Channel prediction method combining deep learning and basis extension model

Abstract

The invention discloses a channel prediction method combining deep learning and a basis extension model in the technical field of wireless communication. The channel prediction method comprises the following steps of: step 1, acquiring a correlation matrix of a channel according to channel information at a historical moment; step 2, carrying out eigenvalue decomposition on the correlation matrix to obtain an optimal primary function; step 3, modeling a channel by using the basis expansion model; step 4, acquiring a basis coefficient estimation value based on historically received pilot signals and an optimal basis function; step 5, constructing a training sample set according to the basis coefficient estimation value; step 6, training a BP neural network by using the training sample set; step 7, acquiring a channel prediction model with an optimal weight and an optimal threshold value; step 8, performing online prediction based on the channel prediction model; and step 9, converting the basis coefficient estimation value into a frequency domain channel matrix. The channel prediction method has low calculation complexity and high prediction precision, and is suitable for efficient acquisition of time-varying channel information in a high-speed mobile environment in the future.

Description

technical field [0001] The invention relates to a channel prediction method. Background technique [0002] In recent years, with the large-scale deployment and operation of high-speed railways and highways, wireless communication in high-speed mobile environments has attracted more and more attention worldwide. Moreover, in the future high-speed mobile scenario (B5G), the vehicle speed will be higher and higher, and the higher vehicle speed will cause a greater Doppler frequency shift, which will lead to rapid time-varying wireless channels, thus This makes the acquisition of channel information more challenging in this scenario. Due to the existence of processing delay, the traditional time-varying channel estimation technology causes the estimated channel to be seriously outdated, and the channel prediction technology is widely used in high-speed mobile scenarios because it can predict the channel in the future based on historical channel information. Efficient acquisiti...

AI Technical Summary

Problems solved by technology

ZHAO Y et al. (ZHAO Y, GAO H, BEAULIEU N C, CHEN Z et al., "Echo state network for fast channel prediction inrice of fading scenarios") gave a channel prediction method based on echo state network, but due to the The method does not update the weight matrix during training, so the prediction accuracy is not high

DING T et al. (DING T, HIROSE A. et al., "Fading channel prediction based on combination of complex-valued neural networks and chirp Z-transform") proposed a channel prediction method based on complex-valued neural networks, which uses The channel estimation obtained by the chirp-Z transform trains the complex-valued neural network, but the computational complexity of this method is high since the computational load of the complex-valued network is much higher than that of the real-valued network

W.Jiang et al., (W.Jiang, H.D.Schotten et al., "Deep Learning for Fading Channel Prediction") aimed at the multipath fading channel in the wireless communication system, constructed two new types based on long short-term memory (Long Short-Term- Term Memory, LSTM) network and Gated Recurrent Unit (Gated Recurrent Unit, GRU) network channel prediction model, these two models send the channel response obtained by continuous sampling into the neural network to realize time-varying channel prediction, but due to The number of network layers and neurons is large, resulting in too much computational complexity

[0005] Compared with traditional prediction methods, neural network-based prediction methods are more suitable for high-speed mobile scenarios. However, these existing channel prediction methods based on deep learning are all direct path-by-path predictions for time-varying channels in the time domain. The time-domain direct prediction of multipath time-varying channels needs to predict a large number of channel tap samples, which will greatly increase the complexity of prediction, thus causing channel outdated

Images