A method and system for estimating time-varying underwater acoustic channel of MIMO-OTFS system based on Kalman filter

Abstract

The application discloses a kind of MIMO-OTFS system time-varying underwater acoustic channel estimation method and system based on Kalman filter, comprising: obtaining signal from underwater acoustic channel, and extracting pilot signal from the signal obtained;Calculate filter error norm, and judge whether filter error norm is greater than the threshold value set, if greater than, pilot signal carries out 3D-SOMP calculation, and obtains the three-dimensional support set updated;If not greater than, according to pilot signal, three-dimensional support set and perception matrix, the underwater acoustic time-varying channel of current time is estimated using Kalman filtering algorithm;The application can avoid multiple iterations and repeated calculation of conventional method when estimating time-varying underwater acoustic channel, and can effectively improve operation speed and estimation accuracy.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic communication technology, specifically a time-varying underwater acoustic channel estimation method and system based on Kalman filter in MIMO-OTFS system. Background Technology

[0002] Underwater acoustic communication is currently the only way to achieve long-distance, high-speed data communication underwater. Underwater acoustic channels are characterized by rapid time-varying properties, large delay spread, severe Doppler effects, and limited available bandwidth. If the underwater acoustic channel cannot be accurately estimated, the reliability of communication will be affected. Therefore, accurate estimation of the underwater acoustic channel is particularly important. However, traditional time-varying underwater acoustic channel estimation suffers from high computational complexity, which limits its practical application. Summary of the Invention

[0003] Purpose of the invention: To address the high complexity of current time-varying underwater acoustic channel estimation methods, this invention proposes a time-varying underwater acoustic channel estimation method and system based on Kalman filters for MIMO-OTFS systems.

[0004] Technical solution: A time-varying underwater acoustic channel estimation method for a MIMO-OTFS system based on Kalman filters, comprising the following steps:

[0005] Step 1: Obtain the received signal from the underwater acoustic channel and extract the pilot signal from the received signal. Based on the pilot signal, generate a sensing matrix that satisfies the constraint equidistant characteristic criterion.

[0006] Step 2: Calculate the filter error norm and determine whether the filter error norm is greater than the set threshold. If it is greater, proceed to step 3; if it is not greater, proceed to step 4.

[0007] Step 3: Perform 3D-SOMP calculations on the received signal to obtain the updated three-dimensional support set, and then proceed to Step 4;

[0008] Step 4: Based on the pilot signal, three-dimensional support set and sensing matrix, the Kalman filter algorithm is used to estimate the underwater acoustic time-varying channel at the current moment to obtain the underwater acoustic channel at the current moment.

[0009] Furthermore, the calculation of the filter error norm includes:

[0010] According to equations (2) to (4), the filter error norm FEN is calculated as follows:

[0011]

[0012]

[0013]

[0014] In the formula, ∑ ic A represents the weighting coefficient. T P represents the transpose of the perception matrix. t|t-1 This represents the state covariance matrix at the current moment. This indicates that the matrices within the parentheses correspond to the rows and columns of T1 and T2, respectively, where T... t =[i1,i2,...,i n ] is a one-dimensional array representing the coordinates of a matrix row or column, A′ T This represents the conjugate transpose of the perceptron transpose. y represents the noise variance. t This indicates the received signal at the current moment. Let A represent the channel estimate at the current moment, and let A represent the sensing matrix. This represents the conjugate transpose of the Kalman update error.

[0015] Furthermore, the aforementioned 3D-SOMP calculation of the pilot signal specifically includes:

[0016] S210: Calculate the residual based on the received signal and the sensing matrix:

[0017] r = y - Ψh (i) (5)

[0018] Where r represents the residual, y represents the received signal, Ψ represents the sensing matrix, and h (i) This represents the estimated i-th path in the channel;

[0019] S220: Calculate the correlation between the perception matrix and the residuals:

[0020] e = Ψ H r (6)

[0021] S230: Based on the results of S220, determine the support sets for time-delay latitude, Doppler latitude, and angular latitude respectively.

[0022]

[0023]

[0024]

[0025] Among them, e r (m), e v (n) represent the residuals obtained by performing 2-norm operations along the time-delay latitude and Doppler latitude, respectively, e θ (g) represents the residual obtained after performing the 2-norm operation on the lifting transformation matrix. These are the support sets for time-delay latitude, Doppler latitude, and angular latitude, respectively.

