78 results about "Matrix group" patented technology

A low density parity check (LDPC) code that is particularly well adapted for hardware implementation of a belief propagation decoder circuit is disclosed. The LDPC code is arranged as a macro matrix (H) whose rows and columns represent block columns and block rows of a corresponding parity check matrix (H_pc). Each non-zero entry corresponds to a permutation matrix, such as a cyclically shifted identity matrix, with the shift corresponding to the position of the permutation matrix entry in the macro matrix. The block columns of the macro matrix are grouped, so that only one column in the macro matrix group contributes to the parity check sum in any given row. The decoder circuitry includes a parity check value estimate memory which may be arranged in banks that can be logically connected in various data widths and depths. A parallel adder generates extrinsic estimates that are applied to parity check update circuitry for generating new parity check value estimates. These parity check value estimates are stored back into the memory, and are forwarded to bit update circuits for updating of probability values for the input nodes. Variations including parallelism, time-sequencing of ultrawide parity check rows, and pairing of circuitry to handle ultrawide code rows, are also disclosed.

An apparatus for generating a precoding matrix codebook includes a matrix group generator to generate a first group of unitary matrices based on a vector-based codebook for precoding of a multiple input multiple output (MIMO) communication scheme; a matrix group extender to extend the first group of unitary matrices to generate a second group of unitary matrices; and a matrix group selector to select a group of columns corresponding to a communication rank from each of the unitary matrices that are elements of the second group so that the columns are optimized based on a distance between the columns corresponding to the communication rank.

The invention discloses a multi-kernel-learning-based multi-mode emotion identification method. According to the method, extraction of emotion features like an expression feature, a voice feature and a physiological feature is carried out on sample data of each mode in a multi-mode emotion database; several different kernel matrixes are constructed for each mode respectively; the kernel matrix groups corresponding to different modes are fused to obtain a fused multi-mode emotion feature; and a multi-kernel support vector machine is used as a classifier to carry out training and identification. Therefore, basic emotions like angering, disgusting, fearing, delighting, upsetting, and surprising and the like can be identified effectively.

The embodiment of the invention discloses a hybrid beam forming transmission method and a network device. According to the invention, a network device determines a digital-domain beam forming weight matrix for a first terminal according to a channel measurement result; on the basis of the digital-domain beam forming weight matrix, a group of beam forming weight matrixes is determined; with a simulated forming method, channel measurement is carried out by the beam forming weight matrix group; a beam forming weight matrix is selected from the beam forming weight matrix group and is used as a simulation-domain beam forming weight matrix of the first terminal; and the network device carries out hybrid beam forming transmission based on the determined digital-domain beam forming weight matrix and the selected simulation-domain beam forming weight matrix. Therefore, the digital-domain beam forming weight matrix and the simulation-domain beam forming weight matrix can be determined for the terminal so as to carry out digital-analog hybrid beam forming transmission with the terminal conveniently.

A method for transmitting and receiving uplink signals using an optimized rank 3 codebook is disclosed. The optimized rank 3 codebook includes 12 precoding matrix groups, which are consisted of 6 Tx antenna power balanced precoding matrix groups and 6 layer power balanced precoding matrix groups. Preferably, the optimized 4Tx rank 3 codebook has 20 precoding matrix, two precoding matrixes are selected from each the 6 Tx antenna power balanced precoding matrix groups, considering chordal distance and the number of precoding matrix. And then 8 precoding matrices are selected from the layer balanced precoding matrix groups.

The invention discloses a non-orthogonal joint diagonalization instantaneous blind source separation method based on double iteration. The method comprises the steps that signal data of repeated observation are recorded; a target matrix group based on observation signals is calculated; an observation signal aliasing model is divided into left and right aliasing matrixes; a blind source separation cost function is established; under the condition of equal left and right aliasing matrix nature, the cost function is reconstructed into a symmetric fitting function; a double iteration method is used to estimate left and right aliasing matrix optimal values required by non-orthogonal diagonalization; and an optimal aliasing model is derived to acquire an isolation signal, and instantaneous aliasing blind separation is realized. High demands of the existing non-orthogonal JD method on a target matrix are avoided, and the application scope of the observation signals is expanded. Compared with the prior art, the method provided by the invention has the advantages that the calculation complexity and the iteration number are significantly reduced; the least time is consumed; and when the same SNR is input, the shortest convergence time, the maximum output SINR and the minimum generalized replacement level are realized, and better separation performance and convergence are realized.

The invention discloses an urban stagger traffic effect analysis method based on RFID data. The method comprises the following steps: acquiring original vehicle running track data from original RFID data; performing OD matrix slicing based on the vehicle running track data; extracting and moveably carrying out a target OD traffic volume of a stagger traffic measure based on an actual stagger policy to generate an initial stagger OD matrix group; iteratively calculating the initial stagger OD matrix group to obtain a predicted stagger OD matrix group; and obtaining a final OD matrix group basedon the iterative result to achieve an effect assessment of the stagger policy. The method provided by the invention can simulate the OD change of the city in the morning and evening peak periods before and after the stagger policy is implemented in the city, so as to obtain the influence of the stagger policy on the urban traffic operation state, thereby providing data support for the formulationand evaluation of the urban traffic policy, and at the same time, providing a strong guarantee for the simulation, planning and evaluation of the transportation system.

