Another high DC Adcom GFA-555

Hi Fred,
As Steve mentioned, setting up a differential pair isn't too difficult. There is the diff pair matcher I posted here before (somewhere). As for thermal tracking, put a touch of thermal compound between the two transistors, then use some heat shrink tubing to hold them together. This also creates a thermal isolation barrier between the input transistors and the environment.

I have included diagrams for the NPN and PNP channel matchers. These work great with J-Fets as well. In fact, i am currently matching some 2SK170 J-Fets right now.

What you do is preselect transistors into rough gain / transconductance groups. From those groups, install pairs that look close and let them settle in. Measure from one collector / drain to the other. The smaller the number in millivolts, the better your match is. Notice there are selectable tail currents available. Test them close to the actual circuit current and supply voltages. I bought 1% resistors and selected matched pairs from those. These determine your accuracy, so don't use carbon film trash here. The tail current generator is fine for carbon film types.

I have also attached a detail photo of one diff pair I had done a while ago. You can see how this pair came together pretty clearly. This pair was installed in a Marantz 240 power amp. It responded well to this bit of detail work.

Hi Steve,
Curve tracers take far too much time to use on the bench for service. The advantage the jig I put together is that it removes temperature as a variable. That is unless the pair is badly mis-matched, in which case you don't want those anyway. Variable case temperature was always my biggest problem in trying to get reproducible results when testing transistors.

-Chris

I disagree on the curve tracer being a time waster. Look at all the time spent on this one amplifier. Happily it appears to be working (YEA) but I contend that given a bunch of transistors of the same type I could have them sorted into gain / voltage breakdown groups less than 30 second each and if stored in groups, you don't need to do it again. This amp was one of the legendary 'tough dogs' and I believe it would have been a much shorter story with the curve tracer. I don't use it frequently but when I do it gives a wealth of information you just can't get any other way. You don't make any 'assumptions' about the devices in terms of leakage and hfe since you can see a whole family of curves at a glance. Your leakage tests were one point on a graph at high voltage in this case and the hfe was at a very low voltage. You have to 'assume' its behaving properly in between and almost always it does but when it DOESN'T, how will you know?

I confess I don't work on consumer gear any more but I used to and once in a while the curve tracer sorted out a 'dog'.

 
Hi Steve,
Curve tracers take far too much time to use on the bench for service. The advantage the jig I put together is that it removes temperature as a variable. That is unless the pair is badly mis-matched, in which case you don't want those anyway. Variable case temperature was always my biggest problem in trying to get reproducible results when testing transistors.

-Chris

I wasn't using the curve tracer for bench work. I used it to sort parts just as stratus46 suggests. Once it is set up, it is just as fast as he says. I never tested less than 200 transistors at a time. You can get some really good matches with all coming from the same batch.
 
I would not test with a load with only one output pair. And I would still have the 100 ohm resistors in at this point in case there is anything is strange with the front end. Actually, I would turn the bias down and see if you can get a full voltage sign wave out, with the 100 ohm resistors still in - keep them in until the output stage is fully populated and tested with bias. Only do full voltage out testing at 100 Hz so that you do not fry the output RC network, do not go deep into clipping.

Next, I would put in one or two amp fuses and run the bias up to the 16 mV, and measure the voltage on each of the emitter resistors to see how they current share. You'd want to have all the outputs installed at this point. Try the full output test again, no load, 100 Hz only so as to not fry the output network.

Next, would be the regular fuses and testing with a load.


Tonight's progress:
Repopulated the output boards by replacing Q17 with the MCM 424 and reconnecting all the case screws. Swapped out the drivers for MJ 32/33. They were an okay match at 122/142hfe. Much closer than what was in there, anyway.

I was able to go full voltage with 16mv bias, no input without any issues. Current limiter resistors still in the fuse holders.

I let it idle 10 minutes, cover off. It was drawing .15 amps and I measured -35mvdc offset at the speaker out.

I think things are still looking good! 😀

Tomorrow I'll try passing a sine wave through with bias on and no load.

Assuming that passes, I'll install fuses and test the current sharing and post those results. I assume I'm looking for numbers that are very close, but how close?
 
With the amp idling at full voltage, current limiters still in, no input, no load, I'm seeing -59mvdc at the input. That doesn't seem right?

I spotted it on the scope when I noticed that the input wave form was amplifying when I turned on the amp.

Is it an issue?

Edit: I'm guessing maybe not, since the other channel is doing it too. Just seems weird to me.

I was able to put my 500mv 100hz sine-wave through, no load, without issue. I got the same scope traces as before, when just one output was connected.

So now I'll go ahead and put in fuses and test the current sharing of the outputs.
 
Last edited:
Remember I told you there would be a very small Ib current for each transistor in the diff pair? Let's work the problem backwards and I'll go from memory you can correct me if I get anything wrong. The input pair has a 2ma current source, which means that Ie is 1 mA for each input device - assuming perfect balance. This is a reasonable assumption since you matched the input pair beta. I recall you measured beta (Ie/Ib) as 400 and we can calculate Ib as:
Ib = 1mA/400 = 2.5 uA
This current passes through the 22.1K base bias resistor from which we can calculate Vb:
Vb = 2.5*10^-6 * 22100 = .055V or 55 mV

I'd say that is normal.
 
Thanks for that, Pete. I think I follow. Do pre-amps block this "reverse" current somehow?

Here are the results for the current sharing tests:
(fuses in, no input, no load, full voltage)

Q17: 27ma Q13: 28ma
Q18: 26ma Q14: 26ma
Q19: 25ma Q15: 28ma
Q20: 27ma Q16: 29ma

For each reading, I let the amp settle for about 5 minutes until the reading was stable. That's pretty close to David's 10% advice. I think I'm ready to test with a load.
 
I have a question about testing with a load. In the manual, for the bias procedure, it recommends running the amp at 66w. I understand that I would need to see 23vac into 8 ohms at the speaker out to be using 66w. What should the amplitude of my 100 hz sine wave input be? Is it as simple as divide by the gain of the amp?

I have so much theory to learn...
 
I have a question about testing with a load. In the manual, for the bias procedure, it recommends running the amp at 66w. I understand that I would need to see 23vac into 8 ohms at the speaker out to be using 66w. What should the amplitude of my 100 hz sine wave input be? Is it as simple as divide by the gain of the amp?

I have so much theory to learn...

Do they mean 23Vrms or 23Vp? If it's rms then you have to multiply the 23V by the square root of 2 to obtain about 32.5Vp, (peak), or 65Vp-p, (peak to peak). This is the conversion for a sine function. Audio is usually expressed in rms.

David.
 
Hi G², Steve,
Using a curve tracer is not something most technicians know how to do. Yes, once you become good with these they are not as much of a bother. Remember, we are talking about someone new at electronics here, not someone like you, Steve or myself. You must match the tool to the skill level of the operator.

Look at all the time spent on this one amplifier.
This has exactly zero to do with what tester is being used, sorry but you're barking up the wrong tree on this one.

This amp was one of the legendary 'tough dogs' and I believe it would have been a much shorter story with the curve tracer.
No.
It's knowledge that leads to tough dogs to be repaired. I would say the same thing about the IT-18 tester being an effective way to weed out iffy parts.

Your leakage tests were one point on a graph at high voltage in this case and the hfe was at a very low voltage. You have to 'assume' its behaving properly in between and almost always it does but when it DOESN'T, how will you know?
Don't forget the low voltage leakage tests performed on the meter. They correlated very well with the high voltage experiment. The IT-18 happens to be excellent at catching leaky parts. Leakage is leakage, any leakage is a bad thing. The best test for transistors would be to test in the circuit they came from. Same for tubes.

I confess I don't work on consumer gear any more but I used to and once in a while the curve tracer sorted out a 'dog'.
Parts is parts. It doesn't matter what class of equipment they are in. I work on all types, including test equipment. Parts is parts.

Hi Steve,
I used it to sort parts just as stratus46 suggests. Once it is set up, it is just as fast as he says. I never tested less than 200 transistors at a time.
No doubt! I also do mass batch testing. I group and categorize the parts as well. I don't use a curve tracer and manage well.

I tend to do batches of 50 to 100 these days. Sometimes as low as 15 pcs. They all get done and I can get matches well in the 1% range using the jig.

Hi Pete,
Your offset figures look good. I also do this when working on a stubborn amplifier were I don't like the DC offset. You did miss one point though. The difference voltage between the bases will be the greatest factor in determining DC offset voltages. If the resistive components from input to common and from feedback to amplifier output are equal, the same DC offset appears on each base. This makes it a common mode signal that is rejected (assuming a current source tail) and you can have a 0.000 VDC offset voltage with both bases sitting at 55 mV.

-Chris
 
Do they mean 23Vrms or 23Vp? If it's rms then you have to multiply the 23V by the square root of 2 to obtain about 32.5Vp, (peak), or 65Vp-p, (peak to peak). This is the conversion for a sine function. Audio is usually expressed in rms.

David.

This is one of those fundamentals that you experienced guys understand, but has tripped me up before. So set your wayback machine to high-school physics, and take a trip with me...

The manual says: "Tests are performed after warm-up at 66 watts into 8 ohms for at least 15 minutes."

Fred says: "Hmm, my DVM can't measure watts. But it can measure volts. Ms. Imler's Physics class taught me that I can use Ohm's Law to calculate voltage: V=square root of (P x R). When I do that math with 8 ohms and 66 watts, I get V=22.97."

