A Novel Pilot-Aided Channel Estimation Scheme Based on RNN for FDD-LTE systems

Abstract

In frequency division duplex (FDD) system, pilot-based channel estimation is very challenging when the channel is complicated and changeable. Conventional algorithms have the disadvantage of the imbalance between accuracy and complexity. To solve this problem, this paper considers the dynamic temporal characteristics for a channel state sequence. We propose a pilot-aided channel estimation scheme based on recurrent neural network (RNN) for FDD-LTE systems. The deep neural network can be regarded as a mapping function without expert knowledge of channel estimation, in which the input data is the known channel state information (CSI) of reference signals (RS) and the output data is the estimated CSI of the whole band. In order to improve estimation accuracy, a bidirectional RNN (Bi- RNN) network structure is introduced to this designed neural network. In addition, simulation results show that the RNN-based scheme can support channel estimation with an infinite sequence in the time domain. At the same time, better performance can be achieved with a relatively low complexity compared to conventional algorithms.