MIMO-OTFS channel estimation method fusing communication and perception

Abstract

The invention discloses a communication and perception fused MIMO-OTFS channel estimation method, and relates to the field of channel estimation. Perception information acquired by a radar is used as an initial support set of an improved algorithm, initial estimation is completed by combining with an optimal support set of delay, Doppler and angle domains, and then accurate estimation is performed by using a linear minimum mean square error method. The influence of noise on the system can be effectively relieved, the pilot frequency overhead is remarkably reduced, and the estimation precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of channel estimation, and specifically relates to a MIMO-OTFS channel estimation method that integrates communication and sensing. Background Art

[0002] Orthogonal time-frequency-space (OTFS) modulation is an emerging multi-carrier modulation technology. Its main advantage lies in the good sparsity, compactness, separability, and stability of the introduced delay-Doppler domain (DD domain), enabling this modulation method to achieve better performance advantages than orthogonal frequency-division multiplexing (OFDM) in high-speed mobile scenarios for 6G. In addition, multiple-input multiple-output (MIMO) technology has become an important pillar of the next-generation communication system due to its high spectral efficiency and large capacity. However, in a MIMO-OTFS system, the pilot overhead of channel estimation increases sharply with the increase in the number of antennas, maximum delay, and Doppler spread, which has become a key issue in the research of MIMO-OTFS key technologies at the present stage.

[0003] Regarding the research on channel estimation of OTFS, the prior art one generalizes it as channel estimation methods based on guard bands, interference cancellation, and other complex systems. The traditional guard-band-based estimation scheme has the advantage of being simple and easy to implement, but the introduction of guard bands also leads to waste of resources. Therefore, subsequent scholars superimposed data and pilots in the frame structure, effectively reducing resource waste. For complex systems, the ideas of sparse Bayesian estimation (SBL) and compressive sensing (CS) can be used to transform the channel estimation problem into a sparse signal reconstruction problem, realizing a simplified description of the estimation problem. The prior art two adds an enhanced rotational invariant subspace to the CS algorithm, effectively reducing the pilot overhead.

[0004] In recent years, with the rise of artificial intelligence technology, deep learning has been widely applied to wireless communication, and the combination of the two can greatly improve the efficiency of signal transmission. Among them, the prior art three uses the adaptive mechanism of the meta-algorithm to effectively utilize the channel correlation and improve the estimation accuracy. Secondly, the neural network can help optimize traditional algorithms and can also introduce the idea of denoising. Regarding the delay-Doppler domain as a 2D image, by regarding the obtained multiple images as multiple noise versions of the same image, and then using the denoising algorithm to improve the estimation performance. However, the above methods all focus on the single-input single-output (SISO-OTFS) tradition. For future 6G high-speed mobile scenarios, most of them are systems that meet MIMO-OTFS. Therefore, simply extending the above methods to the MIMO-OTFS system is not applicable. Therefore, we need to focus on the estimation problem under this system. Summary of the Invention

[0005] In view of the above deficiencies in the prior art, the MIMO-OTFS channel estimation method for integrating communication and sensing provided by the present invention solves the problem that it is difficult for the prior art to meet the channel estimation of the MIMO-OTFS system corresponding to the future 6G high-speed mobile scenario.

[0006] In order to achieve the above invention purpose, the technical solution adopted by the present invention is as follows: Provide a MIMO-OTFS channel estimation method for integrating communication and sensing, which includes: Remove the radar signals corresponding to static objects in the radar data frame to obtain the radar signals with static objects removed; Perform discrete Fourier transforms on the radar signals with static objects removed in three dimensions respectively to obtain the delay information, Doppler information and angle information corresponding to the moving targets; where the three dimensions are the delay dimension, the Doppler dimension and the angle dimension; Perform peak detection through the constant rate false alarm rate algorithm, and extract the corresponding delay information, Doppler information and angle information at each peak; Correspondingly convert the delay information, Doppler information and angle information corresponding to each peak into delay tap indices, Doppler tap indices and angle indices, and then construct an initial support set; Construct a sensing matrix and initialize the residual based on the sensing matrix and the received signal; Calculate the correlation between the current channel and the current residual based on the sensing matrix; Map the correlation in a three-dimensional grid to form a three-dimensional matrix; where the sizes of the three dimensions of the three-dimensional matrix are respectively equal to the delay dimension, the Doppler dimension and the angle dimension; Respectively obtain the strongest support values of the delay dimension, the Doppler dimension and the angle dimension from the three-dimensional matrix; merge the matrix formed by the three strongest support values with the current support set to obtain an updated support set; Based on the latest support set, perform channel estimation through the linear minimum mean square error method, and determine whether the end iteration condition is reached. If so, output the current channel estimation result; otherwise, update the residual and return to the operation of calculating the correlation.

