A novel approach to the design of 2-D linear-phase FIR filters with finite-precision wordlength is presented. The key idea of this approach is the “shaping” of the desired impulse response within time constraint imposed by the filter specification. This approach, named IRS (Impulse Response Shaping), minimizes the objective error function in logical steps by utilizing a property of Fourier series coefficients and a process similar to binary search operation. While searching for the global minima, IRS gradually modifies the error function in order to escape the local minima. The simulation results show that this method always yields much smaller maximum deviation than that of a rounded version of the infinite-precision coefficient filters
Published in:
Circuits and Systems II: Analog and Digital Signal Processing, IEEE Transactions on
(Volume:41
,
Issue:
7
)
Date of Publication: Jul 1994