107 results about "Blind source separation algorithm" patented technology

A system for separation of a user's voice from ambient sound, comprising a device to be worn at the user's ear or at least partly in the user's ear canal comprising a first microphone oriented acoustically outwardly towards the environment and a second microphone oriented acoustically inwardly towards the user's ear canal, and an audio signal processing unit for processing audio signals from the first and second microphone by a blind source separation algorithm adapted to separate the user's voice from ambient sound.

The invention provides a fetus electrocardio blind separation method based on the relative sparsity of time domains of source signals, comprising the following steps of: firstly acquiring mutually overlaid mother-fetus mixed electrocardiosignals of mother and fetus electrocardio in two paths from the body surface of the abdomen of a mother; then carrying out pretreatment on the acquired mother-fetus mixed electrocardiosignals, wherein the pretreatment includes the steps of: correcting the baseline drift of the mother-fetus mixed electrocardiosignals, and filtering the interference of power frequency of 50 Hz, the interference of high-frequency myoelectric signals, and the like; respectively positioning the mother electrocardiosignal and the fetus electrocardiosignal in the pretreated mother-fetus mixed electrocardiosignals; then searching relatively sparse time slices of the mother electrocardio and the fetus electrocardio; and finally separating the mother electrocardiosignal from the fetus electrocardiosignal by utilizing a basic blind source separation algorithm resolved through the general features of a matrix. The invention solves the problem that the time domains and frequency domains of the mother electrocardiosignal and the fetus electrocardiosignal are mutually overlaid and difficult to separate and can efficiently and accurately extract the fetus electrocardiosignal used for medical diagnosis.

Signals are received from audio pickup channels that contain signals from multiple sound sources. The audio pickup channels may include one or more microphones and one or more accelerometers. Signals representative of multiple sound sources are generated using a blind source separation algorithm. It is then determined which of those signals is deemed to be a voice signal and which is deemed to be a noise signal. The output noise signal may be scaled to match a level of the output voice signal, and a clean speech signal is generated based on the output voice signal and the scaled noise signal. Other aspects are described.

The invention provides a method and a system for performing sound source separation on a sound signal acquired by a microphone array. The method comprises the steps of receiving the sound signal and detecting the number n of the sound sources in the sound signal; determining whether n is larger than one, if yes, calculating a direction-of-arrival angle value of each sound source; ordering the direction-of-arrival angle values according to a magnitude sequence, and calculating a difference between two adjacent direction-of-arrival angle values; determining whether all angle intervals is largerthan or equal with a preset angle interval threshold C, if yes, executing a next step, and otherwise, correcting two or more direction-of-arrival angle values, and executing a next step; performing wave beam forming based on the direction-of-arrival angle value and/or the corrected direction-of-arrival angle value for obtaining n enhanced voice signals, and executing a blind source separation algorithm on n enhanced voice signals, and n separated sound signals which correspond with n sound sources. The method and the system according to the invention can realize better sound source separationperformance.

The invention discloses a bridge inhaul cable group cable force synchronous monitoring method and system based on a microwave radar. The system is mainly composed of the microwave radar equipment anda PC, wherein the microwave radar equipment internally emits a microwave signal to a bridge inhaul cable group and receives a reflected echo signal of the inhaul cable group, the echo signal is converted into an inhaul cable group micro time-process signal through a data processor, a display unit of the PC displays the inhaul cable group micro time-process signal, a time-process data analysis unitreads the inhaul cable group micro time-process signal, time-process signals of multiple cables in each measurement unit are separated through an embedded underdetermined time-process signal blind source separation algorithm fusing VMD and time-frequency analysis, and a VMD algorithm is adopted to recognize time-varying cable force of each cable in the inhaul cable group. Through the synchronousmonitoring method and system, an inhaul cable group time-varying cable force measurement solution which is efficient, accurate, high in stability and high in adaptability is provided, the problems that measuring point selection is difficult and test efficiency is low during inhaul cable group cable force measurement are effectively solved, the requirement for evaluation of cable force of a broad range of cable-stayed bridges and suspension bridges is met, and the synchronous monitoring method and system have good engineering application prospects.

