Abstract: A method for compressing a neural network includes: obtaining a neural network including J operation layers; compressing a jth operation layer with Kj compression ratios to generate Kj operation branches; obtaining Kj weighting factors; replacing the jth operation layer with the Kj operation branches weighted by the Kj weighting factors to generate a replacement neural network; performing forward propagation to the replacement neural network, a weighted sum operation being performed on Kj operation results generated by the Kj operation branches with the Kj weighting factors and a result of the weighted sum operation being used as an output of the jth operation layer; performing backward propagation to the replacement neural network, updated values of the Kj weighting factors being calculated based on a model loss; and determining an operation branch corresponding to the maximum value of the updated values of the Kj weighting factors as a compressed jth operation layer.

Abstract: A method for compressing a neural network includes: obtaining a neural network including a plurality of parameters to be compressed; dividing the parameters into J blocks; compressing a jth block with Kj compression ratios to generate Kj operation branches; obtaining Kj weighting factors; replacing the jth block with the Kj operation branches weighted by the Kj weighting factors to generate a replacement neural network; performing forward propagation to the replacement neural network, a weighted sum operation being performed on Kj operation results generated by the Kj operation branches with the Kj weighting factors and a result of the operation being used as an output; performing backward propagation to the replacement neural network, updated values of the Kj weighting factors being calculated based on a model loss; and determining an operation branch corresponding to the maximum value of the updated values of the Kj weighting factors as a compressed jth block.

Abstract: The present application discloses a method and a device for pruning one or more convolutional layer in a neural network. The method includes: obtaining one target convolution layer from the one or more convolution layers in the neural network, the target convolution layer including C filters, each filter including K convolution kernels, and each convolution kernel including M rows and N columns of weight values, where C, K, M and N are positive integers greater than or equal to one; determining a number P of weight values to be pruned for each convolution kernel of the target convolution layer based on a number of weight values M×N in the convolution kernel and a target compression ratio, where P is a positive integer smaller than M×N; and setting P weight values with the smallest absolute values in each convolution kernel of the target convolution layer to zero to form a pruned convolution layer.

Abstract: The present application disclosed a decision feedback equalization processing device and method. The device comprises: a channel estimator for receiving an input signal, and determining, based on the input signal, an input signal autocorrelation matrix Ryy, an input-output signal cross-correlation matrix Ryx and an input-output signal cross-correlation vector ryx; a tap coefficient calculator for receiving Ryy, Ryx and ryx, and calculating a feed-forward equalizer (FFE) tap coefficient vector g and a feedback equalizer (FBE) tap coefficient vector f, wherein at least one of the FFE tap coefficient vector g and the FBE tap coefficient vector f is calculated using a conjugate gradient descent algorithm. The tap coefficient calculator comprises: a circulant matrix construction unit and a fast Fourier transformation (FFT) calculating unit.

Abstract: The present application disclosed a decision feedback equalization processing device and method. The device comprises: a channel estimator for receiving an input signal, and determining, based on the input signal, an input signal autocorrelation matrix Ryy, an input-output signal cross-correlation matrix Ryx and an input-output signal cross-correlation vector ryx; a tap coefficient calculator for receiving Ryy, Ryx and ryx, and calculating a feed-forward equalizer (FFE) tap coefficient vector g and a feedback equalizer (FBE) tap coefficient vector f, wherein at least one of the FFE tap coefficient vector g and the FBE tap coefficient vector f is calculated using a conjugate gradient descent algorithm. The tap coefficient calculator comprises: a circulant matrix construction unit and a fast Fourier transformation (FFT) calculating unit.

Abstract: A digital signal receiver comprises an iterative decoder configured to decode its input signal using an iterative decoding algorithm, a signal quality detector, configured to detect signal quality of the input signal of the iterative decoder; a power consumption monitor, configured to detect a parameter indicating power consumption of the iterative decoder; a voltage regulator, configured to adjust a supply voltage of the iterative decoder to maintain it within a preset supply voltage range, based on the detected parameter indicating the power consumption of the iterative decoder; and an iteration controller, configured to adjust a maximum iteration number of decoding based on the signal quality of the input signal of the iterative decoder.

