Fast DOA Estimation Method for Underwater Acoustic Wideband Signals Based on Space-Time Adaptive Interference Suppression

Abstract

The present invention provides a fast DOA estimation method for underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression. The spatio-temporal processing structure adopted by this method is formed by removing the pre-steering delay unit in the Frost broadband adaptive beamformer. The method includes: Step 1: Orthogonally demodulate the original array element data x m to obtain the corresponding complex envelope signal #imgabs0# Step 2: Evenly divide #imgabs1# into L frequency band complex envelope broadband signals #imgabs2# through a filter bank, and then select the detection snapshot samples by D-fold decimation for each frequency band complex envelope broadband signal #imgabs3#; Step 3: Calculate the data covariance matrix #imgabs4# and the spatio-temporal steering vector a(θ p , f q ) in each frequency band; Step 4: Calculate the azimuth spectrum #imgabs6# of each frequency band according to the data covariance matrix #imgabs5# and the spatio-temporal steering vector a(θ p , f q ); Step 5: Integrate the azimuth spectra of each frequency band into an azimuth spectrum #imgabs7#

Description

Technical Field

[0001] The present invention relates to the technical field of underwater acoustic communication, and particularly relates to a method for fast DOA estimation of underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression. Background Art

[0002] Currently, there are mainly two methods for DOA estimation of underwater acoustic broadband signals based on adaptive interference suppression, namely the DOA estimation method based on broadband space-frequency adaptive beamforming (SFAB) and the DOA estimation method based on broadband space-time adaptive beamforming (STAB). The DOA estimation method based on broadband SFAB uses the Fast Fourier Transform (FFT) to divide the broadband signal into several narrowband signals, and then uses the adaptive beamforming method based on narrowband signals for DOA estimation. The DOA estimation method based on broadband STAB directly constructs the spatio-temporal two-dimensional data covariance matrix of the broadband signal by adding tapped delay lines after each array branch, and uses the Linearly Constrained Minimum Variance (LCMV) to perform the adaptive broadband beamforming algorithm.

[0003] When there are close-angle interferences and finite-time snapshots (such as multi-target interferences and fast-moving sources), the existing broadband DOA estimation methods based on adaptive interference suppression will face great challenges. Under these conditions, for the broadband SFAB based on FFT, the finite-time snapshots make each frequency bin processed by FFT unable to meet the narrowband assumption in the frequency domain, which leads to a decrease in angular resolution. In addition, the sidelobes of each frequency bin will also affect the performance of DOA estimation.

[0004] For wideband STAB, the wideband adaptive beamformer proposed by Frost requires the received signal to be pre-steered to the normal direction of the sensor line through mechanical rotation or electronic signal processing, and its performance is limited by the accuracy of the pre-steering delay. In addition, using pre-steering in beam scanning will increase the hardware cost and computational complexity. Buckley and Ebrahimi respectively proposed using a set of linear constraints defined in the time domain and frequency domain to eliminate the pre-steering delay. However, these multi-linear constraint methods reduce the remaining degrees of freedom of the minimum variance beamformer, reducing the interference suppression ability. At the same time, under multi-linear constraint conditions, the position of the phase center of the array will also affect the interference suppression ability of the beamformer. Therefore, the angular resolution is poor under the condition of a short number of filter taps, while a long number of filter taps will greatly increase the computational complexity of the algorithm and the snapshot length required for detection. Summary of the Invention

[0005] Aiming at the problems of slow data update rate and long algorithm calculation time existing in the existing underwater acoustic wideband signal DOA estimation method based on adaptive interference suppression, the present invention provides a fast DOA estimation method for underwater acoustic wideband signals based on spatio-temporal adaptive interference suppression.

