DC offset in class AB power amp


2017-12-30 12:52 pm
I have a 70's solid state amp on wich i measure about 400mV at the speaker output. I don't think its good for the speakers ?
I managed to get rid of that DC voltage with 2 big electrolytics caps minus to minus, it works but i think i might be loosing some bass (330uF caps ; 2200uF are on order for further testing).

I wonder if there are other ways to reduce that DC ? Maybe a trim pot somewhere ?



  • carte ampli.pdf
    129.4 KB · Views: 123
You can disconnect the transformer center tap from the speaker ground, replacing the 63 v 4700 rail caps with 120 v ones. Then change speaker return to the negative rail. Tiefbassuebert suggested this to me.
Then you can put one 4700 v 63 cap in series with the output. This will reduce the noise I hear at low volume (1/8 W into 8 ohms) from two electrolytic polar caps back to back. It takes two 6800 caps back to back to get decent bass in an 8 ohm speaker ( series caps add to half value).
If there is room in the amp for another board, you can use an op amp to inject a little current in the base of either predriver when the output is off center. This requires averaging of the output down to 1 hz or so. You want to inject current in the base of the predriver opposite the side that has too much DC output. An op amp designed for 60 v supply is the LM675, but you'd have to regulate the supply down some.
It might be easier to replace the boards in this old amp with some boards more modern.
Last edited:
No, it is two 4700 uf 120 v caps connected as the old ones are. You still have a plus and minus side of the amp. The speaker return goes to negative rail to keep the speaker cap charged positively at all times.
I hear the funny noises crossing zero with a minus to minus speaker cap mostly on top octave piano and tinkly bells. Low volume only, base wattage where I keep my amp in my living room on soft passages. If your hearing stops at 8000 hz or your speakers have a lot of distortion 8-14 khz, you my not need to bother with this. Many people think the back to back speaker caps are transparent.
On second thought, check your two rail voltages. One of the old 4700 caps might be leaking more than the other. If rail voltages are balanced, then gain differences between the predriver transistors or drivers is causing the problem. You could use a pot to inject a little current in the base of the predriver side away from the 400 mv. But pots are prone to going open on the wiper after a few years. Better a fixed resistor. And without a DC control feedback circuit, if the DC voltage changes at different temperatures or wattages the fixed resistor can't compensate.
If you want to save $2, the 4700 uf rail caps are pretty generous for a single output transistor pair amp. My ST120 gets away with 3300 uf. I've checked it at 70 w/ch for 5 seconds at a time on 8 ohm speakers.
Last edited:
Strange way to configure DC feedback in that thing.

If it were sitting in front of me, I would tweak the value of R11 to reduce the output offset. You didn't say whether you were seeing + or - 400 mV, so I can't say which way to tweak.

With speakers disconnected, measure the output offset. Now add a 100K resistor across R11 and remeasure the output offset. Did it go in the right direction? If so, pick a parallel resistor that makes it close to zero. If it went in the wrong direction, you will need to add resistance to R11.
Strange way to configure DC feedback in that thing.
True. PSRR considerations, I guess, plus they needed a slight positive DC voltage.

I honestly don't see how you could get 400 mV of DC offset in this. (Spoiler: We'll see about that...)

DC offset is defined by the difference of where Ve(Q6) should be and where it actually is.

Ve(Q6) should be:
Vbe(Q1) + (R10 + R11)/R3 * [ Vbe(Q2) + Vbe(Q4) + R8 * Ic(Q4)]
with Ic(Q4) given by
Ic(Q4) = (+48V - Vbe(Q6) - Vbe(Q8) - Vbe(Q10||Q12)) / (R12 + R13)
~= (+48 V - 1.3 V - 0.55 V) / (3k9 + 15k)
~= 2.44 mA
Ve(Q6) ~= 0.55 V + 0.1018 * [ 0.6 V + 0.65 V + 270R * 2.44 mA]
~= 0.74 V

We also know
Vout = Vbe(Q6) - Vbe(Q8) - Vbe(Q10||Q12)
~= 0.74 V - 1.2 V = -0.46 V.

Oh. Shows what I know, eh? :rolleyes:

(And yes, I used WAGs of Vbe from 0.55 to 0.65 V depending on roughly estimated current and transistor size.)

I can only guess that Ic(Q4) may have changed during the design or that R11 may never have been sized properly in the first place. Looks like it should be more like 33k. Well, maybe 30k or 27k, given that I overestimated DC offset a bit.
Last edited: