A method for constructing and estimating parameters of asynchronous uplink sensing data of a mixed array with CFO / TO suppression

Abstract

The application belongs to the technical field of wireless communication, and particularly relates to a kind of mixed array asynchronous uplink perception data construction and parameter estimation method for inhibiting CFO / TO.The application aims at the problems that spatial sampling freedom is insufficient due to limited radio frequency link in mixed array uplink perception, and that carrier frequency offset and timing offset caused by asynchronous transmitting and receiving clock destroy phase coherence, constructs augmented virtual array snapshot vector to restore spatial sampling dimension by configuring scanning codebook and switching beam in multiple beam switching time slots; equivalent full-digital receiving signal is reconstructed by using scanning codebook orthogonality, the real wave arrival direction of reference path is extracted, and optimal reference beam and scanning beam are designed with the target of maximizing signal-to-interference-and-noise ratio; on this basis, asynchronous interference is eliminated by regularization ratio operation, spatial-frequency-time third-order observation tensor is constructed, and high-resolution estimation and automatic pairing of angle, time delay and Doppler parameter are realized by combining tensor iterative orthogonal matching pursuit.The application effectively inhibits the influence of carrier frequency offset and timing offset under the condition of mixed array and asynchronism, improves the precision and robustness of multi-dimensional parameter joint estimation, and does not need to additionally increase radio frequency link overhead.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, specifically relating to a method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays with suppressed CFO / TO. Background Technology

[0002] Integrated communication and sensing is a crucial evolutionary direction for future mobile networks. Uplink sensing utilizes uplink signals from user terminals to detect the environment, offering advantages such as eliminating the need for full-duplex hardware and avoiding self-interference. Millimeter-wave systems, with their large bandwidth and massive antenna arrays, provide the physical foundation for high-precision angle, delay, and Doppler estimation. However, in practical deployments, to control hardware costs and power consumption, base stations typically employ a hybrid analog-digital array structure. The number of RF links is far less than the number of antenna array elements, resulting in limited spatial sampling degrees of freedom, making it difficult to directly apply traditional high-resolution parameter estimation algorithms. Furthermore, the physical separation of transceivers in the uplink introduces carrier frequency and timing offsets, disrupting the phase coherence of the received signal and further exacerbating the difficulty of joint estimation of multi-dimensional sensing parameters.

[0003] For angle estimation in hybrid arrays, existing technologies mostly employ beam scanning strategies, acquiring spatial observations by switching beams across multiple time slots, trading time resources for spatial sampling dimensionality, and combining subspace methods or compressed sensing techniques to improve multipath resolution. Regarding the clock asynchrony problem in the uplink, existing solutions mainly fall into two categories: one incorporates carrier frequency offset and timing offset as unknown parameters into the joint estimation framework, but this method leads to an expansion of the parameter search dimensionality and is prone to getting trapped in local optima at low signal-to-noise ratios; the other uses cancellation methods, utilizing the phase factor shared by the reference branch and the scanning branch for ratio calculation to eliminate asynchronous interference, but this nonlinear transformation significantly amplifies noise, especially when the reference branch signal amplitude is small, resulting in a heavy-tailed distribution of equivalent noise, causing severe performance degradation of subsequent subspace or tensor estimation algorithms. Furthermore, for the joint estimation of angle, time delay, and Doppler, existing sequential estimation strategies suffer from error accumulation problems, while methods such as tensor decomposition, although preserving multidimensional structures, typically rely on ideal synchronization or Gaussian noise assumptions, making it difficult to directly adapt to observation data with non-Gaussian properties after ratio processing.

[0004] In summary, there is currently a lack of a technical solution that can simultaneously suppress carrier frequency offset and timing offset interference, and achieve stable joint estimation of angle, delay, and Doppler parameters under conditions of limited RF links and asynchronous transmit and receive clocks in hybrid arrays. How to construct high-dimensional equivalent observations under limited RF links, eliminate the influence of asynchronous phase on signal structure, and achieve high-resolution estimation and automatic pairing of multidimensional parameters in a non-Gaussian noise environment after ratio processing are urgent technical problems to be solved in this field. Summary of the Invention

[0005] To address the problems existing in the background art, this invention provides a method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO, comprising:

[0006] S1: Configure the scanning codebook, switch beams in multiple beam switching time slots through the hybrid array receiver, collect raw received data, and construct the raw observation dataset;

[0007] S2: Based on the original observation dataset, construct an augmented virtual array snapshot vector, perform frequency smoothing, estimate the beam arrival angle AoA of the multipath signal, and obtain a candidate set of AoA containing multiple multipaths.

[0008] S3: Reconstruct the equivalent all-digital received signal based on the augmented virtual array snapshot vector, detect the physical angle of the reference path based on the reconstructed equivalent all-digital received signal, and extract the true direction of arrival of the reference path based on the AoA candidate set;

[0009] S4: Based on the actual beam arrival direction of the reference path and the AoA of the non-reference path, construct the interference plus noise covariance matrix, and design the optimal reference beam and scanning beam with the goal of maximizing the signal-to-interference-plus-noise ratio.

[0010] S5: Load the optimal reference beam and scanning beam, construct the reference branch and scanning branch to receive signals, and eliminate asynchronous interference caused by carrier frequency offset CFO and timing offset TO through regularized signal ratio calculation, and construct a third-order observation tensor.

