Thanks for the detailed response!
I have a few comments, if that’s ok.
Thanks! I wondered how those were implemented, and both modes are quite useful. The new peak mode has already proven helpful at quickly identifying high-frequency signals (especially when added to low-frequency signals), and average seems to produce a smoother output than ever.
I agree. However, while I’m not a mathematician, I think the sinc filter may be overkill.
en.wikipedia.org/wiki/Low-pass_filter
Simply weighting the current value with the previous one seems relatively cheap, and very similar to RC filters. A fourth order low pass filter is more than enough for typical audio and biosignal applications. An eighth order filter provides extreme attenuation. For high pass filtering, a single order is often enough. Deep bandpass/stop functionality seems possible with the same IIR filter design.
The properties of most unwanted signals lend themselves to such simple processing quite well. 60Hz noise is the most common annoyance, and it should be easily eliminated with a single-order notch filter. Selecting tones from a noisy environment (eg. amateur radio, power line communication, EMG detection), can be done with a single order bandpass filter. High frequency switching noise (>1khz) can be thoroughly suppressed with relatively low-order low-pass filters. For very low frequency signals (eg. biosignal baseline wandering, infrasonic audio, etc), a single order high pass filter can provide arbitrary cutoff frequencies with strong attenuation of the unwanted signals.
Simple notch filters should be trivial and are are especially important. Implementing a 60Hz notch filter in hardware is virtually impossible. Even using a UAF42 chip (precision thomas-biquad filter) with purpose-ordered precision resistors, 60Hz attenuation is limited to about 5-7x.
A drop in sampling rate while using filters is completely acceptable. 100KS/s should be plenty, far less may be acceptable for many applications.
Switching firmware on the fly is rather easy, especially since user settings seem to persist. I would elect to drop SD card support, calibration, cursors, and perhaps some measurements (duty, pcnt, pwdt) for specialized firmware versions. Also, while I did hear there are no plans to develop your latest firmware as open-source, does that exclude a GCC port?
I tried the FFT firmware, and if it worked well, I might have been happier. However, FFT plots are often noisy and hard to read; that one was no exception. Frankly, filters or waterfall plots are far more useful. Filters are usually best, since the resulting waveform is useful in its own right.
The nano has quite limited capture capability, making this a really slow process. Far, far, far too slow for lab use.