Interpolation and noise reduction in MIMO-OFDM - a complexity driven perspective

We propose a frequency sample interpolation scheme for MIMO-OFDM which can handle arbitrary distributed input samples and can be used for all standard OFDM schemes. We derive how the quality of the channel estimation and the resulting BER degradation can be controlled by the parameters training sequence length Lp and the number of measured frequency samples Dm. The performance improving procedure of interpolation or de-noising of the channel estimates is also feasible with real-valued samples. Furthermore, we compare schemes to calculate MIMO-OFDM filters in terms of absolute and relative computational complexity. For the MIMO detection we assume a linear zero-forcing filter which can be calculated very efficiently for all OFDM-sub-carriers separately or by exploiting matrix wise interpolation of the adjoints and determinants of the channel matrix and Cramer’s rule for the calculation of the inverses.