Multi-Task Learning Based Joint Pulse Detection and Modulation Classification


Akyon F. C. , Nuhoglu M. A. , Alp Y. K. , ARIKAN O.

27th Signal Processing and Communications Applications Conference (SIU), Sivas, Turkey, 24 - 26 April 2019 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • Doi Number: 10.1109/siu.2019.8806285
  • City: Sivas
  • Country: Turkey

Abstract

In this work, a multi-task learning and recurrent neural network based new technique is proposed that performs joint SNR (singal-to-noise ratio) estimation and pulse detection over the pulse based signal samples coming from the digital receivers used in electronic warfare systems, and automatically classifies the modulation present on detected pulses. Proposed technique uses the raw IQ data as input, without any feature extraction. Moreover, usage of separately trained classifiers for different SNR regions is proposed for better classification accuracy. Most suitable classifier is selected based on the estimated SNR level from raw IQ data. Detailed tests show the proposed structure can achieve a detection with %90 accuracy, SNR estimation with 1.5 dB mean absolute error, and classification with 84% accuracy at a very low SNR level as -30 dB.