Optimizing training method of neural network equalizer

Abstract

The invention provides an optimized training method for neural net equalizer, wherein the training time expense of the neural net equalizer under the typical channel condition is first recorded, the optimized restarting point of the neural net equalizer is calculated, training the sequence training neural net equalizer, restarting the training procedure, after the neural net equalizer training contracts, compensating the distorted wireless signal through its output.

Description

Technical field [0001] The invention relates to the field of wireless communication, in particular to an optimized training method of a neural network equalizer. Background technique [0002] Equalization is the core method for wireless communication systems to solve problems such as multipath fading and nonlinear amplification and distortion. In the case that the channel and the transmitter power amplifier do not introduce any amplitude distortion, the linear equalizer based on the mean square error criterion can achieve optimized performance measured by the symbol error probability. However, the complex wireless environment and high-speed wireless modulation technology usually cause significant amplitude distortion, which severely limits the availability of linear equalizers. In this context, nonlinear equalization algorithms have been proposed, such as decision feedback method and maximum likelihood sequence estimation method. However, the above nonlinear methods still have li...

AI Technical Summary

Problems solved by technology

[0004] However, the application of neural networks in nonlinear equilibrium is still restricted by a key factor: the convergence of the training process.

Different from the traditional analytical optimization method, the training process of the neural network mainly utilizes local information; the local optimum in the image space will interfere with the convergence of the training process to the global optimum, which will lead to more inconsistencies in the training process of the neural network. Determinism and variability, that is, the probability distribution of training time overhead has heavy-tailed characteristics

Specifically in the adaptive nonlinear equalization method, the heavy tail of the training time overhead of the neural network equalizer will lead to the training process. In some time periods, the training quality target for the equalizer cannot be successfully achieved, that is, the training process does not converge. During these time periods, the equalizer is in a state of lower performance, which seriously affects the equalization quality

Images