A Novel Online Subcarrier-Wise Extreme Learning Machine Receiver for OFDM Systems

Archive ouverte : Communication dans un congrès

Saideh, Michel | Simon, Eric, | Farah, Joumana

Edité par HAL CCSD ; IEEE

International audience. Recently, Extreme Learning Machine (ELM) started gaining interest among researchers in wireless communications as an online training solution for machine learning based receivers. ELM has proven to provide high training speed and global optimization capabilities. However, the number of needed training pilots is still relatively high and increases rapidly with the number of subcarriers, thus rendering its deployment impractical. In this paper, we propose subcarrier-wise ELM receivers that are robust to the increase in the number of used subcarriers; we then extend them to exploit adjacent channel knowledge, hence providing superior performance in frequency selective channels. In addition, we propose a novel training architecture based on interpolated training that saves more than 50% of the computational and spectral resources of conventional ELM receivers. We show the robustness of the proposed technique in different channel scenarios and OFDM settings by means of both practical channel measurements and numerical simulations.

• Contient
• Sujet
• Description
• Oeuvre