thereby further improving the accuracy of terminal direction finding.

[0068] The conjugate gradient algorithm can directly solve the calibrated spatial spectrum using a simple conjugate gradient algorithm, which can reduce computational costs. However, the CG algorithm is more suitable for wide beamwidths, which affects its reconstruction capability.

[0069] The sparse conjugate gradient algorithm can also be used to solve the calibrated spatial spectrum, which enhances the reconstruction capability of this method and makes the reconstructed spatial spectrum more accurate, thereby further improving the accuracy of terminal direction finding.

[0070] Step 103: Perform a peak search on the reconstructed spatial spectrum and take the angle corresponding to the peak as the angle of arrival.

[0071] Specifically, after determining the spatial spectrum, a spectral peak search is performed, and the angle corresponding to the spectral peak is taken as the angle of arrival.

[0072] The terminal direction finding method provided in this application enhances the estimation capability of Angle of Arrival (AoA) based on the Model-Driven Deep Neural Network (MoD-DNN) method with spatial spectrum correction capability. This invention first constructs an inverse problem from the virtual spatial spectrum to the sparse solution of the overcomplete AoA set. Then, the virtual spatial spectrum is calibrated using a neural network. Based on the output of the neural network, the conjugate gradient (CG) algorithm is used to recover the sparse AoA spectrum with the closed solution, thus obtaining the reconstructed spatial spectrum. This avoids the phase error of angle independence and nonlinearity caused by hardware defects in the actual system, as explained in formula (3). AoA estimation can be performed even in the presence of hardware damage. Furthermore, this application uses a neural network and the CG algorithm to determine the spatial spectrum, which is different from the pure data-driven method. It does not require a large amount of training data, thus improving the accuracy of terminal direction finding.

[0073] In some embodiments, the reference signal transmitted by the terminal is preprocessed to obtain a virtual spatial spectrum, including:

[0074] Determine the covariance matrix of the reference signal transmitted by the terminal;

[0075] Vectorize the covariance matrix to obtain the array signal corresponding to the reference signal;

[0076] The array signal is estimated using array digital beamforming to obtain the virtual spatial spectrum.

[0077] Specifically, for the received signal with phase error mentioned above, this application embodiment establishes a MoD-DNN framework with spatial spectrum correction capability to enhance the estimation performance of AoA. Figure 2 This is a schematic diagram of a model-driven deep neural network with spatial spectrum correction capability provided in an embodiment of this application, as shown below. Figure 2 As shown, MoD-DNN is mainly divided into two modules: a preprocessing module for generating virtual spatial spectra and a spatial spectrum reconstruction module based on digital-analog collaborative driving.

[0078] Figure 3 This is a schematic diagram of the logic flow of a model-driven deep neural network algorithm with spatial spectrum correction capability provided in an embodiment of this application, as shown below. Figure 3 As shown, the preprocessing step first determines the covariance matrix of the reference signal. Considering the discrete direction set θ, the covariance matrix of the received signal h(k) can be expressed by the following expression:

[0079]

[0080] in, η represents the expected value of the variable. l Indicates direction θ l signal power, I represents the power of the AWGN and I represents the identity matrix.

[0081] Furthermore, after determining the covariance matrix of the reference signal, vectorizing the covariance matrix yields the corresponding array signal y, whose expression is as follows:

[0082]

[0083] in, Let A represent an array manifold, 1 ≤ m ≤ M, and A m =[a(θ1)a H (θ1)e m ,…,a(θ L )a H (θ L )e m It is worth noting that η = [η1,…,η] l ,…,η L [It can be considered as a sparse space spectrum, 1≤l≤L,] It is a transition vector. It is a vector where the m-th element is 1 and all other elements are zero. It can be estimated using array digital beamforming to obtain the virtual spatial spectrum, expressed as follows:

[0084]

[0085] Note that formula (6) is The equivalent form, that is, the original problem is equivalent to using the prior projection matrix. From the observed virtual space spectrum The solution η is obtained by solving for η. However, hardware defects in actual communication systems lead to inconsistencies between the actual projection matrix and the ideal projection matrix. To address this issue, embodiments of this application propose a method using deep learning networks and model-driven algorithm iterations for spatial spectrum correction.

