CN107635181B - Multi-address sensing source feedback optimization method based on channel learning - Google Patents
Multi-address sensing source feedback optimization method based on channel learning Download PDFInfo
- Publication number
- CN107635181B CN107635181B CN201710832980.5A CN201710832980A CN107635181B CN 107635181 B CN107635181 B CN 107635181B CN 201710832980 A CN201710832980 A CN 201710832980A CN 107635181 B CN107635181 B CN 107635181B
- Authority
- CN
- China
- Prior art keywords
- channel
- signal
- time
- frequency
- signals
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Abstract
The invention provides a multi-access perception source feedback optimization method based on channel learning. Carrying out feature extraction on the collected white dolphin singing signals to obtain priori knowledge; carrying out quasi-biological signal modeling on the Chinese white dolphin cry; constructing an intelligent marine organism cry signal library and outputting a multi-node biological signal model in real time; estimating a channel by transmitting a channel training signal and utilizing an RLS algorithm to obtain a channel impulse response function; and (4) performing convolution on output signals in the intelligent signal library and time inverse functions of the channels respectively to carry channel information, and obtaining the bionic detection waveform after feedback optimization. The invention utilizes bio-signals to improve concealment in underwater environments. The bionic detection waveform carrying the multi-path information of the channel can be better adapted to a complex ocean channel, and when a receiving end processes echoes, the interference of the multi-path of the channel is inhibited, and more accurate focusing is realized. The method can be well applied to the combined operation of a plurality of underwater UUV.
Description
Technical Field
The invention relates to an underwater sonar signal processing method, in particular to a multi-address perception source feedback optimization method based on channel learning.
Background
The multiple access perception source is a 'biology-like' signal model which is inspired by the cry signal of marine mammals. By utilizing the diversity and identifiability of the biological sound-calling signal on the time-frequency structure, the method can be applied to underwater multi-node sonar detection, can inhibit echo aliasing, and reduce the misjudgment rate during underwater detection.
The channel feedback optimization is originally from the field of radar signal processing, channel characteristics are analyzed by transmitting a channel learning signal before the detection process is started, and subsequently transmitted detection waveforms carry channel multi-path information, so that the transmission waveforms are optimized in a cyclic reciprocating manner. The optimized detection signal is continuously adaptive to a time-varying space-variant ocean channel, a closed-loop feedback mechanism from a receiver to a transmitter can adaptively optimize a sonar transceiving system according to prior and real-time environment, and the detection accuracy is improved.
In the document of the 'underwater hidden sonar waveform construction method using sperm whale to call sound', chinese patent CN105044727A proposes that a plurality of groups of pulse train combinations with a pseudorandom transmission rule are traversed and screened by using sperm whale call sound signals, so as to improve the disguising and hiding capability of an active sonar device, reduce the probability of interception, however, underwater sound channels are complex and changeable, the multipath phenomenon is serious, available frequency band resources are limited, echo signals collected at a receiving end are often distorted seriously, so that the detection precision of the method is greatly influenced, and therefore channel feedback should be emphasized for complex ocean channels, so that the optimized detection waveforms gradually match the channels, and the environmental adaptability is improved.
Disclosure of Invention
The invention aims to provide a multi-address perception source feedback optimization method based on channel learning, which can improve detection accuracy.
The purpose of the invention is realized as follows:
step 1: carrying out feature extraction on collected white dolphin sound signals to obtain priori knowledge, wherein the priori knowledge comprises frequency spectrum distribution and a time-frequency structure;
step 2: performing quasi-biological signal modeling on the dolphin singing by using an over-complete atomic basis signal;
and step 3: constructing an intelligent marine organism cry signal library and outputting a multi-node biological signal model in real time;
and 4, step 4: estimating a channel by transmitting a channel training signal and utilizing an RLS algorithm to obtain a channel impulse response function;
and 5: and (4) performing convolution on output signals in the intelligent signal library and time inverse functions of the channels respectively to carry channel information, and obtaining the bionic detection waveform after feedback optimization.
The invention aims at analyzing the acoustic characteristics of real dolphin sound data to obtain priori knowledge, and selects an efficient signal processing algorithm to model the real dolphin sound data. And establishing a bionic intelligent signal library by utilizing the output of the model, and outputting a detection signal waveform applied to the underwater cluster platform in real time. After the feedback of the channel characteristics is obtained through the channel training sequence, the detection waveform is continuously optimized to gradually adapt to the channel, the distortion of the channel to the detection waveform is overcome, and the detection accuracy is improved.
