Técnicas de Deep Learning enfocadas a la estimación adaptativa de canales en redes de quinta generación

Abstract

In mobile communications, the channel estimation process is one of the main keys to optimize the communication between the transmitter and the receiver. Knowing the channel response is a challenge because there are multiple phenomena like attenuation, multi-path loss, noise, and delays that affect the transmitted signals. Methods based on pilot insertion such as Least Squares (LS) and Minimal Mean Squared Error (MMSE) are commonly used to estimate the channel. Nevertheless, they have issues related with their performance and complexity in varying scenarios. In this document, it is proposed that different Deep Learning (DL) techniques assist a pilot-based channel estimation for a 5G communication system affected by Doppler shift due to the level of mobility. Through simulation modeling including Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) environments in a Tapped Delay Line (TDL) model, it is measured the performance based on the Bit Error Rate (BER), Error Vector Magnitude (EVM), estimation time, and Mean Squared Error (MSE). The results prove that the DL models outperform linear interpolation and practical estimators in a Signal-to-Noise Ratio between 0 dB and 20 dB. Furthermore, the proposed estimation techniques have adaptability to different channel conditions.