Finite wordlength design for FFT/IFFT in UWB-OFDM systems

Orthogonal frequency division multiplexing (OFDM) has been proposed for use in ultra-wideband (UWB) communication systems. In a UWB-OFDM transceiver, resource efficient FFT/IFFT hardware is a necessity due to minimal silicon area and low power requirements. Resource requirements can be reduced by using finite precision arithmetic in the FFT/IFFT algorithms. However, this introduces round-off and overflow noise, resulting in a degradation in BER performance. To address this problem, the finite precision arithmetic and wordlengths should be optimally designed at every stage of the FFT/IFFT. In this paper, we first present a mixed-radix FFT/IFFT algorithm for a UWB-OFDM transceiver with low multiplicative complexity. A round-off noise propagation model is derived and used to determine the optimal wordlength of the outputs and twiddle factors at each stage. The performance of the resulting system is compared to that with infinite precision arithmetic.