The invention relates to a blind image separation method based on frequency-domain sparse component analysis. In the currently-provided blind source separation algorithm, an independent component analysis method with better separation effect has a blind source separation premise that source signals are not in Gaussian distribution, are mutually independent, and can not thoroughly separate sub-Gaussian signals in image signals. The sparse component analysis is a novel blind source separation technology developed in recent years, by applying the technology, source signals are extracted by utilizing the sparse properties of the signals and the better separation effect is obtained. The image signals which do not satisfy sparse conditions can not be separated by applying a traditional sparse component analysis model. In the invention, the images are converted into the frequency domain from the space domain by combining the characteristic that the images are sparse in the frequency domain space and utilizing sparseness algorithms, such as wavelet transform, and the like; a sparse component analysis model is educed in the frequency domain; and a hybrid matrix estimation method and a source signal estimation method based on linear-clustering sparse component analysis are provided; therefore, the source images are extracted. Experiments prove that the method of the invention has the separation precision up to 100 percent and is superior to other separation methods.

The invention relates to railway frequency shift signal anti-interference device and method based on blind source separation. The device comprises a two-channel collection and AD conversion module, a time sequence control module and a digital signal processing module used for executing anti-interference method treatment, wherein the two-channel collection and AD conversion module collects railwaytrack signals and converts the railway track signals to digital signals, and the time sequence control module is respectively connected with the two-channel collection and AD conversion module and the digital signal processing module. The invention can eliminate interference of amplitude modulation signals which are in aliasing with a frequency shift signal time-frequency domain to enhance the signal to noise ratio of locomotive signal detection, utilize a blind source separation algorithm based on geometric transformation to separate out amplitude modulation interference signals or neighboring line interference signals which are in aliasing in frequency shift signals and utilize a wavelet threshold denoising method to lower white noise interference contained in the signals before blind source separation, thereby strengthening the stability of the blind source separation algorithm.

The embodiment of the invention provides a vehicle-mounted voice recognition method. The method comprises the following steps: carrying out voice area signal separation on vehicle-mounted space by using a plurality of microphones, so that the vehicle-mounted space is at least separated into a plurality of voice areas, and forming a distributed microphone network according to the microphones in thevoice areas; collecting audio under the vehicle-mounted environment in real time through the distributed microphone network, and inhibiting vehicle-mounted environment self noise in the audio according to the echo cancellation algorithm; separating voices of a plurality of speakers in the inhibited audio through a blind source separation algorithm; and locating the voice area of each speaker through the distributed microphone network, when no voice made by the speakers is available, collecting the voice as noise estimation, when voice made by the speakers is available, carrying out denoisingaccording to the noise estimation collected by a previous frame, determining clean voice, and carrying out voice recognition. The embodiment of the invention further provides a vehicle-mounted voice recognition system. According to the embodiment of the invention, in the vehicle-mounted noise environment, the relatively high awakening rate and recognition rate are available.

The invention provides a digital signal processor (DSP) implementation system for a two-channel convolution mixed voice signal blind source separation algorithm. The system comprises an embedded processor platform, a microphone access circuit, a power circuit and a signal pre-amplification circuit, wherein the embedded processor platform is used for finishing analogue/digital (A/D) conversion, D/A conversion and analogue signal output; the microphone access circuit comprises two single-channel microphones for acquiring voice signals and a two-channel microphone acquisition circuit; two-channel voice signals acquired by the two single-channel microphones are taken as the left and right channel signal input of a line in audio interface; the signal pre-amplification circuit effectively amplifies the signals of the microphones until gain required by the line in audio interface is achieved. A plurality of functions of microphone acquisition, pre-amplification, voice blind source separation, voice play and storage and the like of the two-channel voice signals of two speakers are realized, and the actually acquired mixed voices of the two speakers can be effectively separated under both an offline condition and an online condition.

