A method for continuous splicing of acoustic signals based on time-domain cross-correlation phase alignment

Abstract

The purpose of this invention is to provide a method for continuous acoustic signal splicing based on time-domain cross-correlation phase alignment, belonging to the field of acoustic signals. The method includes the following steps: receiving the original signal and bandpass filtering it; constructing a computational region for estimating the time delay difference; calculating the time-domain cross-correlation of the signal to estimate the delay; phase alignment and smooth weighted fusion; and recursively splicing segment by segment. After splicing all signals, a final continuous acoustic signal sequence is obtained. This invention ensures no phase distortion in the waveform and accurately captures the time delay, thereby achieving phase alignment. A continuous, smooth, and complete acoustic signal sequence is obtained through recursive segmentation. This invention synthesizes discrete short signal segments into a long signal with continuous phase, thereby obtaining a more granular frequency scale in the Fast Fourier Transform, effectively solving the problems of spectral overlap and insufficient resolution in narrowband signal detection.

Description

Technical Field

[0001] The present invention relates to a signal processing method, specifically a sound signal processing method. Background Technology

[0002] In modern acoustic signal processing, frequency resolution is one of the core indicators for evaluating system performance, directly affecting the accuracy of target detection, identification, and parameter estimation. According to the fundamental properties of the Fourier transform, frequency resolution Δf is inversely proportional to the signal observation duration T, i.e., Δf = 1 / T. This means that to obtain higher precision spectral subdivision, a sufficiently long continuous signal must be acquired.

[0003] However, in practical applications, due to hardware limitations, the acquired acoustic signals sometimes appear as discrete, short-duration, and discontinuous segments. If a Fast Fourier Transform (FFT) is directly performed on these short segments, the spectrum will be severely broadened, causing adjacent narrowband signals to overlap and become difficult to distinguish.

[0004] Existing solutions include zero-filling or simple time-domain splicing. However, zero-filling only interpolates and smooths the spectrum, failing to provide true physical resolution; while simple splicing introduces severe amplitude jumps and phase breaks at the junction. This discontinuity manifests as strong spectral leakage and spurious peaks in the frequency domain, severely masking the true signal characteristics. Therefore, how to reconstruct these fragmented short signals into phase-continuous long signals in the time domain has become a pressing technical challenge for improving the frequency resolution of acoustic signals. Summary of the Invention

[0005] The purpose of this invention is to provide a method for continuous splicing of acoustic signals based on time-domain cross-correlation phase alignment, which can solve the problems of phase discontinuity, amplitude jump and low frequency resolution during splicing of short-time acoustic signal segments.

[0006] The objective of this invention is achieved as follows: This invention provides a method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment, characterized by the following steps: (1) Receive the original signal and perform bandpass filtering on it; (2) Construct a computational domain for estimating the time delay difference; (3) Calculate the time-domain cross-correlation of the signal to estimate the delay; (4) Phase alignment and smooth weighted fusion; (5) Segment by segment, after all signals are spliced ​​together, the final continuous sound signal sequence is obtained.

[0007] The present invention may also include: 1. Step (1) specifically includes: (1.1) Read and obtain the original sequence of M discrete acoustic signals, each with a length of N; (1.2) Construct a Butterworth bandpass filter and set the lower passband frequency, upper passband frequency, and filter order; (1.3) Zero-phase filtering technology is used to process each segment of the signal to eliminate the phase distortion introduced by the filter, so that the waveform characteristics of each segment of the signal in the time domain do not shift.

[0008] 2. In step (1.3), the zero-phase filtering is performed using the FLTFILT algorithm, which includes performing a forward filter on the signal and then a reverse filter.

[0009] 3. The method for constructing the calculation region for estimating the time delay difference in step (2) is as follows: In the M-segment filtered signal sequence obtained in step (1), for any two adjacent signal segments, an overlap ratio R is set. In any two adjacent signal segments, the back round (N·R) of the first segment and the front round (N·R) of the second segment are selected as the calculation region for estimating the time delay difference.

[0010] 4. The method for calculating the time-domain cross-correlation estimation delay in step (3) specifically includes: (3.1) Perform normalized cross-correlation on the sequences of the overlapping region used to estimate the time delay difference: in and The calculation region for estimating the time delay difference is the two segments of signal obtained in step (1.3); (3.2) Find the k corresponding to the maximum value of R(k), which is the time delay lag between the two signal segments. The effective overlap length L is then determined from this. L is calculated as follows: .

[0011] 5. Step (4) phase alignment and smooth weighted fusion includes: in two adjacent signals, selecting the latter L of the first signal and the first L of the second signal, and constructing a weight vector of length L that linearly decreases from 1 to 0. and a weight vector that linearly increases from 0 to 1 : definition These are the last L sampling points of the previously spliced ​​signal sequence. The first L sampling points of the signal to be spliced; the fused signal segment The calculation is as follows: in The Hadamard product represents element-wise multiplication. set up The total sequence after splicing the first i-1 segments of signal is of length i. Let the i-th segment of the signal be... The new sequence after splicing the i-th signal segment Represented as: .

