Adcom 535 MkII

I picked up this little beauty at an audio swap meet.  The original Adcom 535 was designed by Nelson Pass.  The MkII version added triple darlington outputs (AFAIK a plus) and nice double banana binding post outputs (replacing a proprietary speaker connector).  This is the 535 to own!

But why does one need an Adcom at all?  Well, I wanted to check out the early Pass MOSFET designs.  For that purpose, this is a failure.  This is not a MOSFET amplifier.  Instead, the 535 is one of the last commercial bipolar amps designed by Pass.

This is pretty close to Nelson's ultimate "simplicity" thing.  The Adcom bipolar amps are about as simple as a bipolar amp can get and still be objectively good.  To get still simpler, you need MOSFETS, which came in the next generation of Pass designs for Adcom, the ones with two zeroes.

Now, they do have an IC implemented servo (fine) and also a capacitor input (belt and suspenders?) using a polycarbonate cap (about the best you're going to get at this price point).  A large polypropylene would be better but cost a significant fraction of everything else.  Most amps with capacitors use electrolytics, which are far worse electronically than polycarbonate.

John Curl also designed his first Parasounds with IC servos and got (probably undeserved) flak from Stereophile about it.  David Rich wrote an article about this story.  So Curl redesigned the 2200 using a discrete servo circuit.  Adcom wasn't scolded in this way and kept their original IC servo design which is fine.  (I'll have to check if my HCA-1500A has IC servo.)

I measure 87 Watts into 8 ohms, which I think is about spec.  Clipping is calm.  Response is down 1.65dB at 100 Khz, the 3dB down point is 125kHz (this is traditionally called the "bandwidth") and the 6dB down point is 192kHz.

Power is short of what I need for Acoustats, but possibly sufficient for Revels and my musical instrument speaker.

But what this looks best for is a replacement or backup for my supertweeter amplifier.

For the supertweeters, I'm currently using an ATI 1502, which I'd long planned to replace with another Pass design, the F-5 Class A, but who knows when I'll get around to that.  The ATI is curiously bandwidth shy for a supertweeter amp, rolling off at 60kHz IIRC.  But otherwise it was nicely quiet and had low power consumption which was another plus.  The Adcom seems to have about twice the high frequency cutoff, and I don't need the extra power the ATI has (150W) for the tweeters.  The Adcom also seems lower in power consumption than many amps I have (despite claims of "high bias").

It will be tested on Acoustats however, for the record.

But possibly not without bias adjustment.  At 10V output I measure 0.026% THD+N.  That's perfectly fine, though my much older and higher power Dynaco 410 measures better at 0.014%.

But at lower levels, the distortion on the 525 II explodes.  At 1V output, I measure 0.28% THD+N, more than ten times higher.  At 0.3V output it's 1-2% THD.

[See update below.  It's apparently ripple.]

It's not from noise.  The A weighted noise is a mere 22uV, less than half what the Dynaco has.  The 80kHz filter makes hardly any difference (so I left it out).  Strangely, though, the 400 Hz filter made a significant but not decisive difference.

This indicates crossover notch.  The actual crossover notch recorded on the difference output is a very hashy and asymmetrical distortion..  There is none of that at all in my residual.

If my untouched unit has a bias error, what about others?  This amplifier has a reputation for "hard sound."  Well that's exactly what crossover notch is going to do.

I will be checking the bias to see what's going on.

It might be interesting to see if I can hear the distortion too.  Not sure if I want to bother.

Well, sadly there is no room for improvement by increasing the bias.  For one thing, the bias was exactly at spec already (or slightly high in the right channel at 8.5mV instead of the 7 mV specified level).  Both channels measured almost exactly the same in THD+N.  I tried increasing the bias to 20mV, and it made no difference.  Distortion at 1V output still measured 0.26%, higher than I think what a good amplifier should have.

Furthermore it seems to be an awful hashy looking distortion associated with the low voltage peaks.  I'm still assuming this is some kind of weird crossover notch distortion (but maybe not?).  It is not some innocuous looking 2nd harmonic.  Nor is it some ultrasonic ringing.  Applying the 80kHz filter had almost no effect.  Strangely, applying the 400 Hz filter cut the distortion in half.

I still have considerable concern this is due to my flakey analyzer.  (But it measures 0.003% distortion in bypass.)  I'm thinking of bring the Dynaco back to the bench and re-affirming that it measured amazingly low distortion at 1V and lower.  I might also try different analyzers, better grounding, etc.

I am also seeing similar low level distortion which more clearly looks like crossover notch with the Nikko Alpha III that I have also tested recently.  I believe I deliberately lowered the bias on that amplifier and am planning to check if restoring the bias fixes it.  That was the amplifier that the staff at Audio Dimensions once preferred to all other amplifiers.  It was one of the first commercial MOSFET amplifiers.  But the fact that two amplifiers have shown a weird problem does raise the possibility my test setup is flawed.

