A low-cost fast DOA estimation method and device

Abstract

The application discloses a low-cost fast DOA estimation method and device, and relates to the field of communication sensing. The estimation method comprises the following steps: based on a preset total number of fast snapshots and a two-stage channel model, an observation matrix is constructed according to a phase shift vector and a beam pointing direction; an energy spectrum corresponding to the observation matrix is generated, and a DOA coarse estimation result corresponding to each far-field target is generated according to a peak value index corresponding to each spectrum peak in the energy spectrum; based on the DOA coarse estimation result corresponding to each far-field target, an optimal phase offset corresponding to each peak value index is calculated by using a phase offset matrix, and then a high-precision DOA estimation result corresponding to each far-field target is generated. By introducing a phase offset in a peak value neighborhood and establishing a sub-grid offset estimation mechanism, the application overcomes the problem that the existing DOA estimation method is dependent on a discrete angle grid and is prone to estimation deviation or even missed detection when a real DOA deviates from a grid point, and realizes high-precision off-grid DOA estimation.

Description

Technical Field

[0001] This invention relates to the field of communication sensing technology, and specifically to a low-cost, fast DOA estimation method and apparatus. Background Technology

[0002] With the continuous evolution of wireless communication technology, the demand for high-precision target localization and DOA estimation for applications such as intelligent connected vehicles, millimeter-wave / terahertz sensing, and fusion positioning is becoming increasingly urgent. Among existing DOA estimation methods, the MUSIC algorithm based on subspace decomposition has good angular resolution under ideal array and high signal-to-noise ratio conditions, but it is often difficult to work stably in real complex propagation environments: on the one hand, multipath reflection and scattering in dense urban areas will change the statistical characteristics of the received signal, weakening the subspace separation effect under feature decomposition; on the other hand, in non-line-of-sight (NLoS) or occluded scenarios, the direct path is blocked and the effective signal energy decreases, which can easily lead to problems such as spectral peak broadening, angle ambiguity, and estimation offset.

[0003] To improve sensing and localization performance under adverse propagation conditions, reshaping the wireless propagation environment using controllable devices has gradually become an important research direction. Among them, reconfigurable smart surfaces (RIS) provide a new approach for channel modulation and assisted sensing. Compared with traditional passive reflective structures, semi-passive RIS, by configuring receiving and detection capabilities on a small number of units, can achieve programmable control of the reflection phase while maintaining low power consumption and low hardware cost. It can also acquire incident signal information to a certain extent, thus providing an additional observation dimension for DOA estimation under NLoS conditions.

[0004] However, existing RIS-assisted DOA estimation methods still have shortcomings: First, some estimators still use the classical array processing framework and do not fully utilize the structured observation advantages brought by RIS programmable phase modulation, resulting in insufficient robustness under strong multipath and model mismatch conditions; Second, although sparse reconstruction-based schemes (such as atomic norm minimization, ANM) can fully utilize the structured observation advantages of RIS, they usually involve high-dimensional convex optimization or iterative solutions, resulting in large computational and storage overhead, making it difficult to meet real-time processing requirements; Third, the above estimators rely on discrete angle grids for searching, and when the true DOA does not fall on the preset grid points, estimation bias or even missed detection is likely to occur.

[0005] Therefore, there is an urgent need for a target perception and orientation estimation method applicable to complex environments such as occlusion and dense multipath, in order to solve the problems of insufficient accuracy and high computational complexity of existing DOA estimation methods in multipath scattering, NLoS and off-grid scenarios. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a low-cost, fast DOA estimation method and apparatus.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides a low-cost and fast DOA estimation method. A base station, a semi-passive intelligent reflective surface, and a predetermined number of far-field targets are deployed in a two-dimensional plane, with the line-of-sight paths between the base station and each far-field target obstructed. The following steps are used to obtain high-precision DOA estimation results for each far-field target under non-line-of-sight paths:

[0009] Step S1: Initialize the current snapshot time index;

[0010] Step S2: Based on the two-level channel model, calculate the phase shift vector and beam pointing of the passive reflection array at the current snapshot time, and then obtain the beam echo signal received by the active sensing array at the current snapshot time. Determine whether the current snapshot time index is equal to the preset total number of snapshot time indices. If yes, construct the observation matrix and execute step S3. Otherwise, increment the current snapshot time index by 1 and return to execute step S2.

[0011] Step S3: Map the direction of the reflected beam generated by the passive reflection array at each snapshot moment to the pre-constructed virtual space frequency domain, generate the energy spectrum corresponding to the observation matrix using cross-channel incoherent energy detection, and generate the DOA coarse estimation results corresponding to each far-field target according to the peak index corresponding to each peak in the energy spectrum.