[0026] Furthermore, step 4 specifically includes:

[0027] S510: Initialize the state vector h and the state covariance matrix P based on the known system state F and measurement noise Q;

[0028] S520: Based on the system model and the current state vector h k-1 The state transition equation is used to predict the state vector h at the next time step. k and state covariance matrix P k :

[0029]

[0030]

[0031] In the formula, F represents the channel prediction value at time k. k-1 This represents the prediction matrix at time k-1. This represents the channel estimate corrected at time k-1. This represents the state covariance matrix predicted at time k-1. This represents the state covariance matrix corrected at time k-1. Q represents the transpose of the prediction matrix at time k-1. k-1 This represents the noise matrix.

[0032] S530: Obtain the measurement value at the next moment, based on the perception matrix H. k The predicted state vector is mapped into the measurement space, and the residual between the predicted and actual measurements is calculated. Then, the residual is combined with the measurement noise covariance matrix and weighted using the Kalman gain K to obtain the optimal state estimate. and state covariance matrix

[0033]

[0034]

[0035]

[0036] In the formula, y k This represents the received signal at time k.

[0037] This invention discloses a time-varying underwater acoustic channel estimation system based on a Kalman filter for a MIMO-OTFS system, comprising:

[0038] The pilot signal extraction module is used to acquire the received signal from the underwater acoustic channel and extract the pilot signal from the received signal.

[0039] The filter error norm decision module is used to calculate the filter error norm and determine whether the filter error norm is greater than a set threshold. If it is greater, the 3D-SOMP calculation module is called; if it is not greater, the Kalman filter module is called.

[0040] The 3D-SOMP calculation module is used to perform 3D-SOMP calculations on the received signal to obtain the updated three-dimensional support set and call the Kalman filter module.

[0041] The Kalman filter module is used to estimate the time-varying underwater acoustic channel at the current moment using the Kalman filter algorithm based on the pilot signal, the three-dimensional support set, and the sensing matrix.

[0042] Furthermore, in the filter error norm decision module, the following steps are performed:

[0043] According to equations (2) to (4), the filter error norm FEN is calculated as follows:

[0044]

[0045]

[0046]

[0047] In the formula, ∑ ic A represents the weighting coefficient. T P represents the transpose of the perception matrix. t|t-1 This represents the state covariance matrix at the current moment. This indicates that the matrices within the parentheses correspond to the rows and columns of T1 and T2, respectively, where T... t =[i1,i2,...,i n ] is a one-dimensional array representing the coordinates of a matrix row or column, A′ T This represents the conjugate transpose of the perceptron transpose. y represents the noise variance. t This indicates the received signal at the current moment. Let A represent the channel estimate at the current moment, and let A represent the sensing matrix. This represents the conjugate transpose of the Kalman update error.

[0048] Determine if the filter error norm is greater than the set threshold. If it is, call the 3D-SOMP calculation module; if it is not, call the Kalman filter module.

[0049] Furthermore, in the 3D-SOMP calculation module, the following steps are performed:

[0050] S210: Calculate the residual based on the received signal and the sensing matrix:

[0051] r = y - Ψh (i) (5)

[0052] Where r represents the residual, y represents the received signal, Ψ represents the sensing matrix, and h (i) This represents the estimated i-th path in the channel;

[0053] S220: Calculate the correlation between the perception matrix and the residuals:

[0054] e = Ψ H r (6)

[0055] S230: Based on the results of S220, determine the support sets for time-delay latitude, Doppler latitude, and angular latitude respectively.

[0056]

[0057]

[0058]

[0059] Among them, e r (m), e v (n) represent the residuals obtained by performing 2-norm operations along the time-delay latitude and Doppler latitude, respectively, e θ (g) represents the residual obtained after performing the 2-norm operation on the lifting transformation matrix. These are the support sets for time-delay latitude, Doppler latitude, and angular latitude, respectively.

