This details a regulator that is used by the Auteur Audio Lilienfeld's Choir amplifier. It was designed to have both extremely low noise and to be extremely precise. The reason is that the topology of Lilienfeld's Choir, being a non-feedback topology, does not have the power supply rejection witnessed in other topologies. The objective of the amplifier was incredibly low noise and extremely fast response, so a highly performance power supply regulator was a necessity.

The basic notion of regulation has not changed appreciably since the use of a triode and a neon regulator tube. That is to say, regulation has always been an active device comparing the output with a reference and making a correction. However, there is a lot of finesse that is required to do this job well, and there are a lot of ways of achieving this effect that work better for certain scenarios. Lilienfeld's Choir is an amplifier with nearly constant power requirements, which is an affordance for a regulator having some very good qualities at the expense of flexibility.

Long ago in the semiconductor world it was commonplace to see discrete regulators. We all have had a bench supply built this way. Like many commonly used things, it made its way to a specialized integrated circuit. In this case, it was the LM317, released in 1976 by National Semiconductor. Since it solved many problems for many people, it rapidly became a hit. However, like many products that solve something reasonably well for a lot of people, it didn't solve some problems as well as might be desired. This sentiment was the focus of Mike Sulzer's 1980 Audio Amateur article, where the venerable LM317 was updated using discrete components to afford a better output.

Over the years there have been a series of better and better variants on the audio regulator. These have been discussed by many in the field — people like Mike Sulzer, Erno Borbely, Jan Didden, Walt Jung, Gary Galo and others — each building on the lessons of the design predecessor. Warren Young has a very nice overview history of the evolution of the discreet regulator entitled Op-Amp Based Linear Regulators.

The focus of the regulator used by Lilienfeld's Choir is the lowest noise possible. I will attempt to describe how it is different than some of the more commonly seen modern discrete regulators, and why. This regulator, when built properly, will give you a spectral noise on the order of 1.7nV/√Hz, and a voltage stability of better than 100nV over a period of a minute. For those not aware, that is at least 1000 times better on both figures than your average LM317. You typically find this type of performance on a monolithic regulator designed for low power application, like an LNA on a receiver; the figure stated for this amplifier is for an output into the ampere range and an output voltage in the tens of volts. Not bad!

Below is the general circuit:

The most notable feature of this regulator is the op amp being powered on the regulated side. This was a great innovation on the part of Walt Jung's Super Regulator 2 compared with its predecessors. Not only was placement of the op amp an improvement for further reducing power supply coupled noise, but the pass transistor to this day is a fantastic choice for medium-sized regulators. While this regulator has a very novel feedback mechanism using a green LED as as a high bandgap voltage supply in a constant current circuit driving the main pass transistor, in practice this leads to higher noise than other configurations. The regulator appearing in the circuit above omits this by directly sinking from the pass transistor. This has clear limitations, as the sink capacity of the op amp and the gain of the pass transistor set the maximum current the regulator can accommodate. While this works well for the modest current requirements of Lilienfeld's Choir, if your requirements are higher it may be prudent to use a different design.

Like Jung's Super Regulator 2, this regulator makes use of a Zener diode Z2 in the sink path of the drive transistor. I chose a 5.6 volt diode as they tend on average to have the lowest noise. The most important thing is that the Zener gives the op amp U1 plenty of room to develop potential away from the rails. That is to say, for known input and output voltages, the Zener must be selected to abide the needs of the op amp, and secondarily to abide the desire for quietest junction. This is particularly important for many low noise op amps since they often do not have rail-to-rail output. For a reasonable drop of a few volts across the pass transistor Q1, the 5.6 volt Zener is ideal.

The op amp U1 that is used is Linear Technologies LT1128 due to its extremely low noise. In the feedback controlling the LT1128, the two legs of the circuit attempt to be as simple as possible. In both cases there is a practical balance that needs to be made between selection of low-valued resistors (recall the LT1128 has an input-referred noise that is less than a 50 ohm resistor) and realistic choices for decoupling capacitors. The negative input, which forms the reference, is an LM329 that has a subsequent RC filter. R4 is effectively the input resistance to U1, as the node of R3, R4 and Z1, which is the LM329, is essentially a constant voltage. It is advisable to keep the value of R4 under about 400 ohms from a noise perspective. The critical aspects of this part of the circuit are selecting quality components and reducing capacitor microphonics. The positive input, which is where feedback is introduced, is just a voltage divider with decoupling of the top leg to facilitate high frequency suppression. R1 parallel with R2 is the input resistance, giving a little more leeway on the capacitor size for C1 than is available for C2. Given the use of fairly low value resistors, large capacitors are used and an emphasis on microphonics is a must. Do not forget quality snubbers parallel with C1 and C2 for better high frequency performance. Another absolute must is the use of a high quality metal foil potentiometer. If you are going to any effort at all to make an adjustable supply with low noise, you simply must use the best components. The two diodes protecting the inputs of U1 are just 1N914 small signal diodes.

As introduced with the Sulzer-Borbely regulator, the regulator used on Lilienfeld's Choir has a pre-regulator. The LM317 is a great workhorse for this role. This will keep the input voltage constant and give you more controlled dissipation on the pass transistor. It is prudent to provide good bypassing to prevent spurious noise entering the main regulator. It is fairly common to have fairly significant output filter capacitors on a regulator. In this case it is for two reasons. First, the normal reasons, providing additional filtering for spurious current draw and decoupling of high frequencies on the output of the regulator. Second, in the case of this regulator, the output is not stable without sufficient output capacitance. This would often not be tolerated in a regulator, but the objective in this case was the lowest possible noise at all cost, even if it means using redundant output capacitors to reduce the likelihood of output instability in the face of a component failure.

Compared to most of the discrete regulators that are commonly used — Sulzer, Borbely and Jung — the regulator used by Lilienfeld's Choir is remarkably simple. It takes what I perceive to be some of the best features of all of them, sacrificing some of the versatility in exchange for an exquisite noise figure and superb DC stability. The important aspects of making it perform best include choosing the best components and using them properly. Use metal foil whenever possible, particularly for the potentiometer. Decouple the electrolytics with a snubber that works impeccably at high frequencies. Be attentive to component microphonics and set the most critical components in a high quality dissipative material like RTV-162.