A Simple 10W Direct-Coupled Class-A Power Amplifier

[JR.]

Thanks everyone for posting.
I've enjoyed the comments.

THD Residual Analysis

I did some THD analysis of the residual using a conventional HP-334.
The measurement just below clipping, 0.04% THD, (1 kHz, 8R, 22V p-p) is similar to Visual Analyzer's FFT-based calculation.
Below the clipping level the THD measurement from the HP-334 isn't useful since 0.1% is the lowest scale.
The HP334's analyzer output remains very useful.

Near the clipping point with +/-15V rails


Dual Class-A Power Amp, 8 Ohm Load, Just Below Clip with 15V Rails, 10Vdiv output versus THD Signature. THD is approximately 0.04%.

Note that the THD signature above is predominately third harmonic and, when driven into non-linearity or clipping, produces compressive distortion.
At lower levels the distortion products drop rapidly.

At -10dB


Dual Class-A Power Amp, 8 Ohm Load, at -10dB "Just Below Clip," 1Vdiv output versus THD Signature which is also scaled.

The HP-334's "auto" function hasn't worked in years.
The residual scale was also increased to expand it.
You can see quite a bit of fundamental lurking in the above waveform since I am nulling by hand and it's very touchy.

Wayne, that looks like it should sound very good.. 0.04% should not be very audible.

It might be interesting to compare THD at 10 kHz to same conditions at 1kHz... For a standard amplifier relying upon NF the 10kHz distortion should be a bunch worse. If that distortion is just from IxVbe is should be similar at HF (a good thing).

I vaguely recall an esoteric Hifi amp that added an global error correcting circuit to a similar [edit] not really similar amp but was also open loop with separate error correction path [/edit] open-loop amp. IIRC their amp was inverting, so they separately subtracted the output, from the input though resistors feeding a virtual earth summer, and fed the result from that null as a correction signal into the open loop amp.

Kind of a long convoluted way to add simple NF but the hifi tweaks were excited about NF that wasn't really NF or so they thought.

JR

PS: My cooker makes around 200W max, so your amp is not quite there yet...

[mediatechnology]

Thanks John. I'm close to hooking the speakers back up.

I just checked 10 kHz at 22V p-p into 8 Ohms (7.5W) and its 0.076% and predominately third harmonic.
The generator residual at that level and frequency is 0.038%.

[JR.]

My old diy amp (designed in '70s), still working in my living room measured around 0.25% at 20kHz.... FWIW to measure very low distortion in power amps under load you need to use special non-inductive load resistors.

JR

[emrr]

I ASSume class T is based on the Tripath chipset. I was not a fan of those guys when they gave their pitch to Peavey back in the '90s. I asked them reasonable amplifier questions (like duty cycle) and they didn't understand my questions, or didn't want to answer (it appeared like they didn't understand).

Yup, duty cycle does not compute, at least with the provided power supply and a test burst. It gets plenty loud with music. I wonder if a larger PSU capacity improves anything meaningful, or if it's some other mechanism that overloads. I don't know enough about it to guess. I'd assume by now they'd have figured many more fine details out.

[mediatechnology]

Doug - Do you suppose that the toneburst is perceived by the TriPath chipset to be a turn on pop and suppressed?

Added schematic to first post.


Schematic of a Simple 10W Direct-Coupled Class-A Power Amplifier Using a DRV134, 2SA1962 and 2SC5242.

[emrr]

Hmm, hadn't thought of that, could be. It'll tolerate 2 but not 3, and I can see the power LED dim at each burst, then switch off completely.

[mediatechnology]

I had a chance to listen to it yesterday around dinner time and really liked it.
The HF detail and imaging were still there regardless of level.
No grit.

For a comparison I also measured the THD at the 10V peak level of the generic 35W amp. (6.25W 8 Ohm.)
It was about 0.25% THD, predominately third and without switching artifacts, versus the <0.04% THD of the class-A "heater" at the same power level.
At lower levels it dropped some, but not much, and there was a lot of noise in the residual.

What was interesting was the clipping behavior of the class-A circuit.

To monitor the transfer function I put the input into CH-A and the output into CH-B of the 'scope and switched it to X-Y.
This displays linearity and clipping is very obvious due to the change in the line slope.
Until the clipping point the line is a razor sharp diagonal.
When it clips, the slope abruptly changes.
Since the clipping point with +/-18V supplies was a little too loud, I reduced the supply to force clipping.
What I heard was smoother than what I thought I would hear.

So there may be something about the "softness" of Class-A clipping compared to other amp classes.
It does round the waveform before it hits the wall.
Maybe it's the open loop output going Ohmic when saturated and not having feedback attempting to cope with overload.
Maybe it's not having the load part of the feedback network.
I really don't know.

What I gather from this is that 10W/channel may be enough for these speakers and if really pushed into clipping won't sound that bad.
I can monitor at low, medium or high level without grit.
The only downside is that it takes 75W to get 20W.

[JR.]

I had a chance to listen to it yesterday around dinner time and really liked it.
The HF detail and imaging were still there regardless of level.
No grit.

Sounding good is a good thing.. not sure what "grit" is.. back in the day I used two-tone IMD (19/20kHz) to parse out subtle linearity issues that could be audible.

For a comparison I also measured the THD at the 10V peak level of the generic 35W amp. (6.25W 8 Ohm.)
It was about 0.25% THD, predominately third and without switching artifacts, versus the <0.04% THD of the class-A "heater" at the same power level.
At lower levels it dropped some, but not much, and there was a lot of noise in the residual.