[0086] The terminal direction finding method provided in this application adopts the scheme in related technologies. First, it determines the covariance matrix based on the reference signal sent by the terminal. Then, it vectorizes the covariance matrix to obtain the array signal corresponding to the reference signal. Finally, it uses array digital beamforming to estimate the array signal to obtain the virtual spatial spectrum. There is no need to improve the preprocessing method separately, which can reduce the complexity of the system.

[0087] In some embodiments, the virtual spatial spectrum is reconstructed using a neural network and a conjugate gradient algorithm to obtain a reconstructed spatial spectrum, including:

[0088] The virtual spatial spectrum is input into a neural network, and the neural network is used to calibrate the spatial spectrum.

[0089] The conjugate gradient algorithm is used to solve the calibrated spatial spectrum to obtain the reconstructed spatial spectrum of one iteration;

[0090] After multiple iterations of calibration and solution, the final reconstructed spatial spectrum is obtained.

[0091] Specifically, to improve the estimation performance of AoA, spatial spectrum reconstruction can be expressed as the following optimization problem:

[0092]

[0093] in, This is the corrected version of the spatial spectrum, where λ and ε are regularization coefficients associated with the CNN calibrator. w This is the regularization term representing noise and phase error, which can be learned using a CNN with weights w. The optimization problem in equation (7) is equivalent to substitution optimization, which can be expressed by the following expression:

[0094]

[0095]

[0096] The terminal direction finding method provided in this application, after multiple iterations of calibration and solution, yields the final reconstructed spatial spectrum, which can make the obtained reconstructed spatial spectrum more accurate, thereby further improving the accuracy of terminal direction finding.

[0097] Figure 4This is a schematic diagram of the overall iterative structure of a model-driven deep neural network with spatial spectrum correction capability provided in an embodiment of this application, as shown below. Figure 4 As shown, the AoA resolution of the virtual spatial spectrum before iteration is very low, and the peaks have shifted. The model-computer interface (MCI) collaborative spatial spectrum reconstruction consists of two parts: (a) a Sparse Conjugate Gradient (SCG) algorithm module; and (b) a CNN-based spatial spectrum calibration module, forming an iteration of the neural network and conjugate gradient algorithm, i.e., a model-driven deep neural network (MoD-DNN) algorithm. In each iteration, the neural network weights are shared.

[0098] It should be noted that the number of iterations for calibration and solution in the embodiments of this application can be determined as needed, for example, Figure 4 In this method, the number of iterations is set to I. A larger value of I results in a more complex system, which can improve the accuracy of the reconstructed spatial spectrum to some extent, but reduces processing efficiency. Conversely, a smaller value of I leads to higher processing efficiency, but affects the accuracy of the reconstructed spatial spectrum. The specific value of I can be determined based on the equipment configuration and experimental results.

[0099] The terminal direction finding method provided in this application shares the weights of the neural network in each iteration, which can reduce the complexity of the system and improve the efficiency of terminal direction finding.

[0100] In some embodiments, the spatial spectrum reconstruction problem in equation (9) can be solved using a forward model, and the Conjugate Gradient (CG) algorithm can effectively solve this problem. All steps of the CG algorithm have closed-form solutions, indicating that no parameters require additional training, and gradients can be backpropagated from the CG word modules during iteration. The CG algorithm enables MoD-DNN to have modules composed of numerically optimized layers, reducing computational costs. Although the CG algorithm can solve the problem efficiently, the initial form of the normal equation is... in The array is an overcomplete manifold associated with the discrete AoA set θ. This implies that the inverse problem is underdetermined, making the CG algorithm more suitable for wide beams. To further enhance the reconstruction capability of the CG method, this application proposes a sparsity-corrected SCG algorithm. The pseudocode for the SCG-based spatial spectrum reconstruction algorithm is as follows:

[0101]

[0102] In this application embodiment, a neural network structure is used as the spectral corrector in formula (8), for example, a 1D-CNN structure is used. Figure 5 This is a schematic diagram of the structure of the convolutional neural network provided in the embodiments of this application, such as... Figure 5 As shown, 1D-CNN has a total of P layers, and the output of the p-th layer is represented by c. p This is indicated. First, consider the training dataset. The network is trained on the dataset, where D represents the number of training pairs. The training objective is to minimize the mean square error (MSE) between the array signal y and the real value η.

[0103]

[0104] The terminal direction finding method provided in this application uses a one-dimensional convolutional neural network to reduce computational load and improve detection efficiency.