The invention provides an intelligent time-varying blind deconvolution broadband processing method and device. The method comprises: extracting an initial seismic wavelet from an attenuated seismic data; generating a Q value group including a plurality of stratum quality factors Q values randomly; according to the initial seismic wavelet and the Q value group, establishing a time-varying wavelet matrix group; acquiring a plurality of reflection coefficients according to the attenuated seismic data and the time-varying wavelet matrix group; establishing a fitness function; on the basis of the fitness function and the multiple reflection coefficients, acquiring a plurality of fitness values corresponding to the stratum quality factors Q values one by one; according to the multiple fitness values, determining an optimal stratum quality factors Q value. Because the optimal Q value is obtained by intelligent searching instead of manual determination, the work load and error occurrence are reduced, so that an optimal time-varying wavelet is estimated from unsteady data; and an optimal time-varying blind deconvolution result is obtained.

In a method for marking discrepancies of a captured image of an object, an image is captured and compared to a standard image. A discrepant image showing any discrepancies of the captured image is generated, and is separated into an R grayscale image, a G grayscale image, and a B grayscale image. An R channel matrix group, a G channel matrix group, and a B channel matrix group are created. R channel negative matrixes, G channel negative matrixes, and B channel negative matrixes are determined from the RGB channel matrix groups. RGB pixel groups are calculated based on the R channel negative matrixes, the G channel negative matrixes, and the B channel negative matrixes. A target pixel group to be marked is determined by calculating an intersection of the RGB pixel groups. The discrepancies of the digital image are marked out according to the target pixel group.

The present invention provides an outdoor wind-induced snow drift testing device which comprises a fan device for providing wind power, a flow stabilizing device for stabilizing a wind speed, a flow guiding device which is a connection pipe between the fan device and the flow stabilizing device, a vibration device which is arranged at the wind direction part of the flow stabilizing device and is used for simulating a snow environment, and a model device which is arranged I the drift range of snow and is used for simulating a house structure. The outdoor wind-induced snow drift testing device has the advantages that an air refrigeration link is avoided, the operation cost of a test is low, multiple repeated tests can be carried out and a fan matrix group is open, and since the limitation of blocking probability in a wind tunnel is reduced, the scale ratio upper limit of a model component becomes larger.

This invention discloses an encoding method for a low density odd-even check code including: 1, determining code rate R and a code length N and computing the check bit length M, 2, constructing the H to the power p matrix of MxM, 3, determining the structure parameter line weight t of the H to the power d, rank weight q and a sub-matrix dimension u, 4, utilizing the arithmetic sequence to comprise the vector v to constitute matrix D finally, 5, the D matrix and a group of its orthogonal D matrix group constitute matrix H to the power d, 6, evaluating the code sequence c and construct the H matrix, H=H to the power p:H to the power d.

The invention provides an electronic device for realizing pixel jitter calculating based on a graphics processor. A central processor receives the image data, corresponding to a first image code, transmitted by the electronic device, a display displays the image data corresponding to a second image code, the graphics processor converts the image data corresponding to the first image code into the image data corresponding to the second image code and applies a pixel jitter matrix group to generate a corresponding pixel jitter signal, the pixel value of each pixel in each frame is driven to sequentially compare with the threshold of the corresponding matrix of the jitter matrix group, and the intensity of the pixel value is increased when the pixel value is larger than the threshold of the matrix. The invention further provides a method for realizing pixel jitter calculating based on the graphics processor. By the electronic device and the method, the problem that tire marks or blocks exist in LCD image display in the prior art is solved, and display quality is increased.

The invention brings forward a system total interference leakage minimum pre-coding matrix group selection algorithm based on an MIMO. All transformation matrixes in a system are traversed firstly. Characteristic value decomposition is performed on the transformation matrixes, and half of characteristic vectors are selected to act as a pre-coding matrix of one transmitter according to a maximum freedom principle. The pre-coding matrixes of other transmitters are obtained. Total interference leakage of the system is calculated. The corresponding transformation matrix of a pre-coding group which makes total leakage of the system to be the minimum is selected to be the optimal transformation matrix needed to be selected. Characteristic value decomposition is performed on the selected optimal transformation matrix, all the pre-coding matrixes of the transmitters are obtained according to a collaboration interference alignment algorithm and finally an interference inhibition matrix of a receiving terminal is obtained. The speed rate performance of the selection algorithm is better than that of a classic interference alignment algorithm and that of the collaboration interference alignment algorithm so that the selection method is higher in adaptation under statistics channel state information.

