The simple tweak for NFB

There was some discussion regarding pre/post filter NFB with all their advantages and disadvantages.

I once divised a scheme that allows for an intermediate soultion.

I haven't tried it out personally so far but from the math point-of-view and from simulations it should work. I do however not know if it is sonically interesting at all.

It doesn't show the load-independancy of frequency response that post-filter topologies do, but it is a little better than pre-filter NFB in this respect. The simulations show a behaviour a little like the ps-audio HCA-2 amp, maybe they are doing something similar.

The whole idea is based on a transient-perfect crossover that is used to take RF-content from before the filter and AF content from the filter output.
Application is easiest on class-d topologies that use both - a 2nd- order output filter and an inverting/summing integrator.

So which transient-perfect crossover is the most simple one ? The first-order crossover of course !

Stay tuned to see how it is done !


O.K. here we are again.

First the simple part:
We take the NFB from the filter input by a first-order high-pass filter. Because it is feeding the virtual ground represented by the summing node of the integrator we can use an RC series circuit.

The little more complicated one: We want to make a first-order lowpass out of our 2nd-order one. We will therefore have to multiply its output with a term Gain * (1 + sT). With our summing integrator this is again simple: We use a parallel RC circuit.

Below you will see the diagram for this. The original feedback resistor RF is substituted by the four parts RS, RP, CS and CP.
Both caps are of the same value, so are the two resistors.
The two resistors have the same value as the original RF. The caps are caltulated as

C = 1 / (FilterCutoffFrequency * 2 * Pi * RF)

The circuit works best if the output filter has a Q of 0.5 but it should be possible to tweak it for other Q values (which is dependant on load of course). In some cases a Zobel across the output might not be out of place.

There is also a possibility to have an intermediate solution between classic pre-filter NFB and this one. For this purpose RF is left in place but having a higher value. A simple example: Increase all resistors to twice the original RF value, redimension the caps to half and then you will have a 50%/50% classic-pre/mixed NFB solution. The ratio between the solutions can be changed at will by chosing the respective resistor ratios.

As already mentioned this is not a double loop topology but a mixture of pre- and post- filter NFB.




  • simple_tweak.gif
    9.1 KB · Views: 1,986
Charles: Thanks for this suggestion. I wired up a ZAPpulse module to see what happened. Your idea works fine, but it has a flaw, when you start it up without a load, the switching frequency will drop uncontrolled, and a lot of heat is generated in the output choke. Any suggestions?

When you start it up with a normal load, however it works just fine. Didn't listen to it though. THD seems to be unchanged compared to pure post-NFB.
Here is my first impressions: I first used that track 6 of Niels Lan Doky, Dreams album. I tried to listen different passages, and compare the sounds, but it appeared i was hearing two different tracks. Even if it was in fact the same track. But there was so much difference in the sound, that it was hard to believe that i was listening to the same.

The original ZAPpulse made a nice airy top, that really comes out in the room. But the new 'Phase Accurate' modified ZAPpulse, cleared up the room, and drawed a crystal clear picture of the high pitch instruments. The 3D effect is holographic! Some sounds where the floorbass turns slightly, 1:46 minutes into the track, this sound you almost miss with the original ZAPpulse, but with the modified ZAPpulse, it stands out, so you can't miss it! The Bass is right in front of you, while the rest of the band appears to be a few meters further back. Fantastic!
Bass dynamics seem to be untouched by this system, compared to pre NFB. Meaning: Good!

Now that was the good news.

The bad news is that female vocalists get what i call 'the Corrs effect'. This sounds like they are singing into a plastic bowl. Hollow in a way. It has nothing to do with the corrs, only that it is very obvious when you listen to the Corrs. (If you have this effect in your amplifier).

This was only my first impressions, i will continue the work with this and hopefully come up with something that is an improvement in every aspect.

If i measure THD it also seems to be more or less the same like before. (When the amplifier was pre filter NFB'ed).

I once saw an AES paper or patent the showed FB taken from both paths – or it might have been differing positions, not both combined at once – so I got thinking along the lines of Charles…

I played with a similar arrangement – DON’T use ceramic caps for C positions even COG - I made the early mistake of using COG ceramics - although they measured very well, they resulted in a VERY strong sonic signature - a hard edge to piano notes etc - even values like 47pF!!! I wonder if you’re experiencing a similar effect?

Charles suggests trying a 1:1 ratio in the first place, however with a good quality inductor, I found increasing the ratio in favor of Pre-LPF FB improved the audio quality, however insuring enough FB is taken after the inductor to reduce load dependency.

I would be very interested in your sonic evaluations with differing FB ratios splits.

Watch those Caps….



2005-01-09 9:51 am
remain RS, CS, RP and the integrator cap;
retain RF, CP.

this is my simpw mixed feedback scheme: RC integration feedback from switching stage & post-LPF feedback with positive phase shift.

It has a good square wave transient response, but I didn't know if it sounds well. I have no experience on HiFi.


2004-12-25 12:24 pm
Dear Charles,

Does your type of NFB arrangement could be implemented on Carrier Based non- self oscillating Class-D amps .

Hi lars,

Your ZAP module seems to be self-oscillating type, by the way what are your conclusions towards this type NFB with respect to the sonic perspectives...

Best regards,
This is not a double loop as used by B&O. It is much simpler than any double feedback loop topology, although it may not be as powerful in achieving load independancy and low THD as a double NFB loop does.
But it does give a little more freedom to trim the sound to ones liking and to get at least SOME post-filter NFB without much complicated math.

Chances are big that I am not the only one who came up with this idea and the Tripath one has indeed some similarities, although being much more complicated and targeted at RF circuits.
I still wonder why they don't use it on their audio amps.

As long as the first stage is a summing inverter (either just a summer or a summing integrator) the circuit can be used as is - independant of the general working principle (delta sigma, hysteresis controlled self oscillating, carrier based ,.......) as long as it isn't depending on output-filter phase-shift. For all other cases the principle could be applied as well but the circuit would be more complicated.
From the frequency-response point-of-view it could be improved by the use of some active filter(s). The latter could also be used to extend the idea to filter-orders greater than 2. But having active circuits in a feedback loop would be worse in terms of THD and IMD.


I made some more practical experimentation over the weekend. Replacing the capacitors with trimming caps, to more easily find the optimal working points for the Fc's of the feedback filter, and also change the pre / post filter ratio.

By changing the capacitance in the area 1 - 6 pF i could actually completely change the sound of the amplifier.

When more weighted towards post filter NFB, the midrange precision improved, but also the worse got the problem of vocals sounding like they are singing in a plastic bowl. I used Lucie Silvas to pinpoint this. Track 1. With pure post-NFB Placement improved, and everything sounds just great, until you start playing something with a voice in it, then you are not so pleased anymore.....
I try to keep the freq response at something like 100 kHz. I think the sound change when the pre NFB cap increases, because there is a phase shift in the top. (From post filter to pre filter feedback point).

It is very easy to change the ZAPpulse (any version) to operate in post filter NFB mode, so anybody can do that. And even move the feedback point up on the loudspeaker terminals. But i am not sure i want to recommend it, based on sound quality.

From here i will work on with some new ideas and try even more different fedback setups.