Paralleled Class A with Class D

[QED047]

In the solid sate forum I started a topic exploring the use of a switch-mode current sink to load a simple Class A follower....
Switching current source for Class A?
On page 3 of that topic Steven introduced me to a thesis written by Ronan van der Zee:

Originally posted by Steven
Ronan van der Zee showed a similar idea in his thesis, see chapter 6.6.3 (page 96). You can download the complete thesis here: http://doc.utwente.nl/14149/1/t000000d.pdf
or check this: http://purl.org/utwente/14149.

Steven

One topology (the parallel amp) discussed in the above particularly caught my eye and has prompted a detour from the Class A current sink. It seems to go by several different names - I'm not sure what (if anything) it is called around here, and even if it counts as Class D ~ but my best guess is it belongs in the Class D forum.

The original goal I had in mind was to reduce the copius dissipation in a single-ended transistor Class A follower by loading it with an inductor switched to ground and returning the unwanted energy via a flyback diode to the supply. My self-oscillating (hysteretic) control system kept things nice and simple - another key overall objective mine. But this only cut the dissipation in half. Still wishing to retain an analogue signal path (hence not transforming the signal into PWM) I couldn't help being intruiged by paralleling a linear output with a switching half.

So now I have this parallel amp already crudely breadboarded. The lashup I made over the weekend consists of a mosfet follower stage biased to mid-rail and driving a high impedance (100 Ohm) load. An 8 Ohm speaker connects to this via a current sense resistor straddled by a comparator. This generates a bipolar error signal effectively flagging when current is flowing into, or out of, the follower - which really only serves as a voltage reference to be copied at the output. The real "heavy lifting" is acomplished by a half-bridge switched by the error signal and ballasted to the speaker via an inductor.

Attached is a schematic used to simulate the circuit although I have used a different bridge driver to the one shown. The principle of operation is really simple, and it certainly works. Of course the switching frequency is not constant and there are numerous trade-offs in terms of bandwidth and distortion. But the attraction to me is that there is a continuous analogue signal path all the way to the load - the Class A stage could also be a tube stage or AB or whatever but I just love the "no feedback" simplicity of a single transistor follower.

Do we think it would be possible to refine/optimise this configuration to achieve a worthwhile amplifier? Or is it only "all very well in theory"?

[kenpeter]

Have you considered modulating the current source on the basis
of drain current detected above M1? This drain current would be
amplified directly of any error voltage seen under the source. And
if the loop is tight enough, becomes an effective constant current
except what variations demanded by the signal into the load.

In the tube world, the suggested topology might be called a
White Cathode Follower... Search on those three words if you
need the history.

The end result should be mostly the same as the split sense
modulated switching source we discussed in the prior thread.
Except the sense resistor is out the way of the load and less
detriment to damping.

Now you can directly modulate the pulse width of the source
without regard to additional feedbacks, as M1 performs double
duty as the only voltage comparator you will need.

[stocktrader200]

I paralleled a class D and class AB amplifier with good results a few years back. Class A is not required as Class D has no crossover distortion at the audio output.
The input signal may have to be clipped to 5% below maximum amplifier output (limit 100w amp to 95w) to prevent to two amps from fighting each other at the clipping point.

[QED047]

Originally posted by kenpeter
Have you considered modulating the current source on the basis
of drain current detected above M1?

Thanks - I'm not sure but I think that was also covered in the above paper. I'll do the research you suggested.

Originally posted by stocktrader200
I paralleled a class D and class AB amplifier with good results a few years back. Class A is not required as Class D has no crossover distortion at the audio output.

"Good results" sounds promising! When it comes to the linear half of this setup I'm trying to be minimalist - I've got this nutty thing about being able to "see" a direct path for the signal (in the spirit of the perfect wire with gain).

I've already got a Tripath rig that I seldom ever listen to because I much prefer the sound of my Musical Fidelity amps (A1 and A120). But these are a nightmare to feed and care for because of the dissipation. I want something that sounds like the MF but will run all day, every day without cracking up. I'm already close with my home-brew ZEN-like follower loaded with a switching constant current sink (I may well promote this from the breadboard to a finished article) but having had a degree of success with simple parts like this, I'm curious to see just how far I can go with it!

