Modulation Recognition Method Based on Interference Cleaning and Two-Stage Training Convolutional Neural Network Model

Abstract

The invention discloses a modulation recognition method based on interference cleaning and two-stage training convolutional neural network model. The method first uses the sample sequence of the modulation signal collected to generate the original periodic correlation characteristic digital spectrum, and generalizes it Singular value decomposition operation, space division operation, noise elimination operation, and crosstalk suppression operation are processed to obtain the original final cycle-related characteristic digital spectrum, and then two-stage training is performed on the convolutional neural network to obtain the convolutional neural network model to realize the modulation of the input Identify and classify the modulation mode of the signal. Notable features: Improve the accuracy of modulation recognition and classification while reducing complexity; can eliminate additive noise and suppress crosstalk from adjacent channels, enhance the authenticity of training and recognition signal data; improve the generalization ability of modulation recognition.

Description

technical field [0001] The invention relates to the technical field of wireless communication, in particular to a modulation recognition method based on interference cleaning and two-stage training of a convolutional neural network model. Background technique [0002] In my country's existing radio frequency band authorization rules, the 230MHz frequency band is a dynamic shared frequency band allocated to industries such as electric power and water conservancy. Among them, 223-226MHz and 229-233MHz are used for broadband wireless private networks, and 226-228 / 233-235MHz And 228-229MHz for narrowband wireless private network. Widely used modulation modes include BPSK, QPSK, 2FSK, 4FSK, MSK, AM, and FM. [0003] In the prior art, in order to save network construction costs and improve spectrum utilization, it is generally recommended to adopt a shared network construction mode. However, this model has the following drawbacks: [0004] First, there are channel defects such a...

AI Technical Summary

Problems solved by technology

[0004] First, there are channel defects such as multipath fading, frequency drift and phase noise in the frequency band, and the radio signal leakage of adjacent channels will cause crosstalk to the wireless private network;

[0005] Second, the lack of communication and negotiation between the wireless communication industry, coupled with the random and disorderly occupation of various illegal stations, will inevitably cause potential interference hazards to the reliable operation of the wireless private network. It is necessary to explore effective methods to accurately identify the types of interference and Interference traceability, and then take corresponding countermeasures

However, the above feature learning method can only achieve better recognition accuracy under specific conditions, and feature selection requires the design of feature engineering, which relies heavily on experience accumulation, resulting in poor universality and unstable results.

Images