The invention discloses an action feature representation method fused with rotation quantity. The action feature representation method specifically comprises the following steps that step 1, using a Microsoft Kinect2.0 infrared depth sensor for collecting human body skeleton information and human body topological structure information with SpineBase as a root node; step 2, calculating to obtain ahuman body posture matrix group according to the coordinates of the human body skeleton joint points in the step 1; step 3, according to the human body topology structure information and the coordinates of the human body skeleton joint points in the step 2, the rotation amount of the child skeleton joint points relative to the parent skeleton joint points is obtained through quaternion calculation, and 4, combinding the rotation amount in the step 3 with the human body posture matrix set in the step 2, and establishing an action feature representation method fusing the rotation amount. According to the action feature representation method fusing the rotation amount, interference between similar action classes can be effectively avoided, and the purpose of improving the accuracy of action recognition is achieved.

A decoding method, a memory storage device and a memory control circuit unit are provided. The method includes: determining an offset threshold value and a corresponding check matrix; receiving response data from a rewritable non-volatile memory module and performing an iterative decoding process. The check matrix includes at least one sub-matrix group, each sub-matrix of the sub-matrix group has a default dimension, and the offset threshold value is less than a default dimension value corresponding to the default dimension In the iterative decoding process, several default groups in a data set are shifted, so as to obtain first shift groups, while an offset of each first shift group with respect to a corresponding group among the default groups is not over the default threshold value. Therefore, decoding reference data used in the iterative decoding process may be generated more efficiently.

The invention discloses uplink information feedback and downlink data transmission methods and equipment, and aims to solve the problem that relatively severe performance loss is caused when data transmission is performed on the basis of channel state information (CSI) reported by a terminal since CSI information reported by the terminal is not enough to track the change condition of a channel when the moving speed of the terminal is very high. The method comprises the following steps that: a base station acquires an lth level pre-coding matrix group corresponding to a pre-coding matrix of the terminal, wherein the pre-coding matrix of the terminal corresponds to L pre-coding matrix groups, each pre-coding matrix group includes at least one pre-coding matrix in a preset codebook, l is equal to 1, ..., L, and L is a positive integer; and the base station transmits downlink data to the terminal according to the pre-coding matrixes included in the acquired lth level pre-coding matrix group. Through the pre-coding matrix groups, a relatively wide beam coverage range is formed for a user; downlink transmission during high-speed movement of the terminal can be ensured; and the downlink transmission reliability is enhanced.

The invention relates to the technical field of micro-expression recognition, and discloses a video-based micro-expression recognition method, device and equipment and a storage medium. The method comprises the steps of: obtaining a to-be-recognized video in real time; performing face time sequence extraction on a to-be-recognized video to obtain time sequence fragments; performing image time sequence feature extraction on the time sequence segments through a three-dimensional convolutional neural network to obtain a feature vector matrix group; obtaining a region incidence matrix group trained by a graph neural network model, and fusing the region incidence matrix group and the feature vector matrix group to obtain a to-be-classified vector; and performing micro-expression recognition on the to-be-classified vector to obtain a micro-expression result corresponding to the to-be-recognized video. Therefore, the micro-expression type of a person is automatically recognized from the space dimension and the time dimension through the three-dimensional convolutional neural network and through the area incidence matrix group reflecting the incidence relation between key areas in a face, and the micro-expression recognition accuracy and quality are improved.

The invention discloses a double-iteration non-orthogonal joint block diagonalization convolution blind source separation method, which comprises the steps of receiving and recording observation signals of source signals from M sensors, obtaining a second-order time delay correlation matrix group of the observation signals under specified time delay, constructing a convolution symmetric fitting cost function according to the second-order time delay correlation matrix group, and solving the convolution symmetric fitting cost function by using a double iterative algorithm to obtain an estimated value of the source signal. According to the method, the aliasing matrix and the diagonal matrix group are directly iteratively solved in the time domain, the problem of serious arrangement uncertainty existing in a frequency domain algorithm is avoided, the speech recognition accuracy after speech separation can be effectively improved, and the method is stable in convergence, can achieve rapid convergence and is low in time complexity.

The present invention relates to a method for detecting a signal, a device for detecting a signal, and a receiving device. A method for detecting a signal according to the present invention is a method for detecting a signal from a receiving device in a communication system including a multi-output multi-input antenna, including: estimating a channel matrix of a signal received from a plurality of receiving antennas; generating at least one channel matrix groups by aligning the channel matrix and generating at least one transmitting signal group by aligning a transmitting signal to be estimated to correspond thereto; determining a first candidate vector in consideration of all constellation points for a first layer of the at least one transmitting signal group; determining a second candidate vector in consideration of all constellation points for a second layer of the at least one transmitting signal group; generating a final candidate vector including the first candidate vector and the second candidate vector; and calculating a soft output using the final candidate vector.