> I am remembering modeling showing about 68 dB open loop gain which of course may be totally wrong.
Looks odd to me.
http://www.waltzingbear.com/Schematics/RCA/BA-31A.htm
The 6190:402 ratio says closed-loop gain tends toward 1:16.4 or 24.3dB. Q3 Q4 are unity voltage gain, near enuff. The total amp, wired 600R in and out, seems to have 1:6 or 15.5dB transformer gain. Take the iron-gain as given, look at just the amp from point H to point 0.
The 402R is a feedback component. For true open-loop, short it.
Q1 works around 0.2mA so Re is near 150 ohms. Q2 works around 5mA so Re is near 6 ohms. The coupling into Q2 looks messy, but Q2's emitter is substantially grounded. So base resistance of Q2 is near 6*Hfe, say 300 ohms, far smaller than all else in this node, so 90% of Q1 AC current flows to Q2 base and is multiplied by Hfe.
Q2 Hfe is 40 typical, apparently fading to unity at 12MHz (the old data sheets are unclear). So Hfe is flat to ~~300KHz, it's an OK audio device.
Q2 load is 2K in parallel with Q3 base and the 6190 resistor. For various reasons we suspect Q3 base is much more than 2K, so pencil 1.5K here.
Voltage gain is then Hfe*1500/150, or OTOO 400, or 52dB. (Yes, with iron included this is 67.5dB, as you say.)
> minimum 20 dB feedback recommended elsewhere in some articles and papers
Oh, frag that newfangled theory. If we wanted a clean amp, we would not be using four antique transistors. Stick a 990 in there. At gain of 24 it will be clean. Even a good 5532 will be inoffensive. We love the old stuff FOR its imperfections.
Anyway, for many purposes, 20dB is not a lot better than 6dB. You wanna wander out toward 40dB NFB to get high-order residuals below what they are open-loop. And you don't get 40dB margins with essentially 2-stage gain amps.
This is a "current feedback" amp. The NFB impedance is -not- smaller than the inverting input impedance. NFB impedance impacts open-loop gain. To change gain in this design, fiddle R7. That's what RCA did with R4. That's how 90% of modern transistor amps do it. It has the helpful property that the amount of feedback stays near constant as gain is changed over a wide range. No compensation tweaks are needed when changing gain (good thing, because panel-variable compensation adds more parasitics than we can stand). The 1000pFd cap is mostly a 25KHz roll-off (broadcast bandlimiting) not compensation (though it does affect the loop margin).
With R7 shorted, amp gain is around 400, total gain between 600R taps is ~~68dB. Gain is very dependent on Hfe Q2. Distortion at high level will exceed 10% 2nd HD.
Adding R7=150 cuts gain and THD in half, but does not stabilize gain.
NFB from Q2 collector helps stabilize against Q2 variation and drift. Ideally we make R7 less than Re of Q1, say 50R. For a nominal Q2, OL gain is 400 so for 6dB NFB we want CL gain near 200. Actually the 50R will reduce OL gain to say 300, we aim for 150, then R9 could be 7,500R. This is low but not too low compared to Q2 load. 6.19K would be fine also, and suggests ~~40R at R7. Inner amp gain is near 43, total amp gain near 58. Input overload is near 10mV, which is VERY low. You really do not want this much gain.
For lower gain, raise R7. This not only changes the NFB divider ratio, it spoils open-loop gain. For R7=100R, 300R, 1K, nominal inner-amp CL gains of 61, 20, 7, the OL gain is just about 240, 133, 52. NFB margin is 4, 6.6, 7.4... pretty constant, and quite unlikely to change the compensation needed. In fact it is fairly unlikely for a 2-stage amp with just 12dB-17dB NFB to howl unless there is a transformer in the loop. (T101's parasitics may be critical.)
I would bet the "40dB 46dB" numbers are cited for 600R tappings. If you do the "logical" thing and tap the input for 150, you get 6dB more gain.
> these like to see a sub-1K output load
That push-pull output is only push-pull when well loaded. Unloaded it becomes push-push and acts a bit oddly. It is never truly unloaded here, because of C13 and T102. There may be an unconditionally stable set of resistors, but who has time to figure? I'm sure it "wants" to be loaded with nominal 600R resistance (fader or terminated line) at the 600R tapping of T102. Since all modern loads are 10K-22K, tack 680R permanently across the output.
> the iron is a lot of the sound
If so, if there is no Germanium Fanatic in the house, I say a 5532 is quick and easy. It will run fine on the 23V dropped DC. Bias T101 pin L to half the rail. T101 H feeds 5532 input. NFB around the 5332 can be 10K and 50-5K RA pot with 470uFd to common. Put 22R in series with output and use the other half 5532 with 22R as a unity gain booster to get up to 60mA drive into the (assumed) 150R load. Use C12=100uFd to block the DC from the OT, don't use C13. Keep R17=100R to decouple PS noise, but raise C11 to several hundred uFd because 5532 works A-B.
Do note that the "Common", bypassing, and V+ and V- defy convention. (An older version musta been positive ground, and they didn't clean-sheet when NPNs became available.) If you will never build the stock circuit, you may want to alter this to make sense to conventional minds (usually negative ground).
> I noticed the +6db strap on the schemo but what would you do with it?
Change R7 to 2K. Change R4 to 100R. Strap a 0-2K rheostat (2.5K reverse-audio pot) across terminals 46. This gives almost 20dB range of gain, both up and down from the stock gain. Keep pot-leads short. twisted, not shielded.
) vary it quite a bit in both directions. The critical part is the compensation cap in the feedback loop, which is possibly not necessary. If you can lose it, you can get more range than with. I think you could safely get a range of 30-50 dB easily, versus the stock 40-46. 30 will probably sound very different from 50, however.
