A method for underwater positioning navigation multi-scale sampling rate of underwater acoustic signal dynamic splicing processing and adaptive detection
Patent Information
- Application Number
- CN202611424989.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-15
- Publication Date
- 2026-10-23
AI Technical Summary
一是当变采样率导致重叠区点数需求大于标准数据包点数时，传统拼接策略无法从前驱包中获取足够的连续点，导致信号特征边缘截断，引发匹配滤波旁瓣抬高和主峰畸变；
第一，本发明通过根据数据重叠区长度与单包数据标准长度的比较结果，采用尾部截取或前向预测补偿的双分支拼接策略。当数据重叠区长度大于单包数据标准长度时，通过自回归模型反向外推生成预测序列填补空间缺口，有助于减少信号特征边缘截断引发的匹配滤波旁瓣抬高与主峰畸变。
Smart Images
Figure CN122964503A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of underwater acoustic signal detection technology, and in particular to a dynamic splicing and adaptive detection method for underwater acoustic signals with multiple sampling rates in underwater positioning and navigation. Background Technology
[0002] In extremely complex underwater acoustic channel environments such as the open ocean and deep sea, effective detection of underwater acoustic signals is a crucial prerequisite for achieving highly reliable underwater acoustic communication, target localization, and underwater vehicle navigation. Underwater acoustic communication devices typically emit effective signals (such as hyperbolic FM or linear FM signals) with specific time-reversal characteristics or frequency modulation slopes at the transmitting end, serving as the synchronization header for a frame of communication data. Due to multipath effects, severe reverberation, time-varying Doppler shift, and interference from various marine background noises in the underwater environment, the acoustic signals acquired at the receiving end are extremely weak and must be uploaded to a back-end signal processing platform for detection and processing in real time.
[0003] In practical engineering, underwater acoustic arrays need to accommodate different distances, frequency bands, and communication requirements. However, the physical sampling rate of the front-end array elements is not fixed but needs to be dynamically reconstructed and switched across multiple scales based on environmental changes or system commands. To reduce the computational load on the back-end processing and improve the signal-to-noise ratio, the processing end typically uses digital down-conversion and downsampling techniques to shift high-frequency signals to the baseband. However, when the physical sampling rate of the array elements changes abruptly, the number of time-domain points and frequency-domain resolution of the downsampled baseband discrete signal at the processing end will change in a chain reaction. Traditional matched filtering and synchronization algorithms based on fixed sampling rates, fixed frame lengths, and fixed Fourier points will face the following problems: First, when the variable sampling rate causes the number of overlapping area points to be greater than the number of standard data packet points, the traditional splicing strategy cannot obtain enough continuous points from the preceding packet, resulting in the truncation of signal feature edges, which leads to the rise of matched filter sidelobes and distortion of the main peak. Second, if a sudden change in the number of data points in a single packet cannot adaptively trigger the switching of the Fourier points, it will directly cause false alarm interference caused by spectral aliasing. Third, during the signal sampling rate splicing process, discontinuous steps will occur at the splicing boundary; at the same time, the underwater environment contains a large amount of non-Gaussian burst impulse noise, and the traditional mean-based CFAR algorithm will include the impulse energy in the background average power, which will cause the decision threshold to be abnormally raised, making it unable to effectively detect weak underwater acoustic signals. Summary of the Invention
[0004] In view of the above-mentioned prior art, the present invention provides a method for dynamic splicing and adaptive detection of underwater acoustic signals with multi-scale sampling rates in underwater positioning and navigation, which mainly solves the technical problems existing in the background art.
[0005] To achieve the above objectives, the technical solution of this invention is implemented as follows: This invention provides a method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation. The method includes the following steps: Step S1: Obtain the real-time physical sampling rate in the underwater acoustic communication array, and determine the standard length of the downsampled single packet data and the number of feature points of the effective signal at the current downsampling rate based on the preset target downsampling rate; Step S2: Calculate the number of Fast Fourier Transform points based on the standard length of the single packet data and the number of feature points, and calculate the length of the data overlap area between adjacent data packets based on the number of Fast Fourier Transform points; Step S3: Concatenate the currently input single packet data with the previous single packet data to form a basic double packet sequence. Based on the comparison between the length of the data overlap area and the standard length of the single packet data, process the basic double packet sequence into a data packet to be detected with a length equal to the number of points of the Fast Fourier Transform. Step S4: Perform frequency domain transformation on the data packet to be detected to obtain the target frequency domain spectrum; take the conjugate of the local reference signal after frequency domain transformation to obtain the conjugate frequency domain reference spectrum; multiply the target frequency domain spectrum and the conjugate frequency domain reference spectrum point by point in the frequency domain and then perform an inverse transformation to obtain the time-domain correlated signal sequence; Step S5: Based on the length of the data overlap region and the number of feature points, calculate the boundary of the dynamic window interval where the correlation peak generated by the effective signal matching in the time-domain correlation signal sequence is located, and set the effective search window interval based on the boundary of the dynamic window interval; Step S6: Within the effective search window, the background noise is modeled as a composite Gaussian distribution model consisting of speckle and texture components, and the texture components are locally estimated. Based on the estimation results, a generalized likelihood ratio test statistic is constructed. The generalized likelihood ratio test statistic is compared with a preset threshold, and a valid signal is determined based on the comparison results.