[QED047]

As a first step towards reducing the ripple I have added an error amplifier. The ripple is now around 60mv and, as far as simulation goes, is spread out at around 300 ~ 500KHz...



Note the even-order harmonics preserved in all their glory! The results are fairly repeatable on my breadboard, but the switching speeds are a bit strangled by the only H-bridge chip I have to hand (L6203). A LTspice simulation of the circuit incorporating an error amplifier is attached.

[Wavebourn]

I did not build A+D, but A+C works like a charm. I made such amps, also there was a commercial Quad design using such an approach, it was called Current Dumping in the patent.

[payloadde]

Indeed, when I saw your proposal, P.Walker's current dumping from 1975 came to my mind, too.

http://quad405.com/currentdumping.pdf

It's A+C though, and the A is BJT and linearized by heavy negative feedback. In a secondary article the possibility of A+D is discussed, albeit just theoretically.

http://quad405.com/jaes.pdf

I fancy, the balanced bridge could be of use somehow in your project, too. To couple the A and the D part.

I have a question though concerning your project:
let's assume the A be a flawless amp, how accurate can the D follow ? For relaxed listening the output may be just 1 volt or so. If the D should follow this with no less than 0.1% deviation, the error signal would be a mere 1 milli volt. Can the error amplifier handle this ? Or is this a mute point because 5% iss good enough ?

[QED047]

Originally posted by Wavebourn
I did not build A+D, but A+C works like a charm. I made such amps, also there was a commercial Quad design using such an approach, it was called Current Dumping in the patent.

Phew, for a minute there I thought you meant Quad had dipped their toes into Class D! Well, I suppose it is fairly similar if you think of the inductor in my circuit as a ballast ~ preventing an instantaneous current rise as the "heavy duty" transistors are switched to the rails rather than used linearly.

The important (to me) differences between typical Class D and this kind of amp are becoming clearer to me as I play around with it. I'm really quite taken by the results I've got from the initial design (just using a crappy LM393 and H-bridge for motor driving!). By making the Class A section into a modest but fully-fledged amplifier, and reducing the current sense resistor to around 0.1 Ohm the filter section doesn't seem to be all that critical. By this I mean it all seems happy driving a wide range of loads and there appears to be no need for any overall feedback. The reaction of the switches to the current error is the only loop and it appears to be quite well behaved (perhaps it's stabilised by the Class A drive?). Photo of lash-up below.

[QED047]

Originally posted by payloadde
[B]Indeed, when I saw your proposal, P.Walker's current dumping from 1975 came to my mind, too.

Great info on the website quad405.com Many thanks for those links!

I have a question though concerning your project:
let's assume the A be a flawless amp, how accurate can the D follow ? For relaxed listening the output may be just 1 volt or so. If the D should follow this with no less than 0.1% deviation, the error signal would be a mere 1 milli volt. Can the error amplifier handle this ? Or is this a mute point because 5% iss good enough ?

I hadn't thought about this until you mentioned it... although I did wonder about disabling the bridge in the absence of signal (naturally it idles at 50% duty cycle). Maybe this would help.

[QED047]

OMG this is too easy: I'm sat here in almost total disbelief ~ I have a delicious Class A sound emanating from a handful of parts and tepid heatsinks! The bridge chip is supposed to run motors for goodness sake.

I haven't trusted it it anything other than a battle-hardened Tannoy Lynx (3169G - 100W 8Ohm 12" high sensitivity Dual Concentric) so far - but it's producing sounds every bit as musical as my MF amps are capable of (on a good day). The Tannoy's yield 95dB @ 50 - 20kHz for 2.8V so my diminutive 35V supply isn't holding me back from a reasonable audition.

The next step I think is to swap the bridge chip for a pair of IRF540Z's and I'm thinking of letting an LTLT1160 do the driving. I can't see this being any slower than what I have now: "180ns Transition Times Driving 10,000pFn - Adaptive Non-overlapping Gate Drives Prevent
Shoot-Through" Sounds useful. The obvious downside is the 60V top-rail limit. ABS. Max is quoted as 75V ~ in a "protective environment" (i.e. with varistor clamps on the supply rail) I might just stretch it to 70V to get 75Watts into 8Ohms. Even at 60V I will have more poke than the A1 and about the same as the A120...

...Mumble now I'm wondering how to get to 100W to give the Tannoy's something to really chew on.