Real-Time Experimentation of Deep Learning-based RF Signal Classifier on FPGA
Sohraab Soltani, Yalin E. Sagduyu, Raqibul Hasan, Kemal Davaslıoğlu, Hongmei Deng, Tugba Erpek · 2019
We demonstrate the embedded implementation of a deep learning-based RF classifier that is implemented on the Field Programmable Gate Array (FPGA) of the software-defined radio (SDR) platform, DeepRadioTM. Supported by low-power embedded computing, the received signals are classified to different modulation types in real time. The deep neural network that is used for the RF signal classifier runs directly on the FPGA fabric of DeepRadioTM. In the demonstration setup, a USRP radio transmits signals with different modulation types and DeepRadioTMclassifies each received signal (I/Q samples) to its modulation type. Classification results are shown on the mobile app of a smartphone that is connected to DeepRadioTM. Demonstration results show that the classifier achieves high accuracy (close to software implementation) at low latency (microseconds per sample) and low energy consumption (microJoules per sample).