[0006] As a preferred embodiment of the present invention, the specific content of step S1 includes: Obtain the preset physical storage duration of single packet data, and multiply the physical storage duration of single packet data by the target downsampling rate to obtain the standard length of downsampled single packet data; Obtain the preset physical duration of the valid signal, multiply the preset physical duration of the valid signal by the target downsampling rate, and obtain the number of feature points of the valid signal at the current downsampling rate.
[0007] As a preferred embodiment of the present invention, step S2 specifically includes: The specific formula for calculating the number of Fast Fourier Transform points based on the standard length of the single packet data and the number of feature points is as follows:
[0008] in, For the number of points in the Fast Fourier Transform, For standard length of single packet data, This represents the number of feature points at the current downsampling rate. This represents the floor operator; The length of the data overlap area between adjacent data packets is obtained based on the discrete difference between the number of points in the Fast Fourier Transform and the standard length of the single packet data.
[0009] As a preferred embodiment of the present invention, the specific content of step S3, which involves processing the basic double-packet sequence into a data packet to be detected with a length equal to the number of points of the Fast Fourier Transform, includes: Compare the length of the data overlap region with the standard length of the single packet data. If the length of the data overlap region is less than or equal to the standard length of the single packet data, then the last continuous sample points of the basic double packet sequence are truncated to a length equal to the number of points of the fast Fourier transform, and used as the data packet to be detected. If the length of the data overlap area is greater than the standard length of the single packet data, then the difference between the length of the data overlap area and the standard length of the single packet data is calculated to obtain the space compensation length; Extract continuous sample points from the beginning of the previous single-packet data as a fitting set, calculate the linear prediction coefficient of the autoregressive model, use the linear prediction coefficient to perform reverse spatial extrapolation on the previous single-packet data, and generate a prediction sequence with a length equal to the spatial compensation length by predicting point by point. The predicted sequence, the previous single packet data, and the currently input single packet data are concatenated sequentially to form the data packet to be detected.
[0010] As a preferred embodiment of the present invention, the specific content of step S4 includes: Obtain the local reference signal at the current downsampling rate, resample the local reference signal to the target downsampling rate, and perform time-domain inversion on the resampled local reference signal to obtain the inverted reference signal; The data packet to be detected and the inverted reference signal are transformed to the frequency domain respectively to obtain the target frequency domain spectrum and the reference frequency domain spectrum. The conjugate of the reference frequency domain spectrum is then taken to obtain the conjugate frequency domain reference spectrum. The target frequency domain spectrum and the conjugate frequency domain reference spectrum are multiplied in the frequency domain by a Hadamard product to obtain the product frequency domain spectrum; an inverse fast Fourier transform is performed on the product frequency domain spectrum to obtain a time-domain correlated signal sequence.
[0011] As a preferred embodiment of the present invention, the specific content of step S5 includes: Calculate the difference between the length of the data overlap region and the number of feature points, and obtain the lower boundary of the dynamic window interval based on the difference. The specific expression is as follows:
[0012] in, This represents the lower boundary of the dynamic window interval. The length of the data overlap area; The upper boundary of the dynamic window interval is obtained based on the lower boundary of the dynamic window interval and the standard length of the single packet data, and its specific expression is as follows:
[0013] in, This represents the upper boundary of the dynamic window interval; The effective search window interval is obtained based on the lower boundary and the upper boundary of the dynamic window interval.
[0014] As a preferred embodiment of the present invention, step S6, which involves modeling the background noise as a composite Gaussian distribution model composed of speckle components and texture components, specifically includes: With the target unit to be detected as the center, a protection unit and a reference unit are set up within the effective search window range, and background noise samples in the reference unit are extracted to form a background noise sample vector. The background noise sample vector follows a composite Gaussian distribution model consisting of speckle and texture components, wherein the speckle component follows a complex Gaussian distribution and the texture component follows a Gamma distribution; the mathematical expression of the composite Gaussian distribution model is as follows:
[0015] in, The model is a composite Gaussian distribution. For speckle components that follow a complex Gaussian distribution, For texture components that follow a Gamma distribution.
[0016] As a preferred embodiment of the present invention, the specific content of the local estimation of the texture components in step S6 includes: The background noise sample vectors are sorted by amplitude, and the samples at the top and bottom of the sorted set are removed to obtain the purified background noise reference set. Based on the purified background noise reference set, local maximum likelihood estimation is performed on the texture components to obtain local texture power estimates, the specific calculation formula of which is as follows:
[0017] in, This is a local texture power estimate. The effective number of samples in the purified background noise reference set. This serves as a reference set for the purified background noise. The time-domain correlated signal sequence is the first in the purified background noise reference set. Complex amplitude values at each sample point.
[0018] As a preferred embodiment of the present invention, in step S6, the mathematical expression for constructing the generalized likelihood ratio test statistical calculator based on the estimation results is as follows:
[0019] in, For the generalized likelihood ratio test statistical calculation, For the time-domain correlation signal sequence in the current target unit to be detected The complex amplitude value at that location.