[0007] Furthermore, the specific method for removing the radar signals corresponding to static objects in the radar data frame is as follows: Apply clutter removal to the radar data frame to obtain the radar signals with static objects removed, and its expression is:

[0008] where is the radar signal with static objects removed; represents respectively at the delay dimension, the Doppler dimension and the angle dimension Radar data frame in a state is the number of chirp signals in the radar data frame.

[0009] Furthermore, the expressions for performing discrete Fourier transforms on the radar signals with static objects removed in three dimensions are as follows:

[0010]

[0011]

[0012] where 、 and are the delay information, Doppler information, and angle information corresponding to the moving target respectively; is the natural constant; is the imaginary unit; is the pi; k is the index of the frequency component, used to identify different frequency components; 、 and are the total number of states in the delay dimension, the total number of states in the Doppler dimension, and the total number of states in the angle dimension respectively.

[0013] Furthermore, the specific method for peak detection through the constant false alarm rate algorithm is as follows: Sum the delay information, Doppler information, and angle information corresponding to the moving target to obtain the signal after discrete fast Fourier transform in three dimensions ; Perform detection on in the delay dimension and the angle dimension, and then calculate the mean in the Doppler dimension to obtain the signal for peak detection , and its expression is:

[0014] Extract the delay information, Doppler information, and angle information corresponding to each peak, expressed as:

[0015] where P represents the set of delay information, Doppler information, and angle information; represent the delay information, Doppler information, and angle information corresponding to the Jth peak respectively.

[0016] Furthermore, the method for converting the delay information, Doppler information, and angle information corresponding to each peak into delay tap indices, Doppler tap indices, and angle indices respectively is as follows: Convert the delay information into the index of the channel, and obtain the delay normalization result in the propagation path. The expression is:

[0017] where is the delay normalization result in the propagation path; is the delay information corresponding to the j th peak; represents the minimum delay information in the set P; Normalize the Doppler frequency in the propagation path by multiplying the Doppler information by the carrier frequency. The expression is:

[0018] where is the Doppler frequency normalization result in the propagation path; is the Doppler information corresponding to the j th peak; is the carrier frequency; Convert the delay normalization result in the propagation path and the Doppler normalization structure in the propagation path into the delay tap index and the Doppler tap index respectively. The expression is:

[0019]

[0020] where and are the delay tap index and the Doppler tap index respectively; represents the bandwidth of the OTFS system; represents the duration of an OTFS frame; is the floor function; Convert the angle information into the angle index. The expression is:

[0021] where is the that makes the spatial power spectrum function value maximum; is the angle index value; d is the wavelength; is the antenna spacing; is the angle information corresponding to the j th peak; represents taking the angle index value corresponding to the maximum spatial power spectrum function value.

[0022] Furthermore, the expression of the j th element in the initial support set is:

[0023] where is the number of transmitting antennas; is the length of the guard band in the time delay dimension.

[0024] Further, the expression for initializing the residual based on the sensing matrix and the received signal is:

[0025] where is the residual; is the received signal; is the estimated vector of the channel; A is the sensing matrix, , Z is the phase compensation matrix, P is the pilot signal matrix, denotes element-wise multiplication.

[0026] Further, the expression for calculating the correlation between the current channel and the current residual based on the sensing matrix is:

[0027] where z is the correlation; is the sensing matrix A is the conjugate transpose of.