[0105] Figure 6 This is a schematic diagram of a first-order iterative structure of a model-driven deep neural network with spatial spectrum correction capability provided in an embodiment of this application, as shown below. Figure 6 As shown, the virtual space spectrum generated by formula (6) The input is fed into the CNN. Then, using... Figure 5 The calibrated spatial spectrum of the P-layer 1D-CNN network shown is given by c0 = η in the i-th iteration. i-1 The calibrated spatial spectrum is the output z of the P-layer network. i-1 =c P Finally, z i-1 The input is fed into the SCG algorithm to solve the sparse solution η. i After the iterations between the CNN module and the SCG module are complete, the reconstructed spatial spectrum η will be... I As the output of the network.

[0106] The terminal direction finding method provided in this application, after multiple iterations of calibration and solution, yields the final reconstructed spatial spectrum, which can make the obtained reconstructed spatial spectrum more accurate, thereby further improving the accuracy of terminal direction finding.

[0107] The following simulation data illustrates the effectiveness of this application:

[0108] The simulation experiment setup is shown in Table 1. The user equipment (UE) transmits a sounding reference signal (SRS) to the 5G gNB base station at a distance of 10m from the 5G gNB base station, within the AoA range of [-60°, 60°] with an interval of 0.1°. The base station uses a uniform linear array with four antennas, an antenna spacing of half a wavelength, a carrier frequency of 4.8498GHz, and a subcarrier spacing of 60kHz. 60 samples are generated at each location, for a total of 60 × 1201 = 72060 samples, of which 40 × 1201 = 48040 samples are used as training samples. This example demonstrates the effectiveness of the MoD-DNN algorithm in AoA estimation under hardware impairment conditions by conducting simulation experiments using four methods: MUSIC, DeepMUSIC, CNN, and the MoD-DNN algorithm of this invention.

[0109] Table 1 Simulation Experiment Parameter Settings

[0110]

[0111] The spatial spectrum of the CSI generated by the 5G channel simulator is compared with that of the signal-to-noise ratio (SNR) of 10dB. Figure 7 This is a comparison chart of spatial spectrum estimation results provided in the embodiments of this application, such as... Figure 7 As shown, MUSIC, DeepMUSIC, and MoD-DNN can all generate single-peak spatial spectra corresponding to AoA. However, due to hardware impairments, the AoA estimated by MUSIC and DeepMUSIC deviates significantly from the true value. In contrast, the spatial spectrum corrected by MoD-DNN has the sharpest spectral peak and can accurately estimate AoA.

[0112] To further compare the estimation accuracy, the root-mean-square error (RMSE) results of AoA estimation were compared between MUSIC, DeepMUSIC, CNN, and the MoD-DNN of this invention when the SNR ranged from -5dB to 10dB. Figure 8 This is a comparison chart of RMSE of algorithms under different SNRs provided in the embodiments of this application, such as... Figure 8 As shown, the RMSE of the MUSIC and DeepMUSIC algorithms remains almost unchanged under different SNRs, indicating that these two algorithms are statistically ineffective for AoA estimation when hardware impairments exist. In contrast, the RMSE of both CNN and MoD-DNN decreases with increasing SNR, and the estimation performance of the MoD-DNN proposed in this invention is superior to that of CNN. This is because, compared to purely data-driven CNNs, MoD-DNN can utilize the knowledge of the signal model more effectively, and this advantage is more pronounced when the SNR is large.

[0113] To evaluate the effectiveness of the algorithm in defect correction, the RMSE results of AoA estimation were compared between MUSIC, DeepMUSIC, CNN and the MoD-DNN of this invention when the hardware defect coefficient ranged from 0 to 1. Figure 9 This is a comparison chart of RMSE of the algorithm under different damage coefficients provided in the embodiments of this application, such as... Figure 9 As shown, the MUSIC and DeepMUSIC algorithms have low RMSE when the defect coefficient is small, but the RMSE increases with the increase of the defect coefficient. In contrast, the RMSE of the CNN and MoD-DNN algorithms fluctuates only slightly. This is because the MoD-DNN of this invention successfully overcomes the influence of hardware impairment by utilizing the iteration of neural networks and the SCG algorithm, and its performance is superior to that of CNN.

