combining class A and class D

rcownie

Member
2007-11-16 7:37 pm
I had a probably-crazy idea which kept me awake last night.

The idea is to combine the efficiency of class D with the purity of class A
and get almost the best of both.

Roughly, you build two class D output stages which generate roughly
Vout-2V and Vout+2V. Then you use these as the negative and positive
supply for a floating class A output stage.

Now, the class D stages can be pretty rough because they only have
to be close enough for the class A stage to remain properly biased.
And the class A stage can be simple, and can use a transistor with a
low Vds spec. and really fast response.
And most importantly, while the class A stage is passing
all the current, it only has a small voltage drop (e.g. 4V) and thus only
dissipates relatively little power. For example, to get 200Wrms into 8ohm
we need 5Arms, or a peak 7.1A., and a supply voltage of about +/-58V,
so we would have cut the class-A dissipation by about a factor of
(2*58)/4 = 29x.

I'm hoping the very different requirements for the class D stages would
allow the use of much smaller inductors (and perhaps higher frequencies -
obviously the class D stage has to respond quickly enough to maintain
slew rate) thus reducing the EMI problems. But I haven't worked through
that yet.

Apologies if this is an idea that's already out there - I did a search
but didn't find anything quite the same.
 

lumanauw

diyAudio Moderator Emeritus
2002-10-04 1:27 am
Bandung
www.facebook.com
Hi, rcownie

It's really a nice idea :D
The idea of classD amp being made PS is already there as a product, LabGruppen amps. (ClassTD)
I imagine your idea equals biasing LabGruppen amps to classA (originally classB).
Problems will rise when the input signal is quite high frequency. LabGruppen amps seems to limit their response to 15khz or so, while their TD supply is clocked at 600khz. There's something about switching frequency and highest reproducible frequency.
 

rcownie

Member
2007-11-16 7:37 pm
thanks

"It's really a nice idea"

Thanks!

I'm going to play around with some simulations and see if it there's some
gotcha ...

I'm probably most nervous about the need to be really careful that the
two class-D outputs stay in the right range (and roughly the right
relationship).

It seems as though the easiest way to guarantee that would be to make
it self-oscillating, with comparators to check the various conditions
and then some (fairly simple) logic to get each supply going in the
right direction to make things better. But as that involves some delay
(in the comparators, the logic, the transistor switching, and then the LC
response) and a feedback loop there's a danger of instability ...

And your comment about dealing with high frequencies is certainly on
target: if the signal has a high slew rate then the supply amps have
to be able to respond quickly enough before the class-A stage hits
the edge of the operating window. The faster they can respond, the
narrower the voltage window can be (and thus less dissipation in the
class-A and higher efficiency).

I guess I'm starting to understand why class-D has a few pitfalls along the
way.

One especially nice feature - if I can make the whole thing work - is that
in principle the class-A stage, if it's fast enough, should keep all the
high-frequency class-D artefacts completely off f the speaker cable.
That must be a good win from an EMI point of view.

Of course the same topology would work using two class-B amps for the
varying supplies (and that would be asy). But then that seems fairly similar
to class-AB in both quality and efficiency, which is perhaps not so interesting.
 
You are overestimating the performance of a linear output stage operating on only +/-2V. Also you are neglecting the contribution to distortion and powrer losses of the error and voltage amplifiers and driver stage required by such a linear output stage.

Actually a 300Khz self oscillating system can exhibit almost no phase shift at 20Khz.
 

rcownie

Member
2007-11-16 7:37 pm
Eva said:
You are overestimating the performance of a linear output stage operating on only +/-2V. Also you are neglecting the contribution to distortion and powrer losses of the error and voltage amplifiers and driver stage required by such a linear output stage.

Thanks for your comments

The +/-2V figure was just a guess. And I'm mostly thinking MOSFETs (N-chan)
so the gate drive could be well above the +2V allowing low Rds, e.g, 0.2ohma
and thus sufficient current. But if it turned out to need 10V instead of 4V that
would still be a big win.
MOSFETs also ease the driver problem. And as for the voltage gain, the LME49810
seems a very adequate solution, though doubtless an expert - which I'm
definitely not - could suggest alternative approaches.
 

GK

Disabled Account
2006-01-08 4:35 am
rcownie said:


Thanks for your comments

The +/-2V figure was just a guess. And I'm mostly thinking MOSFETs (N-chan)
so the gate drive could be well above the +2V allowing low Rds, e.g, 0.2ohma
and thus sufficient current. But if it turned out to need 10V instead of 4V that
would still be a big win.
MOSFETs also ease the driver problem. And as for the voltage gain, the LME49810
seems a very adequate solution, though doubtless an expert - which I'm
definitely not - could suggest alternative approaches.