The invention selects the acoustic signal of the Chinese white dolphin as an original sample for modeling, and combines the channel multipath information with the dolphin acoustic signal simulating model after transmitting the channel learning signal, thereby improving the environmental suitability.
The invention has the following beneficial effects:
the invention provides a multi-access perception source feedback optimization method based on channel learning. Different from other sonar signal waveform generation methods, the invention adds the impulse response information of the channel into the bionic detection waveform, (1) makes the other party mistakenly think that the received signal comes from the Chinese dolphin community instead of an active sonar device, plays a role of camouflage, and reduces the probability of interception by utilizing the concealment of biological signals in an underwater environment. (2) The bionic detection waveform carrying the multi-path information of the channel can be better adapted to a complex ocean channel, and when a receiving end processes echoes, the interference of the multi-path of the channel is inhibited, and more accurate focusing is realized. The waveform optimization method provided by the invention can be well applied to the combined operation of a plurality of underwater UUV, and provides a new idea for multi-node underwater cooperative detection.
Drawings
Fig. 1 is a block diagram of the overall flow of waveform feedback optimization.
FIG. 2 is a time-frequency spectrogram of six time-frequency structure biological sound signals.
FIG. 3 shows the time-frequency spectrum of the original Dolphin sound signal.
FIG. 4 is a diagram of a dolphin's sound signal time-frequency spectrum with environmental noise filtered.
FIG. 5 is a diagram of a dolphin's sound signal time-frequency spectrum with the frequency domain constant removed.
FIG. 6 is a diagram of a Dolphin sound signal time-frequency spectrum after Gaussian smooth filtering.
FIG. 7 is a time-frequency spectrum of a dolphin's sound signal after threshold filtering.
FIG. 8 is a profile curve of the fundamental frequency of the Dolphin sound signal obtained by peak extraction.
FIG. 9 is a profile curve of the fundamental frequency of the dolphin's ring-tone signal after curve fitting.
Fig. 10 is a diagram of the channel impulse response result estimated by the RLS algorithm.
Fig. 11 is a graph showing the result of matched filtering of the feedback-optimized detection waveform.
Fig. 12 is a graph showing the result of matched filtering of a probe waveform without feedback optimization.
Detailed Description
The invention is described in more detail below by way of example.
The invention discloses a multi-address perception source feedback optimization method based on channel learning, which comprises two parts of establishment of a bionic intelligent signal library and channel feedback optimization waveform:
establishing a bionic intelligent signal library:
step 1: carrying out feature extraction on collected white dolphin sound signals to obtain priori knowledge, wherein the priori knowledge comprises frequency spectrum distribution and a time-frequency structure;
in step 1, it should be mentioned that in the present invention, when the dolphin sound signal features are extracted, a short-time fourier transform is used first. By transforming the size of the window length, a suitable time resolution and frequency resolution is obtained. And then obtaining a smooth fundamental frequency contour curve through high-pass filtering, image smoothing, threshold setting and curve fitting.
Step 2: and performing quasi-biological signal modeling on the dolphin singing by using the overcomplete atomic basis signal.
In the step 2, the overcomplete atomic-based signal is utilized to model the dolphin sound signal, so that the quasi-biological signal can be fully mapped on the complete atomic-based. The operation steps of continuously performing iterative reconstruction model signals in the overcomplete atom library are as follows:
the specific process is summarized as follows:
inputting: sensing matrix phi, measurement vector y and sparsity K;
and (3) outputting: approximation of K sparsity of x
An error vector r;
initialization: the remainder r0Y, the iteration number n is 0;
(1) computing the inner product g of the residuals and each column of the sensing matrix phin＝ΦTrn-1；
(2) Find out gnThe element with the largest absolute value, i.e. k ═ arg max | gn[i]|；i∈{1,2,...,N}。
Where N is the number of columns of the matrix Φ.
(3) Calculating an approximate solution, xn[k]＝xn-1[k]+gn[k]；
(4) Updating margins
Is the kth column of the perceptual matrix Φ.
(5) Whether an iteration stop condition is met, if so, orderOutput of
r; otherwise, turning to the step (1). And step 3: and (3) constructing an intelligent marine organism cry signal library and outputting a multi-node biological signal model in real time.
