I'm interested in implementing an optimized 5x decimation stage. However it seems the resources needed are significantly more than 4x or 8x composed of cascaded 2x stages. For general purpose FIR filters the interpolated (not interpolation) structure makes sense, but from what I can see this efficiency may be lost when doing decimation. I'd be interested on any suggestions or theories about this. My take - for decimation with a normal FIR only every 5th output must be computed - so say a 100 tap filter could be time interleaved to compute 20 taps per clock. However for an IFIR, which could consist of a 30 tap and 15 tap cascade - when decimating the first filter must run at the sample rate - i.e. all 30 taps per clock, and only the second stage can be decimated - or an effective rate of 30 +15/5 = 33 taps per clock.
Is this correct - that IFIR cannot be efficiently used for decimation? Are there efficient alternatives for a 5x case?