[0114] Finally, by comparing the convergence speeds of DeepMUSIC, CNN, and the MoD-DNN method of this invention, standard deviation was used as the evaluation metric. It is worth noting that the convergence values ​​of each algorithm were used as the expected value of the calculated standard deviation. Figure 10 This is a comparison chart of the standard deviation of the algorithm provided in this application as a function of epochs, such as... Figure 10 As shown, the standard deviation convergence speed of the MoD-DNN algorithm of this invention is faster than that of DeepMUSIC and CNN, which demonstrates the advantage of MoD-DNN in training efficiency.

[0115] The terminal direction finding method provided in this application uses neural network spatial spectrum correction and conjugate gradient algorithm to iteratively reconstruct the spatial spectrum, which can perform AoA estimation in the presence of hardware impairment, without the need for a large amount of training data, thus improving the accuracy of terminal direction finding.

[0116] Figure 11 This is a schematic diagram of the structure of a network device provided in an embodiment of this application, such as... Figure 11 As shown, the network device includes a memory 1120, a transceiver 1100, and a processor 1110, wherein:

[0117] The memory 1120 is used to store computer programs; the transceiver 1100 is used to send and receive data under the control of the processor 1110; the processor 1110 is used to read the computer program in the memory 1120 and perform the following operations:

[0118] The reference signal sent by the terminal is preprocessed to obtain the virtual spatial spectrum;

[0119] The virtual spatial spectrum is reconstructed using a neural network and the conjugate gradient algorithm to obtain the reconstructed spatial spectrum;

[0120] A peak search is performed on the reconstructed spatial spectrum, and the angle corresponding to the peak is taken as the angle of arrival.

[0121] Among them, Figure 11 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 1110) and memory (memory 1120). The bus architecture can also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 1100 can be multiple elements, including transmitters and receivers, providing units for communicating with various other devices over transmission media, including wireless channels, wired channels, optical fibers, etc. The processor 1110 is responsible for managing the bus architecture and general processing, and the memory 1120 can store data used by the processor 1110 during operation.

[0122] The processor 1110 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD). The processor can also adopt a multi-core architecture.

[0123] In some embodiments, the reference signal transmitted by the terminal is preprocessed to obtain a virtual spatial spectrum, including:

[0124] Determine the covariance matrix of the reference signal transmitted by the terminal;

[0125] Vectorize the covariance matrix to obtain the array signal corresponding to the reference signal;

[0126] The array signal is estimated using array digital beamforming to obtain the virtual spatial spectrum.

[0127] In some embodiments, the virtual spatial spectrum is reconstructed using a neural network and a conjugate gradient algorithm to obtain a reconstructed spatial spectrum, including:

[0128] The virtual spatial spectrum is input into a neural network, and the neural network is used to calibrate the spatial spectrum.

[0129] The conjugate gradient algorithm is used to solve the calibrated spatial spectrum to obtain the reconstructed spatial spectrum of one iteration;

[0130] After multiple iterations of calibration and solution, the final reconstructed spatial spectrum is obtained.

[0131] In some embodiments, the weights of the neural network are shared in each iteration.

[0132] In some embodiments, it also includes:

[0133] The neural network is trained using a training dataset, with the goal of minimizing the mean square error between the array signal and the actual value of the reconstructed spatial spectrum.

[0134] In some embodiments, the neural network is a one-dimensional convolutional neural network.

[0135] Specifically, the network device provided in this application embodiment can implement all the method steps implemented by the method embodiment with the network device as the execution subject, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0136] Figure 12 This is a schematic diagram of the structure of a terminal direction finding device provided in an embodiment of this application, as shown below. Figure 12 As shown, this application embodiment provides a terminal direction finding device, including:

[0137] The preprocessing module 1201 is used to preprocess the reference signal sent by the terminal to obtain a virtual spatial spectrum;

[0138] Reconstruction Module 1202 is used to reconstruct the virtual spatial spectrum using a neural network and a conjugate gradient algorithm to obtain the reconstructed spatial spectrum.

[0139] The search module 1203 is used to perform spectral peak search on the reconstructed spatial spectrum and take the angle corresponding to the spectral peak as the angle of arrival.

[0140] In some embodiments, the reconstruction module includes:

[0141] The calibration unit is used to input the virtual spatial spectrum into the neural network and use the neural network to calibrate the spatial spectrum.

[0142] The solver unit is used to solve the calibrated spatial spectrum using the conjugate gradient algorithm to obtain the reconstructed spatial spectrum of one iteration;

[0143] The iterative unit is used to obtain the final reconstructed spatial spectrum after multiple iterations of calibration and solution.