The invention discloses a composite fault diagnosis method for a hydraulic AGC system. The method specifically comprises the following steps: S1, mill component dynamic modeling: acquiring the equivalent of each link of a hydraulic AGC system to get a mathematical model thereof; S2, system modeling: simulating a fault, selecting a measuring point, building a multi-signal model of the system, and getting a fault dependence matrix under the multi-signal model; and if the fault is a single fault, getting a fault dependence matrix directly, performing fault diagnosis, and getting a result directly; S3, if the fault is a composite fault, using a blind source separation algorithm to blindly separate measuring point information, getting a fault dependence matrix, and diagnosing the composite fault; and S4, fault diagnosis: after the system fails, getting the information of the measuring point, and locating the fault source by comparing the measuring point information with the fault dependence matrix. Through the diagnosis method of the invention, the purpose of composite fault isolation and location is achieved, and the fault separation rate is improved.

The invention discloses a null space differential operator and blind source separation based bearing combined fault diagnosis method and belongs to the technical field of bearing fault diagnosis. By means of null space differential operators based on bearing fault features, bearing fault signals (random vibration, fault shock components and noises caused by rotation of a normal part of a bearing) are resolved into a series of local narrow-band signals (comprising components of fault features); subsequently, the obtained narrow-band signals and bearing signals are regarded as one group of observation signals; by means of a blind source separation algorithm, the observation signals are subjected to blind source separation to achieve separation of combined faults of the bearing; finally, the signals obtained from separation are subjected to Hilbert demodulation processing, and afterwards, fault features of the bearing are extracted to ultimately achieve combined fault diagnosis of the bearing. In new observation signals, the observation signals are more than source signals, such that a premise required by the blind source separation algorithm can be satisfied, and further, combined fault separation and feature extraction of the bearing are achieved.

Disclosed is a hardware architecture receiving multi-input blind source signals and obtaining multi-output. An apparatus for separating blind source signals includes: a forward process unit receiving a plurality of blind source signal vectors and outputting a plurality of output signal vectors by using a predetermined blind source separation algorithm; an update process unit receiving the plurality of output signal vectors and learning first weighting values used for the predetermined blind source separation algorithm according to a predetermined learning algorithm; and a weight process unit having a matrix operation structure for receiving the first weighting values and converting them into coefficients and second weighting values applicable to the predetermined blind source separation algorithm. The forward process unit includes (L+1) identical processing elements connected in a systolic array structure, where L is the number of sequential delay of blind input signal vectors. The update process unit includes (N²+N)/2×(2L+1) identical updating elements connected in a systolic array structure, where N is the number of the blind input signal vectors. Each cost of the processing elements and the updating elements is initialized by 0 in an initial operation stage.

The invention is applicable to the technical field of transformer signal analysis, and provides a transformer vibration signal separation method and terminal device. The method comprises the followingsteps of acquiring the vibration signal of the transformer; obtaining the core vibration signal and/or winding vibration signal of transformer by processing the vibration signal of transformer with the preset separation model. According to the transformer vibration signal separation method according to an embodiment of the present invention, the preset separation model is used to classify the vibration signal of transformer, so that the separation of core vibration signal and winding vibration signal is realized. Compared with the traditional blind source separation algorithm, the separationresults of this method are significantly improved in reliability and accuracy.

The invention discloses a leakage acoustic wave feature extraction method based on fusion of wavelet transform and the blind source separation algorithm. The method comprises the following steps that an acoustic wave sensor is used for collecting leakage acoustic wave signals and acquiring leakage acoustic wave collection signals; multilayer wavelet decomposition is carried out on the leakage acoustic wave collection signals through wavelet transform, corresponding approximate signals are sequentially obtained through all layers of wavelet decomposition, the leakage acoustic wave collection signals and the approximate signals are used as observation signals, and the observation signals are processed through the blind source separation algorithm to obtain target signals; and the target signals obtained in the second step are evaluated, and compositions of the observation signals are optimized. The method has the beneficial effects that the target signals are evaluated through two evaluation parameters, namely leakage moment sampling point deviation and amplitude loss; by means of the method, the leakage moment can be accurately determined, and meanwhile the compensation effect on the leakage amplitude of weak signals is obvious.

