We propose a two-stage channel estimation technique for high rate ultra-wideband communication system. In the preamble-stage, we perform initial channel estimation using least square approach followed by frequency-domain smoothing operation. Estimated channel at this stage is then windowed by a rectangular window of Hannan-Quinn criteria-optimized length. In the payload-stage, estimated channel from the previous stage is further refined by applying time- and frequency-redundancy in a decision-directed manner followed by smoothing and windowing. Link level simulation shows that the two-stage method with optimized windowing provides a good estimation performance close to that of the minimum-mean-square-estimation (MMSE) scheme with significant reduction in computational cost.