[0060] Furthermore, the following steps are performed in the Kalman filter module:

[0061] S510: Initialize the state vector h and the state covariance matrix P based on the known system state F and measurement noise Q;

[0062] S520: Based on the system model and the current state vector h k-1 The state transition equation is used to predict the state vector h at the next time step. k and state covariance matrix P k :

[0063]

[0064]

[0065] In the formula, F represents the channel prediction value at time k. k-1This represents the prediction matrix at time k-1. This represents the channel estimate corrected at time k-1. This represents the state covariance matrix predicted at time k-1. This represents the state covariance matrix corrected at time k-1. Q represents the transpose of the prediction matrix at time k-1. k-1 This represents the noise matrix.

[0066] S530: Obtain the measurement value at the next moment, based on the perception matrix H. k The predicted state vector is mapped into the measurement space, and the residual between the predicted and actual measurements is calculated. Then, the residual is combined with the measurement noise covariance matrix and weighted using the Kalman gain K to obtain the optimal state estimate. and state covariance matrix

[0067]

[0068]

[0069]

[0070] In the formula, y k This represents the received signal at time k.

[0071] This invention discloses a MIMO-OTFS system, including a transmitter and a receiver;

[0072] The transmitter includes an input signal stream generation module, an orthogonal amplitude modulation module, an inverse symplectic finite Fourier transform module, a Heisenberg transform module, a cyclic prefix insertion module, and a first serial-to-parallel transform module. The input signal stream generated by the input signal stream generation module flows into the orthogonal amplitude modulation module. The input signal stream is a binary bit stream and contains pilot information. After modulation, the signal passes sequentially through the inverse symplectic finite Fourier transform module, the Heisenberg transform module, the cyclic prefix insertion module, and the first serial-to-parallel transform module before being sent into the underwater acoustic channel to complete signal transmission.

[0073] The receiving end includes a time-varying underwater acoustic channel estimation module, which is a time-varying underwater acoustic channel estimation system for a MIMO-OTFS system based on a Kalman filter disclosed above.

[0074] Furthermore, in the quadrature amplitude modulation module, the following steps are performed:

[0075] In the time-delay-Doppler domain, the input signal stream is mapped into M×N orthogonal amplitude modulation symbols;

[0076] M×N quadrature amplitude modulation symbols are placed at sampling intervals T. s (s) and sampling frequency Δf=1 / T s The modulated signal X is obtained from an M×N time-delay Doppler grid divided in units of (Hz). DD :

[0077]

[0078] in, The resolution of the time delay axis. The resolution of the Doppler axis;

[0079] In the inverse symplectic finite Fourier transform module, the following steps are performed:

[0080] Signal X DD Performing an inverse symplectic finite Fourier transform (ISFFT) yields the time-frequency domain signal. Among them, F M Let F be the discrete Fourier matrix at point M. N It is an N-point discrete Fourier matrix;

[0081] In the Heisenberg transform module, the following steps are performed:

[0082] The time-frequency domain signal X FT Perform the Heisenberg transform to obtain the time-domain signal.

[0083] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0084] (1) This invention utilizes the known channel in the MIMO-OTFS communication system to estimate the time-varying channel. First, it generates a MIMO-OTFS communication transmission signal and incorporates pilot data into the transmission signal. Then, it generates a sensing matrix using the pilot data. After receiving the signal, the receiver first uses 3D-SOMP technology to progressively determine the support sets for the three dimensions of time delay, Doppler, and angle. Second, it uses the support sets to select data in the sensing matrix that are correlated with the time-varying channel. Finally, it uses the recursive characteristics of the Kalman filter to continuously track and estimate the time-varying underwater acoustic channel. This invention avoids the multiple iterations and repetitive calculations required by conventional methods when estimating the time-varying underwater acoustic channel, effectively improving computational speed and estimation accuracy.

[0085] (2) The method of the present invention can accurately estimate the time-varying underwater acoustic channel while greatly reducing the computational complexity. Attached Figure Description

[0086] Figure 1This is a system model block diagram for time-varying underwater acoustic channel estimation according to the present invention;

[0087] Figure 2 This is a flowchart of the 3D-SOMP computation process;

[0088] Figure 3 A flowchart for time-varying underwater acoustic channel estimation using a Kalman filter;

[0089] Figure 4 This is a block diagram of a complete MIMO-OTFS communication system. Detailed Implementation

[0090] The technical solution of the present invention will now be further described in conjunction with the accompanying drawings and embodiments.

