Semi-blind OTFS channel estimation based on DPS-BEM

In high-speed mobile scenarios like Internet of vehicles, UAVs, and LEO satellite communications, the large double dispersion dynamic range of continuous Doppler spread channels (CDSC) challenges traditional estimation methods by reducing accuracy and increasing pilot overhead. To address this, we propose a semi-blind channel estimation method based on a discrete prolate spheroidal sequences basic extension model (DPS-BEM). This approach embodies the concept of semi-blind estimation, which leverages a small number of embedded pilot frames alongside the channel’s dynamic characteristics to estimate and predict parameters, allowing most frames to be dedicated to data transmission, thus significantly reducing pilot overhead. A frame structure grouping pilots and data is first designed based on the channel’s quasi-static property. The channel is then modeled in the delay-Doppler (DD) domain using DPS-BEM, transforming high-dimensional parameter estimation into a low-dimensional problem of solving for basis coefficients. To capture the channel’s temporal evolution, a first-order time-varying auto-regressive model (TVARM) is constructed for these coefficients. Dynamic channel estimation is finally achieved using a particle filtering algorithm, which leverages Bayesian theory and the pilot information. Simulations on the MATLAB platform demonstrate that the proposed DPS-BEM-based scheme achieves superior channel estimation and bit error performance. It yields a gain of nearly 2 dB at a BER of $10^{-3}$ compared to traditional methods. Furthermore, with low pilot overhead, its performance is competitive with recently proposed sparse learning-based techniques.