Is that RMS or P-P? I'm not sure. I understand what RMS and P-P mean, at least I think I do, but I'm missing something basic here.

I'm also unsure what the manual means when it specifies a "600-ohm signal source."

I swear, guys, I'm working my way through some books and courses once this repair is done.
 
It's knowledge that leads to tough dogs to be repaired. I would say the same thing about the IT-18 tester being an effective way to weed out iffy parts.

Don't forget the low voltage leakage tests performed on the meter. They correlated very well with the high voltage experiment. The IT-18 happens to be excellent at catching leaky parts. Leakage is leakage, any leakage is a bad thing. The best test for transistors would be to test in the circuit they came from. Same for tubes.

It's not entirely true that the IT-18 tests correlated with the Vceo test. Unlike the bad output TR, the counterfeit TR showed no leakage and appropriate gain on the IT-18, but failed spectacularly when exposed to its rated Vceo.

However, it's also true that I could have had $100k in test gear on my bench that wouldn't have helped since I lacked the knowledge that it was possible for a part to be fake and fatally flawed, but still pass the IT-18 test.

I think my next project will be to build the ultimate transistor tester and matcher... 🙂
 
Hi Pete,
Your offset figures look good. I also do this when working on a stubborn amplifier were I don't like the DC offset. You did miss one point though. The difference voltage between the bases will be the greatest factor in determining DC offset voltages. If the resistive components from input to common and from feedback to amplifier output are equal, the same DC offset appears on each base. This makes it a common mode signal that is rejected (assuming a current source tail) and you can have a 0.000 VDC offset voltage with both bases sitting at 55 mV.

You guys did catch that I'm talking about reading this voltage at the INPUT RCA jack, NOT the speaker output, right?

Such a voltage is also called off-set?

Feeling dumb this morning... 😱
 
Hi Fred,
No, it never occured to me that you would measure DC voltage at the input. This is not usually a useful test. I've only done that measurement a couple times in over 30 years, and only because of a really weird fault that injected a DC voltage there. Most RCA inputs are coupled to the diff pair capacitively (or should be in Krell's case). A straight connection will upset the DC balance and possibly create a DC offset situation where there was none before the capacitor was yanked.

Yes, the IT-18 will miss faults that only show up under higher voltage stresses. However, this is a known limitation of the device and I check for this when the parts are in circuit. A curve tracer will show this defect only if it's capable of the required voltages and currents. Still, this situation is not the rule. Using the IT-18, I've caught far more leaky outputs than I've missed and found via other methods. In fact, it's the IT-18 that normally catches the problems before I reassemble the output stage again. Leaky Vas transistors are even more fun to catch, thank goodness the IT-18 measures leakage in two modes. Most transistor testers do not even measure leakage modes.

Don't bother trying to build the perfect transistor tester. A few jigs, the IT-18 and a curve tracer will serve you extremely well. There is a learning curve involved with reading beta from a curve tracer, leakage is easy to see by eye. You only care if there is leakage, not the actual value of leakage.

-Chris
 
Hi Fred,
No, it never occured to me that you would measure DC voltage at the input. This is not usually a useful test. I've only done that measurement a couple times in over 30 years, and only because of a really weird fault that injected a DC voltage there. Most RCA inputs are coupled to the diff pair capacitively (or should be in Krell's case). A straight connection will upset the DC balance and possibly create a DC offset situation where there was none before the capacitor was yanked.

Well, I only measured it because I noticed on my scope that the input wave-form was changing when I turned on the amp, and I wanted to know why that was happening. (Yes, I know about curiosity and cats...)

So you're saying it's nothing to worry about, then, right?
 
Hi Fred,
Normally, yes. There is no useful information there, in fact I believe you should have shorting jacks installed. That makes sure there are no input signals to upset the steady state conditions that are assumed to exist. I use them, but I don't even think about it anymore.

-Chris
 
Took the next baby step:
Fuses in (amp is now in normal, complete running condition), input shorted (thanks Chris!), connected an 8 ohm dummy load (100w resistor on a heat sink).

Amp went full voltage without issue. I detected no noise or ripple on the scope. I measured 25mvdc offset after it settled for 5 minutes.

I'm ready to put in a 100hz sine wave and, assuming it amplifies that without issue, do the bias adjustment procedure.

I just (stupidly) need to figure out what I should see on my DVM to know I'm getting 66w out.
 
Hi Fred,
Ohms law. You are being told RMS values, so figure it out as if the power figure was for a DC system. Your DVM will read the correct value as long as there isn't much distortion. You would normally use 1 KHz and your 8R dummy load. The power level is not exact, they just want you to get it good and warm. Then you allow it to settle in after it cools off before adjusting. Allow 1/2 hour after each adjustment so everything stabilizes again. It can take a while to do.

Probably somewhere close to 23 Vrms is what you want, a hair shy actually. Close enough.

-Chris