This paper investigates the Signal-to-Noise Ratio (SNR) performance of the Logarithmic Number System (LNS) representation against the SNR performance of the fixed-point representation. Analytic formulas are presented for the evaluation and the comparison of the two aforementioned representations, and the superiority of the LNS representation is demonstrated. It is shown that the base b of the logarithmic representation has a major impact onto the SNR performance, the SNR dependance on the variations of the standard deviation of the analog signal, and the memory requirements for logarithmic addition and subtraction. In addition, step-by-step procedures are introduced to compute the base b that optimizes all these performance measures.
Panagiotis D. Vouzis, Vassilis Paliouras