EMI Signal Encoding based on Deep Auto-encoder Combined with Wavelet Transformation

Abstract

With the development of electronic technology, digital devices are becoming more sensitive to electromagnetic interference. As a result, the recognition methods are under scrutiny. To find the way to improve the accuracy of electromagnetic interference recognition, this paper presents a signal encoding method based on deep auto-encoder and wavelet transformation. We use wavelet de-noising as the first step on our method to reduce the noise interference. Then a deep auto-encoder is trained for extracting EMI signals' feature. Results on evaluation the reconstruction error and recognition accuracy demonstrate that our approach outperforms comparison methods, which indicates that the proposed method could better capture the data structure of the high dimensional EMI signals.