[0012] Step S4: Based on the coarse DOA estimation results corresponding to each far-field target, the optimal phase offset corresponding to each peak index is calculated using the phase offset matrix, and the virtual spatial frequency of each far-field target is refined based on the optimal phase offset. Then, the high-precision DOA estimation results corresponding to each far-field target are obtained through arcsine transformation.

[0013] Furthermore, the passive reflection array and active sensing array equipped in the semi-passive intelligent reflective surface are both uniform linear arrays, and the number of array elements are respectively... and The secondary channel model in step S2 includes the channel model from the base station to the passive reflector array, the channel model from the passive reflector array to the far-field target, and then the channel model from the far-field target to the active sensing array.

[0014] (a) The channel model from the base station to the passive reflection array is as follows:

[0015] ;

[0016] ;

[0017] in, This is the channel model from the base station to the passive reflection array. For the channel loss from the base station to the passive reflector array, The distance between the base station and the semi-passive intelligent reflective surface. The wavelength of the signal transmitted by the base station. The imaginary unit, This is the array response vector of the passive reflection array. The DOA of the base station relative to the passive reflector array. Here, H represents the array response vector of the base station, and H denotes the conjugate transpose. The departure angle of the base station;

[0018] (b) The channel model from the passive reflective array to the far-field target, and then from the far-field target to the active sensing array, is as follows:

[0019] ;

[0020] ;

[0021] in, For far-field targets, For passive reflective array to the first The first far-field target, then the second Channel model from a far-field target to an active sensing array; For passive reflective array to the first The first far-field target, then the second Channel loss from a far-field target to an active sensing array; For the first Radar cross-section of a far-field target Indicates the semi-passive intelligent reflective surface and the first The distance to a distant target. This is the array response vector of the active sensing array. For the first active sensing array The true DOA value of each far-field target.

[0022] Further, step S2 calculates the phase shift vector of the passive reflective array at each snapshot time according to the following formula:

[0023] ;

[0024] in, For indexing quick snapshot moments, For the passive reflective array in the first The phase shift vector at each snapshot moment, Indicates the first in the passive reflection array Phase shift of each array element, This represents the number of elements in the passive reflection array. For the passive reflective array The phase of each array element, is the imaginary unit, and T is the transpose.

[0025] Further, step S2 obtains the beam echo signal received by the active sensing array at each snapshot time according to the following formula:

[0026] ;

[0027] in, For the active sensing array in the first The beam echo signal received at each snapshot moment The amplitude of the transmitted signal is given by , and k is the far-field target. To preset the total number of far-field targets, It is a diagonal array formed by the phase shifts of the elements in a passive reflection array. This is the channel model from the base station to the passive reflection array. Let be the beamforming vector, and satisfy . , express Norm; For unit power transmission signals, This is an additive white Gaussian noise vector, where each element follows a complex normal distribution. , Indicates noise power;

[0028] Furthermore, at each snapshot moment, by aligning the base station's transmitted signal with the semi-passive smart reflective surface, the received signal power of the active sensing array can be maximized:

[0029] ;

[0030] ;

[0031] ;

[0032] in, This refers to the number of antennas at the base station. For the channel loss from the base station to the passive reflector array, The channel loss is from the passive reflective array to the k-th far-field target, and then from the k-th far-field target to the active sensing array. This is the array response vector of the active sensing array. Let DOA be the value of the k-th far-field target on the active sensing array. Here is the array response vector of the passive reflection array, and H represents the conjugate transpose. The DOA of the base station relative to the passive reflector array. Let x be the amplitude of the echo signal of the k-th beam. Let DOA be the virtual spatial frequency domain of the k-th far-field target.

[0033] Furthermore, step S2 constructs an observation matrix by stacking the beam echo signals received by the active reflective array at all snapshot times:

[0034] ;

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] ; ;

[0040] in, To preset the total number of snapshot time indices, For the observation matrix, For active sensor array manifold, This is the overall channel loss matrix. It is a passive reflective array manifold. For the noise matrix, For the measurement matrix, For the passive reflective array in the first The direction of the reflected beam at a quick snapshot moment.

[0041] Further, in step S3, the direction of the reflected beam generated by the passive reflective array at each snapshot moment is mapped to the pre-constructed virtual spatial frequency domain according to the following formula:

[0042] ;

[0043] in, For indexing quick snapshot moments, To preset the total number of snapshot time indices, For the passive reflection array in the virtual space frequency domain at the 1st The direction of the reflected beam at a quick snapshot moment. For the passive reflective array in the first The direction of the reflected beam at a snapshot moment;

[0044] The energy spectrum corresponding to the observation matrix is ​​generated according to the following formula:

[0045] ;

[0046] in, The energy spectrum corresponding to the observation matrix. For the active sensing array in the first The beam echo signal received at each snapshot moment.

