Some time ago, KurtW queried: "Regarding the LM334, this looks like a handy solution for biasing into class A. I noticed the capacitance is much higher than using discrete FETs like the 2N5485, but I assume its still low enough not to be a problem."
Jan replied that the LM334 has only 15pF, and should not affect anything.
I recently started to further upgrade my Corda HA-1 (earlier: changed rectifiers, added power supply capacitance) in a bluish direction. The first thing I did was to replace the 1.5k resistor in the first stage that biases the LM6171 into Class A.
As an aside, I must say that regulation ability of the LM334 is amazing. Seems imune to load, and the variation in current as a function of voltage in the 2-15V range is around 0.1%
I tried 1 mA first and it really had a bad effect, especially on the highs. Moved to 5.5 mA and this was a great improvement. Still seemed as if the highs were a little muted. I then replaced the standalone LM334 with an LM334/single Jfet combination (2n5486) as per the LM334 data sheet. This reduces the capacitance to that of the jFet--around 3 pF. Can't say that the reduction in capacitance was the reason, but the result (at 4.5 mA) was the best of all the combinations I tried.
As another aside, I then added BUF634s in the output stage and then doubled them up. I would not have imagined that doubling the BUF634s would make much difference using rel. high impedance phones like the HD600, but it does. Bass solidity was markedly improved. By the way, I selected a 7 mA bias level for the BUF634s rather than the flat out 15 mA of the high bandwidth mode: the charts show essentially all the benefit is gained by that level of bias. Heatsinked the chips anyway, even though it wasn't necessary.
Norman