Mixed Numerology Interference Recognition Approach for 5G NR

Abstract

The coexistence of multiple numerologies can cause interference and degrade system performance in 5G devices. In this letter, a convolutional neural networks (CNN)-based mixed numerology interference (MNI) recognition approach is devised to resolve the interference for user equipment (UE) to receive new radio (NR) downlink signal. The results are learned from the frequency domain magnitude and phase waveform, for the best precision and efficiency. We use as little sample data as possible to reduce memory/buffer costs and to alleviate the computing resources. Simulations are performed to verify the viability on various signal-noise ratio (SNR), channel condition, modulation, etc. It is shown that the accuracy can reach 97% or higher at varying SNR and fading channels. Furthermore, a dynamic filtering/windowing (DFW) scheme is proposed. The link level simulation results show that the scheme can improve the attainable Signal to Interference plus Noise Ratio (SINR) by 3~6dB for the indicated cases.