This paper studied the application of minimum description length (MDL) criterion for estimating root-mean-squared (RMS) delay spread (RDS) for MIMO OFDM systems. The analytic relationship between the powers and the correlation matrix of multipath components established the feasibility of the application of the MDL criterion to RDS estimation. The estimator presented both the estimate of instantaneous RDS and the estimates of noise variance, channel power and SNR of current channel with low computational complexity. Given the powers of the estimated multipath components, the MDL criterion was adopted to acquire the number of paths and the time delays of each path of current channel without making eigendecomposition of the correlation matrix normally required by MDL criterion, following which the noise variance and the power of each path can be estimated. The power delay profile (PDP) and RDS of the current channel were achieved. Simulation results showed that the proposed estimator was insensitive to variance of SNR and robust against frequency-selectivity.