[0144] Specifically, the terminal direction finding device provided in this application embodiment can implement all the method steps implemented by the method embodiment where the execution subject is a network device, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0145] It should be noted that the division of units / modules in the above embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0146] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0147] In some embodiments, a non-transiently readable storage medium is also provided, the non-transiently readable storage medium storing a computer program for causing a processor to execute the terminal direction finding method provided in the above method embodiments.

[0148] Specifically, the non-transiently readable storage medium provided in this application embodiment can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0149] It should be noted that the non-transiently readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0150] In some embodiments, a processor-readable storage medium is also provided, the processor-readable storage medium storing a computer program for causing a processor to execute the terminal direction finding method provided in the above method embodiments.

[0151] Specifically, the processor-readable storage medium provided in this application embodiment can implement all the method steps implemented in the above method embodiments and achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0152] In some embodiments, a computer-readable storage medium is also provided, the computer-readable storage medium storing a computer program for causing a computer to execute the terminal direction finding method provided in the above method embodiments.

[0153] Specifically, the computer-readable storage medium provided in the embodiments of this application can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0154] In some embodiments, a communication device is also provided, wherein the communication device stores a computer program for causing the communication device to execute the terminal direction finding method provided in the above-described method embodiments.

[0155] Specifically, the communication device provided in this application embodiment can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0156] In some embodiments, a chip product is also provided, wherein the chip product stores a computer program for causing the chip product to execute the terminal direction finding method provided in the above-described method embodiments.

[0157] Specifically, the chip product provided in this application embodiment can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0158] It should also be noted that the terms "first," "second," etc., used in the embodiments of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, and the number of objects is not limited. For example, the first object can be one or more.

[0159] In the embodiments of this application, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0160] In this application's embodiments, "determine B based on A" means that factor A must be considered when determining B. It is not limited to "B can be determined based solely on A," but should also include: "determine B based on A and C," "determine B based on A, C, and E," "determine C based on A, and further determine B based on C," etc. Additionally, it can include using A as a condition for determining B, for example, "when A meets the first condition, determine B using the first method"; another example, "when A meets the second condition, determine B," etc.; another example, "when A meets the third condition, determine B based on the first parameter," etc. Of course, it can also be a condition where A is a factor in determining B, for example, "when A meets the first condition, determine C using the first method, and further determine B based on C," etc.

[0161] In the embodiments of this application, the term "multiple" refers to two or more, and other quantifiers are similar.

[0162] The technical solutions provided in this application can be applied to various systems, especially 5G systems. For example, applicable systems include Global System for Mobile Communication (GSM), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA) General Packet Radio Service (GPRS), Long Term Evolution (LTE), LTE Frequency Division Duplex (FDD), LTE Time Division Duplex (TDD), Long Term Evolution Advanced (LTE-A), Universal Mobile Telecommunication System (UMTS), Worldwide Interoperability for Microwave Access (WiMAX), and 5G New Radio (NR). All of these systems include terminal equipment and network equipment. The systems may also include a core network component, such as Evolved Packet System (EPS) and 5G systems (5GS).

[0163] The terminal devices involved in the embodiments of this application can be devices that provide voice and / or data connectivity to users, handheld devices with wireless connectivity, or other processing devices connected to a wireless modem. The names of the terminal devices may differ in different systems; for example, in a 5G system, a terminal device can be called User Equipment (UE). Wireless terminal devices can communicate with one or more core networks (CNs) via a Radio Access Network (RAN). Wireless terminal devices can be mobile terminal devices, such as mobile phones (or "cellular" phones) and computers with mobile terminal devices, for example, portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile devices that exchange voice and / or data with the RAN. Examples include Personal Communication Service (PCS) phones, cordless phones, Session Initiated Protocol (SIP) phones, Wireless Local Loop (WLL) stations, and Personal Digital Assistants (PDAs). Wireless terminal equipment can also be referred to as a system, subscriber unit, subscriber station, mobile station, mobile station, remote station, access point, remote terminal, access terminal, user terminal, user agent, or user device, but is not limited to these terms in the embodiments of this application.