[0006] The fast DOA estimation method for underwater acoustic wideband signals provided by the present invention adopts a spatio-temporal processing structure formed by removing the pre-steering delay unit in the Frost wideband adaptive beamformer. The method includes:

[0007] Step 1: Orthogonally demodulate the sampled original array element data x m to obtain the corresponding complex envelope signal where m = 1, 2,..., M, and M is the number of array elements;

[0008] Step 2: Divide the complex envelope wideband signal equally spaced into L frequency bands through a filter bank composed of L filters Then, through a decimator, select the detection snapshot samples by decimating each frequency band's complex envelope wideband signal by D times; where l = 1, 2,..., L;

[0009] Step 3: Calculate the data covariance matrix and the spatio-temporal steering vector a(θ p , f q ) in each frequency band, where θ p is the pointing direction of the steering vector, p = 1, 2,..., P; f q is the frequency of the steering vector, q = 1, 2,..., Q; P and Q are respectively the number of angle calculation points and frequency calculation points equally spaced selected in the angle-frequency response calculation grid of the beamformer;

[0010] Step 4: Calculate the azimuth spectrum of each frequency band according to the data covariance matrix and the spatio-temporal steering vector a(θ p , f q ).

[0011] Step 5: Integrate the azimuth spectra of each frequency band into an azimuth spectrum

[0012] Furthermore, in Step 2, the decimation factor D of the decimator does not exceed its maximum value D max : D max = Lf s / B; where f s is the sampling frequency, and B represents the bandwidth of the processed signal of the system;

[0013] And, under the condition of a given number of time-domain taps N, the selection of D and L should make the beamformer satisfy the following two conditions simultaneously:

[0014] Condition 1: The data duration SN / f of the time-domain taps s ≥ T1, where T1 is the upper limit of the data duration of the time-domain taps when meeting the minimum frequency resolution requirement of the system;

[0015] Condition 2: The data duration SN / f of the time-domain taps s should be much greater than the maximum signal delay τ(θ p ) between any array elements.

[0016] Furthermore, the maximum signal delay τ(θ p ) between any array elements is defined as:

[0017]

[0018] Furthermore, when setting the number of time-domain taps N, the set N should satisfy the following two conditions simultaneously:

[0019] Condition 1: For an MN-order data covariance matrix, the number of snapshots K required for estimation should not be less than 2MN; and when the number of sampling points Δ between adjacent snapshots is given, the sample requirement 2MNΔ for estimating the covariance should satisfy: 2MNΔ ≤ V; where V is the sample requirement for estimating the covariance when meeting the minimum data update rate of the system;

[0020] Condition 2: According to the calculation complexity calculated by formula (2), the calculation complexity should satisfy: O STMVDR ≤ O;

[0021] O STMVDR = O{3(ML)3 +PQ[(ML) 2 +2MN]} (2)

[0022] Among them, O is the upper limit of the computational complexity of the algorithm when meeting the minimum running time of the system.

[0023] Furthermore, calculate the data covariance matrix of the l-th frequency band according to formula (3)

[0024]

[0025] Among them, “^” represents estimation, and the superscript [·] H represents the conjugate transpose, K represents the number of snapshots, represents the vector corresponding to the k-th snapshot of the signal in the l-th frequency band.

[0026] Furthermore, the vector corresponding to the k-th snapshot of the signal is defined as:

[0027]

[0028] Among them, is the data at the m-th array element and the n-th tap of the detected snapshot sample after extraction.

[0029] Furthermore, calculate the spatio-temporal steering vector a(θ p , f q ) according to formula (5):

[0030]

[0031] Among them, S s (θ p , f q ) represents an M×1 spatial steering vector, where represents the propagation delay of the m-th sensor relative to the array phase center position when the source azimuth is θ p , S t (f q ) represents an N×1 temporal steering vector, T s is the sampling interval of the array element data, represents the Kronecker product, and D is the decimation factor of the decimator.

[0032] Furthermore, step 4 specifically includes:

[0033] Calculate the azimuth spectrum of the l-th frequency band according to formula (8)

[0034]

[0035] Advantages of the present invention:

[0036] When the spatio-temporal adaptive interference suppression algorithm is applied to the DOA estimation of underwater acoustic broadband signals, since the algorithm frequency resolution depends on the length of the snapshot duration and there is a minimum limit for the snapshot duration in broadband array signal processing, a large number of filter taps are usually required, resulting in a high computational complexity of the algorithm and a large amount of training samples. To solve this problem, the present invention first preprocesses the snapshot data using a filter bank and a decimator, and then performs spatio-temporal adaptive beamforming, which can obtain a high frequency resolution and snapshot duration under the condition of using a small number of filter taps.

