84 results about "Circular convolution" patented technology

The invention discloses a coarse-grained reconfigurable system-oriented convolution neural network loop convolution calculation data reuse system, including a main controller and a connection control module, an input data reuse module, a convolution loop operation processing array, and a data transmission Pathways in four parts. During the convolution cycle operation, the essence is to multiply multiple two-dimensional input data matrices with multiple two-dimensional weight matrices. Generally, these matrices are large in size, and the multiplication takes up most of the time of the entire convolution calculation. The present invention utilizes a coarse-grained reconfigurable array system to complete the convolution calculation process. After receiving the convolution operation request instruction, the register rotation method is used to fully explore the reusability of the input data in the convolution cycle calculation process, which improves the data utilization rate. It also reduces the bandwidth access pressure, and the designed array unit is configurable, and can complete convolution operations with different cyclic convolution scales and step sizes.

The present invention relates to a novel matrix-valued transform framework to process digital signals. According to one aspect of the invention, matrix-valued methods and apparatus are described to transform a vector-valued discrete-time data sequence from time-domain into frequency-domain. In another aspect, matrix-valued methods and apparatus are described to transform a vector-valued data sequence from frequency-domain into time-domain. Furthermore, the new framework disclosed in this invention also provides a plurality of methods and apparatus for basic signal processing functions and operations of a matrix-valued communication system. These functions and operations include but are not limited to matrix-valued fast Fourier transformation, matrix-valued linear and circular convolution, matrix-valued correlation, matrix-valued multiplexing and de-multiplexing, and matrix-valued data coding and decoding.

Audio loudspeaker and headphone virtualizers and methods use room impulse responses with modified individual head-related transfer functions prior to superposition including middle truncation; and perform convolutions in the frequency domain with zero-padded sections to avoid circular convolution overlap.

The invention relates to a method for reducing the PAPR in FRFT-OFDM systems, which belongs to the field of broadband wireless digital communications technology. The method is based on fractional random phase sequence and fractional circular convolution theorem, which can effectively reduce the PAPR of the system. The method of the invention has the advantages of simple system implementation and low computational complexity. In this method, the PAPR of the system can be effectively reduced while maintaining the reliability of the system. When the number of candidate signals is the same, the PAPR performance of the present method was found to be almost the same as that of SLM and better than that of PTS. More importantly, the present method has lower computational complexity than that of SLM and PTS methods.

Channel estimation of block signals containing pilot frequency and balanced method characterizes in containing side channels of force and after of a main channel component in a movable window to determine to get a correct channel estimation interval again to get channel impulse response estimation with the length of N, M or M+2XN, then to take position of home and end of the said window as the positioning information needed by circular convolution of the signal and channel inputs response structure and process the data block via channel transmission as the one after channel distortion via frequency domain balance offset or make time-domain over-sampling channel estimation in the sliding window interval to inerase time-domain resolution of channel estimation. The data block can be IDFT of OFDM or modulated by single carrier.

The primary synchronization signal detection method in an LTE system. The method comprises the following steps: converting cross-correlation process to linear convolution, converting the linear convolution into circular convolution in a zero filling mode, and performing Fourier transform FFT operation to obtain a primary synchronization signal PSS sequence dmf(n) in a local frequency domain; performing the Fourier transform FFT operation on a received signal sequence, and performing point-to-point multiplication on the received signal sequence and the local primary synchronization signal PSS sequence to obtain a cross-correlation power sequence, judging whether the primary synchronization signal PSS is successfully retrieved by judging the peak to average ratio of the cross-correlation power sequence, and if so, calculating a PSS signal position according to the peak position, and calculating a call intra-group identifier; and otherwise, indicating that the retrieval is failed. The primary synchronization signal detection method has the beneficial effects of reducing the complexity of the cross-correlation operation while keeping the wide adaptability of the cross-correlation operation in PSS synchronization, improving the operation accuracy while reducing the single-step operation data size and improving the operation efficiency and accuracy.

Circularly pulse-shaped waveforms for communication systems are disclosed herein including a single carrier modulation in which pulse-shaping is performed using a circular convolution by the transmitter for various modulation schemes. A transmitter, related method, and corresponding receiver are also disclosed for demodulation of the single carrier circularly pulse-shaped signal and data extraction.