[0047] Further, step S3 generates coarse DOA estimation results for each far-field target according to the following steps:

[0048] Step S31, Define Used to represent the energy spectrum The indexes corresponding to each spectral peak are used, and the following formula is used to generate a coarse estimate of the DOA of each far-field target in the virtual spatial frequency domain:

[0049] ;

[0050] ;

[0051] Where k is the far-field target, To preset the total number of far-field targets, Let be the virtual spatial frequency of the k-th far-field target. This is a coarse estimate of the DOA of the k-th far-field target in the virtual space frequency domain.

[0052] Step S32: Based on the mapping relationship between the DOA at the active sensing array and the DOA at the passive reflective array, the following formula is used to generate a coarse estimate of the DOA for each far-field target at the active sensing array:

[0053] ;

[0054] ;

[0055] in, Used to truncate its input to a range ; For the first active sensing array DOA of a far-field target The base station is the DOA relative to the passive reflector array; This is a coarse estimate of the DOA of the k-th far-field target at the active sensing array. Let be the true DOA value of the k-th far-field target in the virtual space frequency domain.

[0056] Further, in step S4, the phase offset matrix, the offset-assisted transformation measurement matrix, and the focusing index are constructed sequentially according to the following formulas:

[0057] ;

[0058] ;

[0059] ;

[0060] in, This is the phase offset matrix. The imaginary unit, For real-valued rotation parameters, This represents the number of elements in the passive reflection array. For the offset-assisted transformation measurement matrix, For the observation matrix, For the measurement matrix, H denotes the conjugate transpose. The number of elements in the active sensing array. To focus on indicators, express No. List;

[0061] The optimal phase offset corresponding to each peak index is then calculated using a one-dimensional search.

[0062] ;

[0063] in, This is the optimal phase offset. This is a preset symmetrical search interval.

[0064] Furthermore, based on the optimal phase offset, step S4 refines the virtual spatial frequency of the k-th target according to the following formula:

[0065] ;

[0066] in, The virtual spatial frequency of the k-th target after refinement;

[0067] And generate high-precision DOA estimation results for each far-field target at the active sensing array according to the following formula:

[0068] ;

[0069] in, High-precision DOA estimation results for each far-field target.

[0070] Another aspect of the present invention provides a low-cost, fast DOA estimation device, comprising:

[0071] Initialization module: Used to set the initial parameters of the system, including the snapshot index and the total number of snapshots;

[0072] Data acquisition module: used to determine the direction of the reflected beam and the echo signal received by the active reflection array frame by frame, and then generate an observation matrix by stacking the received echo signals at all snapshot moments;

[0073] Coarse estimation module: It is used to map the direction of the reflected beam to the virtual space frequency domain, generate the energy spectrum corresponding to the observation matrix by using cross-channel incoherent energy detection, and generate the DOA coarse estimation results for each far-field target according to the peak index corresponding to each peak in the energy spectrum.

[0074] Fine estimation module: Based on the coarse estimation results, it refines and corrects the virtual frequency domain using the phase offset matrix, and outputs the high-precision DOA of each far-field target through arcsine transformation.

[0075] The beneficial effects of adopting the above technical solution are as follows:

[0076] (1) This invention solves the problem of limited estimation accuracy of existing DOA estimation methods in complex environments due to model mismatch and discrete grid search by using structured observation and phase shift mechanism, and significantly improves the DOA estimation accuracy;

[0077] (2) By introducing phase shift in the peak neighborhood and establishing a subgrid offset estimation mechanism, this invention overcomes the shortcomings of existing DOA estimation methods that rely on discrete angle grids and are prone to estimation bias or even missed detection when the real DOA deviates from the grid point, and achieves high-precision off-grid DOA estimation.

[0078] (3) This invention utilizes a semi-passive RIS hardware architecture to effectively reduce hardware costs, and replaces complex matrix decomposition and convex optimization with energy spectrum detection and one-dimensional phase shift search, thereby significantly reducing computational complexity while ensuring estimation accuracy, and achieving low computational complexity and high real-time performance. Attached Figure Description

[0079] Figure 1 This is a scenario diagram for DOA estimation under semi-passive RIS assistance according to the present invention;

[0080] Figure 2 This is a flowchart of the DOA estimation method of the present invention;

[0081] Figure 3 This is a schematic diagram of the normalized energy spectrum of the present invention;

[0082] Figure 4 This is a schematic diagram illustrating the root mean square error of the estimation method of the present invention and the traditional estimation method under different signal-to-noise ratios;