[0037] Moreover, by establishing a minimum variance optimization problem using a single linear constraint condition, the remaining degrees of freedom of the beamformer can be increased and the influence of the array phase center position on the interference suppression performance can be eliminated, thereby improving the angular resolution and robustness of the spatio-temporal adaptive beamforming for DOA estimation under the condition of short filter taps. Description of the Drawings

[0038] Figure 1 It is a schematic flow chart of the method for fast DOA estimation of underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression provided by an embodiment of the present invention;

[0039] Figure 2 It is a schematic diagram of the spatio-temporal processing structure provided by an embodiment of the present invention;

[0040] Figure 3 It is the processing structure of the algorithm at element m provided by an embodiment of the present invention;

[0041] Figure 4 It is the normalized spatial spectrum (SNR = 10 dB) of each method in Table 1 provided by an embodiment of the present invention: (a) is the normalized spatial spectrum of the present invention and the DOA-SFAB method; (b) is the normalized spatial spectrum of the present invention and the DOA-STAB method;

[0042] Figure 5 It is the RMSE of each method in Table 1 provided by an embodiment of the present invention at different signal-to-noise ratios. Detailed Embodiments

[0043] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0044] AsFigure 1 As shown, an embodiment of the present invention provides a fast DOA estimation method for underwater acoustic broadband signals based on space-time adaptive interference suppression. The adopted space-time processing structure is formed by removing the pre-steering delay unit in the Frost broadband adaptive beamformer. In the embodiment of the present invention, as Figure 2 shown, it is assumed that the space-time two-dimensional processor has M array elements, and there is an Nth-order FIR filter behind each array element channel. The input signals of the M channels are complex envelope signals, which are respectively denoted as w mn represents the weight coefficient of space-time processing; x mn is the input signal of each filter tap, n = 1, 2,..., N, m = 1, 2,..., M; M and N are respectively the number of array elements and the number of FIR filter taps; z -D represents the tap delay, D is the decimation factor of the decimator, and y represents the array output.

[0045] The method provided by the embodiment of the present invention includes the following steps:

[0046] S101: Orthogonally demodulate the sampled original array element data x m to obtain the corresponding complex envelope signal m = 1, 2,..., M, where M is the number of array elements;

[0047] S102: Divide the complex envelope broadband signal equally spaced into L frequency bands through a filter bank composed of L filters Then, through the decimator, the complex envelope broadband signal of each frequency band is decimated by D times to select the detection snapshot samples; where, l = 1, 2,..., L;

[0048] Specifically, according to the complex envelope signal of each frequency band the detection samples are selected by downsampling according to the frequency band range, specifically including: setting the decimation factor of the decimator to D, then the tap data of the mth array element in the lth frequency band is

[0049] In addition, it should be noted that the decimation factor D of the decimator does not exceed its maximum value D max : D max = Lf s / B; where, f s is the sampling frequency, and B represents the frequency band width of the processed signal of the system; and, under the condition of a given number of time domain taps N, the selection of D and L should make the beamformer satisfy the following two conditions simultaneously:

[0050] Condition 1: The data duration of the time domain tap SN / f s≥ T1, where T1 is the upper limit of the data duration of the time-domain tap when meeting the system's minimum frequency resolution requirement; this is because: the data duration of the time-domain tap SN / f s determines the frequency-domain resolution of the beamformer (such as the spatio-temporal MVDR beamformer). Therefore, to obtain the desired frequency-domain resolution, it is required that the data duration of the time-domain tap DN / f s ≥ T1.

[0051] Condition 2: The data duration of the time-domain tap DN / f s should be much greater than the maximum signal delay τ(θ p ) between any array elements. This setting of Condition 2 is to ensure that when the beamformer (such as the spatio-temporal MVDR beamformer) processes broadband signals, the snapshot data can propagate through each array element, and each array element can synchronously collect the spatial signal field.

