Near-field terahertz large-scale MIMO target positioning method under non-uniform noise

Abstract

The invention discloses a near-field terahertz large-scale MIMO target positioning method under non-uniform noise, and the method comprises the steps: building a receiving end system model based on an array-of-sub-array (AoSA) mixed structure, and obtaining a signal transmitted by a terahertz signal source; the method comprises the following steps: preprocessing a received signal, and obtaining full-array information from a few radio frequency chains by increasing time slots; calculating a covariance matrix of full array information, and decomposing the data covariance matrix to obtain a diagonal matrix containing noise power; the method comprises the following steps of: selecting a minimum value in diagonal elements to replace other diagonal elements to obtain a reconstructed data covariance matrix, and processing the reconstructed data covariance matrix by using a multi-signal classification (MUSIC) method to obtain an angle and a distance of a terahertz signal source. According to the method, the non-uniform noise interference can be effectively suppressed with relatively low calculation complexity, the positioning precision is improved, and the hardware cost can be reduced by increasing time slots.

Description

Technical Field

[0001] The invention belongs to the technical field of digital communications, and in particular relates to a near-field terahertz large-scale MIMO target positioning method under non-uniform noise. Background Art

[0002] The terahertz band, with its rich spectrum resources, supports ultra-high-speed data transmission and brings great potential to the sixth-generation wireless communication network. The hybrid beamforming structure based on the subarray structure has the advantages of low hardware complexity and high energy efficiency, and has attracted special attention in terahertz massive MIMO systems. For near-field terahertz mMIMO systems, when the transmission distance is shorter than the Fraunhofer distance, the near-field wavefront is spherical, so the array steering vector of the system is not only related to the angle, but also to the communication distance between the transmitter and the receiver. In the near-field environment, high-precision directional arrival angle estimation and distance estimation are key links to ensure the reliability of terahertz sensing, communication, sonar and other electronic reconnaissance.

[0003] Most of the current research is done in a Gaussian noise environment, but in actual scenarios, when the sensor noise power is not the same throughout the array, spatially white non-uniform noise will be generated, which will cause the noise covariance matrix to be in a diagonal form but with different diagonal elements, causing the positioning performance of conventional subspace methods to drop sharply, making it difficult to meet the requirements of electronic reconnaissance reliability. Therefore, it is very important to achieve accurate positioning of terahertz signals in the near field area under a non-uniform noise environment. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a near-field terahertz large-scale MIMO target positioning method under non-uniform noise in view of the deficiencies of the above-mentioned prior art, which can effectively suppress the interference of non-uniform noise with lower computational complexity, improve the positioning accuracy, and reduce the hardware cost by increasing the time slot.

[0005] In order to achieve the above technical purpose, the technical solution adopted by the present invention is:

[0006] A near-field terahertz massive MIMO target positioning method under non-uniform noise, comprising:

[0007] Step 1: Establish a receiving end system model based on the terahertz sub-array hybrid structure to obtain the signal sent by the terahertz signal source;

[0008] Step 2: Preprocess the received signal to obtain full array information from a few RF chains by adding time slots;

[0009] Step 3: Calculate the covariance matrix of the full array information and decompose the data covariance matrix to obtain a diagonal matrix containing the noise power;

[0010] Step 4: Select the minimum value among the diagonal elements to replace the remaining diagonal elements to obtain the reconstructed data covariance matrix, and use the multiple signal classification method to process the reconstructed data covariance matrix to obtain the angle and distance of the terahertz signal source;

[0011] In step 1, a massive MIMO channel model and a near-field model are used to establish a receiving end system model based on a terahertz subarray hybrid structure, wherein the massive MIMO channel model is

[0012]

[0013] Where a(θ l,r [k],r l [k]) and b(θ l,t [k],r l [k]) represent the array steering vectors for reception and transmission, respectively, and α l [f,r,k] represents the complex gain of the lth path at carrier frequency f and communication distance r, A[k] and B[k] are the array manifold matrices of the receiving and transmitting ends respectively, and Λ[k] is the multipath channel complex gain matrix;

[0014] k is the time slot index, θ l,r [k] and θ l,t [k] represents the incident angle and emission angle of the lth path, r l [k] is the communication distance of the lth path, and 1≤l≤L, where L is the multipath number of the terahertz massive MIMO channel.

[0015] The received signal is expressed as:

[0016] Y=W H H[f,r,k]Fs+W H N=W H A[k]X[k]+W H N

[0017] Where X[k] is defined as the equivalent transmitted signal in the kth time slot, and N is the non-uniform noise;

[0018] W is the hybrid beamforming matrix of the AoSA system, and s is the pilot symbol vector sent by the pre-encoder F.

[0019] In step 2, a hybrid beamforming matrix of T time slots randomly generated is used to collect full array data from a few RF chains for estimating near-field signal position parameters. Then, the received signal of the τth time slot is:

[0020]

[0021] Among them, 1≤τ≤T.

