Efficient Two-Dimensional Filters for Doubly-Dispersive Channel Estimation in Time-Frequency Signal Processing

We consider minimum mean squared error (MSE) channel estimators operating on the basis of time-frequency signal observations comprising mixtures of known pilot and unknown data signals. These estimators are useful for a general class of receivers resorting to a time-frequency signal representation of the received signal - obtained by, e.g., a filter bank - which have recently been considered in the context of reconfigurable radio. Taking advantage of the Kronecker product properties we derive two-dimensional filter formulations for doubly-dispersive channel estimation, which afford implementations with moderate complexities. Rank reduction and the use of fast Fourier transform methods are shown to offer further complexity reductions. The different estimator variants are compared in terms of MSE performance and complexity.