The Adcom service manual says that before setting the bias, the amplifier should be heated up by playing 20W for 20 minutes with the cover on.

Starting with the quiescent bias at 8mV, when I ran the amp with cover on for 20 min at 20W  (I actually turned the cover upside down and blocked the sides, because removing the Adcom cover is very difficult, it requires removing all the rear connections first), the bias had at first fallen to 3.66mV as soon as I could measure it.  As the heat sinks cooled off, it rose to 5 mV in 2 minutes, 7 mV in 9 minutes, and more slowly after that finally back to 7.5mV in about 20 minutes.  The Adcom manual says to set the bias when it has "stabilized."  But what is stabilized?  Does that mean as soon as the meter can be read clearly at the highest heatsink temperature?  Or does it mean waiting 20 or more minutes until the heatsinks have cooled down to quiescent level--and in that case why bother to heat them up in the first place?  Or somewhere in between, such as around 9 minutes where's it's barely changing but still slowly falling?

I asked this question on DIYAudio and Nelson Pass himself replied.  He said he designed this amplifier, like most others, according to the 10 second rule.  So the heatsinks heat up to the point where you can keep your hands on them comfortably for 10 seconds (and no more, but he didn't say that in this message).

I was jazzed that Nelson Pass himself replied to my query!

He didn't actually answer my question (and perhaps he didn't know the answer) but what he did give me was a license to experiment!  In fact, any tweak could adjust the bias (it's the big obvious pot on the board) without even using a voltmeter, just hands.  And in fact what they might well do is crank it all the way up because it takes a long time even then just to get warm.

I figured that would probably be OK in an amplifier designed by Nelson Pass and in fact it seemed to be in this case.  I cranked the bias all the way up and at first it measured 19mV at the test points.  Then, with the cover "on" (faked as before) I let it run for 90 minutes (actually, I rechecked a few times to be sure it wasn't getting out of hand).  At 90 minutes the bias voltage had fallen to about 14.5mV.  The heatsinks were now as warm as I could touch with a flat hand for 10 seconds, and no longer.  If you wanted maximum performance this is probably how you would set it.  I might turn it down just a bit for better longevity.   (Anyway, see the update below, I not going to sweat the exact bias setting now...  For now I turned the bias back to 8mV quiescent matching the other channel for now in case I never get back to this project again, as has happened, ouch, so many times before.)

With hours of operation at full on bias (but no signal) in my simulated "cover on" condition, the (unreliable) temp I measured with my IR thermometer was 130F, same temperature as I set my Aragon 8008 BB to.

(I think I adjusted my IR thermometer for heat sinks somehow...at least on the older equipment I look at the numbers seem reasonable and reliable.  Nowadays it works horribly on everything else.  Or maybe it's not actually working well on anything.  I have the feeling that older heatsinks get dusty enough that the high reflectivity of aluminum is not so much a factor anymore.)

In cover off 20 minute testing, 7mV quiescent yielded 93F, 8.5mV yielded 100F.

UPDATE

Finally I am understanding what my Rigol scope is showing me about the distortion output from my ST 1700B analyzer.  The hashy looking "distortion" appears to be related to the power supply at 120 Hz or maybe 180 Hz.  It stays at that same frequency regardless of the frequency regardless of the frequency of the signal the Adcom is amplifying.  So it appears what I thought was "rising low level distortion" is actually power supply ripple.  It apparently does not affect the output much when there is no signal, so I got the very low 22uV noise measurement.  But when the amp starts amplifying a signal, the ripple appears and stays more or less at the same level regardless of output, being swamped by higher outputs.

Meanwhile, the actual distortion, represented by the small wiggles, is actually very small at 1V and below.  This is not a bias problem, it's a power supply problem.  I've ordered two new Vishay 6800 uF capacitors which are rated at 10,000 hours.  These super premium parts appeared to be the best I could get at Mouser to fit the space in the Adcom.  Most filter capacitors are rated only for 3000 hours, which is typical of standard grade electrolytics.  3000 hours was the best I could find now in Rubycon, the brand of the original capacitors in the Adcom, and even they required special order.  (BTW, that means 3000 hours of torture at max temperature and current, so typically it results in 5-10 years of actual use in nice circuits.  10,000 hour capacitors should last 20-30 years in nice circuits.)

Now between the amplifier and just two replacement capacitors I've spent all the money I made selling my Denon DVD player, but I'm getting practice refurbing amplifiers, which is something I have to look forward to for many other amplifiers more important to me now at some point in the future.

After replacing the caps and doing whatever else is needed to fix the apparent power supply problem, I can measure for sure how much the distortion actually changes at low levels as the bias is increased.  Perhaps, in the end, there will be little reason to keep it higher than "7 mV".