0.25% at 1khz is pretty poor, at 20kHz not that unusual. If it is 0.25% at 1 kHz, and a conventional NF amp, it will likely be a lot worse at 20 kHz, or conversely 0.25% at 20kHz will generally clean up at 1kHz (where our hearing is more sensitive). While this is why I liked the two-tone IMD test because it stresses HF linearity, that can audibly sound like grundge (another technical term, but grunge has less HF content, more LF content than grit ).

By switching artifacts are you talking about crossover distortion (spikey looking in THD product)? Class G or H amps can have switching related artifacts when they change PS rails that couple back through the output device Miller capacitance, and I guess an amp with switching PS can have clock frequency noise in the distortion product too. (I suspect your 35W amp uses none of the above.)

What was interesting was the clipping behavior of the class-A circuit.

To monitor the transfer function I put the input into CH-A and the output into CH-B of the 'scope and switched it to X-Y.
This displays linearity and clipping is very obvious due to the change in the line slope.
Until the clipping point the line is a razor sharp diagonal.
When it clips, the slope abruptly changes.
Since the clipping point with +/-18V supplies was a little too loud, I reduced the supply to force clipping.
What I heard was smoother than what I thought I would hear.

So there may be something about the "softness" of Class-A clipping compared to other amp classes.
It does round the waveform before it hits the wall.

Not to be more pedantic than normal, I suspect the clipping/OL behavior is dominated by the open loop output stage, not the class A bias. Class A amps can also use tons of NF and will then clip as suddenly as any other NF amp.

I like to characterize amp recovery from clipping, like how a race car handles after bumping the wall at speed. Important to race care drivers, not so much for careful drivers. At Peavey and in the sound reinforcement industry clipping power amps is more normal than not. That's why pretty much 100% of Peavey amps were designed with clip limiters that managed the behavior when overdriven to sound relatively benign (like a little extra compression).

Maybe it's the open loop output going Ohmic when saturated and not having feedback attempting to cope with overload.
Maybe it's not having the load part of the feedback network.
I really don't know.

I vote for door number 1, while prudent (amp) design anticipates clipping and engineers in fast recovery. One common technique used in power amps is to use anti-saturation clamp diodes in the driver transistors to prevent them from hard saturation. A saturated driver transistor charges up the base region, and this charge needs to be dissipated again before the transistor returns to normal operation. Clamping it to prevent saturation, allows near instantaneous recovery from clipping. Besides the driver stage any-where within the NF loop that loses the NF control when the output saturates can and will charge to some erroneous level, that it must again recover from after the clip event passes before the loop is stable again.

I suspect your front end is well behaved and recovers quickly from clipping, Your output stage not only goes ohmic, but probably is ohmic all the time. While that output behavior changes when one of the two output devices saturates. That might look like a 6dB gain drop between first output clipping and both clipping.

What I gather from this is that 10W/channel may be enough for these speakers and if really pushed into clipping won't sound that bad.
I can monitor at low, medium or high level without grit.
The only downside is that it takes 75W to get 20W.

This design looks interesting for low power and may even be a merchantable esoteric audio product, if you have the stomach to deal with those customers (I don't). I wonder if there is a way to reduce the low distortion even further by adding an AC signal component into the class A bias port. This class A tweak needs to be FW rectified to increase the class A current in both extremes... Of course it should sound fine as is, and the extra complexity will detract from it's elegance that such customers will embrace.

JR

[mediatechnology]

By switching artifacts are you talking about crossover distortion (spikey looking in THD product)?

Yes. And it was absent in the gritty one.

Grunge is also what I mean by grit.
Just dirty - like HF leakage through a pot.
Grunge is probably the better word for it.
(Though I tend to think of "Grunge" more as a genre.
Maybe "Grungre" would be the grunge genre. Did I invent a word?)

Your output stage not only goes ohmic, but probably is ohmic all the time.

I think that it is. While in the linear region maybe approximated by Re/2 plus some change?
What's also invisible are the series base resistors, 50 Ohms, hiding inside the DRV134.

Of course it should sound fine as is, and the extra complexity will detract from it's elegance that such customers will embrace.

Thanks for the compliment.
It seems ultra simple to make from something I already had.
People seem to like the transparency of the headphone amp.
radiance was one of the first to build one that I'm aware of besides me.
I dunno how many people that have bought the boards have actually used one for a line amp but I do sometimes hear from those that have used it for 'phones.
Roger uses it in the FCS Class-A to drive transformers.

The bigger output devices kick up the power up a few notches.
Glad we can get TO-3P devices with good gain and BW.
BTW it slews around 10V/uS with the DRV134 which is a little more than I get with the THAT1646 at around 8V/uS.

[emrr]

So there may be something about the "softness" of Class-A clipping compared to other amp classes.
It does round the waveform before it hits the wall.

Not to be more pedantic than normal, I suspect the clipping/OL behavior is dominated by the open loop output stage, not the class A bias. Class A amps can also use tons of NF and will then clip as suddenly as any other NF amp.

I frequently run class A tube preamps lacking feedback loops into a semi-flatlined waveform when recording electric guitar, and (by accident) human voice occasionally. It acts like compression, and you usually can't hear it. You can use those preamps on drums like limiters, with a definite hard ratio ceiling.

I got a 1932 condenser mic running recently and if you have it 6 feet out from a drum kit, the preamp flatlines loud portions of the waveform, no audible distortion to the ear. In that one the lack of much over 6kHz certainly helps too!

Once you get into mid-1950's class A circuits with 25dB or so NFB, they clip like anything else, kinda nasty and immediately obvious.