It's true that fc and fs are quite off from one another, but the reason behind it is to have a large transition width, therefore the rising slope will have a slow roll-off but lower N and minimized ripples. For example, making fc=20 would have given pre- and post-ringing, resulting in a settling time further away from what can be obtained with fc=1, not to mention much higher N. Also, I have noticed the fc being at ~15Hz, but I thought that has to do with the weighing sinc factor's argument, ft, the average between fc and fs. I forgot to mention that the FIRs in the previous examples are Kaiser, simply because I could specify fc, fs and Att. I am still trying to implement Dolph-Chebyshev in LTspice but the lack of an iterative sum makes this really cumbersome; I would have really liked to compare it to Kaiser with the given signal.
As for the humm induced by the mains, I thought of a notch filter, but it would have been useless extra computation, since I would have had to choose a large band, 45Hz - 55Hz, due to the maximum imposed drift. So, I thought I'll just go for one low-pass that does the job; a higher N than, say, an fs=80Hz, sure, but less than that one plus a Notch.
The FIR with N=40 would sure have a fast response, but I have doubts that it would have the proper frequency domain characteristics; so far N=50 seems to a minimum acceptable for a +/- ~2% ripple -- but this for the given signal. In the end, the main problem is that the input may have an even uglier waveform than the one chosen as an example. In practise, it will be a quadrature signal decomposed from the sensed mains current, multiplied with unit sin/cos and then added. So, similar to the example, but only Lord knows what harmonics it may carry. One may guess based on the existant apparatus, but still... And, as if this weren't enough, the DC level, representing the total active power, may vary from zero to ...whatever maximum can be imposed, and that DC level has to be followed as quickly as possible to avloid the "strain" on the DC link capacitor/inductor or a large, impractical value. Imposing conditions can be a good detective job
Well, as a conclusion, IIRs are bad except Bessel (ringing), analogs -- the same, FIRs only for low N, otherwise they're not good either because of the extra delay. Minimum-phase seem to promise but, from what I've read so far, the price to pay for the reduced delay is ringing, which is unacceptable. Adaptive filters are yet to be discovered, but it seems that they require quite some computation power. So, for now, a thank you for the guide lines, you did help me despite my fuzzy explanaions.
Thank you,
Vlad