ABSTRACT
This chapter demonstrates a number of optimizations that can significantly improve the performance of multidimensional logarithmic number system (MDLNS). It shows that migrating from a two-bit signed system to a one-bit signed system can halve the computation time required to determine the optimal base. The chapter presents two-dimensional logarithmic number system (2DLNS) approach, which does not need additional precision, but the two-digit system does require an accumulation stage to merge the results from the four separate processing channels. The improvements to 2DLNS with an optimal base as well as a one-bit sign target applications where traditional number systems such as fixed-point and floating point binary as well as logarithmic number system (LNS) are used. The LNS offers the best filter performance with 14 or more bits in the fractional exponent; however, this choice may come at the expense of a larger circuit for performing binary-to-LNS-to-binary conversion with 14 fractional bits as well as native LNS addition/subtraction.