[0091] Example 1:

[0092] like Figure 1 As shown, this embodiment discloses a time-varying underwater acoustic channel estimation method for a MIMO-OTFS system based on a Kalman filter, which mainly includes the following steps:

[0093] Step 1: In the time-delay-Doppler domain (DD domain), map the input signal stream into M×N quadrature amplitude modulation (QAM) symbols, and then place the M×N QAM symbols at sampling intervals T. s (s) and sampling frequency Δf=1 / T s The modulated signal X is obtained from an M×N time-delay Doppler grid divided in units of (Hz). DD :

[0094]

[0095] in, The resolution of the time delay axis. The resolution is the Doppler axis.

[0096] The obtained X DD Performing an inverse symplectic finite Fourier transform (ISFFT) yields the time-frequency domain signal. Among them, F M Let F be the discrete Fourier matrix at point M. N Given an N-point discrete Fourier matrix, the time-frequency domain signal X... FT Perform a Heisenberg transform to obtain the time-domain signal. The time-domain signal S is added with a cyclic prefix and then subjected to serial-to-parallel conversion before being sent into the underwater acoustic channel for transmission. The input signal stream in this step is a binary bit stream that includes pilot information.

[0097] Step 2: Acquire the received signal from the underwater acoustic channel. First, remove the cyclic prefix from the received signal to obtain the time-domain signal. Then, extract the pilot signal from the time-domain signal. Perform certain transformations on the pilot signal to generate a sensing matrix that satisfies the Restricted Isometry Property (RIP) criterion. There are many ways to obtain the sensing matrix based on the pilot signal, and it is not unique; it is only necessary to satisfy the Restricted Isometry Property (RIP) criterion.

[0098] Step 3: As Figure 3 As shown, the filter error norm (FEN) is calculated according to equations (2) to (4). This is used to determine whether the channel sparsity has changed and whether the three-dimensional support set required for the Kalman filter algorithm needs to be updated. If the filter error norm (FEN) is greater than the threshold, it indicates that the channel sparsity has changed, and step 4 is executed to update the three-dimensional support set. If FEN is not greater than the threshold, it indicates that the channel sparsity has not changed, and the three-dimensional support set is not updated. Step 5 is executed.

[0099]

[0100]

[0101]

[0102] In the formula, ∑ ic A represents the weighting coefficient. T P represents the transpose of the perception matrix. t|t-1 This represents the state covariance matrix at the current moment. This indicates that the matrices within the parentheses correspond to the rows and columns of T1 and T2, respectively, where T... t =[i1,i2,...,i n ] is a one-dimensional array representing the coordinates of a matrix row or column, A′ T This represents the conjugate transpose of the perceptron transpose. y represents the noise variance. t This indicates the received signal at the current moment. Let A represent the channel estimate at the current moment, and let A represent the sensing matrix. This represents the conjugate transpose of the Kalman update error.

[0103] Step 4: As Figure 2 As shown, 3D-SOMP calculations are performed on the pilot signal to obtain the updated three-dimensional support set;

[0104] The 3D-SOMP mentioned in this step includes the following specific operations:

[0105] S210: Calculate the residual based on the received signal and the sensing matrix:

[0106] r = y - Ψh (i) (5)

[0107] Where r represents the residual, y represents the received signal, Ψ represents the sensing matrix, and h (i) This represents the estimated i-th path in the channel.

[0108] S220: Calculate the correlation between the sensing matrix and the residuals to obtain the location information with strong correlations in the sensing matrix:

[0109] e = Ψ H r (6)

[0110] S230: Based on the results of S220, determine the support sets for time-delay latitude, Doppler latitude, and angular latitude respectively.

[0111]

[0112]

[0113]

[0114] Among them, e r (m), e v (n) are the residuals obtained by performing 2-norm operations along the time-delay latitude and Doppler latitude, respectively, while e θ (g) is the residual obtained by performing a 2-norm operation after the lifting transformation matrix. These are the support sets for time-delay latitude, Doppler latitude, and angular latitude, respectively.

