6db LPF unstable?

Kassen · Posted: Sat Mar 22, 2008 7:29 am Post subject: 6db LPF unstable?

I've started to suspect the NM's 6db/oct LPF has a tendency to react to it's input dropping to a steady zero by (sometimes?) going unstable and outputting either a positive very small value or some sort of noise... I need to to return to zero as soon as the resolution gives instead, at least for CV applications.

Does anybody know more about this? The list archives seem kinda hard to search.

blue hell · Posted: Sun Mar 23, 2008 4:07 pm Post subject:

I don't think I ever heard anyone about this, so I guess you are the expert now Laughing

Kassen · Posted: Sun Mar 23, 2008 5:01 pm Post subject:

blue hell · Posted: Sun Mar 23, 2008 5:06 pm Post subject:

Kassen · Posted: Mon Mar 24, 2008 1:08 pm Post subject:

DrJustice · Posted: Sun Apr 06, 2008 5:29 pm Post subject:

A bit late, but: some filter topologies don't like too short word lengths (low resolution) in the internal filter states, which can lead to the filter getting into an unstable region when you feed it very low frequencies. Not saying that this is the case here, but you may want to test that with some very low frequency signals at a high amplitude that you cut off suddenly - then listen for noises.

DJ
--

Kassen · Posted: Sun Apr 06, 2008 6:31 pm Post subject:

DrJustice · Posted: Sun Apr 06, 2008 7:51 pm Post subject:

It's not the coefficients that are critical, it's the delay elements. I assume that those are 24 bits in the NM, and depending on filter topology that can be too little. (Again I speculate since I know nothing about what happens inside the NM)

I forgot to mention an important criteria (sorry!), and that is for this instability to occur, the cutoff needs to be low, leading to very low frequencies recirculating inside the (IIR) filter. I'm sure you can see where that leads; the recirculation over very many sample periods (ref. low f) means that numerical errors will build up, and if the filter states end up too far from what they "should be" - then the filter can blow up. 1Hz would indeed be considered a low frequency in this case.

Edit: Dang, that was a clumsy formulation - I guess I should rather be asleep now that write stuff like that... Embarassed

DJ
--

Kassen · Posted: Mon Apr 07, 2008 5:58 am Post subject:

Nah, I got it. The cutoff is indeed very low and indeed increasing the cut-off did seem to bring back stability, I'm fairly sure it's NOT "blowing up" though.

Do note I said it's the 6db filter, I kinda suspect a rounding error.