CNN-LSTM multi-branch structure and multi-signal representation-based modulation identification model

Abstract

The invention provides a CNN-LSTM multi-branch structure and multi-signal representation-based modulation identification model, and aims to solve the problem that most of existing modulation recognition methods based on deep learning neglect interaction between different deep learning models and the characteristic that different signal representations can reflect signals at different levels. The method comprises the following steps: firstly, processing a time domain signal in a public data set RML2016.10a into I / Q, A / phi and cyclic spectrogram representation as an input sample of a model; secondly, constructing a CNN-LSTM-based multi-branch model to perform feature extraction on different signal representations, wherein a first branch is responsible for extracting features represented by I / Q signals, a second branch is responsible for extracting features represented by A / phi signals, and a third branch is responsible for extracting features represented by cyclic spectrum signals; and finally, performing outer product on the features extracted from the three branches to obtain a feature matrix, then converting the feature matrix into feature vectors through a flatten layer, and performing classification by using a softmax-based neural network as a classifier. According to the invention, complementarity between different networks and multi-level features reflected by different signal representations are fully considered, related information of the signals in space and time is extracted by using a combined structure of the CNN and the LSTM, the accuracy of modulation identification is improved, the realizability is high, and the model can be well applied to related projects of a non-cooperative communication system.

Description

technical field [0001] The invention relates to algorithms related to deep learning and related theories of signal processing, belonging to the fields of communication signal processing and artificial intelligence. Background technique [0002] The rapid development of wireless communication technology has brought great changes to human life, and has gradually become an indispensable part of human daily life. Under the background of the increasingly prominent information demand, the communication environment is becoming more and more complex, and the modulation methods are becoming more and more complicated and diverse. Various wireless services and wireless products emerge in an endless stream, and the signal density in the surrounding environment is also greatly increased. To ensure that information can be transmitted at high speed in such a complex environment, it is necessary to accurately identify each signal modulation method. Therefore, modulation recognition of sign...

AI Technical Summary

Problems solved by technology

[0005] The current DL-based modulation recognition is based on a single representation of the signal, such as constellation diagram, eye diagram, amplitude / phase, without considering that multiple signal representations can reflect the characteristics of the current modulated signal at different levels, and does not consider the differences between different models. complementary

Images