Hi all,
After seeing the measurements of the famous Topping PA5 amplifier, although I haven't seen an example yet, I thought, "I wonder if a class d composite amplifier can be made?" question stuck.
What do you think about it? Can an audiophile amplifier like the PA5 be designed with a nested feedback based on TI's PFFB TPA32xx structure and using low noise / distortion opamps?
I'm curious about your thoughts on this.
After seeing the measurements of the famous Topping PA5 amplifier, although I haven't seen an example yet, I thought, "I wonder if a class d composite amplifier can be made?" question stuck.
What do you think about it? Can an audiophile amplifier like the PA5 be designed with a nested feedback based on TI's PFFB TPA32xx structure and using low noise / distortion opamps?
I'm curious about your thoughts on this.
Yes definitely, although really tricky to get it to work stable.
Worked on a similar concept some years ago myself.
Worked on a similar concept some years ago myself.
Member @IVX presented a schematic for a composite using TDA8954 : https://www.diyaudio.com/community/...essimistic-review.277130/page-32#post-5527982
Since the LC filter lags close to 180* at high frequencies, it may require a more sophisticated feedback topology (multiple feedback etc.) than those used with linear amplifiers. However, it is still highly unlikely that a add a lot of gain can be added to the error-correction stage.
Thats interesting..
In TI's TPA32xx PFFB reference design its taken from after the LC filter. That ensures far better noise figure I guess.
No, no I wasn't referring to IVX's design in particular. I said that because there would be no point in wrapping up the whole amplifier in an op-amp loop if the filter is going to be left outside. So, yes, post-filter feedback was assumed.
I wasn't under the impression you were, rather I offered it as a counter-example to post-filter feedback. So @IVX's design is pointless in your estimation?
I'll leave that to you, abraxalito. I was under the impression that the purpose of the composite architecture was to correct the errors of the less precise Class-D stage, of which the filter is a part !
Dxvideo, as to whether you can build one, the answer is yes. But it will have to significantly improve performance in order to remain justified. Amplifiers with lesser phase shift at the output might work better.
Dxvideo, as to whether you can build one, the answer is yes. But it will have to significantly improve performance in order to remain justified. Amplifiers with lesser phase shift at the output might work better.
So if we leave aside the phase shift issue and consider the total class d amplifier (included input buffer opamps) as "one opamp", we can make a composite amplifier (which have a better noise and THD values), as I understand.
But the total performance of this system mostly depends on class d stage's phase shift performance.
Am I right?
1) The idea behind the composite is to make the performance significantly better than the bare Class-D amplifier, without introducing any shortcomings.
2) The primary preference is always functionality for which stability is necessary. Performance comes only later, after functionality is strongly established.
3) The total performance of a proper composite amplifier should be closer to the performance of the op-amp rather than that of the Class-D amplifier. Note that this op-amp and the Class-D stage are two separate parts, hence the term composite.
The phase shift of the Class-D amplifier needs to be small in order to prevent stability-related issues, so its closely related to functionality rather than performance. Though the possibility of such an amplifier is not NIL, it is strongly dependent on the skills of the designer. It is worth noting that a non-composite Class-D amplifier itself is difficult to stabilise if the filter is inside the loop. You will thus need to get really clever to make the composite idea work.
2) The primary preference is always functionality for which stability is necessary. Performance comes only later, after functionality is strongly established.
3) The total performance of a proper composite amplifier should be closer to the performance of the op-amp rather than that of the Class-D amplifier. Note that this op-amp and the Class-D stage are two separate parts, hence the term composite.
The phase shift of the Class-D amplifier needs to be small in order to prevent stability-related issues, so its closely related to functionality rather than performance. Though the possibility of such an amplifier is not NIL, it is strongly dependent on the skills of the designer. It is worth noting that a non-composite Class-D amplifier itself is difficult to stabilise if the filter is inside the loop. You will thus need to get really clever to make the composite idea work.
I am not a Papa of course.. 🤐
But if someone lead this kind of project then I can contribute as much as I can.
You could be setting off a lot of side-effects and 'fighting' the topology of the class D without realising it. A high damping factor is a popular assumption, but it's only known to hold true for old designs with constant switching.
Apart from high or low states, an H-bridge based output stage actually has the ability to allow the speaker to float or freewheel (both halves of the H-bridge are OFF), or to short-circuit the voice coil (both halves connected to GND or V+ ). To what extent that is implemented depends on the circuit and the control scheme. For instance, for low signal levels, you could let the VC float a certain % of the time, de facto raising the output resistance. I have not seen this done or documented anywhere, apart from the general idea being presented as power saving modes and reduced RF.
With enough processing power and the ability to detect the impedance curve of the speaker load, the system could be programmed to behave differently at different frequencies. Adding global feedback would defeat that and could introduce a situation of an "internally perfect" amplifier that makes speakers "more revealing" of all their flaws.