The invention discloses a method and a device for decoding an iterated internal code and evaluating a channel for an orthogonal frequency division multiplexing system. The iterating method comprises the following steps: at a transmitting end, carrying out internal code encoding and intraframe interleaving by a code with a long short code; at a receiving end, calculating circular convolution of an OFDM time domain data block and an evaluation value of channel time domain impulse response based on an initial evaluation value of the channel time domain impulse response; completing constellation de-mapping according to the channel evaluation and the result of circular convolution to acquire bit soft information; carrying out intraframe de-interleaving and internal code decoding on the bit soft information to acquire decoded bit soft information, carrying out intraframe interleaving on the decoded bit soft information, and remapping the constellation; and updating the evaluation value of the channel time domain impulse response according to the remapped result, and carrying out channel evaluation post-treatment. The method and the device can improve channel evaluation precision, and reduce system error rate, thereby improving system performance, in particular mobile performance, and ensuring complexity and controllable time delay of the system.

In the invention, the transmitting method comprises: with cyclic shift, letting the peak value of cyclic convolution of circular convolution made for the synchronous signals roughly synchronized at receiving terminal and the local reference signals to be at different positions. The detection method comprises: when making synchronous timing, simultaneously using said positions to detect the cyclic prefix type. By the invention, in the process of getting the synchronous signal timing, the mobile station simultaneously gets the cyclic prefix length type of OFDM symbol so as to have no need to take extra steps used for detecting the cyclic prefix type.

The present invention relates to a novel matrix-valued transform framework to process digital signals. According to one aspect of the invention, matrix-valued methods and apparatus are described to transform a vector-valued discrete-time data sequence from time-domain into frequency-domain. In another aspect, matrix-valued methods and apparatus are described to transform a vector-valued data sequence from frequency-domain into time-domain. Furthermore, the new framework disclosed in this invention also provides a plurality of methods and apparatus for basic signal processing functions and operations of a matrix-valued communication system. These functions and operations include but are not limited to matrix-valued fast Fourier transformation, matrix-valued linear and circular convolution, matrix-valued correlation, matrix-valued multiplexing and de-multiplexing, and matrix-valued data coding and decoding.

The invention discloses a mandarin lip language recognition method based on deep learning. The characteristics of the Chinese language structure are considered, an algorithm framework of an encoder and a decoder is utilized, a convolution neural network is used for extracting video features for universality and expansibility, sub-units of the encoder and the decoder adopt a circular convolution neural network, mandarin labels adopt a word embedding coding mode, and an attention mechanism is added to an encoder-decoder output part for accuracy of lip language algorithm identification. Accordingto the method, mandarin sentence-level lip language recognition is recognized as a research object, restriction factors influencing lip language recognition are analyzed, a targeted solution is addedto a built lip language recognition model, a lip language recognition technology which can be practically applied is obtained, and a theoretical and technical basis is provided for the lip language solution with higher accuracy and higher expansibility.

The invention provides a deep learning method based on a map topological structure and entity text description. In solving the problem of knowledge map completion, the entity to be completed may already exist in the knowledge map (need to be discovered), or may not be in the knowledge map (need to be generated). For the task that needs to be discovered, it can be regarded as a knowledge map completion problem in a closed environment, and the model M1 can 'discover' the entity well; for the tasks that need to be discovered, the model M2 can fully explore the text information with the help of the attention mechanism and the circular convolution network, providing a powerful guarantee for 'generating' the entity. The combination of these two sub-models can solve the problem of complementing the knowledge map of the open world.

Transmitted signals are modified to facilitate the emulation of circular convolution in non-contiguous transmission environments. These modified signals may be derived from well-known signature sequences. In an exemplary method, a tail portion of a final segment of a base signal is prefixed to an initial segment of the base signal, to form a first transmit segment. One or more additional transmit segments are formed by prefixing, to each of the one or more segments of the base signal other than the initial segment, a tail portion of the immediately preceding segment of the base signal. The transmit segments so formed are transmitted in respective ones of the plurality of non-contiguous transmit-time intervals. Corresponding methods for receiving the transmitted segments and reconstructing the base signal are also described, as are corresponding transmitting and receiving apparatuses.

