Efficient Blind Detection for 5G NR Polar-Coded Physical Downlink Control Channels using Deep Learning

In 5G LTE-NR wireless systems, polar codes serve as the channel-coding scheme for physical downlink control channels (PDCCH) due to their efficiency. However, identifying the dynamically changing transmission location of PDCCH poses complexity and latency challenges for receiver design. Recent attempts for latency improvement proposed two-stage blind detection schemes with an initial PDCCH candidate selection stage followed by high-performance decoders, yet the latency of the initial decoding stage remains significant. In this paper, we propose a low-latency scheme leveraging deep neural networks in the candidate selection stage to achieve significant latency reduction. Neural networks are used to recognize the structure of polar codes to efficiently eliminate invalid candidates and apply polar decoding solely to the selected subset. Simulations are performed to demonstrate the achieved latency reduction and detection accuracy. The proposed scheme is shown to reduce the decoding latency significantly at the cost of a slight reduction in the detection accuracy.