[0115] Before obtaining the support set along the angular dimension, the channel burst sparsity along the angular dimension needs to be converted into traditional block sparsity using the lifting transformation method, so that the angular dimension support set can be estimated.

[0116] Step 5: Based on the pilot signal, three-dimensional support set, and sensing matrix, the Kalman filter algorithm is used to estimate the time-varying underwater acoustic channel at the current moment.

[0117] The Kalman filter algorithm mentioned in this step includes the following specific operations:

[0118] S510: Initialize the state vector h and the state covariance matrix P based on the known system state F and measurement noise Q.

[0119] S520: Based on the system model and the current state vector h k-1 The state transition equation is used to predict the state vector h at the next time step. k and state covariance matrix P k:

[0120]

[0121]

[0122] In the formula, F represents the channel prediction value at time k. k-1 This represents the prediction matrix at time k-1. This represents the channel estimate corrected at time k-1. This represents the state covariance matrix predicted at time k-1. This represents the state covariance matrix corrected at time k-1. Q represents the transpose of the prediction matrix at time k-1. k-1 This represents the noise matrix.

[0123] S530: Obtain the measurement value at the next moment, based on the perception matrix H. k The predicted state vector is mapped into the measurement space, and the residual between the predicted and actual measurements is calculated. This residual is then combined with the measurement noise covariance matrix and weighted using a Kalman gain K to obtain the optimal state estimate. and state covariance matrix

[0124]

[0125]

[0126]

[0127] In the formula, y k This represents the received signal at time k.

[0128] The method in this embodiment can be used for signal recovery; see details below. Figure 4 The method of this embodiment is applied to the channel estimation and equalization module. In this module, the pilot signal in the received signal y is extracted, and channel estimation is performed according to the method of this embodiment. The channel after estimation is performed for channel equalization, so that the received signal y can be restored to the state when the signal was transmitted, thereby eliminating channel interference.

[0129] Example 2:

[0130] This embodiment discloses a time-varying underwater acoustic channel estimation system based on a Kalman filter for a MIMO-OTFS system. The MIMO-OTFS system includes a transmitter and a receiver, and the transmitter has N antennas. t ≥Number of receiving antennas N r Pilot overhead is 50%.

[0131] The transmitting end includes an input signal stream generation module, a quadrature amplitude modulation (QAM) modulation module, an inverse symplectic finite Fourier transform (ISFFT) module, a Heisenberg transform module, a cyclic prefix insertion module, and a first serial-to-parallel transform module. The input signal stream generated by the input signal stream generation module flows into the QAM modulation module. The input signal stream is a binary bit stream and includes pilot information. After modulation, the signal sequentially passes through the ISFFT module, the Heisenberg transform module, the cyclic prefix insertion module, and the first serial-to-parallel transform module before being sent into the underwater acoustic channel.

[0132] In the Quadrature Amplitude Modulation (QAM) module, the input signal stream is mapped into M×N QAM symbols in the time-delay-Doppler domain (DD domain), and these M×N QAM symbols are then placed at sampling intervals T. s (s) and sampling frequency Δf=1 / T s The modulated signal X is obtained from an M×N time-delay Doppler grid divided in units of (Hz). DD :

[0133]

[0134] in, The resolution of the time delay axis. The resolution is the Doppler axis.

[0135] In the inverse symplectic finite Fourier transform (ISFFT) module, the obtained X DD Performing an inverse symplectic finite Fourier transform (ISFFT) yields the time-frequency domain signal. Among them, F M Let F be the discrete Fourier matrix at point M. N It is an N-point discrete Fourier matrix.

[0136] In the Heisenberg transform module, the time-frequency domain signal X is transformed... FT Perform a Heisenberg transform to obtain the time-domain signal.

[0137] In the cyclic prefix insertion module, the time-domain signal S is added to the cyclic prefix.

[0138] In the first serial-to-parallel conversion module, the time-domain signal S with a cyclic prefix is ​​converted from serial to parallel, and the signal output by the first serial-to-parallel conversion module is sent into the underwater acoustic channel for transmission.

