Pilot-Symbol-Aided OFDM Channel Estimation in Doubly-Selective Channels

Nowadays, most wireless communication standards utilize Orthogonal Frequency Division Multiplexing (OFDM) as their modulation technique. For coherent detection, the performance of such systems depends strongly on the accuracy of the channel estimation. One possible estimation technique is Pilotsymbol-Aided Channel Estimation (PACE), which allows the reconstruction of the channel by means of interpolation. In this master thesis, I derive a closed-form expression for the Bit Error Probability (BEP) of an OFDM system that utilizes two-dimensional PACE. I assume Rayleigh fading, Gaussian noise and a linear, but other otherwise completely arbitrary, interpolation. For a Signal-to-Interference Ratio (SIR) larger than the signal-to-noise ratio, simulations confirm the analytical results. However, for a lower SIR they exhibit small differences because the Inter-Carrier Interference (ICI) is not Gaussian distributed, violating my assumption. Indeed, analytical calculation of the ICI probability density function (pdf) verifies that even for infinite many subcarriers, the pdf does not approach a Gaussian distribution. It is further shown that the well-known Minimum Mean Squared Error (MMSE) estimation also minimizes the BEP and that, for certain assumptions, the optimal 2D interpolation can be performed in an equivalent way by successively 1D-1D interpolations of the MMSE pilot-symbol estimates. A numerical example then compares different interpolation methods (optimum, linear, spline, and natural neighbor) in terms of BEP. Finally, the analytical BEP is validated by real world measurements, utilizing the Vienna Wireless Testbed in combination with a Rotation Unit, allowing repeatable and controllable measurements at high velocities. 1 in the sense of a straight line