[0012] 6. Step (5) specifically includes: setting an initial sequence. The first segment of the filtered signal is used as an example. Starting from i = 2, the i-th segment of the signal is processed segment by segment. Perform the calculations from steps (2) to (4) above, and continuously update the spliced ​​total sequence. After splicing all M segments of the signal, the final continuous sound signal sequence is obtained. .

[0013] The advantages of this invention are as follows: First, it utilizes a zero-phase Butterworth filter to denoise multiple discrete signal segments, ensuring no phase distortion in the waveform. Then, it performs normalized cross-correlation calculations in the overlapping regions of adjacent signals to accurately capture time delays and achieve phase alignment. Finally, it constructs a linear weighted vector to smooth the overlapping parts, ultimately obtaining a continuous, smooth, and complete acoustic signal sequence through segment-by-segment recursion. This invention synthesizes discrete short signal segments into a long, phase-continuous signal, thereby obtaining a more finely detailed frequency scale in the Fast Fourier Transform, effectively solving the problems of spectral overlap and insufficient resolution in narrowband signal detection. Attached Figure Description

[0014] Figure 1 This is a flowchart of the present invention; Figure 2 This is a comparison of the sound pressure level spectra of the signal after time-domain cross-correlation phase alignment processing and the signal with the beginning and end of the time domain directly connected. Figure 3 The time-domain waveform of the original signal after filtering is directly spliced ​​together from the beginning and end at the first discontinuity point; Figure 4 This is the time-domain waveform of the signal at the first discontinuity after time-domain cross-correlation and phase alignment processing. Detailed Implementation

[0015] The invention will now be described in more detail with reference to the accompanying drawings: Implementation Method 1: The present invention provides a method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment, comprising the following steps: Step 1: Receive the original signal and perform bandpass filtering on it; Step 2: Construct the computational region for estimating the delay difference; Step 3: Calculate the time-domain cross-correlation of the signal to estimate the delay; Step 4: Phase alignment and smooth weighted fusion; Step 5: Segment by segment recursively splice together.

[0016] Implementation Method 2: Based on Implementation Method 1, step 1 includes: Step 1.1: Read and obtain the original sequence of M discrete acoustic signals, each with a length of N.

[0017] Step 1.2: Construct a Butterworth bandpass filter and set the lower passband frequency, upper passband frequency, and filter order.

[0018] Step 1.3: Use zero-phase filtering technology to process each signal segment separately to eliminate the phase distortion introduced by the filter and ensure that the waveform characteristics of each signal segment in the time domain do not shift.

[0019] Implementation Method 3: Based on Implementation Method 2, in step 1.3, zero-phase filtering is performed using the FLTFILT algorithm, which includes performing a forward filter on the signal and then a reverse filter.

[0020] Implementation Method 4: Based on Implementation Method 1, the method for constructing the calculation region for estimating the time delay difference in step 2 is as follows: In the M filtered signal sequences obtained in step 1, for any two adjacent signal segments, an overlap ratio R is set.

[0021] In any two adjacent signal segments, select the latter round(N·R) of the former signal segment and the former round(N·R) of the latter signal segment as the calculation region for estimating the time delay difference.

[0022] Implementation Method 5: Based on Implementation Method 1, the method for calculating the time-domain cross-correlation to estimate the delay in step 3 specifically includes: Step 3.1: Perform normalized cross-correlation on the sequences of the overlapping region used to estimate the time delay difference: in and The calculation area for estimating the time delay difference is the two segments of signal obtained in step 1.3.

[0023] Step 3.2: Find the k corresponding to the maximum value of R(k), which is the time delay lag between the two segments of the signal, and thus determine the effective overlap length L.

[0024] L is calculated as follows: Implementation Method 6: Based on Implementation Method 1, step 4, phase alignment and smooth weighted fusion, includes: In the two adjacent signals mentioned above, the latter L of the former signal is selected, and the former L of the latter signal is selected.

[0025] Construct a weight vector of length L that linearly decreases from 1 to 0. and a weight vector that linearly increases from 0 to 1 : definition These are the last L sampling points of the previously spliced ​​signal sequence. These are the first L sampling points of the signal to be spliced. The fused signal segment. The calculation is as follows: in The Hadamard product represents element-wise multiplication.

[0026] set up The total sequence after splicing the first i-1 segments of signal is of length i. Let the i-th segment of the signal be... The new sequence after splicing the i-th signal segment It can be represented as: Implementation Method 7: Based on Implementation Method 1, step 5 specifically includes: Set initial sequence This is the first filtered signal segment. Starting from i = 2, the i-th signal segment is filtered segment by segment. Perform the calculations from steps 2 to 4 above, and continuously update the spliced ​​total sequence. .

[0027] After splicing together all M signal segments, the final continuous sound signal sequence is obtained. .

[0028] Example: The parameters in this example are as follows: signal frequency is 61.5Hz, sampling rate fs is 50000Hz, signal length N is 50000 points, number of signals M is 30, and overlap ratio R is 0.4.