[0020] In a preferred embodiment of the present invention, step S6, which compares the generalized likelihood ratio test statistic calculator with a preset threshold and determines whether a valid signal is detected based on the comparison result, specifically includes: Determine whether the generalized likelihood ratio test statistic is greater than a preset threshold. If so, determine that the target unit to be detected has a valid signal, and determine the timing index corresponding to the relevant peak value generated by the valid signal matching within the valid search window as the arrival position of the underwater acoustic signal. If not, it is determined that there is no valid signal, and the sliding detection continues.
[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: First, this invention employs a dual-branch splicing strategy of tail truncation or forward prediction compensation based on the comparison between the length of the data overlap region and the standard length of a single packet of data. When the length of the data overlap region is greater than the standard length of a single packet of data, a prediction sequence is generated by back extrapolation through an autoregressive model to fill the spatial gap, which helps to reduce the matched filter sidelobe elevation and main peak distortion caused by signal feature edge truncation.
[0022] Second, this invention adaptively deduces the number of Fast Fourier Transform (FFT) points that meet the condition constraints based on the sum of the standard length of a single packet of data and the number of feature points. Furthermore, the scale of the FFT points automatically changes with the physical sampling rate, which helps to reduce spectral aliasing caused by insufficient frequency domain resolution or signal interference introduced by zero padding.
[0023] Third, the present invention uses a bilateral elimination detection algorithm that combines a composite Gaussian distribution model with a generalized likelihood ratio test to eliminate abnormal samples within the effective search window to purify background noise samples. This can maintain relatively stable detection performance under strong non-Gaussian interference environments and achieve effective detection of underwater acoustic signals in underwater acoustic communication, positioning and navigation. Attached Figure Description
[0024] Figure 1 This is a schematic diagram of the method steps of the present invention.
[0025] Figure 2 This is a schematic diagram of the data stitching timing in a scenario with a physical sampling rate of 125kHz, according to the present invention.
[0026] Figure 3 This is a schematic diagram of the data splicing timing in the scenario where the physical sampling rate suddenly changes to 200kHz.
[0027] Figure 4 This diagram illustrates a performance comparison between the bilateral elimination detection algorithm based on the composite Gaussian distribution model combined with the generalized likelihood ratio test and the traditional cell average CFAR detection algorithm. Detailed Implementation
[0028] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. In the following description, the expression "some embodiments" refers to a subset of all possible embodiments; however, it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.
[0029] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.
[0030] It should be understood that the present invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Furthermore, the terminology used herein is intended only to describe particular embodiments and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “compose” and / or “comprising,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.
[0031] It should also be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "inner," "outer," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.
[0032] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.
[0033] This application provides a method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation. Please refer to the attached document. Figure 1 The method includes the following steps: Step S1: Obtain the real-time physical sampling rate in the underwater acoustic communication array, and determine the standard length of the downsampled single packet data and the number of feature points of the effective signal at the current downsampling rate based on the preset target downsampling rate.
[0034] In this embodiment, the processing end presets the physical storage time for a single packet of data. (Unit: seconds). Physical storage time for the single packet data. With target downsampling rate Multiply to obtain the standard length of the downsampled single packet data. :
[0035] Furthermore, the physical duration of the effective signal emitted by the transmitter is obtained. (Unit: seconds), the physical duration of the effective signal is determined by the synchronization header signal sent by the underwater acoustic communication device and is a fixed value. The physical duration of the effective signal... With the target downsampling rate Multiply to obtain the number of feature points of the effective signal at the current downsampling rate. :
[0036] In practical engineering, the physical duration of an effective signal Typically, this refers to the physical storage time for a single data packet. One-half of, then .
[0037] Step S2: Calculate the number of Fast Fourier Transform points based on the standard length of the single packet data and the number of feature points, and calculate the length of the data overlap area between adjacent data packets based on the number of Fast Fourier Transform points.
[0038] In this embodiment, in order to perform efficient matched filtering in the frequency domain, the number of points in the Fast Fourier Transform (FFT) needs to meet the following constraints: (1) it must be able to completely encompass the discrete total length after packetization; (2) it must meet the butterfly operation structure requirements of the radix-2 FFT. The present invention adopts the following adaptive derivation formula:
[0039] in, For the number of points in the Fast Fourier Transform, For standard length of single packet data, This represents the number of feature points at the current downsampling rate. This represents the floor operator; The length of the data overlap area between adjacent data packets is obtained based on the discrete difference between the number of points in the Fast Fourier Transform and the standard length of a single data packet. The specific calculation formula is as follows:
[0040] in, This represents the length of the data overlap area between adjacent data packets.
[0041] Specifically, when the total length of the data packet to be detected is equal to the number of Fast Fourier Transform points... At that time, the contribution of the currently input single packet data is... Points, remaining The data needs to be truncated from the end of the previous packet. Therefore, the length of the data overlap area... It reflects the degree of dependency between adjacent data packets.
[0042] In this embodiment, the length of the data overlap region The value automatically adjusts as the sampling rate changes: when the sampling rate increases, the standard length of a single packet of data... Increase the number of points in the Fast Fourier Transform. The length of the data overlap area increases synchronously. Corresponding changes; when the sampling rate decreases, the standard length of a single packet of data changes. Reduce the number of points in the Fast Fourier Transform. The length of the data overlap area decreases synchronously. Corresponding changes.