[0083] Figure 5This diagram illustrates the running time of the estimation method of this invention and the traditional estimation method under different signal-to-noise ratios. Detailed Implementation

[0084] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0085] Example 1:

[0086] refer to Figure 1 A semi-passive intelligent reflective surface-assisted perception scenario is constructed within a two-dimensional plane. Specifically, this scenario includes: One far-field target, one equipped with A base station with a uniform linear array of antennas is used, but the line-of-sight path between the base station and the target is blocked. To sense the DOA (Direction of Arrival) of each far-field target, a semi-passive intelligent reflective surface (RIS) is used to construct the non-line-of-sight path. Both the passive reflective array and the active sensing array in the RIS are uniform linear arrays, with the number of array elements being... and .

[0087] refer to Figure 2 For the established perception scenario, this embodiment provides a low-cost and fast DOA estimation method, which obtains high-precision DOA estimation results for each far-field target under non-line-of-sight paths according to the following steps:

[0088] Step S1: Initialize the current snapshot time index;

[0089] Step S2: Based on the two-level channel model, calculate the phase shift vector and beam pointing of the passive reflection array at the current snapshot time, and then obtain the beam echo signal received by the active sensing array at the current snapshot time. Determine whether the current snapshot time index is equal to the preset total number of snapshot time indices. If yes, construct the observation matrix and execute step S3. Otherwise, increment the current snapshot time index by 1 and return to execute step S2.

[0090] Step S3: Map the direction of the reflected beam generated by the passive reflection array at each snapshot moment to the pre-constructed virtual space frequency domain, generate the energy spectrum corresponding to the observation matrix using cross-channel incoherent energy detection, and generate the DOA coarse estimation results corresponding to each far-field target according to the peak index corresponding to each peak in the energy spectrum.

[0091] Step S4: Based on the coarse DOA estimation results corresponding to each far-field target, the optimal phase offset corresponding to each peak index is calculated using the phase offset matrix, and the virtual spatial frequency of each far-field target is refined based on the optimal phase offset. Then, the high-precision DOA estimation results corresponding to each far-field target are obtained through arcsine transformation.

[0092] Furthermore, the secondary channel model in step S2 includes a channel model from the base station to the passive reflection array, a channel model from the passive reflection array to the far-field target, and then a channel model from the far-field target to the active sensing array:

[0093] (a) The channel model from the base station to the passive reflection array is as follows:

[0094] ;

[0095] ;

[0096] in, This is the channel model from the base station to the passive reflection array. For the channel loss from the base station to the passive reflector array, The distance between the base station and the semi-passive intelligent reflective surface. The wavelength of the signal transmitted by the base station. The imaginary unit, This is the array response vector of the passive reflection array. The DOA of the base station relative to the passive reflector array. Here, H represents the array response vector of the base station, and H denotes the conjugate transpose. The departure angle of the base station;

[0097] (b) The channel model from the passive reflective array to the far-field target, and then from the far-field target to the active sensing array, is as follows:

[0098] ;

[0099] ;

[0100] Where k is the far-field target, This is a channel model for the passive reflective array to the k-th far-field target, and then from the k-th far-field target to the active sensing array; The channel loss is from the passive reflective array to the k-th far-field target, and then from the k-th far-field target to the active sensing array. Let be the radar cross-section of the k-th far-field target. This represents the distance between the semi-passive intelligent reflective surface and the k-th far-field target. This is the array response vector of the active sensing array. For the first active sensing array DOA of a far-field target.

[0101] In this embodiment, the element spacing of the passive reflection array, the active sensing array, and the array on the base station is half a wavelength, and the RIS adjusts the beam once at each snapshot. Therefore, step S2 calculates the phase shift vector of the passive reflection array at each snapshot using the following formula:

[0102] ;

[0103] in, For indexing quick snapshot moments, For the passive reflective array in the first The phase shift vector at each snapshot moment, Indicates the first in the passive reflection array Phase shift of each array element, For the passive reflective array The phase of each array element, where T is the transpose.

[0104] Further, step S2 obtains the beam echo signal received by the active sensing array at each snapshot time according to the following formula:

[0105] ;

[0106] in, Let be the beam echo signal received by the active sensing array at the t-th snapshot time. The amplitude of the transmitted signal is given by , and k is the far-field target. The total number of far-field targets. It is a diagonal array formed by the phase shifts of the elements in a passive reflection array. Let be the beamforming vector, and satisfy . , express Norm; For unit power transmission signals, This is an additive white Gaussian noise vector, where each element follows a complex normal distribution. , Indicates noise power;

[0107] Preferably, at each snapshot moment, by aligning the base station's transmitted signal with the semi-passive smart reflective surface, the received signal power of the active sensing array can be maximized.