Apart from high or low states, an H-bridge based output stage actually has the ability to allow the speaker to float or freewheel (both halves of the H-bridge are OFF), or to short-circuit the voice coil (both halves connected to GND or V+ ). To what extent that is implemented depends on the circuit and the control scheme. For instance, for low signal levels, you could let the VC float a certain % of the time, de facto raising the output resistance. I have not seen this done or documented anywhere, apart from the general idea being presented as power saving modes and reduced RF.
With enough processing power and the ability to detect the impedance curve of the speaker load, the system could be programmed to behave differently at different frequencies. Adding global feedback would defeat that and could introduce a situation of an "internally perfect" amplifier that makes speakers "more revealing" of all their flaws.
Hi .Because normally d-class have output filter ,it will introduce phase delay ,lag . So for stability in composite design that must be minimal, which leads to very high switching frequency ,lets say megahertz range . Also , if you need to compare two signals , one of which is delayed , need to delay other too , maybe someone tried delay line or gyrator for that ?
As far as I know, most (but not all) Class D amplifiers use a passive second order (1 inductor, 1 capacitor) LP (Low Pass) output filter. Bridged amps generally use a second order LP filter per side.
There are other ways to make the RFI or EMI that ends up on the speaker leads and speakers less offensive - making the Class D driver chip's clock Spread Spectrum or rate-limiting the Gate drive to the output FETs. Clock dither creates phase noise. Since rate-limiting increases the time the FETs spend in the active region, it increases switching loss and related heating. I'd still include passive LP filtering.
As several of you have properly stated, a second order LP filter's phase lag is eventually asymptotic to 180 degrees. Anyone who has tried, or better-yet, struggled, to include even a single LC or LRC filter inside an inverse feedback loop, knows the pain. It's MUCH easier to take inverse feedback before the output LP filter.
A properly-designed passive, output filter will always present small resistive and reactive
losses. Purists should avoid the nonlinear or hysteretic (BH) effect of Ferromagnetic-cored inductors or the Piezoelectric effect of multilayer ceramic capacitors. Try even making a 1 millihenry air-cored inductor.
My DIY air core 250 microHenry AM Broadcast Band loop antenna (see attached photo) won't quite fit inside an amplifier. Besides, it radiates RF. Someone will complain.
If you're dead set on including the output LC network inside the feedback loop, at least include a lead term, to limit phase lag to somewhere between 90 and 135 degrees, for unconditional, or at least sometimes-conditional loop stability.
You'd think correcting the Class D's output LP filter delay would be easy. Inserting a phase/delay equalizing pre-filter before the audio enters the forward path of the loop seems a solution that doesn't destroy loop stability. The equalizing pre-filter would have to anticipate its own input, which may violate causality.
If the Class D's output LP filter is 3 dB down at, say 50 kHz, that ballparks at (2)(pi)(5E4 Hz) or ~ 314 kiloradians/second, which is [(1 second)/(3.1E5)] or ~ 3 microseconds delay. At the 1,080 ft/second speed of sound in STP air, couldn't you simply move the speaker forward (1/25)?? of an inch, to acoustically correct for the output filter's electronic delay?
Don't ever trust my math. I don't.
Ron
DIY 250 uHenry air core inductor. The RED thing with the GRN antennae, BLK nose and tiny eyeballs is the loop's mascot, a non-Ferrous Christmas candy cane:
Last edited:
See Gerald Stanley's patents for Crown Audio. There are patents going pretty far back using a "current dumping" approach whereby a small linear amplifier is bootstrapped by a PWM amplifier, and even further back using linear techniques.
I still don't get what the OP wqants to achieve. Is it
a) adding an additional post-filter feedback loop around an existing amplifier ? or
b) building a hybrid switchmode/linear amplifier.
a) is possible but the distortion reduction may not be as high as as the increased feedback factor would lead one to expect due to an effect called ripple-aliasing.
b) is also possible, one example is the aforementioned Devialet amplifier. Such amps are also known under the general designation SMALA (Switch Mode Assisted Linear Amplifier).
There was once an interesting paper by a German Prof proposing a feed-forward correction topology that is feeding the correction signal into the "cold end" (i.e actually using an artificial ground) of the output capacitor.
Regards
Charles
a) adding an additional post-filter feedback loop around an existing amplifier ? or
b) building a hybrid switchmode/linear amplifier.
a) is possible but the distortion reduction may not be as high as as the increased feedback factor would lead one to expect due to an effect called ripple-aliasing.
b) is also possible, one example is the aforementioned Devialet amplifier. Such amps are also known under the general designation SMALA (Switch Mode Assisted Linear Amplifier).
There was once an interesting paper by a German Prof proposing a feed-forward correction topology that is feeding the correction signal into the "cold end" (i.e actually using an artificial ground) of the output capacitor.
Regards
Charles
