Reduce LPF phase distortion

No such thing as a free lunch.

If you can, oversampling will reduce the filter requirements. If you can produce your 20-10KHz signal at 200KHz then a pretty simple >10KHz filter will take care of the ultrasonics.

If the output is computed, compute it for smaller time increments. If it is recorded/encoded then resample it at 200KHz (trivial: 20x repeats of each value), digitally filter the result with the ~10KHz low pass filters of your choice and use a simple analog filter at ~10KHz at the output.

A 9-pole elliptical is a PITA in analog, but a much smaller PITA as a digital algorithm.

So you have over/undershoots or even ringing at the output. This can be avoided with a Bessel characteristic of the lowpass. This filter characteristic is especially useful in digitally driven circuits with short rise/fall times.

There are a bunch of switched-capacitor filters out there, but switched-capacitor filters can have aliasing problems. The one I was thinking of is an all-linear one. I was a system engineer on the project that it was used on, and there was a whole bunch of stuff going on around me, so I only paid attention to the fact that it existed, not who made it.

Do you know what poles and zeros are in a circuit design or dynamic systems context? If so, an unstable zero is simply a zero with a positive real part (or who's magnitude is greater than one in a sampled-time system).

If you _don't_ know about poles and zeros and the Laplace domain, an unstable zero makes a system respond to an input with more than the minimum amount of phase shift necessary to get a certain amplitude response -- that's a circular definition given my original assertion, but I can't think of a better way to put it. A system with a single large unstable zero will respond to any initial push by going backwards from the direction it ends up going in response to the push.

In general terms, what is the preferred design approach for minimizing phase distortion when using a second order active LPF?

In this case it is a PWM audio signal from a PIC micro.

Thanks for your reply.

James

What are "unstable zeros"?

That's probably the one I was thinking of then -- IIRC it was a Linear part and it was kinda unique.

Otheres will certainly talk about the analog side. You have a PWM signal from a micro and that may open up another opertunity. If the signals are being made by a PIC, you may be able to shift the phase of some frequencies in the digital side. This could let you take out some of the nonlinearities in the phase, have a sharp cut off and not add stages.

In article , Tim Wescott wrote: [...]

Linear makes one where the frequency can be set to N*10KHz up to 150KHz.

The LT1564 I suggested is continuous time.

Not confusion, just terminology. "Unstable zero" = "Zero in a place where a pole would be unstable". I think it's mostly from control systems usage.

Yes, an unstable zero doesn't make the system unstable, it's just a zero that falls outside of the region of stability for the domain at hand.

Aren't you confusing unstable with non-minimum phase? Last time I looked in Oppenheim and Schafer, zeroes outside the unit circle don't make a system unstable; they make it non-minimum phase.

This isn't what makes a system unstable; the criterion for instability is that the impulse response doesn't vanish for large times after application of the impulse. The polarity of the initial response isn't what determines stability.

I see. You're trying to confuse us filter theory guys.

If you just called it a RHP zero, we'd know what you meant. :-)