Method and system for realizing end-to-end fixed-point fast Fourier transform quantization by neural network

Abstract

The invention discloses a method and a system for realizing end-to-end fixed-point fast Fourier transform quantization through a neural network, and the method comprises the steps: carrying out the quantization processing of floating-point time domain data through the neural network based on deep learning, obtaining corresponding fixed-point time domain data, and further obtaining corresponding vertex frequency domain data through fixed-point fast Fourier transform; and obtaining floating point frequency domain data through the destination vector processing of the neural network based on deep learning. According to the method, the requirement of limited word length is met by using a truncation method, memory resources are conveniently saved, meanwhile, the quantization and de-quantization processes of the signal are jointly optimized through a machine learning method, the quantization work of the signal can be completed without various prior information, and the method is suitable for any linear operation.

Description

technical field [0001] The present invention relates to a technology in the field of signal processing, in particular to a neural network suitable for an OFDM system to realize end-to-end fixed-point fast Fourier transform quantization method and system. Background technique [0002] The modern information and communication technology industry is built on silicon-based solutions. Due to the semiconductivity of silicon, complementary metal-oxide-semiconductor (CMOS) integrated circuits have strong binary information bit representation and computing capabilities, which led to the development of this century. The original information revolution technology. However, perceptual or sampled signals in the real world are usually not binary, and quantization issues have been revealed over the past few decades. Since the resolution of quantization has a strong impact on many aspects of circuit design, such as area, speed, or power consumption, quantization schemes have been extensive...

AI Technical Summary

Problems solved by technology

However, perceptual or sampled signals in the real world are usually not binary, and quantization issues have been revealed over the past few decades

Images