41 results about "Normalized mean square error" patented technology

The invention discloses a non-Gaussian noise 3D-MIMO channel estimation algorithm, which comprises the following steps of obtaining a support set of a channel matrix by utilizing a judgment condition,selecting a dictionary matrix under the support set, and carrying out order selection calculation of a Gaussian mixture model according to the characteristics of a received signal; computing a weightleast square matrix; obtaining a coefficient and a variance of the Gaussian mixture model; estimating the channel matrix column by column in order to obtain an estimated value of the first time; judging whether an iteration result tends to be stable or reaches the number of iterations, and obtaining a channel matrix under the support set; otherwise repeating the steps until the condition is met;and generating a full-zero matrix when the iteration result is met, inserting the channel matrix under the support set into the all-zero matrix line by line according to positions where non-zero elements in the support set are located, making the rest of positions unchanged, and obtaining an actual channel matrix. A normalization mean square error of the estimation algorithm is obviously superiorto that of other algorithms, and an ideal estimation performance is still achieved under a condition that a signal-to-noise ratio is low.

The invention discloses a multi-antenna system channel estimation method based on deep learning, which is suitable for estimating uplink multipath channels. The invention is realized based on a conditional generation confrontation network. The condition generation confrontation network includes two parts: a generator and a discriminator based on a deep learning network. The method includes two parts: offline training and online testing: offline training first generates training samples based on real channel measurements, and then uses the generator to obtain the estimated channel corresponding to the training samples; secondly, the discriminator obtains the discriminant output, and calculates the loss function to update the discriminator and The network parameters of the generator; after the loop iteration is completed, the trained generator neural network is stored in the base station; in the online test phase, the quantized pilot signal and the original pilot signal are input into the trained generator, and the user to all The estimated channel of the antenna. Compared with the prior art, the estimated normalized mean square error can be effectively reduced.

The invention relates to a timing recovery method for a band-limited system based on p-order moments. In the method, in a feedback phase-locked loop, the p-th power of the signal modulus is used to calculate the timing error detection output value, wherein p>0 is a variable parameter , the method of the present invention can be applied in a timing recovery loop to obtain better residual timing error jitter performance, and the normalized mean square error performance of the timing recovery of the method of the present invention is close to the theoretical Cramereau limit under the AWGN channel. The method of the invention is suitable for a bandwidth-limited communication system with a small roll-off factor, and is suitable for a satellite communication system that needs to increase spectrum utilization.

The present invention relates to the field of cognitive radio communication. The present invention provides a multi-layer distributed joint spectrum sensing method based on the Dirichlet process to realize dynamic spectrum sensing, and finds optimal sensing information by fusing sensing data collected by secondary users of multiple hierarchical centers. The Dirichlet process is used to realize the automatic grouping of data. The Bayesian model estimates a shared hyperparameter and the corresponding divergence probability in each group. The standard Viterbi algorithm is used to obtain the hyperparameters, and the hyperparameters are compared with the decision threshold. The comparison is performed to obtain the final spectrum decision result to determine whether the channel is available. The design fully considers the spatial diversity information of compressed sensing data, which reduces the uncertainty of a single secondary user on compressed sensing data, so that the normalized mean square error performance is better, and the algorithm can effectively mine the compressed sensing data information of each hierarchical center , to obtain a higher probability of correct detection and a smaller probability of false alarm, and improve the performance of spectrum sensing.

The invention discloses a combined cooperative multi-satellite weak echo signal time delay and doppler frequency shift estimation method. According to the method, multiple nonstop wave signals in a reference channel are separated; adaptive inhibition for nonstop wave signals and multi-path signals in an echo channel is carried out; fourth-order cyclic cumulant mutual fuzzy function processing based on four weighted fractional Fourier transform for the signals in the echo channel and the different nonstop wave signals is respectively carried out to acquire multiple characteristic vectors; after spectral peak extraction for fuzzy function peak values is carried out, corresponding coordinates are time delay and doppler frequency shift estimates; the multiple sets of time delay and doppler frequency shift estimates are converted into distances from targets to receivers and speeds, data fusion is further carried out, the distances and the speed values are converted into multiple sets of time delay and doppler frequency shift estimates after data fusion, and thereby time delay and doppler frequency shift of cooperative multi-satellite weak echo signals can be estimated; when a signal to noise ratio is greater than 10dB, a normalization mean square error of signal to noise ratio estimation is smaller than 1.

The invention provides a frequency domain serial correlation channel measurement method based on a hypersonic scene, for solving the technical problem of low channel parameter measurement accuracy caused by a limited channel measurement bandwidth in the hypersonic scene of the existing time domain serial correlation channel measurement method. The method comprises the following steps: selecting channel measurement parameters, and generating a frequency domain measurement sequence; performing inverse Fourier transform on the measurement sequence to generate a time domain measurement sequence; selecting a hypersonic channel frequency offset, performing Fourier transform to obtain a hypersonic channel frequency domain parameter, and performing hypersonic channel fading and noise pollution on the time domain measurement sequence to obtain time domain channel measurement data; performing Fourier transform on the time domain channel measurement data to obtain channel measurement frequency domain data, and performing sliding correlation processing on the channel measurement frequency domain data and the frequency domain measurement sequence to obtain channel frequency domain impulse response; and comparing the channel frequency domain impulse response with the hypersonic channel frequency domain parameter to obtain a normalized mean square error of channel measurement.