I have been commisssioned to design and build a one-off audio preamplifier using E88CC / 6922 valves in most of the signal-handling stage. (Don't ask why - it is a strange project but definitely not intended to be a piece of cult audiphool gear)
It has to meet an exacting spec for several parameters, including noise. I was experiencing great problems measuring the noise by means of my workshop bench amplifier because, as I increased the system gain, various forms of instability began to manifest themselves.
At first I thought the problem was caused by V.H.F. instability in two of the stages which were connected as cascodes; in that mode, high gain is maintained to several hundred megacycles and some of my wiring was long enough to have an appreciable effect at those frequencies. I was already using a grid stopper on the bottom triode of the pair, but, unconventionally, I found that another grid stopper on the upper triode was necessary to cure the problem. H.F. oscillation in the HT PSU (which was transistorised) was found to be another cause of some unpredicatable behaviour - but there was still another effect which remained:
This third effect depended on the gain of the whole system, the amplifier on test plus the bench amplifier. The bench amplifier has a monitoring loudspeaker, with an audio gain control independent of the main calibrated measuring attenuator; the effect was found to vary with the setting of this volume control. When the volume was advanced slowly, there was a pause, then the level meter slammed hard over. As the control was turned back, after a few seconds, the meter dropped back. The effect was just like audio feedback in a very high-Q situation.
Eventually I came to the conclusion that this really was audio feedback, but in the ultrasonic range. I found I could alter the oscillation point by covering the loudspeaker with my hand. If I tapped one of the valves with a small hard object, I could send the meter off-scale - although very little sound was audible through the loudspeaker. An oscilloscope showed oscillation building up at 37.5 Kc/s
Presumably the short, robust, electrode structure of these VHF triodes, which were never intended for audio work, has a high-Q mechanical resonance at ultrasonic frequency.