I read through quite a few of the posts in the denoisonator thread and started thinking about a variation that produces similar results but over a wider frequency range. I came up with something that performed pretty well, but realized that the good result could due to my use of a "perfect" voltage source. It will be clear in the schematic below:
V2 is the voltage source in question. So I started thinking about replacing it with a suitably low-noise voltage reference, but it had to be inexpensive. A Vbe multiplier fits that category but has a very bad temperature coefficient. But borrowing some bits from IC design gave me a Vbe miultiplier whose temperature coefficient can be adjusted -- positive, negative, and very low over a selected temperature range. Combining that with the above circuit, here are some sims:
This shows the voltage change over temperature. The multiplier circuit (upper portion of the schematic) has been adjusted to also compensate for Q5. This involved changing the values of R2 and R4 until the output voltage and temperature compensation were close to what I wanted.
The next screenshot shows the PSRR, but ONLY w/regard to the input to the LM317. The Vbe multiplier's power supply is assumed to be very low-noise, probably generated by a second LM317. The power requirement will be low so a ~30-cent SMT style regulator will do fine there.
Finally, the noise voltage:
I haven't built this circuit, this is all theoretical. Anyways, the measurement gear I have is only capable of telling me if the thing is oscillating or not. I expect it will, and taming it will likely degrade the noise performance at higher frequencies.
I don't think this variation of a Vbe multiplier will be all that useful in terms of stabilizing the idle current in class AB output stages but it may be handy for other purposes.
V2 is the voltage source in question. So I started thinking about replacing it with a suitably low-noise voltage reference, but it had to be inexpensive. A Vbe multiplier fits that category but has a very bad temperature coefficient. But borrowing some bits from IC design gave me a Vbe miultiplier whose temperature coefficient can be adjusted -- positive, negative, and very low over a selected temperature range. Combining that with the above circuit, here are some sims:
This shows the voltage change over temperature. The multiplier circuit (upper portion of the schematic) has been adjusted to also compensate for Q5. This involved changing the values of R2 and R4 until the output voltage and temperature compensation were close to what I wanted.
The next screenshot shows the PSRR, but ONLY w/regard to the input to the LM317. The Vbe multiplier's power supply is assumed to be very low-noise, probably generated by a second LM317. The power requirement will be low so a ~30-cent SMT style regulator will do fine there.
Finally, the noise voltage:
I haven't built this circuit, this is all theoretical. Anyways, the measurement gear I have is only capable of telling me if the thing is oscillating or not. I expect it will, and taming it will likely degrade the noise performance at higher frequencies.
I don't think this variation of a Vbe multiplier will be all that useful in terms of stabilizing the idle current in class AB output stages but it may be handy for other purposes.
It could be argued that I replaced a perfect voltage reference with one made with perfectly-matched transistors in the multiplier. So I took a "junk box" approach and used three different small-signal transistors in the LTSpice library -- a 2N3904, a 2N2222 and a BC847B. The temperature coefficient DID degrade....something like 8mv for a 10C change in temperature. The noise performance was pretty much the same.