The invention discloses a multi-feature fusion cognitive underwater acoustic communication space-fast time adaptive processing algorithm. The algorithm includes the following steps: introducing an interference feature analysis-based interference cognitive function on the basis of a traditional space-fast time adaptive array processing algorithm; calculating the number of interference sources by using an array covariance matrix feature decomposition mode, and estimating the direction of a target user and the direction of interference incoming waves by adopting a MUSIC spatial spectrum analysismode; separating interference signals by using a negentropy maximized Fast-ICA based blind source separation algorithm, analyzing time domain features by using an envelope detection mode, analyzing the time-frequency spectrum by using a short-time Fourier transform mode, and further extracting interference features to identify interference types; and finally, selecting appropriate interference samples to perform space-fast time adaptive cancellation according to the interference types, and further improving the convergence speed of the adaptive algorithm. The beneficial effects are that: the dimensionality reduction processing and cancellation sample selection of the space-fast time adaptive algorithm can be realized, and the fast and reliable convergence of the space-fast time adaptive processing can be guaranteed.

The invention belongs to the technical field of blind source separation and specifically relates to an improved natural gradient variable step-size blind source separation algorithm. The algorithm mainly comprises the following steps of randomly generating a hybrid matrix, carrying out linear mixing on source signals, and obtaining three hybrid signals; initializing a separation matrix, an autocorrelation matrix and a learning step, adding an orthogonal constraint, and controlling a variable step size by instantaneous errors and carrying out iteration; calculating an estimation signal; and drawing a recovered source signal waveform according to the estimation signal. The invention provides the natural gradient blind source separation algorithm based on the orthogonal constraint, the algorithm constrains the strength of recovered signals, and the stability of a separation process under an unstable environment is ensured; in addition, the convergence is accelerated, the separation precision is improved, the signal amplitude under the unstable environment constantly changes, the change frequency is relatively high, and a conventional separation method can not perform the separation, so that the related technology of the invention is advanced and prospective.

The invention discloses an inertial sensor aliasing interference signal separation method. The method comprises the following steps that (1) characteristic analysis is conducted on inertial sensor output aliasing signals, and the frequency characteristics of different interference source signals are determined; (2) a threshold function, a wavelet basis and a decomposition level are selected to conduct wavelet filtering on a low frequency section signal in the aliasing signals according to the characteristics of the low frequency section signal; (3) aiming at the high frequency aliasing interference signal subjected to wavelet filtering, N internal empirical mode functions (IMFs) are acquired through self-adapting decomposition of empirical mode decomposition (EEMD); (4) the information content of IMFs is eta, and the number p of the needed IMFs components is calculated and obtained when the eta is larger than or equal to 90 degrees; (5) A PCA algorithm is utilized for dropping N-dimensional IMFs to be P-dimensional; and (6) P-dimensional IMFs is taken as an observation signal, and an appropriate blind source separation algorithm is selected to separate interference source signals at different frequency sections in the high frequency aliasing interference signal. The inertial sensor aliasing interference signal separation method greatly reduces too much man-made experience selections in the signal separation process, and the adaptivity is relatively high.

The invention provides a two-order oscillation particle swarm blind source separation method based on heritable variation optimization and belongs to the technical field of blind signal processing. The method solves the problem that in a traditional blind source separation algorithm, a nonlinear activation function is difficult to select, and mixed signals can be effectively separated under the premise that source signal priori knowledge is not available. Separation signal negative entropy serves as a target function, and the local searching capability and the global searching capability are balanced with fixed inertia weights; population diversity can be maintained under the condition that the number of particles is unchanged by introducing the study factor two-order oscillation link; a heritable variable mechanism is introduced, and thus the situation of convergence rate reduction caused by introduction of two-order oscillation can be easily relieved. It is shown through separation of simulation vibration signals and chaotic mapping signals that the method can be applied to the mechanical signal fault detection field and aspects such as determined noise-like signal processing. By means of the method, the improved type theoretical study of intelligent algorithm blind source separation is supplemented, and great significance is achieved on separation of unknown mixed signals in engineering application.