[0139] The receiving end includes a second serial-to-parallel conversion module, a cyclic prefix removal module, a filter error norm (FEN) decision module, a 3D-SOMP calculation module, and a Kalman filter module. The input signal to the receiving end sequentially passes through the second serial-to-parallel conversion module and the cyclic prefix removal module before being sent to the filter error norm (FEN) decision module. If FEN is greater than a threshold, it is sent to the 3D-SOMP calculation module, and then to the Kalman filter module; if FEN is not greater than the threshold, it is directly sent to the Kalman filter module.

[0140] In the second serial-to-parallel conversion module, the signal from the transmitting end is converted from serial to parallel.

[0141] In the cyclic prefix removal module, the cyclic prefix is ​​removed from the signal to obtain the time-domain signal, and then the pilot signal is extracted from the time-domain signal.

[0142] In the filter error norm (FEN) decision module, the filter error norm (FEN) is calculated according to equations (2) to (4) to determine whether the channel sparsity has changed and whether the three-dimensional support set required for the Kalman filter algorithm needs to be updated. If the filter error norm (FEN) is greater than the threshold, it indicates that the channel sparsity has changed, and the 3D-SOMP calculation module is called to update the three-dimensional support set. If FEN is not greater than the threshold, it indicates that the channel sparsity has not changed, and the three-dimensional support set is not updated. The Kalman filter module is directly entered.

[0143]

[0144]

[0145]

[0146] In the formula, ∑ ic A represents the weighting coefficient. T P represents the transpose of the perception matrix. t|t-1 This represents the state covariance matrix at the current moment. This indicates that the matrices within the parentheses correspond to the rows and columns of T1 and T2, respectively, where T... t =[i1,i2,...,i n ] is a one-dimensional array representing the coordinates of a matrix row or column, A′ T This represents the conjugate transpose of the perceptron transpose. y represents the noise variance. t This indicates the received signal at the current moment. Let A represent the channel estimate at the current moment, and let A represent the sensing matrix. This represents the conjugate transpose of the Kalman update error.

[0147] In the 3D-SOMP calculation module, the pilot signal is subjected to 3D-SOMP calculation to obtain the updated three-dimensional support set, specifically:

[0148] S210: Calculate the residual based on the received signal and the sensing matrix:

[0149] r = y - Ψh (i) (5)

[0150] Where r represents the residual, y represents the received signal, Ψ represents the sensing matrix, and h (i) This represents the estimated i-th path in the channel.

[0151] S220: Calculate the correlation between the sensing matrix and the residuals to obtain the location information with strong correlations in the sensing matrix:

[0152] e = Ψ H r (6)

[0153] S230: Based on the results of S220, determine the support sets for time-delay latitude, Doppler latitude, and angular latitude respectively.

[0154]

[0155]

[0156]

[0157] Among them, e r (m), e v (n) are the residuals obtained by performing 2-norm operations along the time-delay latitude and Doppler latitude, respectively, while e θ (g) is the residual obtained by performing a 2-norm operation after the lifting transformation matrix. These are the support sets for time-delay latitude, Doppler latitude, and angular latitude, respectively.

[0158] Before obtaining the support set along the angular dimension, the channel burst sparsity along the angular dimension needs to be converted into traditional block sparsity using the lifting transformation method, so that the angular dimension support set can be estimated.

[0159] In the Kalman filter module, the Kalman filter algorithm is used to estimate the current underwater acoustic time-varying channel based on the pilot signal, the three-dimensional support set, and the sensing matrix. Specifically:

[0160] S510: Initialize the state vector h and the state covariance matrix P based on the known system state F and measurement noise Q.

[0161] S520: Based on the system model and the current state vector h k-1The state transition equation is used to predict the state vector h at the next time step. k and state covariance matrix P k :

[0162]

[0163]

[0164] In the formula, F represents the channel prediction value at time k. k-1 This represents the prediction matrix at time k-1. This represents the channel estimate corrected at time k-1. This represents the state covariance matrix predicted at time k-1. This represents the state covariance matrix corrected at time k-1. Q represents the transpose of the prediction matrix at time k-1. k-1 This represents the noise matrix.

[0165] S530: Obtain the measurement value at the next moment, based on the perception matrix H. k The predicted state vector is mapped into the measurement space, and the residual between the predicted and actual measurements is calculated. This residual is then combined with the measurement noise covariance matrix and weighted using a Kalman gain K to obtain the optimal state estimate. and state covariance matrix

[0166]

[0167]

[0168]

[0169] In the formula, y k This represents the received signal at time k.