[0022] Set all Y τ Stacking into a matrix gives the total observation matrix Right now:

[0023]

[0024] in,

[0025] The step 3 comprises:

[0026] 1) Calculate the covariance matrix of the total observation matrix:

[0027]

[0028] Where P and Q are the covariance matrices of signal X and noise N respectively;

[0029] 2) Represent the covariance matrix R of the total observation matrix as two matrices R 1 and R 2 sum:

[0030] R=R 1 +R 2

[0031] Among them, the matrix R 2 It is composed of the diagonal elements of the covariance matrix R. 1 It is composed of the remaining elements of the covariance matrix R;

[0032] The step 4 comprises:

[0033] 1) Select the matrix R 2 The minimum element r in min , and use r min Alternative R 2 All the diagonal elements in

[0034]

[0035] 2) Reconstruct the covariance matrix of the total observation matrix:

[0036]

[0037] 3) R w Eigendecomposition:

[0038]

[0039] where Σ S and Σ N They are respectively w The signal and noise diagonal matrix composed of descending eigenvalues, U Sand U N are the subspaces corresponding to the signal and noise diagonal matrices respectively;

[0040] The 2D-MUSIC spectrum function of the terahertz near-field source is obtained as:

[0041]

[0042] Finally, the L highest peaks correspond to the target positions of the near-field source signal.

[0043] The beneficial effects of the present invention are:

[0044] In actual scenarios, when the sensor noise power is not the same throughout the array, spatially white non-uniform noise will be generated, which will cause the noise covariance matrix to be in diagonal form but the diagonal elements are different, resulting in a sharp drop in the positioning performance of the conventional subspace method. To address this problem, the present invention proposes a near-field terahertz large-scale MIMO target positioning method under non-uniform noise. The present invention establishes a receiving end system model based on a hybrid structure of a terahertz subarray structure to obtain a signal sent by a terahertz signal source; pre-processes the received signal, and obtains full array information from a few radio frequency chains by increasing time slots; calculates the covariance matrix of the full array information, and decomposes the data covariance matrix to obtain a diagonal matrix containing noise power; selects the minimum value of the diagonal elements to replace the remaining diagonal elements to obtain a reconstructed data covariance matrix, and uses a multiple signal classification method to process the reconstructed data covariance matrix to obtain the angle and distance of the terahertz signal source. The method of the present invention can effectively suppress non-uniform noise interference with low computational complexity, improve positioning accuracy, and reduce hardware costs by increasing time slots. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a diagram of the architecture of a terahertz mMIMO system with an AoSA hybrid structure for parameter estimation according to the present invention;

[0046] Figure 2 is a flow chart of the method of the present invention;

[0047] Figure 3 This is a curve diagram showing the angle root mean square error versus signal-to-noise ratio variation of the present invention;

[0048] Figure 4 This is a curve diagram showing the change of the distance root mean square error with the signal-to-noise ratio according to the present invention;

[0049] Figure 5 is a curve diagram showing the change of the angle root mean square error with the radio frequency chain and the time slot of the present invention;

[0050] Figure 6This is a curve diagram of the distance root mean square error changing with the radio frequency chain and time slot of the present invention. DETAILED DESCRIPTION

[0051] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0052] The present invention proposes a near-field terahertz large-scale MIMO target positioning method under non-uniform noise, comprising establishing a receiving end system model based on a terahertz subarray structure hybrid structure to obtain a signal sent by a terahertz signal source; preprocessing the received signal to obtain full array information from a small number of radio frequency chains by adding time slots; calculating the covariance matrix of the full array information, and decomposing the data covariance matrix to obtain a diagonal matrix containing noise power; selecting the minimum value of the diagonal elements to replace the remaining diagonal elements to obtain a reconstructed data covariance matrix, and using a multiple signal classification method to process the reconstructed data covariance matrix to obtain the angle and distance of the terahertz signal source.

[0053] like Figure 2 As shown, the target positioning method proposed by the present invention has the following steps:

[0054] Step 1: Define the number of transmitting antennas and receiving antennas as N t and N r , N r ×N t The terahertz time-varying massive MIMO channel matrix H[f,r,k] is:

[0055]

[0056] Where a(θ l,r [k],r l [k]) and b(θ l,t [k],r l [k]) represent the array steering vectors for reception and transmission, respectively, and α l [f,r,k] represents the complex gain of the lth path at carrier frequency f and communication distance r, A[k] and B[k] are the array manifold matrices of the receiving and transmitting ends respectively, and Λ[k] is the multipath channel complex gain matrix;

[0057] k is the time slot index, θ l,r [k] and θ l,t [k] represents the incident angle and emission angle of the lth path, r l [k] is the communication distance of the lth path, and 1≤l≤L, where L is the multipath number of the terahertz massive MIMO channel.