The embodiment of the invention provides a data noise reduction method and system for bridge structure monitoring. The method comprises the following steps: acquiring an original observation signal; performing pole symmetry modal decomposition on the original observation signal to obtain a decomposed intrinsic mode component and a residual component; separating the decomposed intrinsic mode component and the residual component by adopting a blind source separation algorithm to obtain a separation signal; performing frequency domain conversion on the separation signal to obtain a frequency domain conversion result, and obtaining a noise component according to the noise frequency in the frequency domain conversion result; removing a noise component, and performing reverse reconstruction on the decomposed intrinsic mode component and the residual component to obtain a reconstructed signal component; calculating a Spearman coefficient for decomposing the intrinsic mode component and the original observation signal, and determining a preset threshold; and accumulating the reconstructed signal components according to a preset threshold to obtain signal data without noise information. According to the embodiment of the invention, modal decomposition and blind source separation are combined, and decomposition, denoising and reconstruction of the bridge monitoring data are effectively realized.

The invention requests to protect an electroencephalogram classification signal method based on a Gaussian Bernoulli convolution deep belief network. According to the method, a blind source separationalgorithm based on the maximum negative entropy is adopted in the preprocessing stage to remove signal interference of motor imagery electroencephalogram; Selecting frequency and electrode parametersbased on the mutual information; combining the unsupervised learning of the Gaussian Bernoulli restricted Boltzmann machine with the convolutional neural network to carry out feature extraction and classification; A new convolutional deep belief network model based on a Gaussian Bernoulli restricted Boltzmann machine can extract meaningful characteristics from a full-size image through a generated convolutional filter. Compared with the prior art, the method has the advantages that a large number of negative weights are reduced, spatial information can be learned from adjacent image patches more effectively, the accuracy of electroencephalogram category judgment is obviously improved, and the accuracy of electroencephalogram category judgment is greatly improved.

Disclosed is a blind source separation method for mooring rope vibration signals. The blind source separation method comprises the following steps of step 1, selecting any position point on the mooring rope to obtain vibration signals; step 2, carrying out mean-value preprocessing on the collected vibration signals, and performing de-meaning to change observation signals into zero-mean-value vectors; step 3, carrying out whitening treatment on the observation signal data subjected to the de-meaning, so as to remove the correlation between the observation signals; step 4, performing analysis onthe observation signals after the whitening treatment through a blind source separation algorithm to obtain estimation of the mixed matrix and the source signals; and step 5, analyzing the modal response and the vibration type matrix of the mooring rope vibration system according to the estimation of the mixed matrix and the source signals; and adopting a single-mode recognition technology to separate out frequency and damping ratio from the source signal estimation; and ranking the separated signals according to the values, so that the frequency vector, the damping ratio vector and the vibration type matrix of the mooring rope vibration system are obtained. The system parameter identification accuracy is high.

The invention discloses a method and a device for improving recording quality based on blind source separation algorithm. The method includes: performing analog-to-digital conversion on the external voice signals collected by a plurality of microphones to obtain several channels of digital voice signals; according to the blind source separation algorithm, separating the background noise and the reference audio signal for each channel of digital voice signals and filtering the background noise; and packaging and saving each channel of reference audio signal after the each channel of reference audio signal is mixed. According to the invention, as the blind source separation algorithm requires very little on the performance and structural design of the microphones and does not require multiple auxiliary microphones to collect multiple ambient sound signals to meet the analytical requirements of the algorithm, the robustness is very good, thus solving the problem that the existing recording modes have higher requirements on the position and layout of the microphones but the robustness is poor.

The invention discloses an electric vehicle charging recognition method based on an intelligent electric meter. The method comprises the following steps: obtaining electrical parameters required by electric appliance recognition; converting the acquired single-channel current waveform data into a multi-channel to-be-separated characteristic matrix; decomposing the to-be-separated characteristic matrix into characteristic matrixes of a plurality of independent electric appliances by adopting a blind source separation algorithm; and judging whether the power consumption behavior of the electricvehicle exists or not according to the feature matrixes in all the independent electric appliances, If yes, sending alarm information. The invention further provides electronic equipment and a computer readable storage medium. According to the electric vehicle charging recognition method based on the intelligent electric meter, waveform data generated by all electric appliances in the power utilization process are collected through the intelligent electric meter, all the waveform data are analyzed to judge whether an electric vehicle is currently charged or not, and therefore electric vehiclecharging recognition is achieved; according to the method, the electric appliance recognition cost and the implementation difficulty can be reduced, and the detection efficiency is improved.