Claims

1. A method for time-varying underwater acoustic channel estimation of a MIMO-OTFS system based on Kalman filter, characterized in that: The method comprises the following steps: Step 1: obtaining a received signal from an underwater acoustic channel, extracting a pilot signal from the received signal, and generating a sensing matrix satisfying a constraint equidistance characteristic criterion based on the pilot signal; Step 2: calculating a filter error norm, and judging whether the filter error norm is greater than a set threshold value, if yes, executing step 3; if no, executing step 4; Step 3: performing 3D-SOMP calculation on the received signal to obtain an updated three-dimensional support set, and executing step 4; Step 4: estimating the underwater acoustic time-varying channel at the current time by using a Kalman filtering algorithm according to the pilot signal, the three-dimensional support set and the sensing matrix, to obtain the underwater acoustic channel at the current time; The 3D-SOMP calculation on the received signal specifically comprises: S210: calculating a residual according to the received signal and the sensing matrix; （5） wherein, denotes a residual, denotes a received signal, denotes a perception matrix, denotes the i-th path in the estimated channel; S220: calculating the correlation between the sensing matrix and the residual; （6） S230: determining the support sets of the time delay latitude, the Doppler latitude and the angle latitude respectively according to the result of S220; （7） （8） （9） wherein, , are the residuals of the 2-norm operation along the delay latitude, Doppler latitude, respectively, is the residual of the 2-norm operation after the lifting transform matrix, , , are the support sets of the delay latitude, Doppler latitude, angle latitude, respectively.

2. The method of claim 1, wherein the method is based on a Kalman filter-based MIMO-OTFS system time-varying underwater acoustic channel estimation. The calculation of the filter error norm comprises: The filter error norm FEN is calculated according to formulas (2)-(4): （2） （3） （4） wherein denotes a weighting coefficient, denotes the transpose of the perception matrix, denotes the state covariance matrix at the current time instant, denotes the Tth row and Tth column of the matrix in the brackets, denotes the conjugate transpose of the transpose of the perception matrix, denotes the noise variance, denotes the received signal at the current time instant, denotes the channel estimate at the current time instant, denotes the perception matrix, denotes the conjugate transpose of the Kalman update error.

3. The method of claim 1, wherein the method is based on a Kalman filter-based MIMO-OTFS system time-varying underwater acoustic channel estimation method. Step 4 specifically comprises: S510: initialize state vector and measurement noise , initialize state vector and state covariance matrix ; S520: predict the next time state vector and state covariance matrix according to the system model and the current state vector :​​ （10） （11） wherein denotes the channel prediction value at time k, denotes the prediction matrix at time k-1, denotes the corrected channel estimate at time k-1, denotes the predicted state covariance matrix at time k-1, denotes the corrected state covariance matrix at time k-1, denotes the transpose of the prediction matrix at time k-1, denotes the noise matrix; S530: Obtain the measurement value at the next time, according to the perception matrix Map the predicted state vector into the measurement space, and calculate the residual between the predicted measurement value and the actual measurement value; then combine the residual with the measurement noise covariance matrix, and use the Kalman gain Perform weighted averaging to obtain the optimal state estimation value and the state covariance matrix : （12） （13） （14） In the formula, represents the received signal at time k.

4. A Kalman filter based MIMO-OTFS system time varying underwater acoustic channel estimation system, characterized in that: It comprises: A pilot signal extraction module is configured to obtain a received signal from an underwater acoustic channel, and extract a pilot signal from the received signal; A filter error norm judgment module is configured to calculate a filter error norm, and judge whether the filter error norm is greater than a set threshold value, if yes, call a 3D-SOMP calculation module; if no, call a Kalman filtering module; A 3D-SOMP calculation module is configured to perform 3D-SOMP calculation on the received signal to obtain an updated three-dimensional support set, and call the Kalman filtering module; A Kalman filtering module is configured to estimate the underwater acoustic time-varying channel at the current time by using a Kalman filtering algorithm according to the pilot signal, the three-dimensional support set and the sensing matrix; In the 3D-SOMP calculation module, the following steps are executed: S210: calculating a residual according to the received signal and the sensing matrix; （5） wherein, denotes a residual, denotes a received signal, denotes a perception matrix, denotes the i-th path in the estimated channel; S220: calculating the correlation between the sensing matrix and the residual; （6） S230: determining the support sets of the time delay latitude, the Doppler latitude and the angle latitude respectively according to the result of S220; （7） （8） （9） wherein, , are the residuals of the 2-norm operation along the delay latitude, Doppler latitude, respectively, is the residual of the 2-norm operation after the lifting transform matrix, , , are the support sets of the delay latitude, Doppler latitude, angle latitude, respectively.

