Long tail pair - output offset compensation

Thread Starter

atferrari

Joined Jan 6, 2004
4,764
In spite I took the pain to match QA and QB for Vbe and hFE, after testing a just assembled class AB amplifier I found that the output offset is quite high: 125 mV. (The attached .pdf shows the parts of interest).

Based on the measurements I took, with 0V at the input, the reason of the imbalance seems to be the difference in value between Rin and RfA//RfB.

My questions:

1 - Is the above correct? If so, why to bother in matching the pair after all if we have to resort to compensation?

I see several possibilities to reduce the offset voltage to a small value:

2 - Using Pot A to equalize currents. Read that it is bad practice (but I know it works). Not very professional...? In fact it was the first thing I did when playing with a long tail pair, more than 20 years ago.

3 - Using Pot B (or Pot C - one of them only) to equalize the currents in the pair.

4 - Changing the value and ratio of Rf A and Rf B (avoiding to mess with Rin).

5 - Using a compensation in the form of an offset voltage applied to the bottom of Rf B provided by a low Z source.

This idea (totally new to me for a long tail pair) I found it in a project from Dick Cappels, here

I cannot avoid the feeling that matching transistors was a waste of time, vis a vis the necessity of offset compensation.

Yes, my experience in analog designs is very limited and certainly rusty.

Can anyone help with comments?

Gracias.

/Edit Replaced sketch Current source now going to -Vcc /Edit
 

Attachments

Last edited:

#12

Joined Nov 30, 2010
18,224
In my travels, I have found that the only thing that matters is the match between the voltages at the bases because your feedback will have a DC component that will be self-correcting if the base voltages match. Pot A is the preferred method.

I have also seen a transistor pair that was so badly matched that I had to replace Qa and Qb.

I also see that a resistor between each emitter and pot A could be helpful in reducing the effect of transistor mismatch. They will also make pot A have less effect.

I have never seen a pot B or pot C in a practical circuit. All the adjusting is done in the base-emitter section.

There is also temperature matching. That seems irrelevant in an analog stage, but you can glue a pair of TO-92 transistors together quite easily.

Option 4 looks good. It never hurts to have both bases looking at the same resistance.

There's a start.
 

Thread Starter

atferrari

Joined Jan 6, 2004
4,764
Hola #12

Let me tell you that when I replaced the circuit in my inital post I also posted here but my reply vanished. (The same happened about one week ago in another thread).

Today I will try Rin = Rf A // Rf B but I am sure that to get the absolute lowest offset I should use Pot A as well. We shall see.

I will post the outcome.

Gracias for replying.
 

Ron H

Joined Apr 14, 2005
7,063
The rest of the schematic is important. Post all component values.
If the gain of your longtailed pair is low, then the offset of the following stage will be important. Pot A lowers the gain.
 

#12

Joined Nov 30, 2010
18,224
Going along the same lines as RonH, the pair won't balance with any stability unless you have them running closed loop, like an op-amp. A differential pair is the basis for most op-amps. Try thinking in those terms. You're examining a baby op-amp.
 

Thread Starter

atferrari

Joined Jan 6, 2004
4,764
This is the result of a minor modification of a circuit appearing, IIRC, in the January 2011 issue from Elektor.

After assembling, with input to ground I adjusted the output offset to +/- 2mV using the 100 ohms pot. The adjustment seems rather critical (not smooth at all) and the variations do not disappear even after one hour of being turned on.

I can see the offset wandering, at times, more or less slowly between +/- 7 mV.

My questions:

How to judge how bad is this amp regarding the output offset?

Could I play any additional trick to stabilize / reduce the offset?

Any coments are welcome.

Gracias.
 

Attachments

#12

Joined Nov 30, 2010
18,224
You can make the pot easier to adjust by adding resistors in the emitter legs and shunting the 100 ohm pot to reduce the range of the pot.

.004 volts of input offset is very good. The only way to beat that would be to buy a matched pair of transistors in a single package so they both have the same silicon chip that both of them are built on. This is because all of your errors are in the 2 input transistors.

Meanwhile, you can glue the 2 input transistors together so they do not drift because of temperature differences.

Still, this is a good result. Your practical point of view is about how much DC voltage is applied to the output. How much power is wasted with .033 volts applied to the load? The load is 50 ohms plus something. 218 nanowatts maximum.
 

rogs

Joined Aug 28, 2009
279
Could I play any additional trick to stabilize / reduce the offset?