[0028] Further, the method for respectively obtaining the strongest support values in the delay dimension, Doppler dimension, and angle dimension from the three-dimensional matrix; and merging the matrix formed by the three strongest support values with the current support set to obtain the updated support set specifically includes: Sum the three-dimensional matrix in the Doppler dimension and the angle dimension to obtain the projection energy distribution in the delay dimension , and its expression is:

[0029] where respectively refer to the delay dimension, Doppler dimension, and angle dimension; and are respectively the total number of states in the Doppler dimension and the total number of states in the angle dimension; Take the position with the maximum energy value in the projection energy distribution in the delay dimension as the strongest support value in the delay dimension , and its expression is:

[0030] where represents taking the position with the maximum energy value as the strongest support value in the delay dimension; For the delay dimension that has been fixed at the value of the strongest support, the energy value of the angular dimension is statistically calculated to generate the projection distribution of the Doppler dimension, and its expression is:

[0031] where is the projection distribution of the Doppler dimension; is the energy value of the angular dimension when the delay dimension has been fixed at the value of the strongest support; Take the position with the largest energy value in the projection distribution of the Doppler dimension as the strongest support value of the Doppler dimension , and its expression is:

[0032] where means taking the position with the largest energy value as the strongest support value of the Doppler dimension; On the basis that the strongest support value of the delay dimension and the strongest support value of the Doppler dimension have been determined, the projection distribution of the angular dimension is obtained, and its expression is:

[0033] where is the strongest support value and the strongest support value when both have been fixed at the value of the strongest support, and it is the energy value of the angular dimension; Take the position with the largest energy value in the projection distribution of the angular dimension as the strongest support value of the angular dimension, and its expression is:

[0034] where means taking the position with the largest energy value as the strongest support value of the angular dimension; The expression for merging the matrix formed by the three strongest support values with the initial support set is:

[0035] where S is the updated support set; is the current support set, and when the support set is updated for the first time, the current support set is the initial support set.

[0036] Furthermore, the expression for channel estimation by the linear minimum mean square error method is:

[0037] where is the channel estimation vector obtained by the linear minimum mean square error method under the latest support set; is the sensing matrix under the latest support set; is the conjugate transpose of; is the noise power; is the identity matrix; is the received signal.

[0038] The beneficial effects of the present invention are as follows: 1. The present invention uses the sensing information obtained by the radar as the initial support set of the improved algorithm, combines the optimal support sets in the delay, Doppler, and angle domains to complete the initial estimation, and then uses the linear minimum mean square error method (LMMSE algorithm) for accurate estimation. This method significantly reduces the pilot overhead and improves the estimation accuracy.

[0039] 2. The present invention performs three-dimensional structured sparsity on the channel information, improves the channel reconstruction accuracy, and adds the LMMSE algorithm for detailed estimation, further reducing the pilot overhead and improving the estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is the flow schematic diagram of this method; Figure 2 is the sensing model in the embodiment; Figure 3 is the pilot design scheme; Figure 4 is the performance comparison between the traditional algorithm and this method; Figure 5 is the NMSE performance comparison between the traditional algorithm and this method under different signal-to-noise ratios; Figure 6 is the algorithm complexity comparison of different algorithms at ; Figure 7 is the correlation coefficient comparison between the traditional algorithm and this method under the overhead ratio ; Figure 8 is the CCDF (Complementary Cumulative Distribution Function) between various algorithms. DETAILED DESCRIPTION OF THE INVENTION

[0041] The following describes the specific embodiments of the present invention to facilitate those skilled in the art of this technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of this technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions created using the concept of the present invention are within the scope of protection.

[0042] Embodiment 1: As Figure 1 shown, the MIMO-OTFS channel estimation method for integrated communication and sensing includes the following steps: S1. Remove the radar signals corresponding to static objects in the radar data frame to obtain the radar signals with static objects removed; S2. Perform discrete Fourier transforms on the radar signals with static objects removed in three dimensions respectively to obtain the delay information, Doppler information, and angle information corresponding to the moving targets; where the three dimensions are the delay dimension, the Doppler dimension, and the angle dimension respectively; S3. Perform peak detection through the constant false alarm rate algorithm and extract the corresponding delay information, Doppler information, and angle information at each peak; S4. Correspondingly convert the delay information, Doppler information, and angle information corresponding to each peak into delay tap indices, Doppler tap indices, and angle indices, and then construct an initial support set; S5. Construct a sensing matrix and initialize the residual based on the sensing matrix and the received signal; S6. Calculate the correlation between the current channel and the current residual based on the sensing matrix; S7. Map the correlation in a three-dimensional grid to form a three-dimensional matrix; where the sizes of the three dimensions of the three-dimensional matrix are respectively equal to the delay dimension, the Doppler dimension, and the angle dimension; S8. Respectively obtain the strongest support values of the delay dimension, the Doppler dimension, and the angle dimension from the three-dimensional matrix; merge the matrix formed by the three strongest support values with the current support set to obtain an updated support set; S9. Based on the latest support set, perform channel estimation through the linear minimum mean square error method and determine whether the end iteration condition is reached. If so, output the current channel estimation result; otherwise, update the residual and return to the operation of calculating the correlation.