[0108] ;

[0109] ;

[0110] ;

[0111] in, Let x be the amplitude of the echo signal of the k-th beam. Let be the true DOA value of the k-th far-field target in the virtual space frequency domain.

[0112] Furthermore, step S2 constructs an observation matrix by stacking the beam echo signals received by the active reflective array at all snapshot times:

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] ;

[0118] ; ;

[0119] in, To preset the total number of snapshot time indices, For the observation matrix, For active sensor array manifold, This is the overall channel loss matrix. It is a passive reflective array manifold. For the noise matrix, For the measurement matrix, For the passive reflective array in the first The direction of the reflected beam at a quick snapshot moment.

[0120] In this embodiment, a predefined measurement matrix can be used at the base station, and the measurement matrix can be transmitted to the RIS in advance through the RIS controller. Meanwhile, to ensure the accuracy of DOA estimation at the active sensing unit, the measurement matrix in the RIS is configured as a DFT codebook.

[0121] Further, in step S3, the direction of the reflected beam generated by the passive reflective array at each snapshot moment is mapped to the pre-constructed virtual spatial frequency domain according to the following formula:

[0122] ;

[0123] in, For the passive reflection array in the virtual space frequency domain at the 1st The direction of the reflected beam at a quick snapshot moment.

[0124] In this embodiment, the "virtual spatial frequency domain" can be understood as mapping the beam pointing angle to a linear variable u using a sine wave. Peak search and refinement are performed in the virtual spatial frequency domain using a DFT "frequency grid," where the grid points represent discrete directions of the RIS beam pointing in the virtual spatial frequency domain, and the constant interval between adjacent grid points is... Furthermore, the t-th column of the observation matrix can be interpreted as a RIS scan. The detection response obtained at that time.

[0125] The energy spectrum corresponding to the observation matrix is ​​generated according to the following formula:

[0126] ;

[0127] in, This represents the energy spectrum corresponding to the observation matrix.

[0128] Further, step S3 generates coarse DOA estimation results for each far-field target according to the following steps:

[0129] Step S31, Define Used to represent the energy spectrum The indexes corresponding to each spectral peak reflect the approximate location of each far-field target in the virtual space frequency domain, i.e., the grid point number where the far-field target is located. Therefore, using the following formula, a coarse DOA estimate of each far-field target is generated in the virtual space frequency domain:

[0130] ;

[0131] ;

[0132] in, Let be the virtual spatial frequency of the k-th far-field target. This is a coarse estimate of the DOA of the k-th far-field target in the virtual space frequency domain.

[0133] Step S32: Based on the mapping relationship between the DOA at the active sensing array and the DOA at the passive reflective array, the following formula is used to generate a coarse estimate of the DOA for each far-field target at the active sensing array:

[0134] ;

[0135] ;

[0136] in, Used to truncate its input to a range ; This is a coarse estimate of the DOA of the k-th far-field target at the active sensing array.

[0137] When performing a coarse DOA estimation for each far-field target, the resolution is limited by the grid spacing. This means that the resolution is affected by the actual virtual spatial frequency of the k-th far-field target at the passive sensing array. When a beam fails to fall precisely on a DFT grid point, its energy leaks into adjacent beams, causing a deviation in the spectral peak position. Therefore, to mitigate the leakage effect, this embodiment introduces a phase offset matrix into the passive reflection array and calculates the offset that maximizes energy concentration at each peak of the energy spectrum.

[0138] Specifically, the phase offset matrix is ​​first constructed using the following formula:

[0139] ;

[0140] in, This is the phase offset matrix. These are real-valued rotation parameters.

[0141] The offset-assisted transformation measurement matrix is ​​then constructed using the following formula:

[0142] ;

[0143] in, This is a transformation measurement matrix based on offset assistance. In the transformation measurement matrix, firstly, using... Extract passive reflection array domain information; then utilize Perform frequency shift (position) compensation; finally multiply by The compensated response is mapped back to the beam domain to enable a consistent assessment of the energy at a specific beam index.

[0144] Finally, a focusing index for measuring the energy focusing degree of each far-field target is constructed according to the following formula:

[0145] ;

[0146] in, To focus on indicators, express No. List.

[0147] Then, the optimal phase offset corresponding to each peak index is generated through a one-dimensional search:

[0148] ;

[0149] in, This is the optimal phase offset. This is a symmetrical search interval.

[0150] This embodiment sets the symmetric search interval according to existing super-resolution settings. This is used to indicate that half a frequency point of correction can be performed near the spectral peak.