[0058] The use of high terahertz frequencies and large antenna arrays means that plane wave propagation does not hold for short-range targets in the near-field region. In the range of , where D is the array aperture, λ = c / f 0 is the wavelength, and the near-field wavefront is spherical. For a uniform linear array, the array aperture D = (N r -1)d, where the antenna spacing d = λ / 4. In a terahertz system, if the signal distance r l <d F , a near-field signal model must be used.

[0059] like Figure 1 As shown, considering the spherical wave model, in the r The corresponding DOAθ is a symmetrical uniform linear array composed of antennas. l [k] and distance r l [k] defines the near-field array steering vector:

[0060]

[0061] Among them, γ l [k]=-2πdsinθ l [k] / λ,φ l [k] = πd 2 cos 2 θ l [k] / (λr l [k]).

[0062] like Figure 1 As shown. The system configuration N r antennas and N RF RF chains, such that each RF chain is connected to J = N r / N RF The antennas form a subarray. Therefore, the signal vector received at the receiving end can be expressed as:

[0063] Y=W H H[f,r,k]Fs+W H N=W H A[k]X[k]+W H N

[0064] Where X[k] is defined as the equivalent transmitted signal in the kth time slot, N is the non-uniform noise, W is the hybrid beamforming matrix of the AoSA system, and s is the pilot codeword vector sent by the pre-encoder F.

[0065] Step 2: Since N RF <N r After hybrid beamforming, the size of the received signal changes from N r ×1 is reduced to N RF × 1. Therefore, a randomly generated hybrid beamforming matrix of T time slots is used from NRF The RF chain collects full array data to estimate the near-field signal position parameters. In the terahertz system, the symbol time of the signal is in the order of picoseconds, so it can be assumed that the target position remains unchanged within the time slot. The hybrid beamforming matrix is ​​represented as W τ , where 1≤τ≤T, then the received signal of the τth time slot is:

[0066]

[0067] Set all Y τ Stacking into a matrix gives the total observation matrix Right now:

[0068]

[0069] in,

[0070] Step 3: Calculate the covariance matrix of the total observation matrix:

[0071]

[0072] Where P and Q are the covariance matrices of signal X and noise N respectively;

[0073] In order to reduce the impact of non-uniform noise on parameter estimation, the covariance matrix R of the total observation matrix is ​​first expressed as two matrices R 1 and R 2 sum:

[0074] R=R 1 +R 2

[0075] in,

[0076] Step 4: Matrix R 2 It is composed of the diagonal elements of the covariance matrix R and is a real matrix, so the non-uniform noise only affects R 2 In order to reduce this influence, the matrix R is selected 2 The minimum element r in min , and use r min Alternative R 2 All the diagonal elements in , we get:

[0077]

[0078] For the matrix R 2 The reconstruction of is essentially the pre-whitening of the covariance matrix R, which can effectively eliminate the influence of non-uniform noise on parameter estimation. In addition, replacing the diagonal elements with the minimum value can reduce the noise power, thereby equivalently improving the signal-to-noise ratio. The reconstructed diagonal matrix R is obtained2 Then, the covariance matrix of the reconstructed total observation matrix is ​​obtained, that is,

[0079]

[0080] R w Eigendecomposition:

[0081]

[0082] where Σ S and Σ N They are respectively w The signal and noise diagonal matrix composed of descending eigenvalues, U S and U N are the subspaces corresponding to the signal and noise diagonal matrices, respectively, and the signal subspace is orthogonal to the noise subspace, that is, U S ⊥U N ,get:

[0083]

[0084] The 2D-MUSIC spectrum function of the terahertz near-field source is obtained as:

[0085]

[0086] Finally, the L highest peaks correspond to the target positions of the near-field source signal.

[0087] The present invention has completed MATLAB software simulation and verification.

[0088] Set the parameter f c =300GHz, N r=235, d=λ / 4, L=2, and the search step size is set to Δ=0.01. The comparison methods are 2D-MUSIC method (Huang YD, Barkat M. Near-field multiple source localization by passive sensor array [J]. IEEE Transactions on Antennas and Propagation, 1991, 39 (7): 968–975), NCME method (WU Y, HOU C, LIAO G, et al. Direction-of-arrival estimation in the presence of unknown nonuniform noise fields [J]. IEEE Journal of Oceanic Engineering, 2006, 31 (2): 504–510), NISM method (Esfandiari M, Vorobyov SA, Alibani S, et al. Noniterative subspace-based DOA estimation in the presence of nonuniform noise [J]. IEEE Signal Processing Letters, 2019, 26 (6): 848-852) and CRLB.