Any coments are welcome.
I had an application some years ago where any output DC offset was unacceptable, so rather than get involved in exotic thermally controlled servo compensation networks, I simply AC coupled the output!

The same sort of thing (nearly) was done in a rather unusual, but interesting way by QSC, in some of their first series amplifiers.

There's a publically available pdf of the schematic here:

http://www.qscaudio.com/support/library/schems/Discontinued/Series One/1200.pdf

It's certainly a different approach, using the main PSU decoupling capacitors as output capacitiors as well, but it could save you a fortune, if you had an amplifer DC failure, when connected to an expensive loudspeaker!

Not perhaps entirely relevant to your project, but you did say any comments! :)
 

Ron H

Joined Apr 14, 2005
7,063
If the output is drifting ±7mV, that means the input offset is drifting by less than ±1mV, which is probably normal for discrete transistors (especially considering that some of the drift is probably in the second stage). You might be able to reduce this by laying out the circuit so that the input pair can be glued together with epoxy, in order to keep them at the same temperature, as near as possible.
Your input stage gain is only about 7, so drift in the second stage (which has high gain) could also cause problems. Gluing together the two transistors in the second stage might also help.
I would probably try eliminating the pot between the emitters and tie the emitters together, raising input stage gain to ≈20-25. You could adjust offset by adjusting the tail current (vary the value of the resistor that is currently 270Ω). The additional loop gain might result in oscillations, requiring an increase in the value of the compensation cap (currently 10pF).
It would be helpful if you would add reference designators to any schematic you post.
 

Thread Starter

atferrari

Joined Jan 6, 2004
4,764
It would be helpful if you would add reference designators to any schematic you post.
I was conscious of that but composed it in 20 minutes to post it even today.

Next one (the final one?) would have them and will be more easy to read.
 

Thread Starter

atferrari

Joined Jan 6, 2004
4,764
Hola Ron, rogs and #12,

Thanks for the time spent in reading and the suggestions.

After the lady we have to attend tonight I will do some changes to see what comes out probably shunting the pot to app. 50 ohms and all pairs sharing the contact to a common surface. Surely not good looking but worth to try.

Just to have an idea how far I am, what are reasonable offset values for amps like this, implemented with discrete components?

Designed as an audio amp I intend to use it as an output stage for a signal generator up to 16 KHz.

I get more than 12V p to p before clipping with not appreciable distortion (just by watching at the screen...)

Gracias again.
 

#12

Joined Nov 30, 2010
18,224
When I first met an op-amp, almost exactly 40 years ago, the incoming batches were sorted to find the chips with 10 mv, 5 mv, and 2 mv input offset. What you have built is better than 70% of the chips I could buy 40 years ago.

Now, that is not rellevant. You can buy chips with a single microvolt of offset. When you have been building these for 40 years I will expect you to achieve 1 microvolt of offset ;)

After checking 15,428 opamps for sale:
8,455 have less than 4mv of input offset
658 have 4 mv of offset
6,315 have more than 4 mv of offset.

That pretty much puts you in the middle of the quality that you can buy on today's market.

Edit: I just popped back in to say what Ron said in the next post. If the offset seems important, and it doesn't seem important to me, you can use a modern opamp to get less input offset voltage. I'd be perfectly happy with 218 nanowatts of wasted power in an audio amplifier.
 
Last edited:

Ron H

Joined Apr 14, 2005
7,063
Gp ahead and use discrete if you're doing it for the experience.
Attached is a discrete op amp that I designed. It might be more stable thermally than the circuit you posted. Be aware that I have simulated this, but never breadboarded it.:rolleyes:
If you want to optimize performance, use an IC op amp, possibly with a push-pull emitter follower buffer inside the feedback loop.
EDIT: The slew rate may be too low for square waves.:(
 

Attachments

Last edited:

Thread Starter

atferrari

Joined Jan 6, 2004
4,764
I assembled two of these in a small piece of Veroboard. Circuit attached.

The preset (not a POT as shown) is now a 50 Ohms / 20-turns one.

Q6 & Q7 in contact with a small piece of aluminium and Q1 & Q2 in tight contact to each other. Silicon grease smeared to help.

After adjusting them I measured the values shown on the right.

To adjust the offset I found much easier watching at the trace on the screen than reading the fast changing display of my DMM.

When turned on, after being some time off, one of them exhibits an initial offset of about +5mV while the other starts somewhere around -2mV. In few minutes they come to the values shown in the record.

I have the feeling that I reached a practical limit here with this design. Probably, a (tighter) PCB could improve all this.
 

Attachments

Top