[0043] Step S3: Concatenate the currently input single packet data with the previous single packet data to form a basic double packet sequence. Based on the comparison between the length of the data overlap area and the standard length of the single packet data, process the basic double packet sequence into a data packet to be detected with a length equal to the number of points of the Fast Fourier Transform.
[0044] In this embodiment, when the currently input single packet data is the received first packet data (i.e. When there is no previous packet historical data, the constructed length is equal to the standard length of the single packet data. All-zero precursor vector Then, the all-zero precursor vector... Compared with the first packet of data received so far By concatenating the first and last parts, we obtain the basic double-packet sequence with initial values. :
[0045] Furthermore, when the currently input single packet data is the second packet data or subsequent data packets (i.e. When the current input single packet data is received, it is directly passed to the input single packet data. Data from the previous package spliced into the basic double-packet sequence .
[0046] Furthermore, the lengths of the data overlap regions are compared. With the standard length of the single packet data Based on the size relationship, select the corresponding splicing branch according to the comparison result: The first splicing branch, when the length of the data overlap area... Less than or equal to the standard length of the single packet data This indicates that the number of data points at the end of the previous single packet is sufficient to meet the length requirement of the overlapping area. Therefore, from the basic double packet sequence... Starting from the tail, cut forward continuously. 1 sample point, as the data packet to be detected .
[0047] Specifically, taking the data from the first package as an example... For example, the basic double-packet sequence is: Its first packet of data to be detected for: That is, the first packet of data to be detected Depend on One zero-filling point and the first packet of raw data composition.
[0048] Specifically, based on the data in the second package For example, the basic double-packet sequence is: Its second packet of data to be detected for: =[ , That is, the second packet of data to be detected starts from the end of the previous packet. Points and current data packets composition.
[0049] The second splicing branch, when the length of the data overlap area... Greater than the standard length of the single packet data This indicates that the number of data points at the end of the previous single packet is insufficient to meet the length requirement of the overlapping region, and the total length of the basic double packet sequence is also insufficient. Also smaller than ,because ,but There exists a length of Spatial gap.
[0050] To fill this spatial gap, the traditional approach is to directly fill the gap with zeros. However, zero-filling introduces an artificial step that is discontinuous with the signal at the splicing boundary. Therefore, this invention extracts the data from the previous single packet. The continuity of the head For each sample point, calculate using the Burg algorithm. Linear prediction coefficient vector of an autoregressive model The specific process is as follows: Extract the data from the previous single packet. The continuity of the head 1 sample point as the fitting set ,in Need to be greater than Order of an autoregressive model In this embodiment, the value is 2 to 3 times that of the previous embodiment. . The value is usually 1024 or 2048.
[0051] Calculate the sample mean of the fitted set The fitted set is then subjected to mean-removal processing to obtain a zero-mean sequence:
[0052] in, For the first Each sample value.
[0053] Zero-mean sequence These are respectively used as the 0th order forward prediction error sequences. and backward prediction error sequence :
[0054] For order Perform the following recursion: (1) Calculate the first First-order reflection coefficient :
[0055] in, For the first The first order forward prediction error sequence One error value, For the first The first order forward prediction error sequence One error value, For the first The first backward prediction error sequence One error value.
[0056] (2) Using the reflection coefficient Update the forward prediction error sequence:
[0057] (3) Using the reflection coefficient Update backward prediction error:
[0058] in, For the first The first backward prediction error sequence One error value, For the first The first backward prediction error sequence One error value.
[0059] By using the reflection coefficient, each recursion is solved under the constraint of minimizing the sum of the forward and backward prediction error powers, thus ensuring... The stability of the first-order autoregressive model for estimating the linear prediction coefficient vector.
[0060] Complete all After recursion, the reflection coefficients of each order are obtained. The linear prediction coefficient vector is obtained by weighted linear combination of the reflection coefficients. : Using the linear prediction coefficient vector, backward spatial extrapolation is performed on the previous single-packet data, and point-by-point predictions are generated with a length equal to the spatial compensation length. Predicted sequence Its reverse recursive formula is:
[0061] in, Given sample points, For the first Linear prediction coefficients of order 1 This is the order index.
[0062] The predicted sequence The previous single packet data and the currently input single packet data The data packets to be detected are concatenated sequentially. :
[0063] This achieves the goal of increasing the number of points from the spatial filler to the number of points in the Fast Fourier Transform. At the same time, the spatiotemporal continuity and phase smoothness of the signal boundaries are maintained.
[0064] Step S4: Perform frequency domain transformation on the data packet to be detected to obtain the target frequency domain spectrum, take the conjugate of the local reference signal after frequency domain transformation to obtain the conjugate frequency domain reference spectrum, perform point-by-point multiplication of the target frequency domain spectrum and the conjugate frequency domain reference spectrum in the frequency domain and then perform inverse transformation to obtain the time-domain correlated signal sequence.
[0065] In this embodiment, a local reference signal at the current downsampling rate is obtained. The local reference signal is a digital copy of the valid signal emitted by the transmitter and is pre-stored in the processing unit. The local reference signal is resampled to the target downsampling rate so that its discretization point count matches the feature point count at the current sampling rate. Then, a time-domain inversion is performed on the resampled local reference signal to obtain the inverted reference signal.
