The B7 Class-A Discrete Opamp - Updated 7-5-04

[tk@halmi]

Alan,

Great observations! Thanks for your input.

1. Input pair balancing problem
If I wanted to balance the currents better in the input pair I would have added a current mirror. And I think I will do that in the next round of improvements just to reduce the output offset. Many classic designs get away without current mirrors like the Melcor 1731, API2520 and the first generation Forssell opamp. Playing with the values of the collector resistor on Q1 can help achieve balance, but it is not absolutely necessary. It does result in some 2nd harmonic distortion.

2. Output emitter follower biasing
Yes, you may want to select your diodes on the bench so you get at least 10mA going through the output all the time (I will add this to the document). It is cheaper to select the diodes than the transistors. Both the API2520 and the JH990 get away with this and they are not regarded as chopped liver in the pro-audio community. A third diode WILL eliminate the uncertainity and the excercise to swap diodes, but opens up the transistors too much and they start to heat up really fast. Perhaps two silicon and one germanium diode will be less prone to burning up the output stage.

PPS. I believe that D3 and D4 will not do the job you suggest....

Build one and see for yourself :grin:

Cheers,
Tamas

 

[AP]

Re: the diodes...

..ah, I see, the arrangement limits the voltage drop across R9 or R10 to two diode-drops... Ok, that should do it...

Alan

 

[tk@halmi]

I am sorry. My explanations are lacking clarity. The diodes that bias the output transistors are D1 and D2.

 

[PRR]

> Q1 and Q2 are not in balance

No. And this will cause DC offset and higher numeric THD. You can fool with resistors and get it better-balanced. You can add a current-mirror to force the two sides to balance (nearly all chip-amps do). But as Tamas says, many-many-many "classic" designs do it this way. I think the 990 was built both ways: initially with resistor-load, then later with current-mirror (I may be wrong).

> unhappy at the way the output transistors are biassed

I've never liked it either, but it is VERY commonly used.

To get to the core of the matter: you pick transistors with somewhat larger junctions than the diodes. To take an extreme case (and make the math simple): say the 2nd-stage current is 1mA and the diode junction area is 1 "square hair". Use transistors with junction area of 10 "square hairs". Without emitter resistors, the transistors will pass 1mA*10= 10mA. Or if you replaced (for test) an emitter resistor with a 1mA current source, it would show 60mV voltage drop (Shockley's Relation), at 2mA about 40mV drop. Then use a 20 ohm (40mV/2mA) emitter resistor. The idle current will be reasonably stable against temperature variation, and for transistor variation within the same family (die-area, inferred from peak current and dissipation ratings). Believe it or not.

However since the important parameters are not given in data sheets (or not with enough precision), it does mean some experimenting. But it does seem that you can usually find a suitable balance using 1N914 diodes with most 300mA-2A transistors.

Another trick: use diode-connected transistors, from the same lot as your output devices, as the bias-diodes. If you want the output idle current equal or greater than 2nd-stage current, add a resistor between the diodes to get a few more mV. Then the junction-drops nearly cancel, and the emitter resistor voltage tracks the bias-resistor voltage. This works, but a Vbe multiplier is about the same trouble and has the small practical advantage of running cool transistors warmer and hot transistors a bit leaner. It is over-compensated for current regulation but if the transistors run at all warm it is less likely to run-away when hot.

> maybe three diodes

Maybe. But there is another factor. A class AB pair can give very low distortion (even in the crossover from A to B) if the emitter resistors drop 30mV-60mV at idle. The dynamic emitter resistance changes with current. With no resistor (if you could stabilize it) the output impedance drops and gain increases as level rises from Class A to Class B. For two transistors without emitter resistors idling at 1mA and peaking at 30mA: the small signal output impedance is about 14 ohms; on peaks it is about 1 ohm. Small signal gain is 0.977, gain at the peak is 0.998, a 2% change. Adding emitter resistors keeps the gain more constant (does not allow output impedance to drop on peaks). On paper the "ideal" resistance drops 28mV, silicon bandgap voltage. Practical transistors show voltages a little higher than theory, and listening tests always prefer richer rather than poorer, so 50mV is a nice goal. It isn't very critical, which is why you can get away with random-source diodes and transistors as long as they are about the right size.

With three diodes the emitter resistor voltage is up around 300mV. If you stay in Class A, that is fine (though a bit wasteful for high-power use as in loudspeaker amps). But when one of the pair cuts-off, to shift into Class B peaks, the output impedance doubles. This actually gives more distortion than the small-voltage case. It isn't terribly offensive distortion (just "soft peaks") and some designs do work that way. And it sure eliminates your DC stability worries.

 

[clintrubber]

Some more about DC-balance:
using the 'PRR-figure' of 20uA Ib for each input device of the diffpair, the arrangement of your testcircuit would result in an additional & typical 100mV input-offset since for one input that 20uA flows though 5k and the other 20uA flows though 10k. Using 10k i.s.o. 5k would balance this. (The P1 + R2 leg does not matter here since it's decoupled by C1)

I don't have the current gain info of the diffpair devices here though -
20 uA Ib & Ic of 0.5*1.75mA would only mean a beta of 44, which is unlikely low. So likely the contribution mentioned above of the unequal DC-impedances will be a lot less.
But still, about how much offset are we actually talking ?

Nice thread BTW, let's all discuss & learn :thumb:

Bye,

Peter

 

[tk@halmi]

[quote author="clintrubber"]Some more about DC-balance:
using the 'PRR-figure' of 20uA Ib for each input device of the diffpair, the arrangement of your testcircuit would result in an additional & typical 100mV input-offset since for one input that 20uA flows though 5k and the other 20uA flows though 10k. Using 10k i.s.o. 5k would balance this. (The P1 + R2 leg does not matter here since it's decoupled by C1)
[/quote]

Hi Peter

It is nice of you to contribute to the thread. So in general one would want to make sure that the bias resistor for one input matches the feedback resistor to bias both transistors the same way?
How would one pick resistor and pot values for an opamp in general?

Thank You,
Tamas

 

[clintrubber]

Hi,

W.r.t. 'balancing' for equal influences of the DC-input currents, it's about seeing the same impedance (in this case resistance since it's about DC) for each input.

So for your testcircuit it's indeed the feedback-resistor vs the input resistor, but if C1 hadn't been present it would have been (the feedback-res R3 // (P1+R2)) vs (the input-res R1).

So who joins depends on the circuit: determine which resistors are 'present' for DC and provide equal values for each input. In this circuit, keeping all caps and adding AC-coupling before R1 would mean that for a value of 10k for R1 all is done what could be done w.r.t. this contributing mechanism. And you have the benefit that you're done for all gainsettings because of C1.

Bye,

Peter

 

[Winston OBoogie]

[quote author="PRR"]
I think the 990 was built both ways: initially with resistor-load, then later with current-mirror (I may be wrong).
[/quote]

I believe the 990 always had a mirror underneath the resistor loads. I do remember seeing a 1N914 diode initially and then noticing it was changed to a diode connected transistor of the same type as the other side of the mirror on a later schem. Then again, I may be wrong as it's been a while since I looked :?

W.O'B.

 

[Aniol1349]

Tamas,

any chance to re-upload the document again?

Cheers

 

[Sredna]

Voilà!

(Found it through the Wayback machine: http://web.archive.org/)