DL-SIC: Deep Learning Aided Successive Interference Cancellation in Shared Spectrum

With the increasing demand for wireless capacity, multiple wireless technologies will inevitably coexist over shared bands. Successive interference cancellation (SIC) is a promising technique for improving spectrum utilization by utilizing the difference in the powers of concurrently received signals. However, enabling SIC over a shared band faces several challenges, related to the heterogeneity of the coexisting technologies, the unknown powers of received signals, and the uncoordinated and asynchronous nature of transmissions. Traditional SIC (T-SIC) receivers cannot simultaneously achieve low decoding latency and low decoding bit error rate (BER). To address these challenges, we propose DL-SIC, a deep learning approach for accelerating the operation of an SIC receiver. DL-SIC includes a deep learning-based protocol detector for identifying overlapping packets, as well as a deep learning-based SIC classifier for accurate determination of the SIC decoding order in scenarios where the relative strengths of the received signals are unknown. We conduct simulations and over-the-air (OTA) experiments to evaluate DL-SIC, and compare it with two T-SIC approaches, T-SIC₁ and T-SIC₂. Our simulation results clearly indicate that DL-SIC can simultaneously achieve low decoding latency and low decoding BER. Specifically, DL-SIC reduces decoding latency by 75.41% in the worst-case scenario and 84.44% in the best-case scenario compared to T-SIC₁. Furthermore, with a probability of approximately 60%, DL-SIC reduces decoding BER from 10^-1 to 10-⁴ compared to T-SIC₂. Our OTA experiments further confirm the feasibility of DL-SIC.