The multi-node biological signal model based on the marine mammal signals constructed in the step 3 is
In the above formula ai(t) is the amplitude value of the ith harmonic of the subsignal, f0iIs the initial frequency of the ith harmonic of the sub-signal, fiAnd (t) is a function of the frequency change of the sub-signals with different time-frequency structures along with time, and generally comprises upper frequency modulation, lower frequency modulation, hyperbolic frequency modulation, sinusoidal frequency modulation and the like. R is the harmonic frequency, and M is the number of neutron signals in a section of pulse.
Channel feedback optimization waveform:
and 4, step 4: the channel is estimated by transmitting a channel training signal and utilizing an RLS algorithm to obtain a channel impulse response function.
In step 4, a whitening filter is designed, errors are made between the received signal and the transmitted signal each time, and the RLS filter is adjusted by the errors, so that the errors of the filter are gradually reduced each time. This is a fast convergence algorithm that uses the recursive least squares criterion to adjust the coefficients of the equalizer by a weighted sum of the time-mean-square errors.
The tap coefficients of the filter involved in step 4 are adjusted using RLS (recursive least squares) algorithm;
the recursive least square algorithm is to minimize the secondary performance index of the system by directly processing the received information data of the receiving end, i.e. to represent the performance index of the system by time averaging. The cost function of the RLS algorithm can be expressed as follows:
(2) in the formula, eN(t)＝d(t)-WN T(t)YN(t) is the error, λ is the forgetting factor, N is the length of the sequence, with 0 < λ < 1, d (t) is the output signal sequence of the system, Y is the error, andN(t) is an input signal sequence; minimize W from the above equationNAnd (t) is the system optimal weight coefficient. In general, the cost function J is coupled to the weight coefficient WN(t) the derivation is minimized.
The update procedure for the RLS algorithm is summarized as follows:
first of all, the autocorrelation is initializedInverse P of the matrixN(t)＝δINWherein: δ is a constant greater than 0, INIs an identity matrix;
(1) computing output
(2) Calculating error
(3) Computing kalman gain vectors
(4) Updating an inverse of a correlation matrix
(5) Updating the weight coefficients
WN(t)＝WN(t-1)+eN(t)KN(t) (7)
W in the aboveN(t-1)、PN(t-1) represents a weight coefficient WN(t) inverse P of the autocorrelation matrixN(t) the value at the previous moment.
And 5: and (4) performing convolution on output signals in the intelligent signal library and time inverse functions of the channels respectively to carry channel information, and obtaining the bionic detection waveform after feedback optimization.
After the channel impulse response function is obtained in step 5, the bionic transmitting signal is convoluted with the inverse function thereof. The preprocessed bionic detection waveform carrying the multi-path information of the channel is sent out as a transmitting signal again, and the focusing on the multi-path structure of the channel is realized.
The process of processing the transmit waveform using the time-reversal method is as follows:
(1) estimating the channel by using an RLS method to obtain the impulse response function of the channel between the ith platform and the target
(2) Using the time-reversal form of the channel obtained by the last step estimation and the transmitted signal si(t) performing convolution, wherein after the signal passes through the channel again, the signal r (t) received by the receiving platform is:
wherein n isi(t) is the noise received by the platform,
is the time-reversed version of the estimated channel.
(3) From the basic principle of the time reversal method, hi(t)*hi(-t) ═ δ (t). Therefore, the estimation of the channel is assumed to be more accurate, i.e.
Then the result received by the receiver is si(t)+ni(t)*hi(t), and ni(t) and hi(t) is not related, so ni(t)*hi(t) is negligible. It can be seen that the time reversal method achieves adaptive focusing.
The following describes in detail a multiple access sensing source feedback optimization method based on channel learning and its beneficial effects, with reference to specific embodiments.
Fig. 1 is a system design block diagram of a multiple access sensing source feedback optimization method based on channel learning according to the present invention.
Fig. 2 shows a time-frequency spectrum of whistle sounds of white dolphin collected from many sea areas in south China sea in this subject group. From the figure, the fundamental frequency energy of the dolphin sound signal is concentrated in the range of 3-10kHz, belongs to a medium-low frequency signal, and is suitable for medium-long distance detection. Moreover, the time-frequency structure is rich and diverse, the discrimination and the recognition are easy, and the echo separation and the feature extraction can be well carried out in the underwater multi-platform cooperative detection process.
The most important step of the method is to obtain a time-frequency contour curve of the fundamental frequency of the dolphin's ring-tone signal, so as to prepare for the subsequent modeling of the bionic signal. It is necessary to make a detailed explanation of the time-frequency feature extraction of the dolphin's sound with reference to the drawings.