[0043] In the specific implementation process, as Figure 2 shown, the sensing model selects the most common frequency-modulated continuous wave (FMCW) signal to complete the processing of the radar signal. The radar system first transmits this signal, which will interact with the surrounding environment of the user and then be reflected back to the radar system. The specifically transmitted frequency-modulated continuous wave signal is:

[0044] Among them 、 and respectively represent the starting frequency, slope, and duration of the transmitted signal, t represents time.

[0045] To obtain the complete Doppler parameters in a dynamic environment, it is necessary to transmit identical chirp signals to form the entire radar data frame:

[0046] Among them represents the chirp pulse repetition time, represents the duration of the radar data frame. In the radar system, an orthogonal mixer is used for the signal reflected back to the system to mix the transmitted and reflected signals. The in-phase and quadrature components of the output are then passed through a low-pass filter to obtain an intermediate frequency (IF) signal. Assuming w ideal reflectors are the sensing targets, then at the n th root ( ) the antenna can be written as:

[0047] Among them and are the wavelength and antenna spacing of the radar. and are the complex gain and round-trip propagation delay corresponding to the w th reflector, W is the total number of reflectors, exp represents the exponential with the natural constant e as the base, is the complex gain of the radar signal. The received signal at each antenna is sampled by an ADC to form the radar data frame represented by .

[0048] In this embodiment, the specific method for removing the radar signal corresponding to the static object in the radar data frame in step S1 is: Apply clutter removal to the radar data frame to obtain the radar signal with static objects removed, and its expression is:

[0049] Among them is the radar signal with static objects removed; represents the radar data frame in the delay dimension, Doppler dimension, and angle dimension at the state; is the number of chirp signals in the radar data frame.

[0050] In this embodiment, the expressions for performing discrete Fourier transforms on the radar signals from which static objects have been removed in three dimensions are as follows:

[0051]

[0052]

[0053] where 、 and are the delay information, Doppler information, and angle information corresponding to the moving target, respectively; is the natural constant; is the imaginary unit; is the pi; k is the index of the frequency component, used to identify different frequency components; 、 and are the total number of states in the delay dimension, the total number of states in the Doppler dimension, and the total number of states in the angle dimension, respectively.

[0054] In this embodiment, the specific method for peak detection through the constant false alarm rate algorithm includes the following steps: S3-1. Sum the delay information, Doppler information, and angle information corresponding to the moving target to obtain the signal after discrete fast Fourier transform in three dimensions ; S3-2. Detect in the delay dimension and the angle dimension, and then find the mean in the Doppler dimension to obtain the signal for peak detection , and its expression is:

[0055] S3-3. Extract the delay information, Doppler information, and angle information corresponding to each peak, expressed as:

[0056] where P represents the set of delay information, Doppler information, and angle information; respectively represent the delay information, Doppler information, and angle information corresponding to the Jth peak.

[0057] In this embodiment, the method for correspondingly converting the delay information, Doppler information, and angle information corresponding to each peak into delay tap indices, Doppler tap indices, and angle indices includes the following steps: S4-1. Convert the delay information into the index of the channel, and obtain the delay normalization result in the propagation path. Its expression is:

[0058] where is the delay normalization result in the propagation path; is the delay information corresponding to the j th peak; represents the minimum delay information in the set P; S4-2. The normalization based on the shortest radar propagation path can be detected as the peak of P. Since the shortest propagation path may be one of the strongest paths, the Doppler frequency in the propagation path is normalized by the product of the Doppler information and the carrier frequency. Its expression is:

[0059] where is the Doppler frequency normalization result in the propagation path; is the Doppler information corresponding to the j th peak; is the carrier frequency; S4-3. Convert the delay normalization result in the propagation path and the Doppler normalization structure in the propagation path into the delay tap index and the Doppler tap index respectively. Its expression is:

[0060]

[0061] where and are the delay tap index and the Doppler tap index respectively; represents the bandwidth of the OTFS system; represents the time duration of an OTFS frame; is the floor function; S4-4. Convert the angle information into the angle index. Its expression is:

[0062] where is the that makes the value of the spatial power spectrum function the largest; is the angle index value; d is the wavelength; is the antenna spacing; is the angle information corresponding to the j th peak; represents taking the angle index value corresponding to the largest value of the spatial power spectrum function.

[0063] In this embodiment, the expression of the j th element in the initial support set is:

[0064] where is the number of transmit antennas; is the length of the guard band in the time delay dimension.

[0065] In the specific implementation process, to save pilot resources, pilots and data are arranged in a frame structure as Figure 3 . The lengths of the pilots and guard symbols on the time delay axis are and respectively. Among them, the blank guard band is selected in the range of the full Doppler dimension to prevent crosstalk between data and pilots. The length of the guard band needs to satisfy , represents the maximum extension of the system in the delay dimension, and its pilot overhead ratio is measured by .

[0066] Use to represent the pilot index in the delay dimension, is the pilot index in the Doppler dimension, and its received signal can be written as:

[0067] where, represents the number of transmit antennas, represents the length of the guard band in the time delay dimension, is the channel response, , are the transmitted signal and received signal at the index value a at the th antenna respectively. The above formula is written in the form of a matrix and a vector, that is:

[0068] where, and are the vector forms of the received signal and the channel of this system respectively; and represent the phase compensation matrix and the pilot signal matrix respectively. Let , we can get:

[0069] where each element in corresponds to the time delay and Doppler taps in the DD domain at the a th antenna. Due to the sparsity of the DD channel, is a sparse vector, is a noise vector, and the improved compressive sampling matching pursuit algorithm is used to solve it based on the above sparsity to realize the channel estimation of MIMO-OTFS.

[0070] In this embodiment, the expression for initializing the residual based on the sensing matrix and the received signal is:

[0071] where is the residual; is the received signal; is the estimated vector of the channel; A is the sensing matrix, , Z is the phase compensation matrix, P is the pilot signal matrix, denotes element-wise multiplication.

[0072] In this embodiment, the expression for calculating the correlation between the current channel and the current residual based on the sensing matrix is:

[0073] where z is the correlation; is the sensing matrix A is the conjugate transpose of.

[0074] In the specific implementation process, compared with the traditional CoSaMP algorithm, we hope to make full use of the sparsity in the delay, Doppler, and angle dimensions to find the optimal support set. We map the above correlation z in a three-dimensional grid to obtain , denotes reshaping z from a one-dimensional vector into a three-dimensional matrix of size .

[0075] In this embodiment, the specific method for obtaining the strongest support values in the delay dimension, Doppler dimension, and angle dimension from the three-dimensional matrix respectively; merging the matrix formed by the three strongest support values with the current support set to obtain the updated support set includes the following steps: S8-1. Sum the three-dimensional matrix in the Doppler dimension and the angle dimension to obtain the projection energy distribution in the delay dimension , and its expression is:

[0076] where respectively refer to the delay dimension, Doppler dimension, and angle dimension; and are the total number of states in the Doppler dimension and the total number of states in the angle dimension respectively; S8-2. Take the position with the maximum energy value in the projected energy distribution of the delay dimension as the strongest support value of the delay dimension The expression is: , and its expression is:

[0077] where means taking the position with the maximum energy value as the strongest support value of the delay dimension; S8-3. For the values in the delay dimension that have been fixed to the strongest support, count the energy values of the angular dimension to generate the projected distribution of the Doppler dimension. The expression is:

[0078] where is the projected distribution of the Doppler dimension; is the energy value of the angular dimension when the values in the delay dimension have been fixed to the strongest support; S8-4. Take the position with the maximum energy value in the projected distribution of the Doppler dimension as the strongest support value of the Doppler dimension , and its expression is:

