Deep Tree Search Based Pilot Efficiency Semi-Blind Channel Estimation for Massive IoT

As the tremendous increasing of Internet of Thing (IoT) deceives, the channel estimation between devices and base station will be a critical problem due to the pilot contamination. To address this issue, in this work, we provide a new perspective to alleviate pilot contamination for massive IoT system by reducing the command of piolt. A deep tree search (DTS) method is a decision-making algorithm. The basic principle of proposed algorithm is based on DTS and semi-blind channel estimation, which can reduce the number of transmitting pilot sequence and improve transmission efficiency. We first covert the channel state information (CSI) into a ternary quantization form utilizing the sparse channel property of massive IoT system. Then the channel estimation can regarded as a Markov Decision Process (MDP) with the goal of finding the minimal Frobenius norm between estimated CSI and the transmitted signal. We propose a deep neural network (DNN) based tree search method to solve the MDP based CSI estimation problem. Compared with conventional methods, the proposed method exhibits higher estimation efficiency, which reduces by 10% the pilot sequence to achieve the same estimation performance. The experiments show that the SER performance of T-DTS can reach about 5.6 × 10 ^-5 at SNR=6 dB in massive MIMO systems, which outperforms existing methods.