[0011] S6: Decompose the third-order observation tensor into three dimensions: space, frequency, and time. Estimate the parameter candidate set for each dimension. Then, use the tensor iterative orthogonal matching pursuit method to automatically pair the estimated angle of arrival, time delay, and Doppler parameters, and output the final parameter estimation result.

[0012] The present invention has at least the following beneficial effects

[0013] This solution addresses the challenge of multidimensional parameter estimation caused by limited RF links and asynchronous coupling of transmit and receive clocks in hybrid array asynchronous uplink sensing. It proposes a systematic framework: by using time-division beam scanning and augmented virtual array construction, spatial sampling degrees of freedom are effectively restored without adding RF links. Furthermore, by utilizing regularized ratio operations between the reference and scanning branches, common phase interference introduced by carrier frequency and timing offsets is eliminated, and noise amplification in traditional ratio processing is suppressed. The observed data is further organized into a third-order tensor structure, and combined with the tensor iterative orthogonal matched pursuit method, high-resolution estimation and automatic pairing of angle, time delay, and Doppler parameters are achieved. Compared to existing solutions, this method significantly improves the robustness and accuracy of parameter estimation under asynchronous conditions, reduces the computational complexity of multidimensional joint search, and provides a complete technical path for hybrid array uplink sensing that combines anti-asynchronous interference capability and high-dimensional resolution performance. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0015] Figure 2 This is a system architecture diagram of the present invention;

[0016] Figure 3 This is a schematic diagram of the uplink sensing preprocessing framework of the present invention;

[0017] Figure 4 This is a schematic diagram of the overall process framework for uplink sensing in this invention. Detailed Implementation

[0018] 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, and 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.

[0019] Please see Figure 1 In this embodiment, a millimeter-wave uplink using a hybrid antenna array is considered, and the communication sensing scenario is as follows: Figure 1 As shown. Uplink transmission is established on a UE equipped with an omnidirectional antenna, scheduling the same time-frequency resources each time to communicate with a base station equipped with a fully connected hybrid array. This base station's hybrid array consists of two analog subarrays. Subarray and contains a by A uniform linear array (ULA) consisting of three antennas; let... and These represent the signal bandwidth and the number of subcarriers, respectively. Therefore, the subcarrier spacing is... Considering OFDM-based transmission, the corresponding OFDM symbol period is... ,in It is the period of the cyclic prefix (CP). The transmitted signal of each subcarrier can be represented as For base stations, the first The received signal of each subcarrier can be represented as: ;in, It is the additive white Gaussian noise (AWGN) at the base station receiver. It is a channel state information vector.

[0020] In a dual-base uplink configuration, the clock asynchrony caused by the physical separation of the transceiver will inevitably introduce carrier frequency offset. and timing offset Although these hardware offsets are inherently time-varying, the phase drift of the local oscillator is negligible within this window, given the extremely short duration of a single beam scan frame (typically on the order of microseconds). Therefore, this paper assumes that CFO and TO remain constant during a single snapshot observation period.

[0021] Based on the above analysis, let the total number of signal paths generated by the multipath effect be... Among them are static paths and A dynamic path, thus having In millimeter-wave systems, Very small, a static path can typically be defined as having no Doppler shift, while a dynamic path exhibits a Doppler shift. Based on the target's motion characteristics, it is assumed that the target's Doppler shift and propagation delay remain constant within a single coherent processing interval. Accordingly, the first... Effective channel state information vector of each subcarrier It can be modeled as:

[0022]

[0023] in, This indicates that the offsets include both CFO and TO. It is the first The frequency of each subcarrier This is the initial carrier frequency. The AoA of the path is The set of all AoA is denoted as . . No. The complex gain, propagation delay, and Doppler frequency of each path are respectively , and ,in It is the distance the signal travels. It is the radial velocity of the reflector. It's the speed of light. Complex gain. The amplitude characterizes the equivalent scattering intensity of the path, incorporating factors such as propagation loss, antenna gain, and target scattering characteristics. To enhance physical interpretability, the target radar cross-section (RCS) can be equivalently incorporated into... In the bistatic scattering model, there is usually... ,in For the target RCS, This represents the corresponding path loss. Therefore, the difference in RCS for different targets is ultimately equivalent to... The changes in SNR at the receiver affect the detection and estimation performance.

[0024] Furthermore, for the sake of greater conciseness, the following assumptions are made: the index is All paths are static, meaning that for all static paths, there is... The remaining paths are dynamic; to determine the path, the AoA, Doppler, and delay of each path are unique; the index is... The path is defined as the reference path, typically a LosS path in typical millimeter-wave scenarios, while other paths are NLoS paths; however, when the LosS path is blocked, if there is a power-dominant and stably extractable NLoS component, it can also be considered as a reference path. The steering vector of the half-wavelength spacing ULA of the root antenna It is given by the following formula:

[0025]

[0026] Assume the beamforming (BF) matrix of the base station analog subarray is represented as follows: ,and Then the first The received signal vector after each subcarrier passes through the analog subarray can be represented as a vector. :

[0027]

[0028] To address the issues of synchronization error coupling, limited sensing coverage, and scarce pilot resources in uplink sensing, please refer to [link to relevant documentation]. Figure 2 This invention provides a method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO, comprising:

[0029] S1: Configure the scanning codebook, switch beams in multiple beam switching time slots through the hybrid array receiver, collect raw received data, and construct the raw observation dataset;

