open loop gain,how many is enough?

[Johan Potgieter]

It did not appear in the quote - I was referring to Traderbam's post #12.

[Johan Potgieter]

Darkfenriz,

Thanks for your interest (your post #13). But just to get one thing straight first, I do not claim the design as outstanding; it is 10 years old. Others have done better, into the -90-something dB region of distortion reduction. I simply used it as an example of what I regard as ball-park.... OK, perhaps slightly up-tech. I can post the circuit as soon as my scanner is alive again, but a brief description should suffice.

The main aim was the prevention of high order harmonic distortion. Input stage is the usual diff. pair, but with the non-inverting input ground-referenced and feedback-signal mixing with pure resistive adding, to the inverting input. That to obviate any distortion resulting from unequal input transistors. (You would know that any distortion introduced in the mixing process is not cancelled by NFB.)

The Vas is a cascode (CC load), feeding into an emitter-follower driver. This contributed noticeably to linearity; power stage performance suffered when fed from the more usual 10K - 15K Vas collector impedance in contrast to some 200 ohm in this case.

Finally, the output stage is a normal full-complimentary (class-AB, 100mA Iquiesc.), but with the usual 0,22 ohm or so emitter resistors omitted. Again, they were found to cause some high order distortion. This complicated temperature stabilising, but thermal feedback to the Vas CC load does it with the right values, although it takes the amp rather longer to settle after turn-on. But I never switch the amp off.

Transistors are pretty ordinary. I am building this (integrated with control stages) for clients so I am limited to what is readily obtainable here; with the local economy, imports are often a no-no costwise. (TIP35 and TIP36 outputs, BD139 and 140, 2N5416, BF421 and 422, etc.)

Regards.

[mlloyd1]

johan:

thanks for your descriptive summary. did i understand correctly that your output stage is common collector, bipolar outputs with no emitter resistors?

you're a brave man

also, could you say a bit more specifically what distortion was reduced by the removal of those resistors?

maybe i should just hush until the schematic is posted ...

mlloyd1

[Johan Potgieter]

By full-complimentary I meant for the top half: an N-type (BD139) feeding a P- TIP36 in a unity gain topology (E of the BD139 and C of the TIP36 to output - the usual thing), with a reversed polarity pair for the bottom half. (Yes, all bi-polars.)

The emitter resistor situation is interesting. Simply put, at low output when both pairs conduct, those resistors act in parallel between transistor pairs and load. When going to higher output (one half cut off), only one resistor acts, i.e. double the value. This must generate some non-linearity, as these resistors are not included in the unity-feedback pairs. This investigation was done some time ago and I cannot recall why I did not use a topology with an emitter resistor with the power transistors only (i.e. inside the unity gain feedback loop). A Spice simulation shows this up - perhaps I can re-investigate; notes must still be somewhere.

Some will call me crazy rather than brave. But analyses and practical (climate chamber) tests showed that the thermal feedback alone works quite well; one can even get a negative I/T behaviour. The Iquiesc. is not critical. An adjustment is necessary to get it right for specific transistors; in that sense it is not an ideal mass-production amplifier.

[MikeB]

Hi Johan, this is called gm-doubling. With setting the correct bias (~25mv across re) this effect nearly cancels out the unlinearity shown by outputdevices at this low current. (reducing crossover distortion)

Mike

[traderbam]

Because distortion is reduced by the feedback factor, in other words loop gain. E.g. when in an exaggerated case Cdom starts reducing the loop gain (feedback) from say 800 Hz, less feedback will be available at say 5 KHz, thus less reduction of any internally generated products there and higher.

Oh I see what you are saying. If the primary pole is inside the audio band then the feedback factor will be different at different audio frequencies.
Yes. But why burn that extra correction at lower frequencies just to keep the feedback factor constant? IOW what is the benefit of having a constant feedback factor?

[Johan Potgieter]

MikeB,

I suspect we might have a misunderstanding. I know gm-doubling as having to do with both transistors conducting vs. only one, pertaining to the transistors themselves, and there you are correct regarding bias setting. But I was talking about the external 0,22 - 0,33 ohm emitter resistors added for temperature stabilization.

I find cross-over distortion disappearing (in my case) over about 35 mA Ic, where over a limited spread of Ic, 3rd and 5th harmonic are about the same without and with Re=0,22 ohm, with 7th and 9th still higher. Below some 25 mA cross-over distortion sets in rapidly. To me that is too critical for a design basis; I will have to select transistors.

In contrast, at Ic=80 - 100 mA, 3rd harm. is down 2x, 5th is down 4x and 7th is down 11 times without Re compared to with; the latter in the noise. With an Re these high-order products remain. I am trying to get typical spectrum analysis graphs scanned for comparative illustration, if the scanner will only co-operate (never trust electronics).

Traderbam,

Ye-e-e-e-s. I cannot violently disagree with you, except to say that a lot of feedback can itself generate high-order products (mathematical series and such) at frequencies where it is active. Also, all things being equal (which they never are) the more the feedback, the sooner (frequency-wise) you will have to begin to attenuate with Cdom. That means that you will be meeting the lower loop-gain (horizontal) graph on your downward gain-f slope with a phase angle, at a frequency where the lower loop-gain response is still non-reactive and therefore has a larger stability margin - one really needs a diagram here which I cannot do on this PC! ... but I think you will follow. (See typical gain-f and phase-f graphs for op-amps at a frequency where Cdom is already in action.)

It might also impact on low frequency stability requirements (l.f. roll-off) unless you have a d.c. signal path (then again with its own problems - but that is another subject).

Let me stop at that; this is getting long w.r.t. the original theme, although I think that has been commented on as fully as it is possible to.

Regards.

[Johan Potgieter]

Finally, here the harmonic spectra of the 2 output config's mentioned earlier. Iquiesc. was about 90 mA, Pout at 40W both cases.

It will be clear that the test signal was 1 KHz, with the 3rd harmonic in both cases just about 10mV, etc. (It would have been more informative to have had a log amplitude scale; this way it looks nicer!)

Regards.

[Johan Potgieter]

OOOOPS!!!

Sorry folks, that was not a belated April joke; I just scanned without watching again. You will have to turn your monitors upside down or print. (Hope this does not reflect on my state of thinking in audio.)

Regretfully yours.