[0066] Specifically, the purpose of performing time-domain inversion is to realize the frequency-domain conjugation operation of matched filtering. That is, the correlation operation performed in the time domain is equivalent to convolving the inverted signal with the received signal, while in the frequency domain it is represented by taking the conjugation of the reference spectrum.
[0067] Furthermore, the data packets to be detected are respectively... Perform a Fast Fourier Transform on the inverted reference signal to obtain the target frequency domain spectrum and the reference frequency domain spectrum. Take the conjugate of the reference frequency domain spectrum point by point to obtain the conjugate frequency domain reference spectrum.
[0068] Furthermore, the target frequency domain spectrum and the conjugate frequency domain reference spectrum are multiplied in the frequency domain to obtain the product frequency domain spectrum, the specific expression of which is as follows:
[0069] in, The product frequency domain spectrum, For the target frequency domain spectrum, For the conjugate frequency domain reference spectrum, .
[0070] Performing an inverse fast Fourier transform on the product frequency domain spectrum yields a time-domain correlation signal sequence, indicating that when there is an effective signal in the received signal that matches the local reference signal, a significant correlation peak appears at the corresponding time index; the mismatched signal components are suppressed to a lower correlation basis.
[0071] Step S5: Based on the length of the data overlap region and the number of feature points, calculate the dynamic window interval boundary where the correlation peak generated by the effective signal matching in the time-domain correlation signal sequence is located, and set the effective search window interval based on the dynamic window interval boundary.
[0072] In this embodiment, the first part of the time-domain correlated signal sequence The points represent data repetition areas. The starting positions where the truly effective correlation energy might appear need to skip the overlapping areas, while also considering the number of feature points. This determines the broadening interval of the effective signal in the time-domain correlation spectrum. Therefore, by calculating the difference between the length of the data overlap region and the number of feature points, the lower boundary of the dynamic window interval is obtained based on this difference, and its specific expression is as follows:
[0073] in, This represents the lower boundary of the dynamic window interval. The length of the data overlap area; Furthermore, based on the lower boundary of the dynamic window interval and the standard length of the single packet data, the upper boundary of the dynamic window interval is obtained, and its specific expression is as follows:
[0074] in, This represents the upper boundary of the dynamic window interval; The effective search window range is obtained based on the lower and upper boundaries of the dynamic window range. .
[0075] Specifically, in the time-domain correlation signal sequence, only the correlation peaks falling within the effective search window interval correspond to the effective matching energy contributed by the current input single packet data; the correlation peaks outside the window come from the data repetition area, and if they are searched indiscriminately, sidelobe false peaks may be misjudged as effective signals.
[0076] For example, see Figure 2 When the physical sampling rate In scenarios where a target downsampling rate is set... . , .
[0077] The standard length of a single packet of data is calculated to be point; Simultaneously, the number of feature points of the effective hyperbolic frequency modulated signal at the current scale is calculated. for: point; but ; ; ; ; The calculated number of transformation points is 8192. Furthermore, the length of the data overlap area between adjacent data packets is then determined. for:
[0078] Furthermore, determine the current overlap length boundary relationship; currently, it satisfies... Therefore, we choose to execute the first concatenation branch, and the specific process is as follows: Regarding the first data The splicing process is as follows: First, the initial boundary conditions are activated, and then a virtual zero-predecessor vector with a length of 5000 points is generated. .Will With the first data packet currently received By concatenating the first and last parts, a basic double-packet sequence with a length of 10,000 points is formed:
[0079] Subsequently, the first splicing branch was used to directly extract the latter part of the 10,000-point sequence. These discrete points are reassembled into the first data packet to be detected. :
[0080] The temporal spatial topology of the first large packet data to be detected is as follows: the first 3192 discrete points are all-zero data repetition areas, and the last 5000 discrete points are the first data packet measured in practice, with a total length of exactly 8192 points.
[0081] For the subsequent second data packet ( The splicing process: directly splicing the first original data packet... (5000 points) and the second raw data packet (5000 points) concatenated into a standard basic double-package sequence of 10000 points. Using the first splicing branch, tail truncation is performed again to extract the last 8192 consecutive sample points of the 10000-point sequence, constructing the second data packet to be detected. :
[0082] This process is repeated to generate a continuous output of data packets to be detected.
[0083] Then, the current effective search window boundary is calculated:
[0084]
[0085] Furthermore, any of the above-mentioned data packets to be detected The target frequency spectrum is obtained by inputting the 8192-point Fast Fourier Transform (FFT) points into the frequency domain. Simultaneously, the locally pre-stored 40ms physical duration hyperbolic FM signal is discretized at a sampling rate of 62.5kHz, resulting in a 2500-point time-domain digital signal. This signal is then inverted in the time domain, and 8192 points are added to its left. 2500 = 5692 zeros are padded to a length of 8192 points; then a Fourier transform is performed on the 8192 points to obtain the conjugate frequency domain reference spectrum. .