The first step is as follows: high pass filter for processing low frequency noise
Fig. 3 shows the time-frequency spectrum of the collected original dolphin sound signal. We find that the dolphin's sound signal appears very weak due to the strong energy of the ship noise in the test sea area. Although the energy of the ship noise is strong, the frequency is very low and is basically below 1kHz, and a dolphin signal without low-frequency noise interference can be obtained after high-pass filtering processing. As shown in fig. 4, after the ambient noise is filtered, the discernability of the dolphin's sound signal is improved.
The second step is that: time-frequency image smoothing processing based on Gaussian operator
If y isfIs the power spectral energy value at frequency f, and lyfIs its spectral energy value in decibels, i.e./yf＝10log10(yf) Then can pass through btf＝α·lytf+(1-α)·bt-1，fCalculating the background mean b of each time and frequencytf. Where α is a smaller value (default 0.02). The results are given below:
l′ytf＝lytf-bt-1，f(9)
spike like constants can be removed from the spectra. The result of removing the frequency domain constant is shown in fig. 5.
Then, gaussian matrix smoothing is performed on the above results. The following formula:
l″ytf＝l′ytf*G (10)
wherein the content of the first and second substances,
is a gaussian smoothing matrix. The results are shown in FIG. 6.
The effect of removing the frequency domain constant in fig. 5 is obvious, and after gaussian smoothing is performed on fig. 6, a large number of smaller connected domains can be removed, so that the dolphin sound signal with a weak amplitude of about 15kHZ becomes more stable.
The third step: time-frequency contour for setting threshold value to extract fundamental frequency
And setting a threshold value for the smoothed time-frequency spectrum result. Data above the threshold was retained and data points below the threshold (default 130dB) were set to a constant value (default-800 dB). The results are shown in FIG. 7:
for extracting the fundamental frequency, a peak value extraction method is adopted, and the expression is as follows:
f1(m)＝max|Xm[k]| (11)
wherein f is1(m) is the frequency value of the fundamental frequency, Xm[k]Is the amount of energy in the time spectrum. The results are shown in FIG. 8.
The fourth step: and obtaining the instantaneous frequency of the fundamental wave through curve fitting.
The least squares fit of the results of fig. 8 is shown in fig. 9. The straight line in the figure is the original contour before fitting and the "+" line is the curve after fitting. The end result is the instantaneous frequency curve that most closely approximates the fundamental frequency-time profile.
The beneficial effect that the detection waveform after feedback optimization adapts to an underwater complex channel is researched through simulation analysis.
Simulation conditions are as follows: a receiving and transmitting distance: 1000 m; transmitting transducer depth: 20 m; receiving transducer depth: 30 m. Firstly, before the bionic multi-source signal is transmitted, several groups of channel training signals-LFM signals are transmitted in advance. At the receiving end, the result of channel impulse response obtained by RLS algorithm estimation is shown in fig. 10. The time reversal output of the estimated channel in fig. 10 is convolved with the bionic detection waveform to obtain a new detection waveform, the new detection waveform is sent out again through the transmitter, and the matching filtering processing is performed at the receiving end, and the result is shown in fig. 11. If the detection waveform is directly passed through the ocean channel without channel feedback optimization, the matched filtering result is shown in figure 12.
And (3) simulation results: fig. 10 is an impulse response result diagram of an underwater multi-path channel obtained by simulation. From the results in the figure, we can obtain that the channel multipath is severe, and the underwater sound field environment represented by the channel multipath is complex. Comparing fig. 11 and 12, we can find that the detection waveform without feedback optimization generates serious mismatch after being matched and filtered. The results shown in fig. 11 indicate that the detection waveform optimized by the channel feedback proposed by the method can realize focusing on multiple channels.