[0030] Please see Figure 3 In hybrid analog-digital arrays, the receiver can only obtain low-dimensional baseband compressed observations in a single analog-combined state, and cannot directly access the full digital samples element by element. This makes it difficult to directly apply traditional high-resolution AoA estimation algorithms that rely on full-element observations to the raw output. To address this issue, this embodiment designs a time-division beam scanning strategy based on orthogonal codebooks. This strategy trades time resources for spatial observation dimensions, reconstructing equivalent virtual full digital array observations in the digital domain, thus laying the physical foundation for subsequent high-precision AoA estimation. Specifically, it includes:

[0031] S11: Determine the discrete scanning angle index and the corresponding scanning angle set based on the total number of base station antennas, construct the scanning beam vector, and concatenate all scanning beam vectors column by column to form a scanning codebook matrix;

[0032] In this embodiment, the base station is configured as a hybrid array receiver with two analog subarrays, and the total number of physical receiving antennas of the base station is [number missing]. Discrete scan angle index is denoted as And preset the set of scan angles corresponding to each discrete scan angle index. ; ; ;in, Represents the arcsine function; This represents the total number of antennas in a uniform linear array at a base station. Indicates the first A discrete scanning angle; express The normalized space angle; based on the first Discrete scanning angle Construct the first scanning beam vectors :

[0033]

[0034] in, Indicates angle The corresponding array steering vector; Represents the natural base; Represents the imaginary unit; Represents pi; Represents the sine function; This indicates transpose.

[0035] All scan beam vectors are concatenated column-wise to form a scan codebook matrix. :

[0036]

[0037] in, Represents the scan codebook matrix, Represents the field of complex numbers 3D matrix space;

[0038] S12: Set the total number of beam switching time slots within a single sensing frame, and load a dual-array beam matrix containing two adjacent scanning beam vectors in each beam switching time slot;

[0039] In this embodiment, the beam switching time slot index is set to In the Each beam switching time slot loads the corresponding twin-array beam matrix. :

[0040]

[0041] in, Indicates the first The twin-array beam matrix corresponding to each beam switching time slot.

[0042] S13: By repeatedly transmitting multiple sensing frames, the beam is switched sequentially within each snapshot to collect the received signals on all subcarriers, forming the original observation dataset.

[0043] In this embodiment, the snapshot index for repeated collection is set as follows: ,in Indicates the number of S-Frames (sensor frames) that are repeatedly transmitted; in each snapshot Inside, according to arrive The beams are switched sequentially to collect raw received observations on all subcarriers, forming the raw observation dataset:

[0044]

[0045] in, Show the original observation dataset; Indicates the first The first snapshot, the first The beam switching time slot, the first Received signal at each subcarrier; This indicates the total number of system subcarriers.

[0046] In this embodiment, step S1 involves configuring an orthogonal scanning codebook and sequentially switching beams in multiple beam switching time slots to collect raw received data. Under the limited condition that the number of RF links in the hybrid array is much less than the number of antenna array elements, time resources are exchanged for spatial sampling dimensions, and a raw observation dataset containing rich spatial information is successfully constructed. This lays a complete observation foundation for the subsequent construction of augmented virtual array snapshot vectors and high-resolution angle estimation, effectively solving the problem of limited spatial sampling degrees of freedom caused by insufficient RF links.

[0047] S2: Based on the original observation dataset, construct an augmented virtual array snapshot vector, perform frequency smoothing, estimate the waveguide arrival angle AoA of the multipath signal, and obtain a candidate set of AoA containing multiple multipaths.

[0048] Preferably, step S2 specifically includes:

[0049] S21: Vertically stack the received signals of all beam switching time slots within the same snapshot to construct an augmented virtual array snapshot vector;

[0050] In this embodiment, the first The first snapshot, the first The beam switching time slot, the first The received signal of each subcarrier is modeled as follows:

[0051]

[0052] in, Indicates the first The first snapshot, the first The beam switching time slot, the first Received signal at each subcarrier; This represents the common asynchronous phase term introduced by both carrier frequency offset and timing offset; Indicates the first Each beam switching time slot corresponds to the dual-array beam matrix. The conjugate transpose of; Representation space dictionary matrix; Indicates the first The first snapshot, the first The effective channel vectors corresponding to each subcarrier; This represents an additive white Gaussian noise vector. This indicates the conjugate transpose.

[0053] The spatial dictionary matrix and effective channel vector They are represented as follows:

[0054]

[0055]

[0056]

[0057] in, Indicates the first The array guide vector corresponding to each path; Indicates the first The direction of beam arrival corresponding to each path; Indicates the total number of paths; Indicates the first Complex gain of the path; Indicates the first The propagation delay of each path; Indicates the first Doppler parameters of the path; Indicates the frequency spacing between adjacent subcarriers; Indicates the slow time sampling interval between adjacent snapshots; Indicates the subcarrier index; Indicates a snapshot index; Indicates conjugate transpose; Indicates conjugate transpose; It represents pi (π).

[0058] For the same snapshot The received signals from all beam switching time slots within the lower S-Frame are vertically stacked to construct an augmented virtual array snapshot vector. :

[0059]

[0060]

[0061] in, Indicates the first The first snapshot, the first The augmented virtual array snapshot vector corresponding to each subcarrier; This represents a common asynchronous phase term that is approximately shared within a single S-Frame; Represents the scan codebook matrix; This represents the noise vector of the stacked codebook matrix. Represents the field of complex numbers A column vector space; k represents the subcarrier index.