[0052] In the embodiment of the present invention, the maximum signal delay τ(θ p ) between any array elements is defined as:

[0053]

[0054] In addition, when setting the number N of the time-domain taps, the set N should simultaneously meet the following two conditions:

[0055] Condition 1: For an MN-order data covariance matrix, the required number of snapshots K for estimation should be no less than 2MN; the purpose of such a limitation is to ensure that according to the RMB criterion, the average loss of the signal-to-noise ratio gain caused by the estimation error of the data covariance matrix is less than 3 dB;

[0056] And when the number of sampling points Δ of the data interval between adjacent snapshots is given, the sample demand 2MNΔ for estimating the covariance should satisfy: 2MNΔ ≤ V, so as to ensure the data update rate of the algorithm; where V is the sample demand for estimating the covariance when meeting the system's minimum data update rate;

[0057] Condition 2: According to the calculation complexity calculated by formula (2), the calculation complexity should satisfy: O STMVDR ≤ O, so as to ensure the running time of the algorithm;

[0058] O STMVDR = O{3(ML) 3 + PQ[(ML) 2 + 2MN]} (2)

[0059] where O is the upper limit of the calculation complexity of the algorithm when meeting the system's minimum running time.

[0060] S103: Calculate the data covariance matrix in each frequency band and the spatio-temporal steering vector a(θ p , f q ), where θ p is the pointing direction of the steering vector, p = 1, 2,..., P; f q is the frequency of the steering vector, q = 1, 2,..., Q; P and Q are the number of angular calculation points and the number of frequency calculation points respectively selected at equal intervals in the angular-frequency response calculation grid of the beamformer.

[0061] Specifically, calculate the covariance matrix of the data (the "data" here includes the desired signal, interference, and noise) in the l-th frequency band according to formula (3)

[0062]

[0063] where "^" represents estimation, and the superscript [·] H represents conjugate transpose, K represents the number of snapshots, represents the vector corresponding to the k-th signal snapshot, defined as:

[0064]

[0065] where is the data at the m-th array element and the n-th tap after decimation of the detected snapshot samples.

[0066] Calculate the spatio-temporal steering vector a(θ p , f q ) according to formula (5):

[0067]

[0068] where S s (θ p , f q ) represents an M×1 spatial steering vector, represents the propagation delay of the m-th sensor relative to the array phase center position when the source azimuth is θ p (for a uniform linear array, d is the spacing between adjacent array elements), S t (f q ) represents an N×1 temporal steering vector, T s is the sampling interval of the array element data, represents the Kronecker product, and D is the decimation factor of the decimator.

[0069] S104: Calculate the azimuth spectrum of each frequency band according to the data covariance matrix and the spatio-temporal steering vector a(θ p , f q )

[0070] Specifically, since the space-time processing structure adopted by the present invention does not include a pre-steering delay unit, new constraint conditions need to be designed. In the embodiment of the present invention, the single linear constraint space-time adaptive beamforming (SLC-STAB) can be described as the following optimization problem, as shown in formula (6):

[0071]

[0072] where w is the space-time processing weight vector, which can be expressed as: w = [w 11 , w 12 ,..., w 1N , w 21 ,..., w 2N ,..., w M1 ,..., w MN T ; where [·] T represents the transpose of the vector.

[0073] Using the Lagrange multiplier method, the optimal weight value can be obtained as

[0074]

[0075] Correspondingly, the Capon spectral estimation output based on SLC-STAB can be modeled as

[0076]

[0077] Then, based on the spectral outputs of all frequencies at azimuth θ p and summing over all azimuths, the wideband fast DOA estimation formula based on space-time adaptivity can be expressed as

[0078]

[0079] Correspondingly, based on formula (9), the calculation formula for the azimuth spectrum

[0080] of the l-th frequency band is:

[0081] S105: Integrate the azimuth spectra of each frequency band into an azimuth spectrum

[0082] ​In the embodiments of the present invention, by replacing the multi-linear constraint condition in the wideband STAB with a single linear constraint condition, the remaining degrees of freedom of the minimum variance beamformer can be increased, and the influence of the array phase center position design on the beamforming interference suppression performance can be eliminated, thereby improving the angular resolution of the DOA estimation method based on STAB under the condition of short filter taps. Reducing the number of filter taps can reduce the computational complexity of the algorithm and the time snapshot length required for detection, improve the data update rate, and reduce the running time. Secondly, in the embodiments of the present invention, by using a filter bank and a decimator to increase the time length of a single snapshot without changing the number of filter taps, a higher frequency resolution and snapshot duration can be obtained under the condition of using a small number of filter taps, solving the problem that there is a minimum limit on the snapshot duration for frequency resolution and wideband array signal processing. The present invention can adaptively suppress interference and improve angular resolution, and can be applied to the azimuth detection of underwater acoustic broadband targets under multi-target interference conditions.