A region segmentation method for naturally placing grape bunch pedicel by machine vision is provided. A binocular vision system is constructed by adopting a multi-degree-of-freedom bracket, and that irradiation mode of diffuse reflective light source is designed to obtain grape bunch image, Then, the image is decomposed in HSI model, the circular convolution kernel with radius of 3 pixels is designed, the median filter is carried out pixel by pixel along the grape string image, and the LOG kernel function is used to sharpen the edge of the grape string image. Then, based on the distribution characteristics of grape edge region, the minimum distance between disconnected domains in the minimum neighborhood is used to represent the edge distance of grape, According to the distribution histogram of edge distance, the segmentation points of stem edge and kernel edge of grape bunch are obtained, and the morphological kernels of open and close are designed by using the segmentation points ofedge distance, and the morphological operation is carried out by using the adaptive convolution check image. Finally, the high precision and high efficiency segmentation of naturally placing grape bunch pedicel region is realized by machine vision.

The present invention relates to a novel matrix-valued transform framework to process digital signals. According to one aspect of the invention, matrix-valued methods and apparatus are described to transform a vector-valued discrete-time data sequence from time-domain into frequency-domain. In another aspect, matrix-valued methods and apparatus are described to transform a vector-valued data sequence from frequency-domain into time-domain. Furthermore, the new framework disclosed in this invention also provides a plurality of methods and apparatus for basic signal processing functions and operations of a matrix-valued communication system. These functions and operations include but are not limited to matrix-valued fast Fourier transformation, matrix-valued linear and circular convolution, matrix-valued correlation, matrix-valued multiplexing and de-multiplexing, and matrix-valued data coding and decoding.

The invention provides a signal spectrum zooming method based on linear frequency modulation z transform. A symmetrical linear system h(n) is obtained by changing the expression form of a Bluestein equation and carrying out zero filling and periodic extending on the linear system h(n); the method for calculating the Fourier transform of an L point linear system sequence h(n) through a stored L / 2+1 point linear system sequence is put forward; the result q(n) of the circular convolution is horizontally moved leftwards by N0 units, the front M values obtained after horizontal movement are multiplied with k (k=0,1,..., M-1), and then the signal partial spectrum with the improved frequency resolution can be obtained. By means of the scheme, the storage data size of the linear system sequence and the Fourier transform of the linear system sequence can be reduced, the storage space is saved, the Fourier transform of an L point linear system can be calculated through the stored L / 2+1 point linear system data, the calculated quantity of an FFT is reduced, and the calculating efficiency is improved.

The invention relates to a method that low complexity suppression of PAPR in FRFT-OFDM system, which belongs to the field of broadband wireless digital communications technology and can be used to reduce the PAPR in FRFT-OFDM system. The method is based on fractional random phase sequence and fractional circular convolution theorem, which can effectively reduce the PAPR of system. The method of the invention has the advantages of simple system implementation and low computational complexity. In this method, the PAPR of the system can be effectively reduced while keeping the system reliability. When the number of candidate signals is the same, the PAPR performance of the proposed method was found to be almost the same as that of SLM and better than that of PTS. More importantly, the proposed method has lower computational complexity than that of SLM and PTS.

The invention discloses a high-speed parallel equalizer and an equalizing method thereof, relating to the field of a digital receiver. The method comprises the following steps of: changing a carrier wave-recovered data which is sent by a front end receiver into parallel 2n lines; converting the data by means of frequency domain; filtering the wave of the data by means of equalizing in the frequency domain; outputting the wave-filtered data after being converted into frequency domain; selecting the latter n lines of output data according to the principle of circular convolution; extracting error information; converting the error information by means of frequency domain; and completing the update of a weight coefficient in the frequency domain, thereby completing a data-equalizing and a weight coefficient-updating process. The method is quite suitable for processing high-speed digital communication and conquers the limitation of hardware resources.

