A spatio-temporal-frequency domain joint high gain line spectrum enhancement method

Abstract

The application discloses a space-time-frequency domain joint high-gain line spectrum enhancement method, which comprises the following steps: firstly, calculating the output of a primary multi-channel time domain filtering algorithm, and then calculating the output of a secondary multi-channel time domain filtering algorithm; next, generating a robust super-directivity weighting vector and generating beam output data; finally, performing frequency domain high-gain processing on the time domain data of the beam output.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to a spatiotemporal frequency domain joint high-gain line spectrum enhancement method. Background Technology

[0002] Low-frequency line spectrum signals radiated by underwater vehicles are the primary target of passive sonar. However, with the development of ship vibration reduction and noise reduction technologies worldwide and the increase in ocean noise levels caused by human activities, low-frequency line spectra are gradually being "submerged" in background noise. The signal-to-noise ratio of the line spectrum received by passive sonar is declining year by year. At the same time, the complex marine environment makes the line spectrum signals blurred, unstable, and mixed with many non-target interference signals, which brings great difficulties to the detection and identification of underwater acoustic low-frequency line spectrum signals. Existing line spectrum signal processing methods have limited processing gain and cannot effectively detect weak underwater line spectrum signals. Therefore, it is necessary to study high-gain processing methods for underwater acoustic low-frequency line spectrum signals to improve the detection performance of passive sonar.

[0003] Conventional ALE methods only utilize single-channel information, resulting in limited processing gain. References 1 and 2 mention that when the input signal-to-noise ratio is low, the processing effect is poor, making it impossible to detect underwater weak line spectrum targets. Reference 3 proposes a spatiotemporal domain joint method for detecting underwater unknown line spectrum targets, which to some extent achieves the detection of underwater weak line spectrum signals in the spatiotemporal domain. However, this method does not consider the frequency domain processing gain of the signal. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides a spatiotemporal frequency domain joint high-gain line spectrum enhancement method. First, the output of a first-level multi-channel time-domain filtering algorithm is calculated, and then the output of a second-level multi-channel time-domain filtering algorithm is calculated. Next, a robust super-directivity weighted vector is generated, and beam output data is generated. Finally, the time-domain data of the beam output is subjected to frequency domain high-gain processing.

[0005] The technical solution adopted by this invention to solve its technical problem is as follows:

[0006] An arbitrary-shaped sensor array is used to receive signals emitted by a sound source, and only horizontal incidence is considered. Assuming H line spectrum signals from different targets in space are incident on the sensor receiving array as far-field plane waves, the signal of the m-th element is represented as:

[0007]

[0008] Among them, A h f is the amplitude of the h-th signal. h f is the frequency of the h-th signal, n is the discrete time index, and f s Sampling frequency, φ represents the phase of the m-th array element relative to the reference point when it receives the h-th signal. h Let be the initial phase of the h-th signal, which follows a uniform distribution;

[0009] Step 1: Calculate the output of the first-level multi-channel time-domain filtering algorithm;

[0010] The output of the i-th multi-channel time-domain filtering algorithm is expressed as:

[0011]

[0012] Among them, y i (n) represents the output of the i-th multi-channel time-domain filtering algorithm, where n is the time series, i is the channel series, L is the number of taps, M is the number of array elements, w is the weight, and Δ is the number of pre-delay points; w l,m (·) represents the weight coefficients of the multi-channel time-domain filtering algorithm;

[0013] The error function is expressed as:

[0014] ε i (n)=x i (n)-y i (n) (3)

[0015] The weight coefficient iteration formula for the multi-channel time-domain filtering algorithm is derived from the LMS algorithm:

[0016]

[0017] Where μ is the adaptive iteration step size;

[0018] Step 2: Calculate the output of the two-stage multi-channel time-domain filtering algorithm;

[0019] The output of the first-level multi-channel time-domain filtering algorithm is then used to calculate the output of the second-level multi-channel time-domain filtering algorithm:

[0020]

[0021] Where zi (n) represents the output of the i-th secondary multi-channel time-domain filtering algorithm, y m (·) represents the output of the m-th first-level multi-channel time-domain filtering algorithm;

[0022] The error function is expressed as:

[0023] ε i (n)=y i (n)-z i (n) (6)

[0024] The iterative formula for the weight coefficients is:

[0025]

[0026] Step 3: Generate robust super-directional weighted vectors and generate beam output data;

