Quote:
Originally Posted by
pepar
where I can read about the shortcomings like the limited impulse response of FIRs?
The shortcomings of course depend on the application.
FIR is by definition finite impulse response filter (for example
http://en.wikipedia.org/wiki/Finite_impulse_response). The memory FIR has (the delay line) equals the number of filter order + 1, the result of the filter is a sum of the values in the delay lines times their respective filter coefficients. Any sample that is fed to FIR can only affect the result while it is in the delay line.
There are plenty of applications that FIR suits well. One good thing about FIR's is exactly that "finite", because possible calculation errors do not stay in the system for long. DSP's are also very efficient in calculating FIR's (multiply-accumulate instruction, two data fetches per cycle, loop-hardware). But if you need narrow filters, or good suppression in the stop-band, or you require good ripple characteristics on the pass-band, the filter order needs to be increased, which in other words increase the length of the impulse response. The narrower filter bandwidth or the steeper the transition bands, the longer the required impulse response. After a certain point the length of the FIR becomes a memory and calculation bottle-neck, and it also introduces delay to the signal (normally the filter order divided by two).
In these cases IIR (infinite impulse response) is usually the answer. The filter can affect the signal for a much longer time with fairly little calculation, although the calculation must be performed with better precision than with FIR.
Of course I may have interpreted something in the wrong way in the promo sheet, as it really does not say much about the technology. They use "sophisticated FIR-based methods", but FIR itself doesn't make them sophisticated. Anyway, that's marketing.