[0151] Furthermore, based on the optimal phase offset, step S4 refines the virtual spatial frequency of the k-th target according to the following formula:

[0152] ;

[0153] in, The virtual spatial frequency of the k-th target after refinement;

[0154] In this embodiment, the mapping item Consistent with the symmetric search interval and grid spacing, that is, when exist When internal changes occur, the correction term The range of values ​​is This is equivalent to half the correction of a DFT frequency point interval in the virtual space frequency domain.

[0155] Then, using the following formula, high-precision DOA estimation results are generated for each far-field target at the active sensing array:

[0156] ;

[0157] in, High-precision DOA estimation results for each far-field target.

[0158] The technical effects of this embodiment will be further explained in detail below with reference to simulation experiments. The simulation experiments are based on MATLAB R2024a software, running on a Windows 11 operating system, with a hardware configuration of an i7-10700F processor and 64 GB of memory. For the semidefinite programming problem defined in ANM, the widely used MATLAB toolkit CVX is used for solution.

[0159] The simulation content and results analysis are as follows:

[0160] Figure 3 This is a schematic diagram of the normalized energy spectrum. The preset number of far-field targets is 12, the active sensor array is equipped with 8 elements, and the passive reflective array is equipped with 200 elements; the total number of snapshots is set to 200, and the signal-to-noise ratio is 10dB. From... Figure 3 It can be seen that even in multi-target scenarios, the DOA estimation method provided in this embodiment can still obtain clear and sharp spectral peaks in all real-world vicinities, and the positions of the spectral peaks are consistent with... Figure 3The reference markings for the true frequencies are basically consistent. This indicates that the DOA estimation method provided in this embodiment can effectively utilize the additional degrees of freedom provided by RIS, thereby achieving target resolution even when the number of resolvable targets exceeds the number of active sensor array elements. At the same time, the amplitude of spectral lines far from the spectral peaks remains at a level close to the noise floor (approximately −20 dB), indicating low energy leakage and strong multi-target separability and robustness.

[0161] Specifically, the signal-to-noise ratio is calculated using the following formula during the simulation process:

[0162] ;

[0163] in, This represents the F-norm.

[0164] Figure 4 This diagram illustrates the root mean square error of the estimation method of this invention and traditional estimation methods under different signal-to-noise ratios. The traditional estimation methods include the MUSIC estimation method and the ANM estimation method. The specific simulation parameters are: signal... 28GHz carrier frequency , , , , m, dBsm, target DOA is The distance from RIS to the target is In the ANM algorithm, the error tolerance is set to 1000. The grid search range is The simulation was run 500 times. Figure 4 It can be seen that traditional ANM estimation methods and MUSIC estimation methods have reached their performance limits because they cannot solve off-grid errors. In contrast, the DOA estimation method provided in this embodiment can achieve sub-grid point accuracy as the signal-to-noise ratio increases, and is close to the performance limit (RCRB).

[0165] Specifically, the simulation process calculates the root mean square error according to the following formula:

[0166] ;

[0167] in, For the number of Monte Carlo experiments, Indicates the first The far-field target in the first The estimated DOA value in this simulation experiment.

[0168] Figure 5This diagram illustrates the running time of the estimation method of this invention and traditional estimation methods under different signal-to-noise ratios. The traditional estimation methods include the MUSIC estimation method and the ANM estimation method. Simulation conditions and... Figure 4 The average run time was consistent across 500 Monte Carlo trials, presented as mean ± standard deviation. Figure 5 As can be seen, the runtime curves of each method remain relatively flat within the examined signal-to-noise ratio range, indicating that the computational cost is mainly dominated by the problem dimension and the internal search / optimization process, rather than by the noise level. Specifically, due to the need for iterative optimization, the ANM method consistently has the longest runtime; the MUSIC method has the shortest runtime and the fastest computation speed. The proposed method's runtime falls between that of the MUSIC and ANM methods, thus achieving a favorable trade-off between estimation accuracy and computational complexity.

[0169] In summary, the analysis of the simulation results shows that the DOA estimation method provided in this embodiment can effectively estimate the DOA of targets far from the grid. Furthermore, this method fully utilizes the characteristics of RIS (Reference-Induced Allocation), improving DOA estimation accuracy compared to traditional MUSIC methods. Compared to the ANM method, it achieves sub-grid-point accuracy while maintaining lower computational complexity.

