CNN-Attention-DNN Design for CFO Estimation of Non-Pilot-Assisted OFDM System
Mahesh Shamrao Chaudhari, Sudhan Majhi, Sandesh Jain
Abstract
In this letter, we propose a novel attention mechanism inspired convolutional neural network (CNN)-deep neural network (DNN) called CAD model architecture for carrier frequency offset (CFO) estimation in orthogonal frequency division multiplexing (OFDM) systems. The hyperparameters used in the proposed CAD model are optimized using the Bayesian approach. The proposed CFO estimation scheme is robust to small symbol timing offset errors and does not require apriori knowledge of modulation format, channel, and transmitted OFDM signal parameters. Simulations performed over the realistic frequency-selective channel models with and without power amplifier non-linearity indicate that the proposed CAD architecture outperforms the conventional statistical-based and recurrent neural network (RNN)-based models. Lastly, the proposed CAD model is validated on the radio frequency testbed.