Abstract: A circuit for predistortion, comprising a digital predistorter configured to generate a modified digital signal by modifying an input baseband signal with a predistortion coefficient; a DAC configured to generate an analog signal by converting the modified digital signal; a PA configured to generate an amplified signal by amplifying the analog signal; an attenuator configured to generate an attenuated signal by attenuating the amplified signal; an ADC configured to generate a attenuated digital signal; a timing synchronizer configured to generate a synchronized amplified signal by synchronizing the attenuated digital signal with the modified digital signal; a statistics generator configured to generate a probability density function of a plurality of samples of the input baseband signal; an orthogonal basis function generator configured to generate a set of normalized orthogonal basis functions; a predistortion coefficient training circuit configured to generate an updated predistortion coefficient.

Abstract: A circuit for predistortion, comprising a digital predistorter configured to generate a modified digital signal by modifying an input baseband signal with a predistortion coefficient; a DAC configured to generate an analog signal by converting the modified digital signal; a PA configured to generate an amplified signal by amplifying the analog signal; an attenuator configured to generate an attenuated signal by attenuating the amplified signal; an ADC configured to generate a attenuated digital signal; a timing synchronizer configured to generate a synchronized amplified signal by synchronizing the attenuated digital signal with the modified digital signal; a statistics generator configured to generate a probability density function of a plurality of samples of the input baseband signal; an orthogonal basis function generator configured to generate a set of normalized orthogonal basis functions; a predistortion coefficient training circuit configured to generate an updated predistortion coefficient.

Abstract: A digital signal receiver comprises an iterative decoder configured to decode its input signal using an iterative decoding algorithm, a signal quality detector, configured to detect signal quality of the input signal of the iterative decoder; a power consumption monitor, configured to detect a parameter indicating power consumption of the iterative decoder; a voltage regulator, configured to adjust a supply voltage of the iterative decoder to maintain it within a preset supply voltage range, based on the detected parameter indicating the power consumption of the iterative decoder; and an iteration controller, configured to adjust a maximum iteration number of decoding based on the signal quality of the input signal of the iterative decoder.

Abstract: The present invention relates to a method and system for channel estimation. First, pilot signals are extracted from a received multi-path signal, in which each pilot signal includes a first pilot and a second pilot. Then, an initially estimated channel frequency-domain response is obtained based on the extracted first pilot. Afterward, a frequency-domain response estimate of each pilot frequency in the second pilot is calculated according to the obtained initially estimated channel frequency-domain response, an actual value of each pilot frequency in the second pilot is obtained based on the extracted second pilot, and a deviation between the frequency-domain response estimate and the actual value of each pilot frequency is calculated. When it is detected that aliasing components exist in the initially estimated channel frequency-domain response, a center of each aliasing component is determined according to the deviation, so as to determine an estimated position of each aliasing component.

Abstract: A channel estimation method and a channel estimation system based on time-domain training sequence are provided. The channel estimation system first acquires an initial channel vector used for channel estimation of a current frame, and calculates an algorithm initial vector based on a training sequence vector contained in a received receiving signal vector and the receiving signal vector, then performs estimation based on the initial channel vector and the algorithm initial vector and using a conjugate gradient method based on a preprocessing matrix to acquire each channel estimation value, and eventually perform noise reduction processing on each channel estimation value based on a first predetermined threshold value to acquire a final channel estimation value. Compared with the existing channel estimation methods, the present invention has a shorter delay and lower calculation complexity, and thus can be easily implemented.

Abstract: Data rate conversion devices and methods are provided. A method for converting a first digital signal having a first sampling rate into a second digital signal having a sampling rate close to a predetermined second sampling rate comprises the following operations: when the ratio of the first sampling rate to the second sampling rate is a repeating infinite decimal, calculate at least two calibrating coefficient values and output the calibrating coefficient values according to a predetermined rule; conduct overflow operation on the output calibrating coefficient; and interpolate the first digital signal using the output calibrating coefficient and the result of the overflow operation to obtain the second digital signal such that during any period of a certain length along time axis, sampling times of the second digital signal equals to sampling times of the second sampling rate.

Abstract: The present invention relates to a blind adaptive filter for narrowband interference cancellation, which includes an adaptive filter, a delay unit coupled to the adaptive filter for generating a delayed signal with a predetermined delay length from the output signal of the adaptive filter, and an error calculation unit coupled to the adaptive filter and the delay unit. The error calculation unit compares the output signal from the adaptive filter and the delayed signal from the delay unit to extract error information, and feedback the first error information to the adaptive filter. The first error information is formed of a transfer function including a number of coefficients, and used to adjust the adaptive filter and remove interference in the next input signal. The disclosed technique is also applicable in wideband receivers, as well as resisting multiple strong narrowband interferences having a frequency sweep rate of tens of milliseconds.