[0079] where means taking the position with the maximum energy value as the strongest support value of the Doppler dimension; S8-5. On the basis that the strongest support value of the delay dimension and the strongest support value of the Doppler dimension have been determined, the projected distribution of the angular dimension is obtained. The expression is:

[0080] where is the strongest support value and the strongest support value is the energy value of the angular dimension when they have been fixed to the strongest support; S8-6. Take the position with the maximum energy value in the projected distribution of the angular dimension as the strongest support value of the angular dimension , and its expression is:

[0081] where means taking the position with the maximum energy value as the strongest support value of the angular dimension; S8-7. The expression for merging the matrix formed by the three strongest support values with the initial support set is:

[0082] where S is the updated support set; is the current support set, and when the support set is updated for the first time, the current support set is the initial support set.

[0083] In this embodiment, the expression for channel estimation by the linear minimum mean square error method is:

[0084] where is the channel estimation vector obtained by the linear minimum mean square error method under the latest support set; is the sensing matrix under the latest support set; is the conjugate transpose of; is the noise power; is the identity matrix; is the received signal.

[0085] In the specific implementation process, the update expression of the residual is:

[0086] where is the updated residual.

[0087] When the residual is small enough, it means that the product of the estimated channel vector and the sensing matrix is closer to the received signal, that is, the estimation performance is the best; at the same time, to avoid the method falling into a dead loop, the iteration condition in this embodiment is that the residual value is less than the residual threshold or the number of iterations is less than the iteration threshold. When the residual value reaches the residual threshold or the number of iterations reaches the iteration threshold, it means that the iteration termination condition is reached.

[0088] In this embodiment, a simulation experiment is carried out on this method (R-TDCoSaMP). In order to verify the advantages of this method, the parameter settings are shown in Table 1.

[0089] Table 1

[0090] This embodiment compares the normalized mean square error (NMSE) performance of different channel estimation algorithms from different pilot overhead ratios and signal-to-noise ratios.

[0091]

[0092] Secondly, compare the correlation of the channel frequency response between different algorithms, and this correlation can measure the estimated channel frequency response and the true channel frequency response The similarity degree between them. It reflects whether the estimated channel can effectively reproduce the characteristics of the real channel. The closer the correlation is to 1, the more consistent the estimated channel is with the real channel. We select the correlation coefficient (Correlation Coefficient) to represent this correlation, where represents the Frobenius norm:

[0093] To obtain an effective estimation result, the simulation experiment of this embodiment is mainly divided into the following parts: (1) Perform three-dimensional optimization on the traditional CoSaMP algorithm and add an LMMSE module to achieve accurate estimation, and compare the performance between the optimized algorithm and the traditional algorithm. (2) Use the perception information obtained by the radar as the initial support set of the optimized algorithm, and further improve the estimation performance by using the interaction information between the user and the environment. (3) Compare the NMSE of different algorithms from the perspective of pilot overhead and signal-to-noise ratio, as well as their performance in terms of channel frequency response and complementary cumulative distribution function (CCDF).

[0094] Then we compared the performance between the traditional algorithm and this method, as Figure 4 shown. The signal-to-noise ratio here is set to 10 dB. It can be observed that the NMSE of our proposed R-TDCoSaMP is better than that of the traditional OMP, ROMP, and CoSaMP algorithms under different conditions. In particular, compared with the NMSE used in the traditional algorithm CoSaMP, when , the R-TDCoSaMP method obtains similar NMSE performance; compared with the traditional algorithm ROMP and the traditional algorithm OMP in the same state, the R-TDCoSaMP algorithm can obtain lower pilot overhead. This shows that this method reduces the pilot overhead by about 12.5% compared with the traditional CoSaMP.

[0095] In Figure 5 , the comparison of the NMSE performance between the traditional algorithm and this method under different signal-to-noise ratios is given, and the number of antennas is set to 32 at this time. In this case setting , it can be observed that the NMSE performance of these algorithms improves with the increase of the signal-to-noise ratio, and the performance of this method is better than that of the traditional algorithm. In particular, compared with CoSaMP, R-TDCoSaMP obtains similar NMSE performance when the signal-to-noise ratio is 3 dB lower, and is better than the traditional OMP and ROMP. Therefore, it can be seen from the simulation diagram that the channel estimation with radar perception information added and the LMMSE algorithm added in the three-dimensional selection of the support set can improve the estimation performance to a certain extent, and obtains a gain of about 3 dB compared with the traditional algorithm.

