This originally was an article in the LLDIY Newsletter Issue 9 April 1993. Following is an update on some of the issues/modifications covered since.

A Power Supply Regulator Design

These designs are aimed mainly at tube circuits. The ideas could probably be used with solid state circuits but the output impedance isn't as low as designs that incorporate feedback. This I don't believe a problem in tube circuits with their inherent high impedances and low currents. I have little experience of solid state designs. Tube circuits, being on the whole single ended, have very poor PSRR's therefore the quality of the PSU and its sound will make a large contribution to the sound of an amplifier. What I believe is important for tube circuits is low noise, high attenuation and a low, constant output impedance (relative).

I have been recently working on '0 feedback' designs that I believe fulfil these requirements and which have now replaced previous feedback designs of high repute after comparative listening tests. A design with a high degree of feedback will always be playing catch-up with the amplifying circuit's varying current draw. My method is to use a stable voltage reference followed by a simple buffer feeding a storage device (a capacitor). The reference provides the attenuation, the buffer provides constant output impedance and the capacitor, instant supply of current for the amplifier. Both designs were modelled to confirm design goals. High feedback designs offer greater attenuation and have a lower but rising output impedance yet my designs seem to perform better by the listening test, in my own design of pre-amp. How they perform with other designs I cannot say.

PART 1 - Stage Buffer

The purpose of this buffer is to isolate each amplification stage from others. A low PSRR stage requires 'protection' from another stage's varying current requirements. This design can only follow a main regulator that has a stable output. I use 6 buffers.

The reference is a simple resistor divider, bypassed by a capacitor to present a lowish impedance for the AC content - other stages' modulation of the power supply. The FET is used for current amplification and dictates the output impedance, which approximates l/gm. The output capacitor provides the energy storage and lowers output impedance further above about 10kHz. A minimum of 22uF is recommended.

The 'sound' of the design is determined by capacitor quality so I can only recommend that the best be used. I prefer to use a WIMA FKP1 for the reference cap and an ANSAR output cap which is bypassed. I also use WIMA MKP10s, bypassed, and would think a Sidereal would work well also. Do not use an ANSAR for the reference cap - it sucks the life out of the music. Whether this is due to the size I tried, 2uF or the caps quality I don't know.

PART2 - Main Regulator

The main regulator has to provide ripple rejection and a stable output voltage. This design is based upon the simplest of regulators but with each component replaced by what I think is the best way of implementing that component's function.

Maximum attenuation will occur if R is large and VREF is small. Voltage references work best if fed from a constant current source. Constant current sources should have high impedance. I used the 'ring of 2' configuration because of its high tolerance of input voltage variation and its good wide frequency performance. The higher the hfe of the transistors, the better the performance. The readily available MJE350 performs well.

Zeners are inherently noisy and have very poor temperature coefficients. The output voltage will vary noticeably with variations of airflow within an amplifier chassis. They also do not present a very low impedance. Voltage references have low noise and impedance but do not go up to 300 volts. I have used a variation of the Vbe multiplier used in power amp bias networks in conjunction with a voltage reference to provide a stable output voltage and low impedance. I use the (again easily obtained) TL431, which is adjustable. The higher the transconductance of the reference FET, the lower the reference impedance. I use the IRF series of power FETs. Like the buffer, current gain is provided by a FET and current storage by a capacitor.

Zeners are used to provide protection only. Pass element is on remote heatsink. Grid resistor should be as close as possible to the FET.

R varies output voltage with pre-set at centre position. Pre-set alters 0/P +/- 50V. (600k + R)/R X 15.6V (Vref + Vgs) = Vout (approx). Values shown gives 300V O/P Vin should be 50V min. greater than the highest Vout.

The effects of C1 & C2 have not been tried.

T1 & T3 dissipate about 3/4W so I don't heatsink them. T2's heat-sinking requirements depends on the In/Out differential and the current draw of the following circuits.

I have built these designs on PCBs and if anyone is interested, they can have a copy of the design. I would like to add that the buffer design has been in service for about 4 months but the main regulator only a week. Its components have not yet been optimised or proven over time.

They are unfortunately not cheap to build due to their reliance on high quality components however their success, I believe, to be worth it. If nothing else I have experienced a vast improvement in that hard to define area of musicality. Andy and I hope to conduct some comparison experiments between the various regulators we have at hand (Hi feedback solid state, low feedback tube and 0 feedback solid state). If anyone wants to discuss these designs further, give me a call.
 

What's Happened Since?

We've been using these designs for 8 years now with no failures of any kind. There have been a few changes and some of the issues raised in the article have since been covered. See here for latest schematic

C1 was initially as shown across the TL431 reference. We also tried it across the upper arm of the TL431 feedback resistors - lower noise on the scope but not better sound. C2 was added and proved not to be beneficial either.

The buffers are no longer used as I now prefer RC decoupled stages.