Joint Channel Estimation and Signal Detection for LoRa Systems Using Convolutional Neural Network

In long-range (LoRa) systems, the performance of signal detection is significantly affected by channel fading and the same or different spreading factor (co-SF/inter-SF) interference. Although coherent detection in LoRa systems can achieve more robust performance than the non-coherent detection, the channel estimation is required. Due to the co-SF/inter-SF interference, the estimated channel state information (CSI) is inaccurate, thus further resulting in performance degradation. To address this problem, a convolutional neural network based joint channel estimation and signal detection (CNN-JCESD) structure for the LoRa systems is proposed. Specifically, we construct a new frame structure to obtain more accurate CSI under the co-SF/inter-SF interference. Then, we utilize layer normalization technique in data pre-processing to obtain better performance. Moreover, we design a new CNN for the LoRa systems to achieve jointly channel estimation and signal detection. Simulation results show that the proposed CNN-JCESD structure has better performance and more robustness compared to the existing detectors over Rayleigh block-fading channel and the co-SF/inter-SF interference.