[0062] S22: Calculate the sample covariance matrix of the augmented virtual array snapshot vectors for all subcarriers and all snapshots, and perform frequency smoothing;

[0063] In this embodiment, the sample covariance matrix is ​​calculated from the augmented virtual array snapshot vectors of all subcarriers and all snapshots to obtain the frequency-smoothed sample covariance matrix. :

[0064]

[0065] in, This represents the sample covariance matrix after frequency smoothing. Indicates the total number of system subcarriers; This represents the total number of snapshots created by repeated transmissions.

[0066] S23: Perform eigenvalue decomposition on the frequency-smoothed covariance matrix to obtain the noise subspace; construct the MUSIC spatial spectrum function based on the noise subspace, and obtain a candidate set containing L multipath AoA through spectral peak search.

[0067] In this embodiment, the frequency-smoothed covariance matrix Perform EVD eigenvalue decomposition to decompose it into signal subspace. and noise subspace :

[0068]

[0069]

[0070]

[0071]

[0072] Among them, the former Large eigenvalues The table shows the effects of multipath propagation. A multipath signal characteristic value; This represents the noise variance of additive white Gaussian noise;

[0073] Set the angle search range Construct the MUSIC spatial spectral function By traversing and searching the peak positions of the music spectrum, a candidate set containing L multipath AoA values ​​is obtained. :

[0074]

[0075]

[0076] in, express The conjugate transpose of . , The guiding vector of a uniform linear array; Indicates the first Candidate AoA values ​​for each path; It represents the square of the 2-norm.

[0077] In this embodiment, step S2 vertically stacks the received signals from all beam-switching time slots within the same snapshot, integrating low-dimensional observations scattered across multiple time slots into an augmented virtual array snapshot vector. This effectively compensates for the lack of spatial sampling degrees of freedom caused by insufficient RF links in the hybrid array. Based on this, frequency smoothing is performed to eliminate the frequency selectivity effects between subcarriers, making the multipath signals exhibit coherent and distinguishable characteristics in the spatial dimension. Then, the MUSIC algorithm is used to stably estimate the candidate set containing all multipath angles of arrival, providing reliable spatial prior information for subsequent reference path extraction and beam design.

[0078] S3: Reconstruct the equivalent all-digital received signal based on the augmented virtual array snapshot vector, detect the physical angle of the reference path based on the reconstructed equivalent all-digital received signal, and extract the true direction of arrival of the reference path based on the AoA candidate set;

[0079] Preferably, step S3 specifically includes:

[0080] S31: Utilize the orthogonality of the scan codebook matrix to perform an inverse transformation on the augmented virtual array snapshot vector and reconstruct the equivalent all-digital received signal;

[0081] In this embodiment, the beam arrival direction of the reference path is set as follows: Constructing beamforming vectors based on beam arrival direction :

[0082]

[0083] Using the scan codebook matrix The DFT orthogonality of augmented virtual array snapshot vectors Perform an inverse transform to obtain the equivalent all-digital received signal. :

[0084]

[0085] in, Indicates the first The first snapshot, the first The equivalent all-digital received signal corresponding to each subcarrier.

[0086] S32: Zero-padding and fast Fourier transform are performed on the reconstructed equivalent all-digital received signal to obtain the angle detection spectrum, and the physical angle estimate is calculated based on the peak position;

[0087] In this embodiment, after removing the cyclic prefix (CP) and pilot signals, the reconstructed equivalent all-digital received signal is... Establish a received signal model containing L multipath paths:

[0088]

[0089] in, Indicates the first Complex gain of each signal path; Indicates the first Doppler parameters of the path; Indicates the first The propagation delay of each signal path; This represents the time interval between adjacent slow-time samples; Indicates the frequency spacing between adjacent subcarriers; This represents the reconstructed noise vector.

[0090] Reconstruct the signal Fill to zero Dimension, to obtain the reconstructed signal after zero padding. ,in The reconstructed signal after zero padding Perform a fast Fourier transform to obtain the angle detection spectrum. The peak index of the angle detection spectrum is denoted as . According to the peak index Calculating the normalized space angle using zero-padding length Then, the normalized spatial angle is mapped to a physical angle estimate using the arcsine function. :

[0091]

[0092]

[0093]

[0094]

[0095] in, Indicates the reconstructed signal The signal obtained after zero padding; This indicates the number of points in the Fast Fourier Transform after zero-padding; Represents the angle detection spectrum; Indicates the angle detection spectrum In the Spectral values ​​at each frequency point; Indicates Fast Fourier Transform; Represents the sine function; Represents the arcsine function; Indicates the peak index of the angle detection spectrum; This represents the normalized spatial angle calculated based on the FFT peak position and zero-padding length. This represents the estimated physical angle obtained from the reference path detection; This represents the variable that maximizes the objective function; It represents the modulus of a complex number.

[0096] S33: Calculate the absolute deviation between the physical angle estimate and each candidate angle in the AoA candidate set, and take the candidate angle with the smallest deviation as the true direction of arrival of the reference path.

[0097] In this embodiment, the physical angle estimate is calculated. With AoA candidate set Various candidate angles absolute deviation Take the angle with the smallest absolute angular deviation. The actual beam arrival direction as a reference path :

[0098]

[0099]

[0100] in, This represents the candidate set of AoA obtained in step S26; Represents the first in the candidate set of AoA One candidate angle; Represents the estimated value of the physical angle. With the AoA candidate values The absolute angular deviation between them; This represents the candidate path index corresponding to the minimum absolute angle deviation; This represents the variable that minimizes the objective function.

