Noise Interpolation for Unique Word OFDM

Unique word-orthogonal frequency division multiplexing (UW-OFDM) is known to feature an excellent bit error ratio performance when compared to conventional OFDM using cyclic prefixes (CP). In recent papers classical and Bayesian linear data estimators as well as the non-linear sphere decoding (SD) approach have been investigated in the UW-OFDM context. In general, the remaining error samples after a linear data estimation in UW-OFDM are correlated. In this work, noise interpolation (NI) is proposed, which uses preliminary data decisions together with Wiener interpolation to exploit these correlations. It turns out that the selection of samples to be used for NI is crucial. With the sample selection policy suggested in this work, NI clearly outperforms LMMSE data estimation, and due to its lower complexity compared to SD, it can be regarded as an attractive compromise for UW-OFDM reception.