[0096] InFigure 6 compares the algorithm complexities of different algorithms at and calculates the running times of various algorithms. Select for analysis. It can be seen from the figure that the OMP algorithm takes a relatively long time of 0.90265 s; the ROMP algorithm takes a relatively long time of 0.07928 s; the CoSaMP algorithm takes a relatively long time of 0.04442 s; the time of R-TDCoSaMP is shortened to 0.03633 s. From the above results, it can be seen that the proposed method obtains lower running times at different signal-to-noise ratios. In particular, R-TDCoSaMP improves by 0.866 compared with OMP, by 0.043 s compared with ROMP, and by 0.008 s compared with CoSaMP.

[0097] Figure 7 compares the correlation coefficients of the traditional algorithm and the proposed method at the overhead ratio . It can be seen from the figure that the proposed R-TDCoSaMP algorithm can achieve better performance, and the channel frequency response of this algorithm is closest to 1.

[0098] Finally, we compare the CCDFs between algorithms, as Figure 8 shown. We compare the relationship between the NMSE of different algorithms directly and the threshold of NMSE, that is , where represents the threshold corresponding to the NMSE of the algorithm. It can be seen from the figure that the proposed R-TOCoSaMP algorithm can exhibit better robustness in the small error range and the large error range.

Claims

1. A MIMO-OTFS channel estimation method integrating communication and perception, characterized in that: include: Removing radar signals corresponding to static objects from the radar data frame to obtain radar signals from which the static objects have been removed; Performing discrete Fourier transform on the radar signal with static objects removed in three dimensions respectively to obtain delay information, Doppler information and angle information corresponding to the moving target; the three dimensions are delay dimension, Doppler dimension and angle dimension; Peak detection is performed using a constant rate false alarm rate algorithm, and the delay information, Doppler information, and angle information corresponding to each peak are extracted; The delay information, Doppler information and angle information corresponding to each peak are converted into delay tap index, Doppler tap index and angle index, and then the initial support set is constructed; Construct a sensing matrix and initialize the residual based on the sensing matrix and the received signal; Calculate the correlation between the current channel and the current residual based on the perception matrix; Mapping the correlation in a three-dimensional grid to form a three-dimensional matrix; wherein the sizes of the three dimensions of the three-dimensional matrix are respectively equal to the delay dimension, the Doppler dimension and the angle dimension; The strongest support values ​​of the delay dimension, the Doppler dimension and the angle dimension are obtained from the three-dimensional matrix respectively; the matrix formed by the three strongest support values ​​is merged with the current support set to obtain an updated support set; Based on the latest support set, the linear minimum mean square error method is used to perform channel estimation and determine whether the conditions for ending the iteration are met. If so, the current channel estimation result is output; Otherwise update the residual and return to the compute correlation operation.

2. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 1, characterized in that: The specific method of removing the radar signal corresponding to the static object in the radar data frame is: Applying clutter removal to the radar data frame, we get the radar signal with static objects removed, which is expressed as: in To remove the radar signal of static objects; It means that in the delay dimension, Doppler dimension and angle dimension, Radar data frame in status; is the number of linear frequency modulation signals in the radar data frame.

3. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 2, characterized in that: The expressions for discrete Fourier transform of radar signals with static objects removed in three dimensions are: in , and They are respectively the delay information, Doppler information and angle information corresponding to the moving target; is a natural constant; is an imaginary unit; is the ratio of pi; k is the index of the frequency component, used to identify different frequency components; , and They are the total number of states in the delay dimension, the total number of states in the Doppler dimension, and the total number of states in the angle dimension.

4. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 3, characterized in that: The specific method of peak detection using the constant rate false alarm rate algorithm is: The delay information, Doppler information and angle information corresponding to the moving target are summed to obtain the signal after discrete fast Fourier transform in three dimensions. ; right Detection is performed in the delay dimension and the angle dimension, and then the average is calculated in the Doppler dimension to obtain the signal for peak detection. , whose expression is: extract The delay information, Doppler information and angle information corresponding to each peak are expressed as: Where P represents the set of delay information, Doppler information and angle information; They respectively represent the delay information, Doppler information and angle information corresponding to the J-th peak.

5. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 4, characterized in that: The method of converting the delay information, Doppler information and angle information corresponding to each peak into the delay tap index, Doppler tap index and angle index is: The delay information is converted into the channel index to obtain the delay normalization result in the propagation path, which is expressed as: in is the delay normalization result in the propagation path; For the j The delay information corresponding to the peak value; Represents the minimum delay information in the set P; The Doppler frequency in the propagation path is normalized by the product of the Doppler information and the carrier frequency, and its expression is: in is the normalized result of Doppler frequency in the propagation path; For the j The Doppler information corresponding to the peak; is the carrier frequency; The delay normalization result in the propagation path and the Doppler normalization structure in the propagation path are converted into a delay tap index and a Doppler tap index respectively, and their expressions are: in and are the delay tap index and the Doppler tap index respectively; represents the bandwidth of the OTFS system; Indicates the duration of an OTFS frame; is the floor rounding function; Convert the angle information into an angle index, the expression is: in To maximize the value of the spatial power spectrum function ; is the angle index value; d is the wavelength; is the antenna spacing; For the j The angle information corresponding to the peak value; is the number of transmitting antennas; Indicates the angle index value corresponding to the maximum value of the spatial power spectrum function.

6. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 5, characterized in that: Initial support set j Elements The expression is: in is the length of the guard band in the delay dimension.

7. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 1, characterized in that: The expression of the residual initialized based on the sensing matrix and the received signal is: in is the residual; To receive the signal; is the estimated vector of the channel; A is the perception matrix, , Z is the phase compensation matrix, P is the pilot signal matrix, Represents element-wise multiplication.

8. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 7, characterized in that: The expression for calculating the correlation between the current channel and the current residual based on the perception matrix is: Where z is the correlation; The perception matrix A The conjugate transpose of .

9. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 1, characterized in that: Obtain the strongest support values ​​of the delay dimension, Doppler dimension and angle dimension from the three-dimensional matrix respectively; The specific method of merging the matrix formed by the three strongest support values ​​with the current support set to obtain the updated support set includes: For a three-dimensional matrix The projection energy distribution in the delay dimension is obtained by summing the Doppler dimension and the angle dimension. , whose expression is: in They refer to the delay dimension, Doppler dimension, and angle dimension respectively; and are the total number of states in the Doppler dimension and the total number of states in the angle dimension respectively; The projected energy distribution of the delay dimension The position with the largest energy value is taken as the strongest support value of the delay dimension , whose expression is: in It means taking the position with the largest energy value as the strongest support value of the delay dimension; The delay dimension has been fixed to the value under the strongest support, and the energy value of the angle dimension is counted to generate the projection distribution of the Doppler dimension, which is expressed as: in is the projection distribution in Doppler dimension; is the energy value of the angle dimension when the delay dimension has been fixed to the value under the strongest support; The projection distribution of the Doppler dimension The position with the largest energy value is taken as the strongest support value of the Doppler dimension , whose expression is: in It means taking the position with the largest energy value as the strongest support value of the Doppler dimension; The strongest support value in the delay dimension and the strongest support value of the Doppler dimension Based on the determined basis, the projection distribution of the angle dimension is obtained , whose expression is: in The strongest support value And the strongest support value The energy value of the angle dimension when it has been fixed to the value under the strongest support; The projection distribution of the angle dimension The position with the largest energy value is taken as the strongest support value in the angle dimension , whose expression is: in It means taking the position with the largest energy value as the strongest support value in the angle dimension; The expression for merging the matrix formed by the three strongest support values ​​with the initial support set is: Where S is the updated support set; is the current support set. When the support set is updated for the first time, the current support set is the initial support set.

10. The MIMO-OTFS channel estimation method integrating communication and perception according to claim 1, characterized in that: The expression for channel estimation using the linear minimum mean square error method is: in is the channel estimation vector obtained by the linear minimum mean square error method under the latest support set; is the perception matrix under the latest support set; for The conjugate transpose of ; is the noise power; is the identity matrix; To receive the signal.

Images