[0029] In this embodiment, a second-order Butterworth bandpass filter is constructed with an upper passband frequency of 100Hz and a lower passband frequency of 20Hz.

[0030] Figure 2 shows a comparison of the sound pressure level spectra of the signal after time-domain cross-correlation phase alignment processing and the signal directly connected end-to-end in the time domain. In the case of direct splicing, the spectral peak at 61.5 Hz does not appear; instead, a peak near 61.5 Hz is present. The spurious spectral peaks at integer multiples of frequency (i.e., spurious peaks at integer multiples of frequency 1 Hz) were identified. After time-domain cross-correlation phase alignment processing, a main peak appeared at 61.5 Hz, and its position coincided with the signal frequency, as shown in the figure. This indicates that compared to directly connecting the beginning and end of the time domain signal, this method effectively increases the signal duration, thereby achieving higher frequency resolution.

[0031] Figure 3 The image shows the time-domain waveform of the original signal after filtering, spliced ​​directly from the beginning and end at the first discontinuity. Figure 4 The time-domain waveform of the signal at the first discontinuity point is shown after time-domain cross-correlation phase alignment processing. It can be seen that when the original signal is directly spliced ​​together, a significant jump can be observed at the signal junction point. This discontinuity is the root cause of frequency domain spectral leakage. After time-domain cross-correlation phase alignment processing, the jump between signal segments is eliminated.

[0032] Therefore, this invention solves the problems of phase discontinuity, amplitude jump and low frequency resolution when splicing short-time acoustic signal segments.

[0033] The present invention provides a method for continuous splicing of acoustic signals based on phase alignment of time-domain cross-correlation. First, a zero-phase Butterworth filter is used to denoise multiple discrete signals to ensure that the waveform has no phase distortion. Then, normalized cross-correlation is performed in the overlapping area of ​​adjacent signals to accurately capture the time delay and achieve phase alignment. Finally, a linear weighted vector is constructed to perform smoothing processing on the overlapping part, and finally a continuous and smooth complete acoustic signal sequence is obtained by recursively tracing segment by segment.

Claims

1. A method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment, characterized in that: Includes the following steps: (1) Receive the original signal and perform bandpass filtering on it; (2) Construct a computational domain for estimating the time delay difference; (3) Calculate the time-domain cross-correlation of the signal to estimate the delay; (4) Phase alignment and smooth weighted fusion; (5) Segment by segment, after all signals are spliced ​​together, the final continuous sound signal sequence is obtained.

2. The method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment according to claim 1, characterized in that: Step (1) specifically includes: (1.1) Read and obtain the original sequence of M discrete acoustic signals, each with a length of N; (1.2) Construct a Butterworth bandpass filter and set the lower passband frequency, upper passband frequency, and filter order; (1.3) Zero-phase filtering technology is used to process each segment of the signal to eliminate the phase distortion introduced by the filter, so that the waveform characteristics of each segment of the signal in the time domain do not shift.

3. The method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment according to claim 2, characterized in that: In step (1.3), the zero-phase filtering is performed using the FLTFILT algorithm, which includes performing a forward filter on the signal and then a reverse filter.

4. The method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment according to claim 1, characterized in that: The method for constructing the calculation region for estimating the time delay difference in step (2) is as follows: In the M-segment filtered signal sequence obtained in step (1), for any two adjacent signal segments, an overlap ratio R is set. In any two adjacent signal segments, the back round (N·R) of the first segment and the front round (N·R) of the second segment are selected as the calculation region for estimating the time delay difference.

5. The method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment according to claim 1, characterized in that: The method for calculating the time-domain cross-correlation estimation delay in step (3) specifically includes: (3.1) Perform normalized cross-correlation on the sequences of the overlapping region used to estimate the time delay difference: in and The calculation region for estimating the time delay difference is the two segments of signal obtained in step (1.3); (3.2) Find the k corresponding to the maximum value of R(k), which is the time delay lag between the two signal segments. The effective overlap length L is then determined from this. L is calculated as follows: 。 6. The method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment according to claim 1, characterized in that: Step (4) phase alignment and smooth weighted fusion includes: in two adjacent signals, selecting the latter L of the first signal and the first L of the second signal, and constructing a weight vector of length L that linearly decreases from 1 to 0. and a weight vector that linearly increases from 0 to 1 : definition These are the last L sampling points of the previously spliced ​​signal sequence. The first L sampling points of the signal to be spliced; the fused signal segment The calculation is as follows: in The Hadamard product represents element-wise multiplication. set up The total sequence after splicing the first i-1 segments of signal is of length i. Let the i-th segment of the signal be... The new sequence after splicing the i-th signal segment Represented as: 。 7. The method for continuous splicing of acoustic signals based on temporal cross-correlation phase alignment according to claim 1, characterized in that: Step (5) specifically includes: setting an initial sequence The first segment of the filtered signal is used as an example. Starting from i = 2, the i-th segment of the signal is processed segment by segment. Perform the calculations from steps (2) to (4) above, and continuously update the spliced ​​total sequence. After splicing all M signal segments, the final continuous sound signal sequence is obtained. .

Images