Fixed the problem, but a disconcerting state to leave the amp in (who knows what kind of a bad influence the faulty opamp might be on its twin eventually). So I ordered a few 4558 chips as replacement and wanted to swap out the older ones: I mean the old ones were 35 years old or so (I think the amp is from 1987 but a 1983 model, no XLR input) so I'd expect current versions of it to fare better, and they are cheap. Problem was that the "modern" ones turned out distinctly noisier and I prefer my amps to be quiet when I don't do anything so I don't have to keep switching them on and off between activities (like an acoustic set in between or discussions or some pause or such). So I moved things around in order to just replace one original 4558, outside of the input stages so that its noise level would not likely matter.
Still, I wouldn't mind less noise (you might ask why I bothered fixing the reverb at all since its use does not quite fit in with the goal of maximum quietness; I don't have a really good answer for that but don't like half-broken things even when I am just using the other half). And given the experience with the modern replacements, the noise level appears to be not a mere consequence of the overall design but relevantly dependent on opamp performance/quality.
Now there's been positively a deluge of opamps in that form factor released since that time and swapping them out is straightforward as they are socketed. But the attractive ones appear to contain some significant differences:
- JFET input stages: very high input impedance which is generally nice but not necessarily the lowest-noise fit. TL072 or successors.
- diode-clamped inputs which make me a bit worried they will behave strangely when overdriven. NE5532
- PNP input stages. MC33078, mentioned as "for improved performance" in the LM833 specifications, where the latter in turn is missing both a circuit diagram (so no idea about the input circuitry) in the ON Semi data sheet as well as some core specs.
I may be overthinking this, but am partially worried about using stuff out of specs. On the other hand, the circuit diagrams of the KB300 show discrete clamping diodes on the opamp inputs to the ±16V rails which would be quite insufficient for keeping the original 4558 out of its absolute maximum ratings. So maybe I should worry less and experiment more. But it's possible that someone else already made some more systematic experiments, and since I have to mail-order for any experiment, minimising the number of experiments reduces cost and waste (some parts are cheap enough that one can order just tentatively, at least for small numbers).
It's also likely that it's nonsensical to replace anything past the instrument input stage chips (and reverb input stage?) with regard to the noise performance, but low distortion should theoretically be useful also in the higher-level stages should one make use of the line-out for recording or something. Though very much pointless otherwise when the speakers are high-efficiency instrument ones (here Black Widow, and piezo tweeters). I certainly won't get a wonderfully uncolored and distortion-free auditory response, but I do have a probably exaggerated appreciation of low-noise quiescence. Not a fan of "comfort noise" for addressing the "is this thing on?" question.
Thanks for any insights!
PostScriptum: I find that the JFET input variants, at least TL072, tend to have distinctly higher noise voltage numbers (15nV/sqrt(Hz) rather than about 5nv/sqrt(Hz)) but negligible noise current. For the 220kOhm termination of the open input, that would deliver by far the quietest results and would make sense for directly connecting a high impedance source like a pedal-less guitar. But this is essentially a keyboard amp so it probably was a mistake to evaluate noise levels without anything connected: the actual use case would be more like a 2k source to which the input circuitry adds 4k7. The LM4562 breaks even with noise voltage/current at about 2k so I'd fare better with something like the MC33078 (or LM833) which have their break-even noise point at about 9k, meaning that the voltage noise figure will actually be relevant in the input circuitry of the KB300. That's for connecting something with reasonably low impedance. And to be honest, quieting a guitar input is sort of a fool's errand, and 220k is sort of a lowish input impedance for guitar anyway. So I'll probably not shoot for one of the JFET opamps since my main objective is for quieter operation for my use cases.
And another PostScriptum: it would appear that at a temperature of 300K, a 5k resistance alone is already good for 9nV/sqrt(Hz) of noise. Which kind of explains the low input resistance (before negative feedback) of the "good" opamps and means I should not expect too much of an improvement just from swapping opamps around. If I interpret stuff correctly.