[0086] Within the frequency domain processor, the target frequency domain spectrum With conjugate frequency domain reference spectrum Perform the Hadamard product, and then inversely transform the frequency domain spectrum of the product back to the time domain to reconstruct the time-domain correlated signal sequence.
[0087] For example, see Figure 3 When the physical sampling rate In this scenario, the target downsampling rate is set to... . , .
[0088] The standard length of a single packet of data is calculated to be point; Simultaneously, the number of feature points of the effective hyperbolic frequency modulated signal at the current scale is calculated. for: ; but ; ; ; ; As the number of discrete points increases, the frequency domain matched filter scale is adaptively upgraded from 8192 points to 16384 points to ensure sufficient frequency domain resolution and complete signal coverage. Subsequently, the required data overlap length between adjacent data packets at this high sampling rate is automatically calculated. for:
[0089] Furthermore, determine the current data overlap length boundary relationship. At this point, the length of the data overlap area has exceeded the standard length of a single data packet. This means that if the first splicing branch is still used, the total length will only be 8000 + 8000 = 16000 points, which still falls short of 16384 points, i.e., there is a problem. Discrete spatial gaps at points.
[0090] Therefore, in this example, the second splicing branch is selected for execution, and the specific process is as follows: Calculate the number of points that need zero-space compensation. :
[0091] Suppose the current input is the mutated i-th... Data packets The preceding data packet is In the second splicing branch, a discrete space of 384 elements is constructed to extract the previous data packet. Head Then, using the Burg algorithm to quickly solve for consecutive samples, the results are obtained. The linear prediction coefficient vector of the autoregressive model is obtained, and a sequence of 384 predicted sample points is iteratively derived in reverse. These 384 predicted points are then concatenated at the beginning of the basic sequence to obtain a data packet containing 16,384 points to be detected. This ensures the continuity of the waveform derivative at the concatenation point, effectively avoiding the high-frequency glitches and spectral breaks caused by traditional direct zero-padding.
[0092] Then, the current effective search window boundary is calculated:
[0093]
[0094] Step S6: Within the effective search window, the background noise is modeled as a composite Gaussian distribution model consisting of speckle and texture components, and the texture components are locally estimated. Based on the estimation results, a generalized likelihood ratio test statistic is constructed. The generalized likelihood ratio test statistic is compared with a preset threshold, and a valid signal is determined based on the comparison results.
[0095] Furthermore, in step S6, the specific content of modeling the background noise as a composite Gaussian distribution model composed of speckle components and texture components includes: Centered on the target unit to be detected, a protection unit and a reference unit are set up within the effective search window. The protection unit is located on both sides of the target unit to prevent signal energy leakage from contaminating the effective search window. The reference unit is located outside the protection unit to collect background noise samples.
[0096] Specifically, background noise samples are extracted from the reference unit to form a background noise sample vector. In complex underwater acoustic environments, background noise is not ideal Gaussian white noise, but a non-stationary random process modulated by rapidly fluctuating local scattering interference effects and slowly time-varying reverberation intensity envelope. Therefore, this invention models the background noise sample vector as a composite Gaussian distribution model composed of speckle and texture components, wherein the speckle component follows a complex Gaussian distribution, and the texture component follows a Gamma distribution; the expression of the composite Gaussian distribution model is as follows:
[0097] in, The model is a composite Gaussian distribution. For speckle components that follow a complex Gaussian distribution, For texture components that follow a Gamma distribution.
[0098] Furthermore, in step S6, the specific content of performing local estimation of the texture components includes: The background noise sample vectors are sorted by amplitude, and samples at the top and bottom of the sorted set are removed to obtain a purified background noise reference set. This ensures that the samples accurately reflect the intensity level of the background noise, rather than the power of the signal components. Let the number of effective samples be... Based on the purified background noise reference set, local maximum likelihood estimation is performed on the texture components to obtain local texture power estimates, the specific calculation formula of which is as follows:
[0099] in, This is a local texture power estimate. The effective number of samples in the purified background noise reference set. This serves as a reference set for the purified background noise. The time-domain correlated signal sequence is the first in the purified background noise reference set. Complex amplitude values at each sample point.
[0100] Furthermore, in step S6, the specific mathematical expression for constructing the generalized likelihood ratio test statistic calculator based on the estimation results is as follows:
[0101] in, For the generalized likelihood ratio test statistical calculation, For the time-domain correlation signal sequence in the current target unit to be detected The complex amplitude value at that location.
[0102] For details, see Figure 4 This figure reflects the performance comparison curves of the bilateral elimination detection algorithm based on the composite Gaussian distribution model combined with the generalized likelihood ratio test (i.e., bilateral elimination + GLRT in the figure) and the traditional cell average CFAR detection algorithm (i.e., traditional CA-CFAR in the figure) under a composite Gaussian channel. The horizontal axis in the figure represents the input signal-to-noise ratio (dB), the vertical axis represents the detection probability, and the horizontal dashed line marks the detection probability of 0.9 as the engineering reliability detection baseline.
[0103] like Figure 4 It can be seen that in the low signal-to-noise ratio range (-5dB to 5dB), the composite Gaussian noise and multipath clutter jointly suppress the target signal, and the detection probability of both algorithms is close to 0, with no significant difference in performance.