[0101] In this embodiment, step S3 utilizes the orthogonality of the scan codebook matrix to perform an inverse transformation on the augmented virtual array snapshot vector, successfully reconstructing the equivalent all-digital received signal. This breaks the physical limitation that the hybrid array can only acquire compressed observations, providing a complete array element domain data foundation for subsequent high-precision parameter estimation. Based on this, a fast coarse detection of the reference path angle is achieved through Fast Fourier Transform, and minimum deviation matching is performed by combining the accurate AoA candidate set obtained in step S2. This effectively solves the problem of accurate identification of the reference path and extraction of the true direction of arrival in a multipath environment, providing highly reliable angle prior information for the optimal design of the subsequent reference beam.

[0102] S4: Based on the true direction of arrival of the reference path and the AoA of the non-reference path, construct the interference plus noise covariance matrix, and design the optimal reference beam and scanning beam with the goal of maximizing the signal-to-interference-plus-noise ratio.

[0103] Preferably, step S4 specifically includes:

[0104] S41: Based on the true direction of arrival of the reference path and the AoA of the non-reference path, construct the full path space dictionary matrix and the non-reference path space dictionary matrix respectively.

[0105] In this embodiment, the actual beam arrival direction is based on the reference path. Construct the full path space dictionary matrix with the other AoA candidate angles. Non-reference pathspace dictionary matrix

[0106]

[0107]

[0108] in, Represents the dictionary matrix of the entire path space; Represents the dictionary matrix of the non-reference pathspace.

[0109] S42: The multipath complex gain is obtained by least squares estimation, and the reference path complex gain and the non-reference path complex gain are separated.

[0110] In this embodiment, the full path space dictionary matrix is ​​used. The reconstructed equivalent all-digital received signal Perform LS estimation to obtain the multipath complex gain estimation vector. and separate the reference path complex gain Non-reference path complex gain vector ;

[0111]

[0112]

[0113] in, Represents the dictionary matrix of the total path space Moore–Penrose pseudo-inverse; This represents the reference path complex gain estimate; This represents the non-reference path complex gain estimate.

[0114] S43: Construct a reference signal covariance matrix based on the reference path complex gain, and construct an interference plus noise covariance matrix based on the non-reference path complex gain and the estimated noise power;

[0115] In this embodiment, the sample covariance matrix of the reconstructed received signal is calculated. And by analyzing the covariance matrix Eigenvalue decomposition of EVD is performed to estimate noise power, and the noise covariance matrix is ​​calculated. :

[0116]

[0117]

[0118]

[0119]

[0120] in, This represents the sample covariance matrix of the reconstructed received signal; Indicates the first The first snapshot, the first The equivalent all-digital received signal corresponding to each subcarrier; Representation matrix eigenvector matrix; Representation matrix The eigenvalue diagonal matrix; Indicates the first One eigenvalue; This represents the estimated noise power. Represents the noise covariance matrix; express 3D identity matrix; This represents a diagonal matrix consisting of the elements within the brackets;

[0121] Based on the reference path complex gain estimate Construct the reference signal covariance matrix Based on non-reference path complex gain vector and noise covariance matrix Constructing the interference plus noise covariance matrix :

[0122]

[0123]

[0124]

[0125] in, This represents the reference path complex gain estimate; Represents the complex gain vector of the non-reference path; Indicates the actual direction of arrival of the reference path; Represents the dictionary matrix of the non-reference pathspace; Represents the reference signal covariance matrix; Represents the interference plus noise covariance matrix; It represents the modulus of a complex number.

[0126] S44: To maximize the signal-to-interference-plus-noise ratio, solve the generalized Rayleigh quotient optimization problem, and use the eigenvector corresponding to the largest eigenvalue as the optimal reference beam.

[0127] In this embodiment, the generalized Rayleigh quotient optimization problem is constructed with the goal of maximizing the signal-to-interference-plus-noise ratio (SINR):

[0128]

[0129] Constructing a matrix And perform EVD eigenvalue decomposition on it:

[0130]

[0131] The normalized eigenvector corresponding to the largest eigenvalue is taken as the optimal reference beam:

[0132]

[0133] in, This represents the beam vector to be optimized; Indicates the optimal reference beam; This represents the variable that maximizes the objective function; Indicates constraints; Representation matrix The inverse matrix; Represents the eigenvector matrix; Represents an eigenvalue diagonal matrix; Represents the largest eigenvalue; Represents the largest eigenvalue The corresponding normalized feature vector.

[0134] In this embodiment, step S4 constructs the full path and non-reference path spatial dictionary matrices based on the true direction of arrival of the reference path and the accurate AoA candidate set of the non-reference path. The complex gain of the reference path and the non-reference path is separated by least squares estimation, and then the reference signal covariance matrix and the interference plus noise covariance matrix are constructed. On this basis, with the goal of maximizing the signal-to-interference-plus-noise ratio, the optimal reference beam is obtained by solving the generalized Rayleigh quotient optimization problem. This effectively improves the signal reception quality of the target path and suppresses the influence of multipath interference and noise. It provides a reference branch input with a high signal-to-interference-plus-noise ratio for subsequent regularization ratio calculation, and significantly enhances the stability of asynchronous interference cancellation and the robustness of parameter estimation.

[0135] S5: Load the optimal reference beam and scanning beam, construct the reference branch and scanning branch to receive signals, and eliminate asynchronous interference caused by carrier frequency offset CFO and timing offset TO through regularized signal ratio calculation, and construct a third-order observation tensor.