5. The Kalman filter based MIMO-OTFS system time varying underwater acoustic channel estimation system according to claim 4, characterized in that: In the filter error norm judgment module, the following steps are executed: The filter error norm FEN is calculated according to formulas (2)-(4): （2） （3） （4） wherein denotes a weighting coefficient, denotes the transpose of the perception matrix, denotes the state covariance matrix at the current time instant, denotes the Tth row and Tth column of the matrix in the brackets, denotes the conjugate transpose of the transpose of the perception matrix, denotes the noise variance, denotes the received signal at the current time instant, denotes the channel estimate at the current time instant, denotes the perception matrix, denotes the conjugate transpose of the Kalman update error; Judge whether the filter error norm is greater than a set threshold value, if yes, call the 3D-SOMP calculation module; if no, call the Kalman filtering module.

6. The Kalman filter based MIMO-OTFS system time varying underwater acoustic channel estimation system according to claim 4, characterized in that: In the Kalman filtering module, the following steps are executed: S510: initialize state vector and measurement noise , initialize state vector and state covariance matrix ; S520: predict the next time state vector and state covariance matrix according to the system model and the current state vector , using the state transition equation and state covariance matrix : （10） （11） wherein denotes the channel prediction value at time k, denotes the prediction matrix at time k-1, denotes the corrected channel estimate at time k-1, denotes the predicted state covariance matrix at time k-1, denotes the corrected state covariance matrix at time k-1, denotes the transpose of the prediction matrix at time k-1, denotes the noise matrix; S530: Obtain the measurement value at the next time, according to the perception matrix Map the predicted state vector into the measurement space, and calculate the residual between the predicted measurement value and the actual measurement value; then combine the residual with the measurement noise covariance matrix, and use the Kalman gain Perform weighted averaging to obtain the optimal state estimation value and the state covariance matrix : （12） （13） （14） In the formula, represents the received signal at time k.

7. A MIMO-OTFS system, characterized by: It comprises a transmitting end and a receiving end; The transmitting end comprises an input signal stream generation module, a quadrature amplitude modulation module, an inverse symplectic finite Fourier transform module, a Heisenberg transform module, a cyclic prefix insertion module and a first serial-parallel conversion module; the input signal stream generated by the input signal stream generation module flows into the quadrature amplitude modulation module, the input signal stream is a binary bit stream and contains pilot information; the signal after modulation successively passes through the inverse symplectic finite Fourier transform module, the Heisenberg transform module, the cyclic prefix insertion module and the first serial-parallel conversion module and is then sent into the underwater acoustic channel to complete signal transmission; The receiving end comprises a time-varying underwater acoustic channel estimation module, which is a time-varying underwater acoustic channel estimation system of a MIMO-OTFS system based on a Kalman filter according to any one of claims 4 to 6.

8. The MIMO-OTFS system of claim 7, wherein: In the quadrature amplitude modulation module, the following steps are performed: In the delay-Doppler domain, the input signal stream is mapped into quadrature amplitude modulation symbols; one quadrature amplitude modulation symbol is placed in a time delay-Doppler grid divided in units of and sampling frequency (Hz) , resulting in a modulated signal :​ （1） wherein, is the resolution of the delay axis, is the resolution of the Doppler axis; In the inverse symplectic finite Fourier transform module, the following steps are performed: signal performing an inverse symplectic Fourier transform (ISFFT) to obtain a time-frequency domain signal wherein is a discrete Fourier matrix of size is a discrete Fourier matrix of size In the Heisenberg transform module, the following steps are performed: transforming the time-frequency domain signal performing a Heisenberg transform to obtain a time domain signal .

Images