I'm actually building a high performance ~1ppm THD-20 1kW per channel class A amp using tracking rails. I started the thread on this project on the SS forum, but I haven't been updating it with progress reports due to the fact that the thread was a magnet for keyboard warriors and BS artists.
I'm using a class AB amp to drive the tracking supply however, but some points of relevance:
A class A linear output stage (either biploar or MOSFET) with enough output devices will have better open loop linearity than just about any class D output stage (<0.1% THD easy).
Tracking rails further enhances the linearity of the output stage due to both the lack of non linear base current (gate voltage) modulation via the collector-base (drain-gate) capacitance [which varies in a nonlinear fashion with Vce (Vds)] and slope distortion/transconductance variation.
However, you still need a reasonable supply voltage for the class A output stage. Below about 5V Vce (Vds) the collector-base (drain-gate) capacitance of power BJT's (MOSFETs) goes through the roof and device fT drops dramatically. I'm running +/-7.5V as a compromise between fT and power dissipation.
Also, the above mentioned capacitance is a real bother when it comes to the class D switching component (and its harmonics) on the supply rails modualting the base current (gate voltage) and appearing on the output.
A pair of modern RET BJT's will have a Cob of about 500pF for the PNP and 250pF for the NPN. Perforated emitter technology BJT's such as the MJL21193/mjl21194 pairs have double the Cob (1nF/500pF). MOSFET's have similar capacitances. This makes the output stage far from a pefect shield against class D switching artifacts on the rails.
A driver stage designed to have a very low high fequency output impedance (powered from clean, fixed high voltage rails) is mandatory here to address the problem.
 
The carrier residual being fed through Miller capacitance itself is not the problem, carriers are not a bad thing, the problem is that Miller capacitance is highly non linear, and in the case of bipolars, current gain and Vbe are also strongly dependent on Vce in the low region and they get modulated by residuals in a non-linear fashion.

I prefer the plain class D approach. You can get half power THD in the 0.0xx range with proper design, and you can build 200W amplifiers with a pair of TO-220 MOSFETs without heatsink :cool:
 

GK

Disabled Account
2006-01-08 4:35 am
Eva said:
The carrier residual being fed through Miller capacitance itself is not the problem, carriers are not a bad thing, the problem is that Miller capacitance is highly non linear, and in the case of bipolars, current gain and Vbe are also strongly dependent on Vce in the low region and they get modulated by residuals in a non-linear fashion.


Carrier feed-through is a problem if (for the class-A stage) you want to employ conventional linear design techniques utilising large amounts of global negative feedback and the highest unity loop gain crossover frequency (conflicting with an adequate phase margin) for adequate loop gain at 20kHz to get a degree of linearity and IMD performance typical of a well designed conventional class A amplifier.
If you are not going to address this issue, or design along such lines, then you might as well forget about bothering with a class-A add-on for you class D amplifier altogether because the net result will be a very (if at all) marginal improvement indeed.
And as I already mentioned, in a rail tracking amplifier the nonlinearly of the collector-base (drain-gate) capacitance is largely mitigated due to the fact that the ratio of variation of Vce (Vds) and the degree of Cob variation is minimal (compared to a conventional amplifier) - provided that Vce (Vds) is kept above the threshold at which fT rapidly degrades.
 
The idea was eventually implemented in the Audio Physic Strada mono blocks some years ago. I don´t know what happened to these amps that were quite expensive and got were good reviews.
I remember that they used silk spun litz wire for the output chokes though as they claimed the sound of the amps benefited from it...;)
http://forum.audiogon.com/cgi-bin/fr.pl?aamps&1022672181&openusid&zzDinos&4&5
 
lumanauw said:
Glen, EVA,

Thanks for the input here.

For efficiency, what do you think about classG such can be seen on TDA7294 datasheet? The dissipation is about halved, without the difficulty of making classD part (classD is not easy :D)

"Chopping" the rails in that way is not particularly advantageous in terms of distortion or circuit complexity. In the end it's not that hard to go the plain class D way.
 
this subject was the title of a couple of AES papers in Munich and amsterdam in the late 90's when Karsten Neilson was doing the Mecc papers, IIRC people from phillips and a japanese concern were presenting on it.
it all seemed to die a death due to the increased performance of direct class d and digital in amplifiers with bitstream to delta sigma or pwm direct conversion in dsp.
I will try and dig the papers out, maybe some of the problems are solved with faster and better fets available now?
 
I notice that it's been a while since the last post here, but I guess some progress has been made in exploring ideas/desings concerning Class A-D desings. I recently started reading about it.

If anyone by chance has a complete schematic of the Audio Physic Strada digital mono amp, I would be very interested. I contacted them today, but the project has been abandoned so they could concentrate on their core business; speakers. They also do not have the schematics of the Strada anymore, sadly.

Any other similar or improved design combining the best of both worlds is also welcome.

Thanks in advance.
 
IMO, Eva completely covered the topic by stating how simple a class D amp can be while giving you adequate (if not similar or better) audio performance. Why in the world would you possibly want to design 3 (OK, 2.5 power converters instead of just 1?!? Sure, it was a cool idea many years ago but class D has advanced since then and this idea is just plain outdated. I wholeheartedly agree with Bruno when he said something to the effect that anyone who can design a good class D amp can design a great linear amp. So why not just cut right to the chase? Again, just my opinion.

But don't let me deter you as a learning experience. There's a lot of good stuff to learn by doing a project like this, if you're up to it.