[0027] Represent the array output data obtained in step 2 as a matrix:

[0028]

[0029] Perform an H-point DFT (Discrete Fourier Transform) on the metadata of each matrix, i.e.:

[0030]

[0031] The frequency domain data of each array element in subband k can be written in vector form:

[0032]

[0033] The narrowband form of the robust hyperdirectional weighted vector design is as follows:

[0034] w(f k )=[w1(f k ),w2(f k ),…w M (f k )] T (11)

[0035] Among them, w1(f k ),w2(f k ),…w M (f k ) represent the weighted vector values ​​of each array element;

[0036] Synthesize the sub-band beams:

[0037]

[0038] Among them, w H(f k ) indicates a frequency of f k The weighted vector, The frequency is f k The conjugate value of the m-th term of the weighted vector;

[0039] Perform inverse DFT on the output data of each sub-band beam

[0040] y (n) (h), (h=0,1,…,H-1)=IDFT[Y (n) (k),(k=0,1,…,H-1)] (13)

[0041] The above process is repeated for each segment of array data, and finally the time domain data of each segment are spliced ​​together to form the long-term output sequence y(i) of the beam;

[0042] Step 4: Perform high-gain frequency domain processing on the time-domain data of the beam output;

[0043] The beam output time-domain data obtained in step 3 is segmented, assuming each segment is N in length and there are L segments in total. An N-point FFT is performed on the segmented data to obtain the amplitude information |Y| at the frequency point k0 of the l-th segment. l (k0)|;

[0044] Construct a corrected phase term

[0045]

[0046] Where θ l-2 θ l-1 θ l These represent the phase of a certain frequency point in the DFT processing results of data segments l-2, l-1, and l, respectively.

[0047] Calculate the output of the high-gain frequency domain processing method at k0:

[0048]

[0049] Preferably, the beamforming method is a super-directional beamforming method.

[0050] The beneficial effects of this invention are as follows:

[0051] This invention proposes a time-space-frequency domain high-gain processing method for low-frequency line spectrum signals. It fully utilizes the time, spatial, and frequency domain information of the signal to extract processing gains in these domains. The proposed high-gain processing method effectively improves the processing gain of underwater acoustic low-frequency line spectrum signals, enabling better detection of weak underwater line spectrum signals. Attached Figure Description

[0052] Figure 1 This is a basic block diagram of the algorithm of this invention.

[0053] Figure 2 The time-frequency plot results of the spatiotemporal domain joint processing method are shown in the following figures: (a) original array elements; (b) array element results of the first-level multi-channel time-domain filtering algorithm; (c) array element results of the second-level multi-channel time-domain filtering algorithm; and (d) output results of the spatiotemporal domain joint processing method.

[0054] Figure 3 The spectrum results are from the spatiotemporal frequency domain joint processing method.

[0055] Figure 4 The curves show the gain variation of the spatiotemporal domain joint processing method: (a) the output signal-to-noise ratio (SNR) versus the input SNR; and (b) the gain versus the input SNR.

[0056] Figure 5 The curves show the gain variation of the spatiotemporal frequency domain joint processing method: (a) the output signal-to-noise ratio (SNR) versus the input SNR; and (b) the gain versus the input SNR. Detailed Implementation

[0057] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0058] This invention comprises three processing stages: time-domain, spatial-domain, and frequency-domain. The time-domain stage uses a two-stage multi-channel time-domain filtering algorithm, the spatial-domain stage employs a robust superdirectional beamforming method, and the frequency-domain stage utilizes a DFT-based high-gain frequency-domain processing method. This method fully leverages the time-domain, spatial-domain, and frequency-domain information of the signal to extract processing gains in these three domains. The multi-channel time-domain filter yields array element-domain data, where spatially isotropic noise retains strong correlation after passing through the filter. Therefore, superdirectional beamforming can be performed on the output of the two-stage multi-channel time-domain filter, achieving higher spatial-domain processing gains than conventional methods. After obtaining the beam-domain signal, the frequency-domain processing gain can be further obtained using the DFT-based high-gain frequency-domain processing method. The overall structural block diagram is shown below. Figure 1 Place Show.