[0104] In the mid-signal-to-noise ratio range (5dB to 10dB), this invention preprocesses samples within the effective search window through bilateral extremum removal, filtering out multipath spikes and deep fading singularities to obtain a cleaned reference set. It then performs local estimation of texture components and constructs a generalized likelihood ratio test calculus for decision-making. This significantly improves the stability of background noise sample estimation and rapidly increases the detection probability, reaching 0.6 near 10dB. In contrast, the traditional unit-averaged CFAR detection algorithm directly raises the noise mean due to multipath anomaly energy, resulting in an excessively raised decision threshold and a detection probability consistently below 0.05, essentially losing its detection capability.
[0105] In the high signal-to-noise ratio range (10dB to 15dB), the algorithm of this invention achieves a detection probability of 0.9 at approximately 10.5dB and converges quickly to 1; while the traditional cell average CFAR detection algorithm only achieves the same detection performance at approximately 14dB, requiring an additional signal-to-noise ratio margin of approximately 3.5dB compared to the algorithm of this invention.
[0106] Therefore, this invention completes the purification of pre-noise samples by bilateral extreme value removal, and constructs a generalized likelihood ratio test calculus for decision-making after estimating the texture components based on the purified reference set. It improves the robustness of the detector from two dimensions: noise estimation and decision criteria, and effectively reduces the signal-to-noise ratio threshold for reliable detection.
[0107] Furthermore, in step S6, the comparison of the generalized likelihood ratio test statistic calculator with a preset threshold, and the determination of whether a valid signal is detected based on the comparison result, specifically includes: Determine whether the generalized likelihood ratio test statistic is greater than a preset threshold. If so, determine that the target unit to be detected has a valid signal, and determine the timing index corresponding to the relevant peak value generated by the valid signal matching within the valid search window as the arrival position of the underwater acoustic signal. If not, it is determined that there is no valid signal, and the sliding detection continues.
[0108] In this embodiment, the generalized likelihood ratio test statistic is set as the ratio of the instantaneous power of the target unit to the estimated local texture power. The larger the ratio, the higher the probability that the target unit contains a valid target signal; the smaller the ratio, the more likely the target unit contains only background noise and no valid target signal.
[0109] Specifically, once the texture components of the background noise are accurately estimated, in an impulse noise environment, because this invention sorts the effective search window samples by amplitude and removes extreme values before texture estimation, the impulse components will not affect the local texture power estimate. The calculation, and thus the local texture power estimate. It will not rise abnormally with the appearance of pulses, thus ensuring that the generalized likelihood ratio test statistic can still effectively detect weak signals.
[0110] In this embodiment, the preset threshold The setting method is as follows: based on the preset false alarm probability The corresponding preset threshold value is determined through Monte Carlo simulation. For background noise following a composite Gaussian distribution, the distribution under the no-signal assumption can be approximated by a Gamma distribution, and its preset threshold... Determined by the following formula:
[0111] in, No signal condition The probability density function of the substatistic is calculated. This is a statistical computing unit operating under no-signal conditions. In practical engineering, to adapt to different sea conditions, the preset threshold... The adjustment range is generally ±20% of the center value.
[0112] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation, characterized in that, The method includes the following steps: Step S1: Obtain the real-time physical sampling rate in the underwater acoustic communication array, and determine the standard length of the downsampled single packet data and the number of feature points of the effective signal at the current downsampling rate according to the preset target downsampling rate; Step S2: Calculate the number of Fast Fourier Transform points based on the standard length of the single packet data and the number of feature points, and calculate the length of the data overlap area between adjacent data packets based on the number of Fast Fourier Transform points; Step S3: Concatenate the currently input single packet data with the previous single packet data to form a basic double packet sequence. Based on the comparison between the length of the data overlap area and the standard length of the single packet data, process the basic double packet sequence into a data packet to be detected with a length equal to the number of points of the Fast Fourier Transform. Step S4: Perform frequency domain transformation on the data packet to be detected to obtain the target frequency domain spectrum; take the conjugate of the local reference signal after frequency domain transformation to obtain the conjugate frequency domain reference spectrum; multiply the target frequency domain spectrum and the conjugate frequency domain reference spectrum point by point in the frequency domain and then perform an inverse transformation to obtain the time-domain correlated signal sequence; Step S5: Based on the length of the data overlap region and the number of feature points, calculate the boundary of the dynamic window interval where the correlation peak generated by the effective signal matching in the time-domain correlation signal sequence is located, and set the effective search window interval based on the boundary of the dynamic window interval; Step S6: Within the effective search window, the background noise is modeled as a composite Gaussian distribution model consisting of speckle and texture components, and the texture components are locally estimated. Based on the estimation results, a generalized likelihood ratio test statistic is constructed. The generalized likelihood ratio test statistic is compared with a preset threshold, and a valid signal is determined based on the comparison results.
2. The method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation according to claim 1, characterized in that, The specific content of step S1 includes: Obtain the preset physical storage duration of single packet data, and multiply the physical storage duration of single packet data by the target downsampling rate to obtain the standard length of downsampled single packet data; Obtain the preset physical duration of the valid signal, multiply the preset physical duration of the valid signal by the target downsampling rate, and obtain the number of feature points of the valid signal at the current downsampling rate.
