This is most probably because of too little cooling on the heater regulator - it gets too hot and powers down. Try a (much) bigger heat sink.
Lets make this the one and only
"Official" GPultec help thread...
Attached a .JPG image with a simplified explanation of the filter workings - if you need to figure out what parts do what job:
In rough outlines: The pultec filter can be thought of as a simple 10:1 or 20dB attenuator - the 10K ohm of the "boost high" potentiometer and then the 1K ohm of the "cut high" to ground. This is also the bypass condition. So this 10K/1K forms the basis of all our filtering, and we put in approximately that amount of makeup gain after it, to level things out.
All the filter functions can be thought of as various sorts of partial bypassing of this attenuator:
The high boost, e.g. simply bypasses the 10K resistor with a frequency-selective short. Depending on how you set high-boost pot, you partially short out part of the 10K in the 10K/1K divider, thereby NOT attenuating as much in that frequency area (which comes out as boost after the linear makeup gain stage). Frequency select is just a LC filter that has low impedance at resonance freq*, shorting specific frequency bands more than others across the 10K part of the divider. The Q pot simply adds a series resistance to the LC filter, thereby reducing resonance steepness.
The high cut is very much the same, if you look closer: the 1K part of the 10K/1K divider is being partially shorted by a capacitor to ground, meaning that higher-than-cutoff-frequencies signal is attenuated EVEN MORE than the 20dB (because it reduces the 1K impedance above the capacitor's take-over frequency). Frequency simply depends on capacitor value*.
On a side note, you can see that we can present a VERY low load impedance to the input transformer if we boost highs AND cut highs at the same time. This why input transformer must be relatively low-impedance and low-resistance: otherwise it will limit amount of boost available by introducing a virtual resistance in series with the 10K. This is why I suggest e.g. the LL5402 as input transformer.
The low boost is slightly different, as it is a circuit that is put in series with the 1K part of the attenuator, meddling with (enlarging) its apparent resistance at low frequencies. Think of low boost as a filter that sits between the ground end of the 10K/1K attenuator and the actual ground. When "boost low" control is at minimum function, the potentiometer is simply shorted, and the filter is exactly like we described above. But this short to ground (below the 1K part of the 10K/1K attenuator) can be gradually replaced with a capacitor in stead. What this does is to gradually introduce a higher impedance to low frequencies, making the 1K part of the attenuator in effect seem to be HIGHER resistance at lower frequencies, thus attenuating LESS against the 10K (and less attenuation, again, equals boosting when seen after the makeup amplifier)
Last - the low cut - is different from all the above, as it does not mess with our 10K/1K divider: In stead it takes the output signal from our previous divider and forms a high-pass filter consisting of a low-cut capacitor against the 10K resistor (or against the interstage transformer ~10K impedance). In order to have non-permanent but variable low cut, we simply short said capacitor with a 100K potentiometer: when potentiometer is at full resistance 100K, the capacitor dominates and attenuates below its corner frequency, but as potentiometer resistance is reduced towards zero, the capacitor is gradually shorted and taken out of the equation until when all the way down the capacitor shorted, there is no longer any attenuation or filter..
..All this just a quick description, but should be informative enough to let most of the supposed magic out of that circuit.. :-)
Jakob E.