[0089] Experiment 1: Set the number of snapshots to 200, the number of RF chains to 15, and the two signal source parameters to (θ 1 ,r 1 )=(5.576°,0.212m) and (θ 2 ,r 2 )=(15.428°,0.536m). Figure 3 and Figure 4 The relationship between the angle and distance root mean square error of the received signal and the signal-to-noise ratio when using 1 time slot is shown respectively. Figure 3 and Figure 4 It can be seen that the RMS errors of the four methods decrease with the increase of SNR, and the method proposed in this invention is obviously better than the other three methods and close to CRLB.

[0090] Experiment 2: Figure 5 and Figure 6The following are the relationship between the angle and distance root mean square error and the signal-to-noise ratio under different numbers of radio frequency chains and different time slots according to the method proposed in the present invention, where the number of snapshots is 200. The results show that a smaller root mean square error can be obtained by increasing the number of radio frequency chains or the number of time slots. Specifically, when SNR = 5dB and T = 8, the number of radio frequency chains N RF From 9 to 15, the RMS error of DOA estimation is reduced by 0.94dB, and the RMS error of distance estimation is reduced by 0.72dB; when SNR = 10dB, N RF = 9, the time slot T increases from 2 to 8, the RMS error of angle estimation decreases by 4.13 dB, and the RMS error of distance estimation decreases by 3.81 dB. Therefore, for the same positioning accuracy, the hardware overhead can be reduced by increasing the number of time slots and reducing the number of RF chains.

[0091] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A near-field terahertz massive MIMO target positioning method under non-uniform noise, characterized in that: include: Step 1: Establish a receiving end system model based on the terahertz sub-array hybrid structure to obtain the signal sent by the terahertz signal source; Step 2: Preprocess the received signal to obtain full array information from a few RF chains by adding time slots; Step 3: Calculate the covariance matrix of the full array information and decompose the data covariance matrix to obtain a diagonal matrix containing the noise power; Step 4: Select the minimum value among the diagonal elements to replace the remaining diagonal elements to obtain the reconstructed data covariance matrix. Use the multiple signal classification method to process the reconstructed data covariance matrix to obtain the angle and distance of the terahertz signal source.

2. The near-field terahertz massive MIMO target positioning method under non-uniform noise according to claim 1, characterized in that: In step 1, a massive MIMO channel model and a near-field model are used to establish a receiving end system model based on a terahertz subarray hybrid structure, wherein the massive MIMO channel model is Where a(θ l,r [k],r l [k]) and b(θ l,t [k],r l [k]) represent the array steering vectors for reception and transmission, respectively, and α l [f,r,k] represents the complex gain of the lth path at carrier frequency f and communication distance r, A[k] and B[k] are the array manifold matrices of the receiving and transmitting ends respectively, and Λ[k] is the multipath channel complex gain matrix; k is the time slot index, θ l,r [k] and θ l,t [k] represents the incident angle and emission angle of the lth path, r l [k] is the communication distance of the lth path, and 1≤l≤L, L is the multipath number of the terahertz massive MIMO channel, The received signal is expressed as: Y=W H H[f,r,k]Fs+W H N=W H A[k]X[k]+W H N Where X[k] is defined as the equivalent transmitted signal in the kth time slot, and N is the non-uniform noise; W is the hybrid beamforming matrix of the AoSA system, and s is the pilot symbol vector sent by the pre-encoder F.

3. The near-field terahertz massive MIMO target positioning method under non-uniform noise according to claim 2, characterized in that: In step 2, a hybrid beamforming matrix of T time slots randomly generated is used to collect full array data from a few RF chains for estimating near-field signal position parameters. Then, the received signal of the τth time slot is: Among them, 1≤τ≤T, Set all Y τ Stacking into a matrix gives the total observation matrix Right now in, 4. The near-field terahertz massive MIMO target positioning method under non-uniform noise according to claim 3, characterized in that: The step 3 comprises: 1) Calculate the covariance matrix of the total observation matrix: Where P and Q are the covariance matrices of signal X and noise N respectively; 2) Express the covariance matrix R of the total observation matrix as the sum of two matrices R1 and R2: R=R1+R2 Among them, matrix R2 is composed of the diagonal elements of the covariance matrix R, and matrix R1 is composed of the remaining elements of the covariance matrix R.

5. The near-field terahertz massive MIMO target positioning method under non-uniform noise according to claim 4, characterized in that: The step 4 comprises: 1) Select the minimum element r in matrix R2 min , and use r min Substituting all diagonal elements in R2, we get 2) Reconstruct the covariance matrix of the total observation matrix: 3) R w Eigendecomposition: where Σ S and Σ N They are respectively w The signal and noise diagonal matrix composed of descending eigenvalues, U S and U N are the subspaces corresponding to the signal and noise diagonal matrices respectively; The 2D-MUSIC spectrum function of the terahertz near-field source is obtained as: Finally, the L highest peaks correspond to the target positions of the near-field source signal.

Images