[0136] Preferably, the construction of the third-order observation tensor includes:

[0137] S51: Load the optimal reference beam onto the first analog subarray, and load the scanning beam onto the second analog subarray to form a combined beam matrix;

[0138] In this embodiment, the subcarrier index is set to The communication frame index is The OFDM symbol index within each communication frame is And define a unified slow time index. for:

[0139]

[0140]

[0141] in, Indicates the total number of communication frames; This represents the total number of OFDM symbols within each communication frame; Represents a uniform slow-time index; This represents the total number of slow time samples;

[0142] Loading the optimal reference beam in the first analog subarray In the second analog subarray according to the scan codebook matrix Loading scanning beam Thus constructing a combined beam matrix :

[0143]

[0144] S52: Based on the combined beam matrix, the reference branch received signal and the scanning branch received signal are obtained;

[0145] In this embodiment, the first The subcarrier, the first The reference branch received signal at the slow time sampling point and scan branch received signal vector :

[0146]

[0147]

[0148]

[0149]

[0150] in, Indicates the first The subcarrier, the first The scalar received signal of the reference branch under slow time sampling; Indicates the first The subcarrier, the first Vector received signal of a slow-time scanning branch; Indicates the first The subcarrier, the first A common multiplicative term consisting of unknown transmitted data symbols and asynchronous phase terms under a slow-time sampling condition; Indicates the first The subcarrier, the first The unknown transmitted data symbol corresponding to the slow time sample; Indicates the carrier frequency offset at the 1st A common phase rotation term introduced on a slow-time sampling; Indicates the timing offset at the 1st A common phase rotation term introduced on each subcarrier; This is the equivalent virtual array space response vector; and They represent the first The subcarrier, the first Additive white Gaussian noise vectors for the reference branch and the scan branch under slow-time sampling; Indicates a complex Gaussian distribution; Represents the variance of the noise; Indicates the complex gain of the reference path; Indicates the first Complex gain of multiple paths; Indicates the subcarrier spacing; Indicates the propagation delay of the reference path; This indicates the Doppler frequency shift of the signal along the reference path; Indicates the duration of the OFDM symbol; Indicates the first The subcarrier, the first An equivalent channel vector containing all signal paths under slow-time sampling;

[0151] S53: Calculate the average power of the received signal in the reference branch, and perform a regularized ratio operation on the conjugate of the received signal in the scanned branch and the received signal in the reference branch to obtain a regularized signal ratio vector.

[0152] In this embodiment, the average power of the received signal in the reference branch is calculated:

[0153]

[0154]

[0155] in, This represents the regularization intensity adjustment coefficient; The SNR represents the average power of the signal received by the reference branch; the SNR represents the signal-to-noise ratio of a single baseband observation at the receiver.

[0156] Constructing a regularized signal ratio vector :

[0157]

[0158] in, Indicates the first The subcarrier, the first A vector of regularized signal ratios corresponding to each slow-time sample; Represents the complex conjugate operation; Represents the modulus of a complex number; This represents the regularization factor.

[0159] S54: Reorganize the regularized signal ratio vector of all subcarriers and all slow-time samples into a third-order observation tensor according to the three dimensions of space, frequency and time.

[0160] In this embodiment, the regularization ratio operation is repeatedly performed on all subcarriers and all slow-time samples, and the resulting regularized signal ratio vector is reorganized into a third-order observation tensor according to the three dimensions of space, frequency, and time. :

[0161]

[0162]

[0163] in, Represents the spatial-frequency-temporal third-order observation tensor; Tensor In the The frequency index and the first Spatial slices at time indices; the spatial dimension corresponds to virtual array spatial sampling, the frequency dimension corresponds to subcarrier index, and the time dimension corresponds to slow time sampling index.

[0164] In this embodiment, step S5 constructs the received signals of the reference branch and the scanning branch by loading the optimal reference beam designed in step S4 onto the first analog subarray and the scanning beam onto the second analog subarray. Based on this, a regularization factor is introduced to adaptively adjust the average power of the reference branch signal. The common asynchronous phase term introduced by the carrier frequency offset and timing offset is effectively eliminated through regularized ratio calculation, while suppressing the noise amplification problem caused by the small amplitude of the denominator in the traditional ratio processing. Furthermore, the ratio vector of all subcarriers and slow-time sampling is reorganized into a third-order observation tensor according to the three dimensions of space, frequency and time. The multipath information after the asynchronous interference is eliminated is completely preserved in a structured form, providing a high-quality data foundation for subsequent high-resolution parameter estimation and automatic pairing in the tensor domain.

[0165] S6: Decompose the third-order observation tensor into three dimensions: space, frequency, and time. Estimate the parameter candidate set for each dimension. Then, use the tensor iterative orthogonal matching pursuit method to automatically pair the estimated angle of arrival, time delay, and Doppler parameters, and output the final parameter estimation result.

[0166] Please see Figure 3 In this embodiment, the core mathematical model and solution algorithm of the RATE-aware architecture are constructed based on the regularized signal ratio tensor, specifically including:

[0167] S61: Expand the third-order observation tensor along the spatial mode, frequency mode and time mode respectively, and construct the covariance matrix of each mode;

[0168] In this embodiment, a spatial dimension equivalent virtual array steering vector is defined. Frequency dimension guided vector and time-dimensional oriented vector :

[0169]

[0170]

[0171]

[0172] in, Indicates the first The guide vector of the path in the virtual array space dimension; Indicates the first The guiding vector of a path in the frequency dimension; Indicates the first The path's guidance vector in the time dimension; Represents the scan codebook matrix; Indicates angle The corresponding array steering vector; Indicates the total number of subcarriers; This represents the total number of slow time samples; Indicates the subcarrier spacing; This indicates the OFDM symbol sampling interval.

