I never understood the fascination of FNDR - the "not in the signal path" heuristic is a obvious fail
the input common mode Z of op amps is nonlinear - and its outside of the feedback loop's ability to correct
on the plus side you can get lower Q sensitivity to op amp GBW and reduced component value spread with the higher number of amps per section
but multiple feedback low pass biquads are well protected from high frequency DAC glitch/image/error rectification at the op amp input by the RC lowpass up front in the loop and don't have any Vcm swing at the op amp input
the input common mode Z of op amps is nonlinear - and its outside of the feedback loop's ability to correct
on the plus side you can get lower Q sensitivity to op amp GBW and reduced component value spread with the higher number of amps per section
but multiple feedback low pass biquads are well protected from high frequency DAC glitch/image/error rectification at the op amp input by the RC lowpass up front in the loop and don't have any Vcm swing at the op amp input
Last edited:
kinku:
maybe this will help.
there are many different filter topologies.
FDNR was popular in CDPs and DACs for a while for some of the reasons jcx mentioned, but eventually fell out of favor also because of some of the other reasons jcx mentioned.
so, the bottom line is you might be better off using a different filter topology.
keep in mind that generally speaking, op amp circuits have better distortion performance when used in inverting configurations where the + input is grounded (minimal common mode swing). FDNR filters don't do this.
but, this is DIY after all, so it is always up to you ...
mlloyd1
maybe this will help.
there are many different filter topologies.
FDNR was popular in CDPs and DACs for a while for some of the reasons jcx mentioned, but eventually fell out of favor also because of some of the other reasons jcx mentioned.
so, the bottom line is you might be better off using a different filter topology.
keep in mind that generally speaking, op amp circuits have better distortion performance when used in inverting configurations where the + input is grounded (minimal common mode swing). FDNR filters don't do this.
but, this is DIY after all, so it is always up to you ...
mlloyd1
my usual suggestion in answer to this question is to find some products you have heard that you like, study their filters and copy/modify as appropriate.
also, jcx gave a hint if you wanted to pursue that path:
"... but multiple feedback low pass biquads are well protected from high frequency DAC glitch/image/error rectification at the op amp input by the RC lowpass up front in the loop and don't have any Vcm swing at the op amp input "
regarding further questions on FDNR based filters, a google search on something like "distortion in FDNR filters" links to some very useful reading material, some of which is even understandable by non-EEs.
All that being said, I have fond memories of my long gone DX-5700 cd player that used FNDR filters (after PCM58 dacs). With cleaned up grounds and an improved clock, it sounded pretty good to me.
best regards,
mlloyd1
also, jcx gave a hint if you wanted to pursue that path:
"... but multiple feedback low pass biquads are well protected from high frequency DAC glitch/image/error rectification at the op amp input by the RC lowpass up front in the loop and don't have any Vcm swing at the op amp input "
regarding further questions on FDNR based filters, a google search on something like "distortion in FDNR filters" links to some very useful reading material, some of which is even understandable by non-EEs.
All that being said, I have fond memories of my long gone DX-5700 cd player that used FNDR filters (after PCM58 dacs). With cleaned up grounds and an improved clock, it sounded pretty good to me.
best regards,
mlloyd1
This whole thing is basically GIC filters. In my experience(which is not trivial), GIC filters NEVER sound as good as they are meant to, and never as harmless as they appear. I have upgraded MANY players and dacs that used GIC's, trying all manner of opamps, way-overkill power supply support, etc., and in EVERY CASE, the sound was better with the whole GIC section disconnected. I am firmly of the opinion that the whole concept is fatally flawed. I believe even BurrBrown, who were pushing GIC's really hard for a while, suddenly stopped showing them in the app circuits of their dac chips' datasheets, I think either when they brought out the PCM1702 or PCM1704, after which they never showed GIC's again. I am not an NOS guy, so maybe they do more good than harm vs the needfully-aggressive alternatives, but I would think that GIC's problems would be FAR more audible with the filter freq so very close to the audio band.
Thanks Abrax, I think I have read this paper (it may even have been you that pointed me to it before), and it did help me somewhat originally, but I think I also had other (breadboard spaghetti) layout issues as well. I should re-read it now that I have a proper pcb. I hope that I have the solution now (just a bigger compensation cap) but I didn't get a chance to try yet. maybe tonight. The oscillation is interesting, in that it doesn't show on the output (except as noise where the signal is significantly attenuated) but it shows up quite prominently at the inverting inputs of the opamps (where they are tied together). In my case approx 1Mhz and about 1V p2p! It wasn't till I started poking around different parts of the circuit I found it.
Kinku, I found a filter calculator for butterworth filters up to 10th order, I can see why the suggestion of cascading third order filters instead though. a ninth order FDNR will require many opamps and they may not play well together. Cascading FDNR's (possibly staggered in frequency a bit) may work as each FDNR can be a single 3rd order filter block.
I still think that the slope you are after is impossibly steep. It's doable by the nyquist freq but not by 21.1Khz that I can see.
I'm all for treading the path less trodden, sometimes though you will find that there is a good reason it is so (as alluded to by various posters above). Occasionally though you may discover something worthwhile. It's more about the journey and the learning than the end result. If the end result is good all the better, if it is not, you have hopefully learnt something along the way
I also did the google search suggested, and found nothing negative. low distortion is actually one of the touted benefits of the FDNR. For my application, the biggest benefit is that the FDNR allows me to translate directly a passive filter to an active one, for a speaker crossover that is very useful, allowing me to use the optimization feautures of passive crossover design tools to hit my target response and then simply implement that electronically.
