Ensemble Extreme Learning Machine Based Equalizers for OFDM Systems

Abstract

Extreme Learning Machine (ELM) technology has started gaining interest in the channel estimation and equalization aspects of wireless communications systems. This is due to its fast training and global optimization capabilities that might allow the ELM-based receivers to be deployed in an online mode while facing the channel scenario at hand. However, ELM still needs a relatively large amount of training samples, thus causing important losses in spectral resources. In this work, we make use of the ensemble learning theory to propose different ensemble learning-based ELM equalizers that need much less spectral resources, while achieving better performance accuracy. Also, we verify the robustness of our proposed equalizers within different communication settings and channel scenarios by conducting different Monte Carlo simulations.

Artificial intelligence for channel estimation in multicarrier systems for B5G/6G communications: a survey

“…However, note that after training, the AI model can be less complex than the MMSE, which requires regular estimates of channel parameters (e.g., noise variance). However, dimensionality reduction [174], Bayesian learning [223], ELM networks [257], k-mean clustering [325], metalearning [324], LS-based ML algorithms [326], and DNN [296] have been investigated as solution candidates for reducing training sequences. Further research can consider reducing the computational work required by each AI-aided channel estimation.…”

“…This learning technique has been applied in the channel estimation field for OFDM and MIMO-OFDM systems. The evaluated ELM networks are a single-hidden layer with an implementation based on the AMBCE [255], ABCE [256][257][258][259][260][261][262][263][264], and ABCEx [265,266] approaches. The referred works employ a network comprising p input and m output neurons, as shown in Fig.…”

“…A technique to reduce the number of training pilots was developed based on the ensemble learning theory [257]. This method generates and combines different models to find an optimal predictive model.…”