3. The method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation according to claim 2, characterized in that, The specific content of step S2 includes: The specific formula for calculating the number of Fast Fourier Transform points based on the standard length of the single packet data and the number of feature points is as follows: in, For the number of points in the Fast Fourier Transform, For standard length of single packet data, This represents the number of feature points at the current downsampling rate. This represents the floor operator; The length of the data overlap area between adjacent data packets is obtained based on the discrete difference between the number of points in the Fast Fourier Transform and the standard length of the single packet data.
4. The method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation according to claim 3, characterized in that, In step S3, the specific content of processing the basic double-packet sequence into a data packet to be detected with a length equal to the number of points of the Fast Fourier Transform includes: Compare the length of the data overlap region with the standard length of the single packet data. If the length of the data overlap region is less than or equal to the standard length of the single packet data, then the last continuous sample points of the basic double packet sequence are truncated to a length equal to the number of points of the fast Fourier transform, and used as the data packet to be detected. If the length of the data overlap area is greater than the standard length of the single packet data, then the difference between the length of the data overlap area and the standard length of the single packet data is calculated to obtain the space compensation length; Extract continuous sample points from the beginning of the previous single-packet data as a fitting set, calculate the linear prediction coefficient of the autoregressive model, use the linear prediction coefficient to perform reverse spatial extrapolation on the previous single-packet data, and generate a prediction sequence with a length equal to the spatial compensation length by predicting point by point. The predicted sequence, the previous single packet data, and the currently input single packet data are concatenated sequentially to form the data packet to be detected.
5. A method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation according to claim 4, characterized in that, The specific content of step S4 includes: Obtain the local reference signal at the current downsampling rate, resample the local reference signal to the target downsampling rate, and perform time-domain inversion on the resampled local reference signal to obtain the inverted reference signal; The data packet to be detected and the inverted reference signal are transformed to the frequency domain respectively to obtain the target frequency domain spectrum and the reference frequency domain spectrum. The conjugate of the reference frequency domain spectrum is then taken to obtain the conjugate frequency domain reference spectrum. The target frequency domain spectrum and the conjugate frequency domain reference spectrum are multiplied in the frequency domain by a Hadamard product to obtain the product frequency domain spectrum; an inverse fast Fourier transform is performed on the product frequency domain spectrum to obtain a time-domain correlated signal sequence.
6. A method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation according to claim 5, characterized in that, The specific content of step S5 includes: Calculate the difference between the length of the data overlap region and the number of feature points, and obtain the lower boundary of the dynamic window interval based on the difference. The specific expression is as follows: in, This represents the lower boundary of the dynamic window interval. The length of the data overlap area; The upper boundary of the dynamic window interval is obtained based on the lower boundary of the dynamic window interval and the standard length of the single packet data, and its specific expression is as follows: in, This represents the upper boundary of the dynamic window interval; The effective search window interval is obtained based on the lower boundary and the upper boundary of the dynamic window interval.
7. A method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation according to claim 6, characterized in that, In step S6, the specific content of modeling the background noise as a composite Gaussian distribution model composed of speckle components and texture components includes: With the target unit to be detected as the center, a protection unit and a reference unit are set up within the effective search window range, and background noise samples in the reference unit are extracted to form a background noise sample vector. The background noise sample vector follows a composite Gaussian distribution model consisting of speckle and texture components, wherein the speckle component follows a complex Gaussian distribution and the texture component follows a Gamma distribution; the mathematical expression of the composite Gaussian distribution model is as follows: in, The model is a composite Gaussian distribution. For speckle components that follow a complex Gaussian distribution, For texture components that follow a Gamma distribution.
8. A method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation according to claim 7, characterized in that, In step S6, the specific content of performing local estimation of the texture components includes: The background noise sample vectors are sorted by amplitude, and the samples at the top and bottom of the sorted set are removed to obtain the purified background noise reference set. Based on the purified background noise reference set, local maximum likelihood estimation is performed on the texture components to obtain local texture power estimates, the specific calculation formula of which is as follows: in, This is a local texture power estimate. The effective number of samples in the purified background noise reference set. This serves as a reference set for the purified background noise. The time-domain correlated signal sequence is the first in the purified background noise reference set. Complex amplitude values at each sample point.
9. A method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation, as described in claim 8, is characterized in that... In step S6, the mathematical expression for constructing the generalized likelihood ratio test statistician based on the estimation results is as follows: in, For the generalized likelihood ratio test statistical calculation, For the time-domain correlation signal sequence in the current target unit to be detected The complex amplitude value at that location.
10. A method for dynamic stitching and adaptive detection of underwater acoustic signals with multiple sampling rates in underwater positioning and navigation, as described in claim 9, is characterized in that... In step S6, comparing the generalized likelihood ratio test statistic calculator with a preset threshold and determining whether a valid signal is detected based on the comparison result includes the following specific steps: Determine whether the generalized likelihood ratio test statistic is greater than a preset threshold. If so, determine that the target unit to be detected has a valid signal, and determine the timing index corresponding to the relevant peak value generated by the valid signal matching within the valid search window as the arrival position of the underwater acoustic signal. If not, it is determined that there is no valid signal, and the sliding detection continues.