The invention relates to a low-complexity peak-to-average ratio (PAPR) inhibition method of an FRFT-OFDM (fractional Fourier transform-orthogonal frequency division multiplexing) system, belongs to the technical field of broadband wireless digital communication, and can be used for reducing the PAPR of the FRFT-OFDM system. The method is based on the Fractional order random phase sequence and the Fractional order circular convolution theorem, and the operation complexity is effectively reduced. The method provided by the invention has the advantages that the system realization is simple, and the calculation complexity is low. According to the method, the reliability of the system is maintained, meanwhile, the PAPR of the system can be effectively reduced, and when the alternative signal numbers are identical, the method has the same PAPR inhibition performance as an SLM (selection mapping) method and has the better PAPR inhibition performance than a PTS (partial transmission sequence) method. In addition, compared with that of the SLM method and the PAPR method, the calculation complexity of the low-complexity PAPR inhibition method is greatly reduced.

The invention discloses an offshore oil-gas platform detection method based on a structural recurrent neural network. The method comprises: according to serialization structure characteristics of a space-time image, a structural recurrent neural network model of a seaborne target is established; parameter optimization adjustment is carried out on the established structural recurrent neural networkmodel and then a high-order space-time characteristic model of the seaborne target is established; on the basis of the high-order space-time characteristic model of the seaborne target, a high-orderspace-time characteristic of the seaborne target is obtained; a bi-directional circular convolution layer is constructed based on the high-order space-time characteristic of the seaborne target and abi-directional circular convolutional neural network with fusion of front and rear time phase features is established; and on the basis of parallel optimization adjustment of the bi-directional circular convolution layer, accurate detection of the seaborne moving target under the single time phase is realized. With the deep learning, the capability of obtaining the target feature is enhanced; andwith full utilization of the time-space features of the target, the target in the complex background interference can be detected. Moreover, the universality of the algorithm is enhanced.

The invention discloses a signal channel estimation method of a transmission diversity system and belongs to the technical field of digital information transmission. The method comprises the following steps: a space time code designed according to the matrix orthogonality is taken as a diversity training sequence; a sender fills the diversity training sequence in a data frame and sends the data frame out; a receiver extracts the training sequence filled in the data frame and the linear convolutions of channel impulse response from the corresponding sender to the receiver; the sum of the linear convolutions is reconstructed for acquiring a circular convolution; the frequency domain form of the circular convolution is computed; the frequency domain estimation of the channel impulse response is computed according to the frequency domain form; the channel estimation from the sender to the receiver is computed according to the frequency domain estimation of the channel impulse response. The invention uses the designed space time code as the diversity training sequences, and selects the diversity training sequences with constant modulus in the frequency domains or the sum of frequency domains, thus effectively eliminating interference among the diversity training sequences and simplifying the channel estimation of the transmission diversity system.

The invention provides a circular convolution hexagonal multicarrier transmission method. The method comprises the following steps: determining the number of sub symbols contained in a sent data block; determining the number of to-be-sent data symbols according to the number of the sub symbols and the number of subcarriers; carrying out shift and zero fill on each to-be-sent data symbol according to the position of a belonging sub symbol on the time to obtain a corresponding sending vector; separately carrying out circular convolution operation on each sending vector and a Gauss pulse shaping function, and carrying out frequency modulation on a circular convolution result according to the sub symbols and the subcarriers belonging to the sending vector to obtain a pulse signal carrying the data symbol information; accumulating the pulse signals to obtain a multicarrier signal; and adding a cyclic prefix for the multicarrier signal according to the maximum delay extension of a channel to form a to-be-send baseband signal. The circular convolution hexagonal multicarrier transmission method provided by the embodiment of the invention has very strong double-selection channel resistance, realizes higher spectrum effectiveness of the to-be-send signal, and meanwhile reduces the peak-to-average power ratio.