[0083] To verify the effectiveness and advantages of the method of the present invention, the present invention is illustrated by three groups of simulation results.

[0084] This experiment considers a uniform linear array composed of 16 omnidirectional elements. Assume the sampling frequency is f s , and the element spacing is selected as half of the wavelength of f s / 2. In the simulation, the processing frequency band range is [0.25f s , 0.45f s , and uncorrelated Gaussian white noise is used to simulate the underwater acoustic broadband signal; the processing structure of a single element channel is as Figure 3 shown, and the passband ranges of the band-pass filters H1 and H2 are [0.25f s , 0.35f s and [0.35f s , 0.45f s respectively. The decimation factor D = 8 is selected. The time interval between adjacent snapshots is selected as 10 sampling points. The number of snapshots required for detection satisfies the RMB criterion and is selected as 2MN.

[0085] First, we evaluated the training sample requirements and computational complexity of the traditional method and the method of the present invention. This evaluation characterized the training sample requirements and computational complexity by the number of data points required to estimate the data covariance matrix each time and the running time on the same computer using Matlab R2018b and a 2.6GHz CPU. For the DOA estimation based on broadband SFAB (DOA-SFAB), 128-point DFT, 512-point DFT, and 1024-point DFT were selected; for the DOA estimation based on broadband STAB (DOA-STAB), N = 5, N = 95, and N = 100 were selected. As shown in Table 1, both the training sample requirements and computational complexity of the method of the present invention are lower than those of the DOA-SFAB method with three different snapshot lengths and the DOA-STAB method with two long filter taps. Compared with the DOA-STAB method using the same number of filter taps, the method of the present invention has a slightly increase in terms of training sample requirements and computational complexity.

[0086] Table 1 Training sample requirements and running time

[0087] DOA Estimation Method Training Sample Requirement (sec) Computational Complexity (sec) DOA-SFAB (128-point DFT) 2112 1.57 DOA-SFAB (512-point DFT) 8448 4.80 DOA-SFAB (1024-point DFT) 16896 9.28 DOA-STAB (N = 5) 800 0.48 DOA-STAB (N = 95) 15290 25.40 DOA-STAB (N = 100) 16095 28.82 The Present Invention (N = 5) 1630 0.82

[0088] Secondly, in the presence of strong near-angle interference, the angular resolution of the traditional method and the method of the present invention was evaluated by comparing the normalized spatial spectra of the methods in Table 1. Assume that a broadband source with a signal-to-noise ratio of 10 dB and an uncorrelated broadband interference with a signal-to-noise ratio of -20 dB impinge on the array from 40° and 30° points respectively. The comparison of the normalized spatial spectra of the method of the present invention and the DOA-SFAB method with different snapshot lengths is as Figure 4 (a) shown. It can be seen that the angular resolution of the DOA-SFAB method increases with the increase of the DFT points, and the angular resolution of the method of the present invention is the best. Figure 4 (b) shows the normalized spatial spectra of the DOA-STAB method with different numbers of filter taps and the method of the present invention. It can be found that the DOA-STAB method with N = 5 can no longer distinguish the two angles, and the angular resolution of the method of the present invention is between the DOA-STAB method with N = 95 and the DOA-STAB method with N = 100. However, the method of the present invention has the lowest requirements for training samples and computational complexity.

[0089] Finally, the root mean square error (RMSE) of the DOA estimation of all the methods in Table 1 at different signal-to-noise ratios was evaluated. The RMSE of the DOA estimation was estimated through 300 independent Monte Carlo simulations. The signal-to-noise ratio of the broadband light source was -10 dB to 20 dB within a range of 45°. The results are as Figure 5As shown. Compared with the DOA-SFAB method, the RMSE of the method of the present invention is lower than that of the 128-point DFT DOA-SFAB method only when the signal-to-noise ratio is greater than 10 dB. Compared with the DOA-STAB method, the root mean square error (RMSE) of the method of the present invention is much lower than that of the DOA-STAB method with the same number of filter taps, but higher than that of the DOA-STAB method with longer filter taps. In addition, as the signal-to-noise ratio increases, the decreasing trend of the root mean square error of the DOA-SFAB method tends to level off, while the decreasing trend of the root mean square error of the method of the present invention remains almost unchanged.