[0173] The clean observation tensor It can be expressed as the sum of the tensor cross products of all multipath components:

[0174]

[0175] And vectorize it as:

[0176]

[0177] in, Represents the cross product of vectors; Represents the tensor noise term; Indicates vectorized operations; Represents the Khatri–Rao product; Tensor The vectorized result; Represents the path complex gain vector; Represents the path complex gain vector; , , These represent the response matrices in the spatial dimension, frequency dimension, and time dimension, respectively;

[0178] Response matrices in each dimension , , and path gain vector Defined as: S62: Estimate the number of effective sources in the three modes using the AIC criterion, and determine the global model order;

[0179]

[0180]

[0181]

[0182]

[0183] in, Represents the spatial response matrix; Represents the frequency-dimensional response matrix; Represents the time-dimensional response matrix;

[0184] The third-order observation tensor Expanded along the spatial mode, frequency mode, and time mode respectively as follows , and And construct the covariance matrix of each mode. , and ;

[0185]

[0186]

[0187]

[0188] in, Represents the spatial modular expansion matrix; Represents the frequency modulus expansion matrix; Represents the time-modulus expansion matrix; , , These represent the sample covariance matrices for the three modes, respectively.

[0189] S62: Use the AIC criterion to estimate the number of effective sources in the three modes respectively, and determine the global model order;

[0190] In this embodiment, the AIC criterion is used to estimate the number of effective sources in the three modes. And determine the global model order. :

[0191]

[0192] S63: Execute the Root-MUSIC algorithm on the three modes respectively to obtain the AoA candidate set, the time delay candidate set, and the Doppler candidate set;

[0193] Determining the global model order Then, the Root-MUSIC algorithm is executed based on the spatial mode, frequency mode, and time mode respectively to obtain the AoA candidate set, the time delay candidate set, and the Doppler candidate set:

[0194]

[0195]

[0196]

[0197] in, Indicates modal index; Indicates the first The number of effective sources obtained from modal estimation; Indicates the global model order; Represents the candidate set of AoA; Represents the candidate set of delays; This represents the Doppler candidate set.

[0198] S64: Initialize the residual tensor and support set. Using the tensor iterative orthogonal matching pursuit method, calculate the multidimensional correlation tensor between the current residual tensor and the three modal estimation response matrices in each iteration. Select the index with the maximum correlation to update the support set. Reconstruct the observation tensor and update the residual tensor through least squares back substitution until the number of iterations reaches the global model order. Output the automatically paired parameter set and the corresponding path complex gain.

[0199] In this embodiment, an estimated response matrix is ​​constructed based on the AoA candidate set, the time delay candidate set, and the Doppler candidate set. , , and initialize the residual tensor. and support set :

[0200]

[0201] in, , , These represent the spatial, frequency, and time-dimensional estimated response matrices constructed from the candidate parameters, respectively. Represents the initial residual tensor; Represents the initial support set; Indicates the empty set;

[0202] In the In each iteration, the current residual tensor is projected onto the estimated response matrices of the three modes to construct a multidimensional correlation tensor, and the three-dimensional index with the highest correlation is selected to update the support set.

[0203]

[0204]

[0205]

[0206] in, Indicates the first Modular tensor product, where , , These represent tensor products along the first, second, and third modules, respectively; Indicates the first The multidimensional correlation tensor obtained from rounds of iteration; Indicates the first The selected 3D index in the round of iteration; This represents the updated support set; This indicates that the square is taken modulo the element.

[0207] Least square back substitution is performed on the submatrix corresponding to the current support set to obtain the core tensor estimate. And reconstruct the observation tensor and update the residual tensor:

[0208]

[0209]

[0210] in, Indicates the first Core tensor estimation obtained from rounds of iteration; Indicates the first The low-rank reconstructed tensor obtained from rounds of iteration; This represents the updated residual tensor.

[0211] Repeat the iterative process of step S64 until the number of iterations reaches the global model order. Output the automatically paired parameter set and the corresponding path complex gain:

[0212]

[0213] in, Indicates the relationship with the first The complex gain estimate corresponding to the paired path parameters.

[0214] In this embodiment, step S6 expands the third-order observation tensor along the spatial, frequency, and temporal dimensions to construct the covariance matrix. Combined with the AIC criterion, the global model order is accurately estimated. The Root-MUSIC algorithm is used to independently estimate the candidate sets of angle of arrival, time delay, and Doppler in each dimension, effectively avoiding the problem of error propagation in sequential estimation. On this basis, the tensor iterative orthogonal matching pursuit method is adopted to calculate the multidimensional correlation tensor between the residual tensor and the three modal estimation response matrices in each iteration. The support set is updated by automatically selecting the maximum correlation index, and the observation tensor is accurately reconstructed by least squares back substitution. Finally, high-resolution joint estimation and automatic pairing of angle of arrival, time delay, and Doppler parameters are achieved, solving the problem of parameter matching difficulties in non-Gaussian noise environments after ratio processing in traditional methods, and significantly improving the estimation accuracy and output reliability of multidimensional sensing parameters.