[0164] The network device involved in this application embodiment can be a base station, which may include multiple cells providing services to terminals. Depending on the specific application, a base station may also be called an access point, or a device in an access network that communicates with a wireless terminal device through one or more sectors on the air interface, or other names. The network device can be used to exchange received air frames with Internet Protocol (IP) packets, acting as a router between the wireless terminal device and the rest of the access network, where the rest of the access network may include an Internet Protocol (IP) communication network. The network device can also coordinate the attribute management of the air interface. For example, the network equipment involved in the embodiments of this application can be a base transceiver station (BTS) in a Global System for Mobile communications (GSM) or Code Division Multiple Access (CDMA), a NodeB in a Wide-band Code Division Multiple Access (WCDMA) system, an evolved Node B (eNB or e-NodeB) in a long term evolution (LTE) system, a 5G base station (gNB) in a next generation system, a Home evolved Node B (HeNB), a relay node, a femto, a pico, etc., and is not limited in the embodiments of this application. In some network structures, the network equipment may include centralized unit (CU) nodes and distributed unit (DU) nodes, and the centralized unit and distributed unit may be geographically separated.

[0165] Network devices and terminal devices can each use one or more antennas for multiple-input multiple-output (MIMO) transmission. MIMO transmission can be single-user MIMO (SU-MIMO) or multiple-user MIMO (MU-MIMO). Depending on the configuration and number of antenna combinations, MIMO transmission can be 2D-MIMO, 3D-MIMO, FD-MIMO, or massive-MIMO, and can also be diversity transmission, precoding transmission, or beamforming transmission, etc.

[0166] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0167] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-executable instructions. These computer-executable instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0168] These processor-executable instructions may also be stored in a processor-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the processor-readable memory produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0169] These processors can execute instructions that can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0170] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A terminal direction finding method, characterized in that, include: The reference signal sent by the terminal is preprocessed to obtain the virtual spatial spectrum; The virtual spatial spectrum is reconstructed using a neural network and the conjugate gradient algorithm to obtain the reconstructed spatial spectrum; A peak search is performed on the reconstructed spatial spectrum, and the angle corresponding to the peak is taken as the angle of arrival. The virtual spatial spectrum is reconstructed using a neural network and the conjugate gradient algorithm to obtain the reconstructed spatial spectrum, including: The virtual spatial spectrum is input into a neural network, and the neural network is used to calibrate the spatial spectrum. The conjugate gradient algorithm is used to solve the calibrated spatial spectrum to obtain the reconstructed spatial spectrum of one iteration; After multiple iterations of calibration and solution, the final reconstructed spatial spectrum is obtained.

2. The terminal direction finding method according to claim 1, characterized in that, The reference signal transmitted by the terminal is preprocessed to obtain a virtual spatial spectrum, including: Determine the covariance matrix of the reference signal transmitted by the terminal; Vectorize the covariance matrix to obtain the array signal corresponding to the reference signal; The array signal is estimated using array digital beamforming to obtain the virtual spatial spectrum.

3. The terminal direction finding method according to claim 1, characterized in that, In each iteration, the weights of the neural network are shared.

4. The terminal direction finding method according to claim 1, characterized in that, Also includes: The neural network is trained using a training dataset, with the goal of minimizing the mean square error between the array signal and the actual value of the reconstructed spatial spectrum.

5. The terminal direction finding method according to claim 4, characterized in that, The neural network is a one-dimensional convolutional neural network.

6. A terminal direction finding device, characterized in that, include: The preprocessing module is used to preprocess the reference signal sent by the terminal to obtain the virtual spatial spectrum; The reconstruction module is used to reconstruct the virtual spatial spectrum using a neural network and a conjugate gradient algorithm to obtain the reconstructed spatial spectrum. The search module is used to search for spectral peaks in the reconstructed spatial spectrum and take the angle corresponding to the spectral peak as the angle of arrival. The reconstruction module includes: The calibration unit is used to input the virtual spatial spectrum into the neural network and use the neural network to calibrate the spatial spectrum. The solver unit is used to solve the calibrated spatial spectrum using the conjugate gradient algorithm to obtain the reconstructed spatial spectrum of one iteration; The iterative unit is used to obtain the final reconstructed spatial spectrum after multiple iterations of calibration and solution.

7. A network device, characterized in that, Includes memory, transceiver, and processor; Memory, used to store computer programs; Transceiver, used to send and receive data under the control of the processor; A processor for reading a computer program from the memory and executing the terminal direction finding method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that causes a computer to perform the terminal direction finding method according to any one of claims 1 to 5.

Images