Abstract: The present invention relates to a method and system for channel estimation. First, pilot signals are extracted from a received multi-path signal, in which each pilot signal includes a first pilot and a second pilot. Then, an initially estimated channel frequency-domain response is obtained based on the extracted first pilot. Afterward, a frequency-domain response estimate of each pilot frequency in the second pilot is calculated according to the obtained initially estimated channel frequency-domain response, an actual value of each pilot frequency in the second pilot is obtained based on the extracted second pilot, and a deviation between the frequency-domain response estimate and the actual value of each pilot frequency is calculated. When it is detected that aliasing components exist in the initially estimated channel frequency-domain response, a center of each aliasing component is determined according to the deviation, so as to determine an estimated position of each aliasing component.

Abstract: A channel estimation method and a channel estimation system based on time-domain training sequence are provided. The channel estimation system first acquires an initial channel vector used for channel estimation of a current frame, and calculates an algorithm initial vector based on a training sequence vector contained in a received receiving signal vector and the receiving signal vector, then performs estimation based on the initial channel vector and the algorithm initial vector and using a conjugate gradient method based on a preprocessing matrix to acquire each channel estimation value, and eventually perform noise reduction processing on each channel estimation value based on a first predetermined threshold value to acquire a final channel estimation value. Compared with the existing channel estimation methods, the present invention has a shorter delay and lower calculation complexity, and thus can be easily implemented.

Abstract: An IMPLEMENTATION-ORIENTED METHOD of Bit Interleaved Coded Modulation (BICM) based on a Low-Density Parity-Check (LDPC) check matrix including constructing an LDPC code having a block check matrix or providing an existing LDPC, where the block check matrix is divided into one or more sub-matrixes Hij with the size being B*B, constructing a BICM structure; in the BICM structure mapping an ith bit sequence [bitps(i,0), . . . , bitps(i,m?1)] with a length being m to obtain a mapping symbol s(i), where the size of a set of the mapping symbols s(i) is 2m; enabling mapping bits of all the mapping symbols s(i) of a mapping symbol subset S, =[s,(0), . . . s(i), . . . s(B?1)] to correspond to m check sub-matrixes. The mapping bits at the same position correspond to the same check sub-matrix, the number of which is m.

Abstract: A method and system for canceling an interference signal are provided. In the method, a received signal is transformed from a time domain to a frequency domain at relative low precision to obtain an estimated frequency and an estimated amplitude of the interference signal. A reference signal is generated using the estimated frequency and the estimated amplitude. An error signal is generated using the received signal and the reference signal based on minimum correlation criteria. The reference signal is adjusted using the error signal based on an adaptive algorithm. The reference signal is subtracted from the received signal to obtain the useful signal.

Abstract: A method and apparatus for measuring parameters of a receiver having a mixer for generating an I signal and a Q signal using an input signal, an I channel circuit for processing the I signal, and a Q channel circuit for processing the Q signal. The method includes feeding the receiver a first testing signal before the mixer. The method includes feeding the receiver a second testing signal on the I channel circuit. The method includes feeding the receiver a third testing signal on the Q channel circuit. The method includes measuring I/Q quadrature deviation and I/Q delay imbalance of the receiver using the first, the second, and the third testing signals. This separates the I/Q quadrature deviation and I/Q delay imbalance.

Abstract: The present invention relates to a blind adaptive filter for narrowband interference cancellation, which includes an adaptive filter, a delay unit coupled to the adaptive filter for generating a delayed signal with a predetermined delay length from the output signal of the adaptive filter, and an error calculation unit coupled to the adaptive filter and the delay unit. The error calculation unit compares the output signal from the adaptive filter and the delayed signal from the delay unit to extract error information, and feedback the first error information to the adaptive filter. The first error information is formed of a transfer function including a number of coefficients, and used to adjust the adaptive filter and remove interference in the next input signal. The disclosed technique is also applicable in wideband receivers, as well as resisting multiple strong narrowband interferences having a frequency sweep rate of tens of milliseconds.

Abstract: Data rate conversion devices and methods are provided. A method for converting a first digital signal having a first sampling rate into a second digital signal having a sampling rate close to a predetermined second sampling rate comprises the following operations: when the ratio of the first sampling rate to the second sampling rate is a repeating infinite decimal, calculate at least two calibrating coefficient values and output the calibrating coefficient values according to a predetermined rule; conduct overflow operation on the output calibrating coefficient; and interpolate the first digital signal using the output calibrating coefficient and the result of the overflow operation to obtain the second digital signal such that during any period of a certain length along time axis, sampling times of the second digital signal equals to sampling times of the second sampling rate.