A Modulation Pattern Recognition Method Based on Spatial-Temporal Feature Extraction and Deep Learning

Abstract

The invention discloses a modulation pattern recognition method based on deep learning of spatio-temporal feature extraction. The method includes collecting signals of modulation patterns to be identified; constructing an automatic modulation recognition deep learning model including a parameter estimation module, a parameter change module and a spatio-temporal feature extraction module, and training it; using the trained automatic modulation recognition deep learning model to The collected signal is subjected to modulation pattern recognition. In order to solve the problem that the modulation recognition model in the prior art has high complexity and it is difficult to achieve high recognition accuracy under the condition of low model complexity, the present invention proposes an automatic modulation mode based on parameter estimation and transformation for spatio-temporal feature extraction Identify the deep learning model, the number of parameters for modulation pattern recognition using this model is less than the existing deep learning-based automatic modulation recognition method, and the training cost is lower than other methods with the same level of recognition accuracy.

Description

technical field [0001] The invention relates to a modulation pattern recognition method, in particular to a modulation pattern recognition method based on spatio-temporal feature extraction deep learning. Background technique [0002] During the transmission process, communication signals are usually affected by unfavorable factors in the channel, such as noise, multipath fading, shadow fading, center frequency offset, sampling rate offset, etc., which will cause amplitude attenuation and carrier frequency of the received signal. and phase shift. Automatic Modulation Recognition (AMR) is an important step between signal detection and demodulation, which provides the basic functionality to detect modulation schemes. With the rapid development of wireless communication, signal modulation schemes will become more complex and diverse to meet the needs of increasingly complex communication scenarios, so it is urgent to design effective AMR models. [0003] Traditional AMR resea...

AI Technical Summary

Problems solved by technology

Someone proposed a similar model that replaces the LSTM network with a gated recurrent unit (GRU). Although the recognition accuracy is reduced, the model complexity is also reduced.

Someone proposed a spatio-temporal multi-channel learning model, using complementary information from the I / Q channel, I channel and Q channel data, and using the spatial and temporal attributes existing in the signal, to achieve automatic modulation recognition, which is so far, The model with the highest recognition accuracy in the field of automatic modulation recognition, but the number of parameters and complexity of the model are relatively high

[0005] At present, the automatic modulation recognition model based on deep learning has high complexity, and it is difficult to achieve high recognition accuracy with low model complexity.

Currently there are some models with high recognition accuracy that can run in offline tasks, but AMR usually needs to be processed online, and if the model is too complex, it will encounter excessive delays

High computational complexity also prevents deployment on resource-constrained devices, such as many Internet of Things (IoT) devices with limited memory, computing power, and energy

Images