Neural Network Based Data Estimation for Unique Word OFDM

Model-based methods have been employed for data estimation for several decades. Due to the incredible success of data-driven machine learning methods, efforts have recently been made to utilize neural networks (NNs) for data estimation in general multiple-input multiple-output (MIMO) communication systems. In this paper, NN-based data estimation is conducted for a communication system employing the unique word orthogonal frequency division multiplexing (UW-OFDM) signaling scheme. In particular, we utilize the so-called DetNet, an NN that has been proposed for data estimation in a general MIMO system. However, to achieve satisfying results for data estimation in a UW-OFDM system an appropriate pre-processing of the input data of DetNet has to be introduced. We investigate its bit error ratio performance in indoor frequency selective environments, we conduct a brief complexity analysis, and we highlight its partially peculiar estimation characteristics.