[0170] Example 2:

[0171] This embodiment provides a low-cost, fast DOA estimation device, including:

[0172] Initialization module: Used to set the initial parameters of the system, including the snapshot index and the total number of snapshots;

[0173] Data acquisition module: used to determine the direction of the reflected beam and the echo signal received by the active reflection array frame by frame, and then generate an observation matrix by stacking the received echo signals at all snapshot moments;

[0174] Coarse estimation module: It is used to map the direction of the reflected beam to the virtual space frequency domain, generate the energy spectrum corresponding to the observation matrix by using cross-channel incoherent energy detection, and generate the DOA coarse estimation results for each far-field target according to the peak index corresponding to each peak in the energy spectrum.

[0175] Fine estimation module: Based on the coarse estimation results, it refines and corrects the virtual frequency domain using the phase offset matrix, and outputs the high-precision DOA of each far-field target through arcsine transformation.

[0176] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A low-cost, fast DOA estimation method, characterized in that, A base station, a semi-passive intelligent reflective surface, and a predetermined number of far-field targets are deployed in a two-dimensional plane, with the line-of-sight paths between the base station and each far-field target obstructed. The high-precision DOA estimation results for each far-field target under non-line-of-sight paths are obtained by following these steps: Step S1: Initialize the current snapshot time index; Step S2: Based on the two-level channel model, calculate the phase shift vector and beam pointing of the passive reflection array at the current snapshot time, and then obtain the beam echo signal received by the active sensing array at the current snapshot time. Determine whether the current snapshot time index is equal to the preset total number of snapshot time indices. If yes, construct the observation matrix and execute step S3. Otherwise, increment the current snapshot time index by 1 and return to execute step S2. Step S3: Map the direction of the reflected beam generated by the passive reflection array at each snapshot moment to the pre-constructed virtual spatial frequency domain, generate the energy spectrum corresponding to the observation matrix using cross-channel incoherent energy detection, and generate the DOA coarse estimation results for each far-field target according to the peak index corresponding to each spectral peak in the energy spectrum. Step S4: Based on the coarse DOA estimation results corresponding to each far-field target, the optimal phase offset corresponding to each peak index is calculated using the phase offset matrix, and the virtual spatial frequency of each far-field target is refined based on the optimal phase offset. Then, the high-precision DOA estimation results corresponding to each far-field target are obtained through arcsine transformation.

2. The low-cost, fast DOA estimation method according to claim 1, characterized in that, The passive reflection array and active sensing array equipped in the semi-passive intelligent reflective surface are both uniform linear arrays, and the number of array elements are respectively... and The secondary channel model in step S2 includes the channel model from the base station to the passive reflector array, the channel model from the passive reflector array to the far-field target, and then the channel model from the far-field target to the active sensing array. (a) The channel model from the base station to the passive reflection array is as follows: ； ； in, This is the channel model from the base station to the passive reflection array. For the channel loss from the base station to the passive reflector array, The distance between the base station and the semi-passive intelligent reflective surface. The wavelength of the signal transmitted by the base station. The imaginary unit, This is the array response vector of the passive reflection array. The DOA of the base station relative to the passive reflector array. Here, H represents the array response vector of the base station, and H denotes the conjugate transpose. The departure angle of the base station; (b) The channel model from the passive reflective array to the far-field target, and then from the far-field target to the active sensing array, is as follows: ； ； in, For far-field targets, For passive reflective array to the first The first far-field target, then the second Channel model from a far-field target to an active sensing array; For passive reflective array to the first The first far-field target, then the second Channel loss from a far-field target to an active sensing array; For the first Radar cross-section of a far-field target Indicates the semi-passive intelligent reflective surface and the first The distance to a distant target. This is the array response vector of the active sensing array. For the first active sensing array DOA of a far-field target.

3. The low-cost, fast DOA estimation method according to claim 1, characterized in that, Step S2 calculates the phase shift vector of the passive reflective array at each snapshot time according to the following formula: ； in, For indexing quick snapshot moments, For the passive reflective array in the first The phase shift vector at each snapshot moment, Indicates the first in the passive reflection array Phase shift of each array element, This represents the number of elements in the passive reflection array. For the passive reflective array The phase of each array element, is the imaginary unit, and T is the transpose.

4. The low-cost, fast DOA estimation method according to claim 3, characterized in that, Step S2 obtains the beam echo signal received by the active sensing array at each snapshot time according to the following formula: ； in, For the active sensing array in the first The beam echo signal received at each snapshot moment The amplitude of the transmitted signal, For far-field targets, To preset the total number of far-field targets, It is a diagonal array formed by the phase shifts of the elements in a passive reflection array. This is the channel model from the base station to the passive reflection array. Let be the beamforming vector, and satisfy . , express Norm; For unit power transmission signals, This is an additive white Gaussian noise vector, where each element follows a complex normal distribution. , Indicates noise power; Furthermore, at each snapshot moment, by aligning the base station's transmitted signal with the semi-passive smart reflective surface, the received signal power of the active sensing array can be maximized: ； ； ； in, This refers to the number of antennas at the base station. For the channel loss from the base station to the passive reflector array, For passive reflective array to the first The first far-field target, then the second Channel loss from a far-field target to an active sensing array; This is the array response vector of the active sensing array. For the first active sensing array DOA of a far-field target Here is the array response vector of the passive reflection array, and H represents the conjugate transpose. The DOA of the base station relative to the passive reflector array. For the first The amplitude of the beam echo signal, For the first The true DOA value of a far-field target in the virtual space frequency domain.