[0215] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO, characterized in that, include: S1: Configure the scanning codebook, switch beams in multiple beam switching time slots through the hybrid array receiver, collect raw received data, and construct the raw observation dataset; S2: Based on the original observation dataset, construct an augmented virtual array snapshot vector, perform frequency smoothing, estimate the beam arrival angle AoA of the multipath signal, and obtain a candidate set of AoA containing multiple multipaths. S3: Reconstruct the equivalent all-digital received signal based on the augmented virtual array snapshot vector, detect the physical angle of the reference path based on the reconstructed equivalent all-digital received signal, and extract the true direction of arrival of the reference path based on the AoA candidate set; S4: Based on the true direction of arrival of the reference path and the AoA of the non-reference path, construct the interference plus noise covariance matrix, and design the optimal reference beam and scanning beam with the goal of maximizing the signal-to-interference-plus-noise ratio. S5: Load the optimal reference beam and scanning beam, construct the reference branch and scanning branch to receive signals, and eliminate asynchronous interference caused by carrier frequency offset CFO and timing offset TO through regularized signal ratio calculation, and construct a third-order observation tensor. S6: Decompose the third-order observation tensor into three dimensions: space, frequency, and time. Estimate the parameter candidate set for each dimension. Then, use the tensor iterative orthogonal matching pursuit method to automatically pair the estimated angle of arrival, time delay, and Doppler parameters, and output the final parameter estimation result.

2. The method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO according to claim 1, characterized in that, Step S1 specifically includes: S11: Determine the discrete scanning angle index and the corresponding scanning angle set based on the total number of base station antennas, construct the scanning beam vector, and concatenate all scanning beam vectors column by column to form a scanning codebook matrix; S12: Set the total number of beam switching time slots within a single sensing frame, and load a dual-array beam matrix containing two adjacent scanning beam vectors in each beam switching time slot; S13: By repeatedly transmitting multiple sensing frames, the beam is switched sequentially within each snapshot to collect the received signals on all subcarriers, forming the original observation dataset.

3. The method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO according to claim 1, characterized in that, Step S2 specifically includes: S21: Vertically stack the received signals of all beam switching time slots within the same snapshot to construct an augmented virtual array snapshot vector; S22: Calculate the sample covariance matrix of the augmented virtual array snapshot vectors for all subcarriers and all snapshots, and perform frequency smoothing; S23: Perform eigenvalue decomposition on the frequency-smoothed covariance matrix to obtain the noise subspace; construct the MUSIC spatial spectrum function based on the noise subspace, and obtain a candidate set containing L multipath AoA through spectral peak search.

4. The method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO according to claim 1, characterized in that, Step S3 specifically includes: S31: Utilize the orthogonality of the scan codebook matrix to perform an inverse transformation on the augmented virtual array snapshot vector and reconstruct the equivalent all-digital received signal; S32: Zero-padding and fast Fourier transform are performed on the reconstructed equivalent all-digital received signal to obtain the angle detection spectrum, and the physical angle estimate is calculated based on the peak position; S33: Calculate the absolute deviation between the physical angle estimate and each candidate angle in the AoA candidate set, and take the candidate angle with the smallest deviation as the true direction of arrival of the reference path.

5. The method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO according to claim 1, characterized in that, Step S4 specifically includes: S41: Based on the true direction of arrival of the reference path and the AoA of the non-reference path, construct the full path space dictionary matrix and the non-reference path space dictionary matrix respectively. S42: The multipath complex gain is obtained by least squares estimation, and the reference path complex gain and the non-reference path complex gain are separated. S43: Construct a reference signal covariance matrix based on the reference path complex gain, and construct an interference plus noise covariance matrix based on the non-reference path complex gain and the estimated noise power; S44: To maximize the signal-to-interference-plus-noise ratio, solve the generalized Rayleigh quotient optimization problem, and use the eigenvector corresponding to the largest eigenvalue as the optimal reference beam.

6. The method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO according to claim 1, characterized in that, The construction of the third-order observation tensor includes: S51: Load the optimal reference beam onto the first analog subarray, and load the scanning beam onto the second analog subarray to form a combined beam matrix; S52: Based on the combined beam matrix, the reference branch received signal and the scanning branch received signal are obtained; S53: Calculate the average power of the received signal in the reference branch, and perform a regularized ratio operation on the conjugate of the received signal in the scanned branch and the received signal in the reference branch to obtain a regularized signal ratio vector. S54: Reorganize the regularized signal ratio vector of all subcarriers and all slow-time samples into a third-order observation tensor according to the three dimensions of space, frequency and time.

7. The method for constructing and estimating parameters of asynchronous uplink sensing data for hybrid arrays to suppress CFO / TO according to claim 1, characterized in that, Step S6, parameter estimation and automatic pairing, includes: S61: Expand the third-order observation tensor along the spatial mode, frequency mode and time mode respectively, and construct the covariance matrix of each mode; S62: Use the AIC criterion to estimate the number of effective sources in the three modes respectively, and determine the global model order; S63: Execute the Root-MUSIC algorithm on the three modes respectively to obtain the AoA candidate set, the time delay candidate set, and the Doppler candidate set; S64: Initialize the residual tensor and support set. Using the tensor iterative orthogonal matching pursuit method, calculate the multidimensional correlation tensor between the current residual tensor and the three modal estimation response matrices in each iteration. Select the index with the maximum correlation to update the support set. Reconstruct the observation tensor and update the residual tensor through least squares back substitution until the number of iterations reaches the global model order. Output the automatically paired parameter set and the corresponding path complex gain.

Images