Claims (3)
1. A multi-address perception source feedback optimization method based on channel learning is characterized by comprising the following steps:
step 1: carrying out feature extraction on collected white dolphin sound signals to obtain priori knowledge, wherein the priori knowledge comprises frequency spectrum distribution and a time-frequency structure;
step 2: performing quasi-biological signal modeling on the dolphin singing by using an over-complete atomic basis signal;
and step 3: constructing an intelligent marine organism cry signal library and outputting a multi-node biological signal model in real time;
and 4, step 4: estimating a channel by transmitting a channel training signal and utilizing an RLS algorithm to obtain a channel impulse response function;
and 5: output signals in the intelligent signal library are respectively convolved with time inverse functions of channels, carry channel information and obtain bionic detection waveforms after feedback optimization;
the biological signal modeling of the dolphin singing by using the overcomplete atomic-based signals specifically comprises the following steps:
inputting: sensing matrix phi, measurement vector y and sparsity K;
and (3) outputting: approximation of K sparsity of x
An error vector r;
initialization: the remainder r0Y, the iteration number n is 0;
(1) computing the inner product g of the residuals and each column of the sensing matrix phin＝ΦTrn-1；
(2) Find out gnThe element with the largest absolute value, i.e. k ═ arg max | gn[i]|；i∈{1,2,...,N}，
Where N is the number of columns of the matrix Φ;
(3) calculating an approximate solution, xn[k]＝xn-1[k]+gn[k]；
(4) Updating margins
Is the kth column of the sensing matrix Φ;
(5) whether an iteration stop condition is met, if so, orderr＝rnOutput of
r; otherwise, turning to the step (1); constructing an intelligent marine organism cry signal library, and constructing six intelligent marine mammal cry signal libraries; the multi-node biological signal model comprises the following steps:
wherein: a isi(t) is the amplitude value of the ith harmonic of the subsignal, f0iIs the initial frequency of the ith harmonic of the sub-signal, fi(t) is a function of the frequency change of different time-frequency structures of the sub-signals along with time, R is the harmonic frequency, and M is the number of the sub-signals in a section of pulse.
2. The method as claimed in claim 1, wherein the estimating the channel by using the RLS algorithm specifically comprises:
first, a cost function of the RLS algorithm is obtained, which is expressed as follows:
in the formula, eN(t)＝d(t)-WN T(t)YN(t) is the error, λ is the forgetting factor, N is the length of the sequence, 0 < λ < 1, d (t) is the output signal sequence of the system, Y is the error, andN(t) is an input signal sequence; minimize W from the above equationN(t) is the optimal weight coefficient of the system, and the cost function J is the weight coefficient WN(t) minimizing the derivative;
the inverse P of the initialized autocorrelation matrix is then updatedN(t)＝δINWherein: δ is a constant greater than 0, INIs an identity matrix;
(1) computing output
(2) Calculating error
(3) Computing kalman gain vectors
(4) Updating an inverse of a correlation matrix
(5) Updating the weight coefficients
WN(t)＝WN(t-1)+eN(t)KN(t)；
W in the aboveN(t-1)、PN(t-1) represents a weight coefficient WN(t) inverse P of the autocorrelation matrixN(t) the value at the previous moment.
3. The method of claim 2, wherein the inverse form of the transmitted waveform and the channel time is convolved, and the process is as follows:
(1) estimating the channel by using an RLS method to obtain the impulse response function of the channel between the ith platform and the target
(2) Using the time-reversal form of the channel obtained by the last step estimation and the transmitted signal si(t) performing convolution, wherein after the signal passes through the channel again, the signal r (t) received by the receiving platform is:
wherein n isi(t) is the noise received by the platform,
is the time-reversed form of the estimated channel;
(3) obtained by the basic principle of the time reversal method, hi(t)*hiδ (t), so the estimation of the channel is assumed to be accurate, i.e. to be accurate
Then the result received by the receiver is si(t)+ni(t)*hi(t), and ni(t) and hi(t) is not related, so ni(t)*hi(t) is ignored.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710832980.5A CN107635181B (en) | 2017-09-15 | 2017-09-15 | Multi-address sensing source feedback optimization method based on channel learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710832980.5A CN107635181B (en) | 2017-09-15 | 2017-09-15 | Multi-address sensing source feedback optimization method based on channel learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107635181A CN107635181A (en) | 2018-01-26 |
| CN107635181B true CN107635181B (en) | 2020-01-17 |
Family
ID=61101371