[0090] From the simulation results in the above three aspects, under the conditions of shorter operation time and the snapshot length required for detection, the DOA estimation of the present invention has higher angular resolution, narrower nulls for interference suppression, and lower estimated mean square error.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fast DOA estimation method for underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression, characterized in that The adopted space-time processing structure is formed by removing the pre-steering delay unit in the Frost broadband adaptive beamformer. The method includes: Step 1: Orthogonally demodulate the original array element data x after sampling m to obtain the corresponding complex envelope signal where M is the number of array elements; Step 2: Use a filter bank consisting of L filters to divide the complex envelope wideband signal that is equally spaced into L frequency bands Then, use a decimator to select detection snapshot samples by decimating the complex envelope wideband signal of each frequency band by a factor of D; where l = 1, 2, …, L; Step 3: Calculate the data covariance matrix in each frequency band and the spatio-temporal steering vector a(θ p , f q ), where θ p is the pointing direction of the steering vector, p = 1, 2, …, P; f q is the frequency of the steering vector, q = 1, 2, …, Q; P and Q are the number of angle calculation points and the number of frequency calculation points respectively selected at equal intervals in the angle-frequency response calculation grid of the beamformer; calculate the spatio-temporal steering vector a(θ p , f q ) according to formula (5): Among them, S s (θ p , f q ) represents an M×1 spatial steering vector, where represents the propagation delay of the m-th sensor relative to the array phase center position when the source azimuth is θ p , S t (f q ) represents an N×1 temporal steering vector, T s is the sampling interval of the array element data, represents the Kronecker product, and D is the decimation factor of the decimator; Step 4: According to the data covariance matrix and the space-time steering vector a(θ p , f q ), calculate the azimuth spectrum of each frequency band Specifically, it includes: calculating the azimuth spectrum of the l-th frequency band according to formula (8) Step 5: Integrate the azimuth spectra of each frequency band into an azimuth spectrum 2. The fast DOA estimation method for underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression according to claim 1, characterized in that In step 2, the extraction multiple D of the extractor does not exceed its maximum value D max : D max = Lf s / B; where f s is the sampling frequency, and B represents the bandwidth of the processed signal of the system; Moreover, under the condition of a given number of time-domain taps N, the selection of D and L should enable the beamformer to satisfy the following two conditions simultaneously: Condition 1: The data duration of the time-domain tap is DN / f s ≥ T1, where T1 is the upper limit of the data duration of the time-domain tap when meeting the requirements of the system's minimum frequency resolution; Condition 2: The data duration of the time-domain tap, DN / f s should be much greater than the maximum signal delay τ(θ p ) between any array elements.

3. The fast DOA estimation method for underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression according to claim 2, characterized in that The maximum signal delay τ(θ p ) between any array elements is defined as:

4. The fast DOA estimation method for underwater acoustic broadband signals based on space-time adaptive interference suppression according to claim 2, characterized in that When setting the number of time-domain taps N, the set N should satisfy the following two conditions simultaneously: Condition 1: For an MN-order data covariance matrix, the required number of snapshots K for estimation should be no less than 2MN; and when the number of sampling points △ in the data interval between adjacent snapshots is given, the sample demand 2MN△ for estimating the covariance should satisfy: 2MN△ ≤ V; where V is the sample demand for estimating the covariance when meeting the minimum data update rate of the system. Condition 2: Calculate the computational complexity according to formula (2), and the computational complexity should satisfy: Where O is the upper limit of the computational complexity of the algorithm when meeting the minimum running time of the system.

5. The fast DOA estimation method for underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression according to claim 1, characterized in that Calculate the data covariance matrix of the l-th frequency band according to formula (3). where "^" represents estimation and the superscript [·] H represents conjugate transpose, K represents the number of snapshots, denotes the vector corresponding to the k-th signal snapshot in the l-th frequency band.

6. The fast DOA estimation method for underwater acoustic broadband signals based on spatio-temporal adaptive interference suppression according to claim 5, characterized in that The vector corresponding to the k-th signal snapshot is defined as: Among them, is the data of the detected snapshot sample after extraction at the m-th array element and the n-th tap.

Images