[0059] Using an arbitrary-shaped sensor array to receive signals emitted by a sound source, and considering only horizontal incidence, assuming that H line spectrum signals from different targets in space are incident on the sensor receiving array in the form of far-field plane waves, the signal of the m-th array element can be expressed as:

[0060]

[0061] Among them, A h f is the amplitude of the h-th signal. h f is the frequency of the h-th signal, n is the discrete time index, and fs Sampling frequency, φ represents the phase of the m-th array element relative to the reference point when it receives the h-th signal. h Let be the initial phase of the h-th signal, which follows a uniform distribution.

[0062] Step 1: Calculate the output of the first-level multi-channel time-domain filtering algorithm;

[0063] The output of the i-th multi-channel time-domain filtering algorithm can be expressed as:

[0064]

[0065] Among them, y i (n) represents the output of the i-th multi-channel time-domain filtering algorithm, where n is the time series, i is the channel series, L is the number of taps, M is the number of array elements, w is the weight, and Δ is the number of pre-delay points.

[0066] The error function can be expressed as:

[0067] ε i (n)=x i (n)-y i (n) (18)

[0068] The weight coefficient iteration formula for the multi-channel time-domain filtering algorithm is derived from the LMS algorithm:

[0069]

[0070] Where μ is the adaptive iteration step size.

[0071] Step 2: Calculate the output of the two-stage multi-channel time-domain filtering algorithm;

[0072] The output of the first-level multi-channel time-domain filtering algorithm is then used to calculate the output of the second-level multi-channel time-domain filtering algorithm:

[0073]

[0074] Where z i (n) represents the output of the i-th secondary multi-channel time-domain filtering algorithm, y m (n) represents the output of the m-th level multi-channel time-domain filtering algorithm.

[0075] The error function can be expressed as:

[0076] ε i (n)=y i (n)-z i (n) (21)

[0077] The iterative formula for the weight coefficients is:

[0078]

[0079] Where μ is the adaptive iteration step size.

[0080] Step 3: Generate a robust super-directional weighted vector and generate beam output data;

[0081] Represent the array output data obtained in step two as a matrix:

[0082]

[0083] Perform an H-point DFT on the metadata of each matrix, i.e.:

[0084]

[0085] The frequency domain data of each array element in subband k can be written in vector form:

[0086]

[0087] The narrowband form of a well-designed robust hyper-directional weighted vector can be expressed as:

[0088] w(f k )=[w1(f k ),w2(f k ),…w M (f k )] T ,k=0,1,…H-1 (26)

[0089] Synthesize the sub-band beams:

[0090]

[0091] Perform inverse DFT on the output data of each sub-band beam:

[0092] y (n) (h), (h=0,1,…,H-1)=IDFT[Y (n) (k),(k=0,1,…,H-1)] (28)

[0093] The above process is repeated for each segment of array data, and finally the time domain data of each segment are spliced ​​together to form the long-term output sequence y(i) of the beam.

[0094] Step 4: Perform high-gain frequency domain processing on the time-domain data of the beam output;

[0095] The beam output time-domain data obtained in step three is segmented, assuming each segment is N in length, for a total of L segments. An N-point FFT is performed on the segmented data to obtain the amplitude information |Y| at frequency point k0 of the l-th segment. l (k0)|.

[0096] Construct a corrected phase term

[0097]

[0098] Where θ l-2 ,θ l-1 ,θ l These represent the phase of a certain frequency point in the DFT processing results of data segments l-2, l-1, and l, respectively.

[0099] Calculate the output of the high-gain frequency domain processing method at k0:

[0100]

[0101] Example:

[0102] The simulation signal frequency is 350Hz, the sampling frequency is 5kHz, the signal duration is 10s, the broadband noise frequency range is [300Hz-500Hz], and the signal-to-noise ratio is 0dB. The multi-channel time-domain filtering algorithm has 1000 taps and a step size of 4×10⁻⁶. -12 The pre-delay time is 0.01s. The array used is a 12-element uniform circular array with a radius of 2.5m.

[0103] Figure 2 (a) The time-frequency diagram of the original array elements is given. Figure 2 (b), (c), and (d) are time-frequency diagrams of the array element outputs of the first-level multi-channel time-domain filtering algorithm, the second-level multi-channel time-domain filtering algorithm, and the spatiotemporal joint processing method, respectively. It can be seen that the spatiotemporal joint processing method, which combines the second-level multi-channel time-domain filtering algorithm and the super-directional beamforming method, achieves the best processing effect.