For your situation where you are dealing with standard slopes, that doesn't apply. Other filter topologies may actually provide better objective performance, where it comes to the filter slope continuing to attenuate into the MHz rather than just the 100's of Khz (whether this is important or not I don't know, once you get down to a certain level of attenuation I suspect not).
Tony.
Kinku, I found a filter calculator for butterworth filters up to 10th order, I can see why the suggestion of cascading third order filters instead though. a ninth order FDNR will require many opamps and they may not play well together. Cascading FDNR's (possibly staggered in frequency a bit) may work as each FDNR can be a single 3rd order filter block.
I still think that the slope you are after is impossibly steep. It's doable by the nyquist freq but not by 21.1Khz that I can see.
I'm all for treading the path less trodden, sometimes though you will find that there is a good reason it is so (as alluded to by various posters above). Occasionally though you may discover something worthwhile. It's more about the journey and the learning than the end result. If the end result is good all the better, if it is not, you have hopefully learnt something along the way
I also did the google search suggested, and found nothing negative. low distortion is actually one of the touted benefits of the FDNR. For my application, the biggest benefit is that the FDNR allows me to translate directly a passive filter to an active one, for a speaker crossover that is very useful, allowing me to use the optimization feautures of passive crossover design tools to hit my target response and then simply implement that electronically.
For your situation where you are dealing with standard slopes, that doesn't apply. Other filter topologies may actually provide better objective performance, where it comes to the filter slope continuing to attenuate into the MHz rather than just the 100's of Khz (whether this is important or not I don't know, once you get down to a certain level of attenuation I suspect not).
Tony.
Last edited:
Thanks JCX, my filter theory is very lacking, I rely on calculators. The math does my head in.
Also my comment above once I look at it is not correct. a 9th order FDNR implementation would only use 1 more opamp than three cascaded 3rd orders (which doesn't sound like it would work anyway). Kinku mentioned an appnote that suggested cascading 3rd order filters which is why I commented on it.
Tony.
Also my comment above once I look at it is not correct. a 9th order FDNR implementation would only use 1 more opamp than three cascaded 3rd orders (which doesn't sound like it would work anyway). Kinku mentioned an appnote that suggested cascading 3rd order filters which is why I commented on it.
Tony.
Now you mention it, I think it probably was. My poor memory! - apologies for being like a stuck record. At that time I was also looking at transforming my passive filter into an active one and the FDNR route was the path of least resistance, at least initially. But then I found another topology which used fewer opamps and without the stability issues, but it required a ground-up redesign of the filter.
Now I reckon FDNRs are best kept for low frequency work where there's plenty of opamp GBW available. In addition non-linear common-mode input capacitances are less of a worry at LF. Opamp power supplies will probably the limiting factor as regards SQ (which for me these days means dynamics).
As I said, I'm not an NOS believer, and also not an opamp believer(anymore). My approach now to digital playback is oversampling into ladder dac, then resistor i-v and discrete jfet or tube voltage gain stage. I've found that, with 8x OS, just the gate/grid capacitance is totally adequate to give what little LPF is needed.
The thing with GIC/FDNR is that they measure great with repetitive waveforms, yet sound bad, just like a 1980's yamaha power amp. Sine & squarewaves just don't show you how the circuit actually behaves(doesn't behave) with music.
Tony the app note said cascading work better than creating an Nth order without cascading due to GBW limitation.
It seems like I have to live with the first image after 20KHz and obtaining a steep cutoff within that narrow passband is difficult.
Abrax what is the low frequency that you mentioned the FDNR are good for? Will less than 20K qualifies?
So far AADE filter design program seems really good for passive filter design. Which one is the new program you found Tony?
Abrax is using a passive design with Chebyshev filter that can cut steeply, but there will be phase issues and ripple in the pass band.But they are the one that give the optimum of everything I needed for this application.
As far going with NOS ,sorry if that is the case then there is no need for this discussion and there is plenty to found too. Just want to try the NOS sound and hey if I don't like it ,I believe you. But want to make sure I did everything possible before I do the A-B comparison.
Hi I just found an online calculator which did up to 10th order butterworth. Nothing special I don't even know how well it works (and based on the sim below I suspect it is out by a reasonable margin). I simulated a 9th order passive and fdnr, and they gave quite similar results. Neither looked particularly inspiring!
what is very interesting is how pathetic the passive filter is if you put in some realistic parasitics. The curve shown is with unrealisticly low DCR for the coils and ESR for the caps (with NO esl) Just enough resistance to tame the resonance (at the unfortunate frequency of around 40Khz).
I think you are best of looking at app notes for NOS dac chips and seeing what they recommend. I've not looked at this stuff before but it is becoming apparent very quickly why oversampling was introduced!
Tony.
I don't even know how well it works (and based on the sim below I suspect it is out by a reasonable margin). I simulated a 9th order passive and fdnr, and they gave quite similar results. Neither looked particularly inspiring! what is very interesting is how pathetic the passive filter is if you put in some realistic parasitics. The curve shown is with unrealisticly low DCR for the coils and ESR for the caps (with NO esl) Just enough resistance to tame the resonance (at the unfortunate frequency of around 40Khz).
I think you are best of looking at app notes for NOS dac chips and seeing what they recommend. I've not looked at this stuff before but it is becoming apparent very quickly why oversampling was introduced!
Tony.
Attachments
I used this one to design my NOS passive filter - works very well up to 11th order : Chebyshev Pi LC Low Pass Filter Calculator
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.