Numerology Selection for OFDM Systems Based on Deep Neural Networks

TL;DR

This paper introduces a deep neural network approach to optimize numerology selection in OFDM systems, improving performance by adapting to channel conditions such as delay spread, velocity, and noise.

Contribution

It presents a novel DNN-based method for numerology selection in OFDM that considers channel features to minimize SNR loss, outperforming existing techniques.

Findings

DNN achieves better performance than existing methods

The approach adapts numerology based on channel characteristics

Decision boundaries illustrate application ranges

Abstract

In order to support diverse scenarios and deployments, the numerology of orthogonal frequency division multiplexing (OFDM) is defined for the parametrization of subcarrier spacing and cyclic prefix (CP). The time-frequency dispersion of mobile radio channels and the channel noise result in different performance deterioration in different numerologies. In this letter, we propose a deep neutral network (DNN) approach for numerology selection of OFDM systems. Considering the inter-symbol interference (ISI), inter-carrier interference (ICI) and noise level, the SNR loss is established as the objective to be minimized. We extract the power delay profile, mobile velocity and noise power as the input features to the DNN. The proposed DNN learns from the channel characteristics to obtain the optimal numerology selection. Simulation results show that the proposed DNN achieves better performance than the existing methods. The decision boundaries of different numerologies are also illustrated to show the application range according to the channel characteristics.