The invention provides a deep learning super-resolution method based on an enhanced upsampling and discrimination fusion mechanism, which comprises the following steps of: directly extracting original features of a low-resolution input image by using a single layer based on a deep learning convolutional layer in a residual error branch; using 6 cascaded multi-layer feature fusion-based circular convolution units to extract deep features, and enabling the deep features output by each circular convolution unit to be subjected to up-sampling through 6 deconvolution layers; fusing the up-sampled high-resolution features by applying a feature fusion mode with a discrimination mechanism, and performing dimensionality reduction on the up-sampled features by using a single-layer convolutional layer to obtain a residual error of the high-resolution image; carrying out up-sampling on the low-resolution image by using a bicubic interpolation method in a mapping branch; and adding the mapping and the residual of the high-resolution image pixel by pixel to obtain a final high-resolution image. According to the invention, an up-sampling opportunity is provided for low-resolution characteristics of each stage, and higher reconstruction accuracy is obtained.

The invention discloses a correlated peak position information-based time synchronization method for underwater acoustic spread spectrum communication. The method comprises the following steps: 1, initializing parameters of a receiving terminal; 2, acquiring a signal by using the receiving terminal; 3, performing correlation operation on a second processed signal block variable and a local synchronization reference signal; 4, calculating the peak position of a correlation operation result of the second processed signal block variable and the local synchronization reference signal; 5, judging whether a difference value between pos_1 and pos_2 is less than a certain permissible deviation range or not; 6, extracting a section of the signal s1(k) from the first processed signal block variable; 7, performing circular convolution calculation on the s1(k) and an inverted sequence sflip1r(k) of a local demodulation reference signal; 8, judging whether the deviation of the peak position from the starting or ending position of the signal is within a certain permissible deviation range or not. According to the method, the addition of a signal head is avoided, the easiness in networking is ensured, the requirements on the calculation capability of a platform are low, the method is convenient to operate, and the pseudo triggering or triggering omission rate is reduced.

The invention discloses a method for forming guard intervals of a frame structure in a wireless communication emission system. An initial state of a linear feedback shift register (LFSR) for generating a current frame of circular and continuous frame guard intervals is a final state of the LFSR for generating a previous frame of circular and continuous guard intervals. The invention also discloses a method and a device for reconstructing a signal channel and emitting an information data circular convolution. A required data sequence r1 in the current frame of the circular and continuous guard intervals, the whole information data sequence r2 behind the current frame of the circular and continuous guard intervals and a receiving data sequence r3 with a start length of a maximal predicted delay spread of a next frame of the circular and continuous guard intervals are respectively intercepted from receiving data r, and the obtained three data sequences are aligned and calculated so as to obtain a reconstructed signal channel and a circular convolution of emitted information data. The invention can reduce the calculation complexity and obtain higher accuracy.

An FBMC equalization and demodulation unit and corresponding method to process an FBMC signal includes FBMC symbols, each FBMC symbol comprising data mapped over M subcarriers, oversampled by a factor K, filtered by a prototype filter and transposed in the time-domain, comprising: a frequency domain transposition unit, configured to transpose a block of P*KM samples comprising at least one FBMC symbol into frequency domain samples, where P is an integer greater than one, an equalizer unit configured to multiply the frequency domain samples by one or more coefficients computed from a propagation channel estimate, at least one circular convolution unit, configured to perform P circular convolutions between subsets of the equalized samples and a frequency domain response of a frequency shifted version of the prototype filter, and adders, to sum corresponding outputs of each of the P circular convolutions.

A method for measuring a frequency characteristic of a system for measurement, in which the length of an impulse response of the system is greater than the length N of a TSP (time stretched pulse) signal, is provided. The method includes the steps of supplying the TSP signal to the system continuously for a predetermined number of times, adding and averaging output signals each having the length N, and performing circular convolution on a value obtained by the adding and averaging so that the frequency characteristic of the system is obtained.

Equipment for a MIMO differential SC-FED communication system that includes a transmitter and a receiver is provided. The transmitter has a differential block encoder module for receiving a plurality of data block pairs and performing a circular convolution operation on the data blocks to produce a plurality of coded data blocks in a space-time block coding (STBC) unit. The STBC unit will perform a space-time block encoding process on the output from the differential block encoder module to obtain a plurality of transmitting data blocks. A plurality of frame generators receives the respective transmitting data blocks and adds a cyclic prefix to the corresponding transmitting data blocks to generate a plurality of block frames. Then, the frame generators send the block frames to the receiver of the present invention via a corresponding transmitting antenna.