Efficient Modulation Recognition by Predictive Correction and Double GRUs via a Lightweight Framework

Abstract

Automatic modulation recognition (AMR) provides excellent performance advantages over conventional algorithms and plays a key role in modern communication. However, a general challenge is that the channel errors greatly deteriorate the classification capability, and the computational complexity is extremely high. To reduce the offset error of the signal and ensure fewer parameters to save training resources, we demonstrated an efficient modulation recognition scheme combined predictive correction with double Gate Recurrent Unit (GRU), thus realizing a lightweight neural framework. Predictive correction reduces channel errors, and double GRUs are better at capturing long-term dependencies. The results show that when the signal-to-noise ratio is around 18 dB, the highest recognition accuracy can be achieved, and the computational complexity is significantly reduced. The proposed scheme exhibits a tradeoff between the accuracy and the computational complexity, providing an attractive method for modulation recognition.

Keywords

• modulation recognition
• deep learning
• predictive correction