Quantisation effects in FIR filters: fixed versus floating point

Floating point implementations are more complex than fixed point (by a factor of up to three for typical hardware realisations), but are generally assumed to offer reduced quantisation errors. In this contribution a comparison is made between the use of fixed and floating point arithmetic for FIR filters. It is argued that in many applications the performance of the simpler fixed point arithmetic can equal or exceed that of floating point by means of a modest increase in wordlength