5. The low-cost, fast DOA estimation method according to claim 4, characterized in that, Step S2 involves constructing an observation matrix using the beam echo signals received by the stacked active reflective array at all snapshot times. ； ； ； ； ； ； ； in, To preset the total number of snapshot time indices, For the observation matrix, For active sensor array manifold, This is the overall channel loss matrix. It is a passive reflective array manifold. For the noise matrix, For the measurement matrix, For the passive reflective array in the first The direction of the reflected beam at a quick snapshot moment.

6. The low-cost, fast DOA estimation method according to claim 1, characterized in that, Step S3 maps the direction of the reflected beam generated by the passive reflective array at each snapshot moment to the pre-constructed virtual spatial frequency domain according to the following formula: ； in, For indexing quick snapshot moments, To preset the total number of snapshot time indices, For the passive reflection array in the virtual space frequency domain at the 1st The direction of the reflected beam at a quick snapshot moment. For the passive reflective array in the first The direction of the reflected beam at a snapshot moment; The energy spectrum corresponding to the observation matrix is ​​generated according to the following formula: ； in, The energy spectrum corresponding to the observation matrix. For the active sensing array in the first The beam echo signal received at each snapshot moment.

7. The low-cost, fast DOA estimation method according to claim 6, characterized in that, Step S3 generates coarse DOA estimation results for each far-field target according to the following steps: Step S31, Define Used to represent the energy spectrum The indexes corresponding to each spectral peak are used, and the following formula is used to generate a coarse estimate of the DOA of each far-field target in the virtual spatial frequency domain: ； ； in, For far-field targets, To preset the total number of far-field targets, For the first Virtual spatial frequency of a far-field target In the virtual space frequency domain, the first Coarse DOA estimation results for several far-field targets; Step S32: Based on the mapping relationship between the DOA at the active sensing array and the DOA at the passive reflective array, the following formula is used to generate a coarse estimate of the DOA for each far-field target at the active sensing array: ； ； in, Used to truncate its input to a range ; For the first active sensing array DOA of a far-field target The base station is the DOA relative to the passive reflector array; For the active sensing array at the first Coarse DOA estimation results for several far-field targets For the first The true DOA value of a far-field target in the virtual space frequency domain.

8. The low-cost, fast DOA estimation method according to claim 7, characterized in that, Step S4 constructs the phase offset matrix, the offset-assisted transformation measurement matrix, and the focusing index sequentially according to the following formulas: ； ； ； in, This is the phase offset matrix. The imaginary unit, For real-valued rotation parameters, This represents the number of elements in the passive reflection array. For the offset-assisted transformation measurement matrix, For the observation matrix, For the measurement matrix, H denotes the conjugate transpose. The number of elements in the active sensing array. To focus on indicators, express No. List; The optimal phase offset corresponding to each peak index is then calculated using a one-dimensional search. ； in, This is the optimal phase offset. This is a preset symmetrical search interval.

9. The low-cost, fast DOA estimation method according to claim 8, characterized in that, Based on the optimal phase offset, step S4 is refined according to the following formula: Virtual space frequency of each target: ； in, For the refined first The virtual space frequency of each target; And generate high-precision DOA estimation results for each far-field target at the active sensing array according to the following formula: ； in, High-precision DOA estimation results for each far-field target.

10. A low-cost, fast DOA estimation device, characterized in that, include: Initialization module: Used to set the initial parameters of the system, including the snapshot index and the total number of snapshots; Data acquisition module: used to determine the direction of the reflected beam and the echo signal received by the active reflection array frame by frame, and then generate an observation matrix by stacking the received echo signals at all snapshot moments; Coarse estimation module: It is used to map the direction of the reflected beam to the virtual space frequency domain, generate the energy spectrum corresponding to the observation matrix by using cross-channel incoherent energy detection, and generate the DOA coarse estimation results for each far-field target according to the peak index corresponding to each peak in the energy spectrum. Fine estimation module: Based on the coarse estimation results, it refines and corrects the virtual frequency domain using the phase offset matrix, and outputs the high-precision DOA of each far-field target through arcsine transformation.

Images