— The Complex Logarithmic Number System (CLNS) uses log/polar complex representation. We propose CLNS for the 128-point Fast Fourier Transform (FFT) in Orthogonal Frequency Division Multiplexing (OFDM) demodulation. We show the cost and quality of the FFT depends on the CLNS base b. Memory requirements decrease as b increases. Experimentally, when optimized for Signal-to-Noise Ratio (SNR), b > 2 results in memory reduction up to 36% compared to b = 2. Dependence of the SNR on the input’s standard deviation variations is weaker for the CLNS than for fixed-point and decreases as b increases.
Panagiotis D. Vouzis, Mark G. Arnold, Vassilis Pal