10/10/2013

Intervenant(s) : Silviu-Ioan FILIP (ENS de Lyon)

The problem of designing finite impulse response (FIR) filters is one that is central in many engineering applications. In this talk, I will discuss a novel way of quantizing such filters based on a powerful tool from Algorithmic Number Theory (euclidean lattices) and a classical algorithm due to Parks and McClellan. The results that we get are encouraging, for example in designing a length 221 low-pass FIR filter with 30-bit fixed-point coefficients, we were able to get a 12.16dB increase in accuracy over a filter obtained by naively rounding the coefficients obtained by using the algorithm of Parks and McClellan. This is the result of work done under the supervision of N.Brisebarre and G. Hanrot.