Contents I. Introduction
With current 5G wireless technology gradually expanding its deployment, conventional modulation schemes such as OFDM will have difficulties for some high-mobility user application scenarios such as high-speed railways, autonomous vehicles, unmanned aerial vehicles (UAVs), etc., when a high Doppler shift can generate severe inter-carrier interference and degrade the overall system performance. In order to eliminate the undesired effects introduced by Doppler shift and improve the overall system performance, a recently proposed new modulation scheme named orthogonal time frequency space (OTFS) [1] serves as a promising technique that has been shown to be more robust in combating the Doppler effect experienced in doubly-selective wireless channels. In OTFS, the information symbols are modulated in the delay-Doppler domain, where the time-varying wireless channel can be regarded as almost time invariant in the delay-Doppler domain, within a stationary region. Additionally, OTFS modulates information symbol over two dimensional (2D) basis function that spreads across the entire time-frequency space, which in return can potentially achieve full time-frequency diversity [2]–[4]. One drawback for OTFS is that it cannot adopt a simple single-tap equalizer for the receive signals. In regards to this issue, tailored detection methods are thus required to be designed on the receiver side in order to unleash its full potentials. Several signal detection techniques have been addressed in the literature. To name a few, non-linear receivers such as the maximum a posteriori (MAP) detection in [5] can achieve the optimal performance, but at a cost of high complexity which increases exponentially with OTFS frame length. Other message passing based detections such as the message passing algorithm (MPA) in [2] has a relatively lower complexity but compromised the performance from MAP. The authors in [6] proposed a unitary approximate message passing (UAMP) that shows a superior performance. In [7], the author proposed a variational Bayes (VB) detector that can obtain the global optimum of the approximate message passing (AMP) regardless whether the graph is loopy or not. In [8], the authors considered a cross domain signal detection and exchange extrinsic soft information iteratively in different domains, which can also be exploited in our proposed system model. Other non-linear detectors also have been proposed by authors in [9]–[12]. Linear receivers such as LMMSE in [13] generally has a simpler algorithm. In [14], the authors applied block circular matrices properties to simplify matrix inversion. The authors in [15] and [16] applied the successive and parallel interference cancellation (SIC, PIC) for the MMSE signal detection.