[0104] In frequency domain processing, each signal segment is 1 second long, and the data segment overlap rate is 50%. To ensure fair comparison, the average power method is used to normalize the superdirectional beam output of the secondary multi-channel time-domain filtering, and the same operation is performed on the spatiotemporal frequency domain joint high-gain processing method. The processing results are as follows: Figure 3 As shown, the spatiotemporal frequency domain joint high-gain processing method further reduces background noise and narrows the "main lobe". The spatiotemporal frequency domain joint high-gain processing method has better line spectrum detection effect.

[0105] The gain calculation results of the spatiotemporal domain joint processing method are as follows: Figure 4 As shown. The gain calculation results of the spatiotemporal frequency domain joint processing method are as follows. Figure 5 As shown, the spatiotemporal frequency domain joint processing method proposed in this invention achieves the highest gain and has the best processing effect.

Claims

1. A spatiotemporal frequency domain joint high-gain line spectrum enhancement method, characterized in that, Includes the following steps: An arbitrary-shaped sensor array is used to receive signals emitted by a sound source, and only horizontal incidence is considered. Assuming H line spectrum signals from different targets in space are incident on the sensor receiving array as far-field plane waves, the signal of the m-th element is represented as: Among them, A h f is the amplitude of the h-th signal. h f is the frequency of the h-th signal, n is the discrete time index, and f s Sampling frequency, φ represents the phase of the m-th array element relative to the reference point when it receives the h-th signal. h Let be the initial phase of the h-th signal, which follows a uniform distribution; Step 1: Calculate the output of the first-level multi-channel time-domain filtering algorithm; The output of the i-th multi-channel time-domain filtering algorithm is expressed as: Among them, y i (n) represents the output of the i-th multi-channel time-domain filtering algorithm, where n is the time series, i is the channel series, L is the number of taps, M is the number of array elements, w is the weight, and Δ is the number of pre-delay points; w l,m (·) represents the weight coefficients of the multi-channel time-domain filtering algorithm; The error function is expressed as: ε i (n)=x i (n)-y i (n) (3) The weight coefficient iteration formula for the multi-channel time-domain filtering algorithm is derived from the LMS algorithm: Where μ is the adaptive iteration step size; Step 2: Calculate the output of the two-stage multi-channel time-domain filtering algorithm; The output of the first-level multi-channel time-domain filtering algorithm is then used to calculate the output of the second-level multi-channel time-domain filtering algorithm: Where z i (n) represents the output of the i-th secondary multi-channel time-domain filtering algorithm, y m (·) represents the output of the m-th first-level multi-channel time-domain filtering algorithm; The error function is expressed as: ε i (n)=y i (n)-z i (n) (6) The iterative formula for the weight coefficients is: Step 3: Generate robust super-directional weighted vectors and generate beam output data; Represent the array output data obtained in step 2 as a matrix: Perform an H-point DFT (Discrete Fourier Transform) on the metadata of each matrix, i.e.: The frequency domain data of each array element in subband k can be written in vector form: The narrowband form of the robust hyperdirectional weighted vector design is as follows: w(f k )=[w1(f k ),w2(f k ),…w M (f k )] T (11) Among them, w1(f k ),w2(f k ),…w M (f k ) represent the weighted vector values ​​of each array element; Synthesize the sub-band beams: Among them, w H (f k ) indicates a frequency of f k The weighted vector, The frequency is f k The conjugate value of the m-th term of the weighted vector; Perform inverse DFT on the output data of each sub-band beam y (n) (h),(h=0,1,…,H-1)=IDFT[Y (n) (k),(k=0,1,…,H-1)] (13) The above process is repeated for each segment of array data, and finally the time domain data of each segment are spliced ​​together to form the long-term output sequence y(i) of the beam; Step 4: Perform high-gain frequency domain processing on the time-domain data of the beam output; The beam output time-domain data obtained in step 3 is segmented, assuming each segment is N in length and there are L segments in total. An N-point FFT is performed on the segmented data to obtain the amplitude information |Y| at the frequency point k0 of the l-th segment. l (k0)|; Construct a corrected phase term Where θ l-2 θ l-1 θ l These represent the phase of a certain frequency point in the DFT processing results of data segments l-2, l-1, and l, respectively. Calculate the output of the high-gain frequency domain processing method at k0:

2. The spatiotemporal frequency domain joint high-gain line spectrum enhancement method according to claim 1, characterized in that, The beam formation method is a super-directional beamforming method.

Images