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710832980.5A Active CN107635181B (en) | 2017-09-15 | 2017-09-15 | Multi-address sensing source feedback optimization method based on channel learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107635181B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110008630B (en) * | 2019-04-18 | 2022-07-15 | 哈尔滨工程大学 | Underwater unmanned vehicle cluster simulation platform |
| CN110568439B (en) * | 2019-09-25 | 2023-08-22 | 中南大学 | Deconvolution-based impulse type through-wall radar antenna ringing suppression method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2215305C2 (en) * | 2001-07-05 | 2003-10-27 | Федеральное государственное унитарное предприятие "Центральный научно-исследовательский институт "Морфизприбор" | Method disguising underwater noise of fishing craft and disguising facility |
| CN105044725A (en) * | 2015-07-27 | 2015-11-11 | 天津大学 | Underwater target hidden distance measurement method and device by utilizing sound emitted from sperm whale |
| CN105391500A (en) * | 2015-10-13 | 2016-03-09 | 哈尔滨工程大学 | Dolphin click simulating underwater acoustic communication method based on ultra-wideband signal |
| CN106653032A (en) * | 2016-11-23 | 2017-05-10 | 福州大学 | Animal sound detecting method based on multiband energy distribution in low signal-to-noise-ratio environment |
-
2017
- 2017-09-15 CN CN201710832980.5A patent/CN107635181B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2215305C2 (en) * | 2001-07-05 | 2003-10-27 | Федеральное государственное унитарное предприятие "Центральный научно-исследовательский институт "Морфизприбор" | Method disguising underwater noise of fishing craft and disguising facility |
| CN105044725A (en) * | 2015-07-27 | 2015-11-11 | 天津大学 | Underwater target hidden distance measurement method and device by utilizing sound emitted from sperm whale |
| CN105391500A (en) * | 2015-10-13 | 2016-03-09 | 哈尔滨工程大学 | Dolphin click simulating underwater acoustic communication method based on ultra-wideband signal |
| CN106653032A (en) * | 2016-11-23 | 2017-05-10 | 福州大学 | Animal sound detecting method based on multiband energy distribution in low signal-to-noise-ratio environment |
Non-Patent Citations (2)
| Title |
|---|
| 一种利用海豚叫声的仿生水声通信方法;刘凇佐，乔钢，尹艳玲;《物理学报》;20130528;144303-1-144303-7页 * |
| 仿海豚叫声水声通信技术研究;刘冰洁;《万方学位论文》;20160505;第22页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107635181A (en) | 2018-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110867181B (en) | Multi-target speech enhancement method based on SCNN and TCNN joint estimation | |
| CN106597408B (en) | High-order PPS signal parameter estimation method based on time-frequency analysis and instantaneous frequency curve fitting | |
| CN108768543B (en) | Multi-feature fusion cognitive underwater acoustic communication space-time and fast-time adaptive processing algorithm | |
| CN110109058A (en) | A kind of planar array deconvolution identification of sound source method | |
| CN112466320B (en) | Underwater sound signal noise reduction method based on generation countermeasure network | |
| CN107635181B (en) | Multi-address sensing source feedback optimization method based on channel learning | |
| CN108091345A (en) | A kind of ears speech separating method based on support vector machines | |
| CN110808057A (en) | Voice enhancement method for generating confrontation network based on constraint naive | |
| CN111816200B (en) | Multi-channel speech enhancement method based on time-frequency domain binary mask | |
| Jiang et al. | A sonar-embedded disguised communication strategy by combining sonar waveforms and whale call pulses for underwater sensor platforms | |
| CN113780521B (en) | Radiation source individual identification method based on deep learning | |
| Liu et al. | Few-shot learning with data enhancement and transfer learning for underwater target recognition | |
| CN105429720B (en) | The Time Delay Estimation Based reconstructed based on EMD | |
| CN104665875A (en) | Ultrasonic Doppler envelope and heart rate detection method | |
| CN108564962A (en) | Unmanned plane voice signal Enhancement Method based on tetrahedron microphone array | |
| CN113822363B (en) | Convolutional neural network time delay estimation method based on quadratic rational kernel function | |
| CN116449350A (en) | Reverberation suppression method, system and storage medium based on self-supervision learning | |
| CN115932808A (en) | Passive sonar intelligent detection method based on multi-feature fusion | |
| CN109658944A (en) | Helicopter acoustic signal Enhancement Method and device | |
| Oliinyk et al. | Center weighted median filter application to time delay estimation in non-Gaussian noise environment | |
| Kim et al. | HMM-based underwater target classification with synthesized active sonar signals | |
| CN113176541B (en) | Self-adaptive anti-spectrum dispersion interference method and system | |
| CN112713907B (en) | Marine CSEM noise reduction method and system based on dictionary learning | |
| CN114330441A (en) | Underwater sound JANUS signal identification method and system based on time-frequency spectrum and transfer learning | |
| CN112784412A (en) | Single hydrophone normal wave modal separation method and system based on compressed sensing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |