PAPERS

Audio Power Supplies
Audio Timbre: A walk in the forest

Introduction

Here is a brief tour of audio amplifier power supplies, which are much in the news at present with the DIY Melbourne Audio Club members. The influence of the power supply on the sound of an amp is not intuitive, and I will air a few ideas here. Very often the power supply of a preamp (and even a power amp) is simple and inexpensive to modify, and the results can be most gratifying.
The topic of power supplies is very broad. Power for audio circuits is derived almost invariably from the mains, and there lies the rub. The mains is notoriously 'dirty', with spikes created by heavy current switching and troublesome voltage peaks and troughs corresponding to rises and falls in demand. In fact, batteries are much better. Moreover, power supplies boast switching spikes introduced by the rectifier diodes, which produce 'hash' as they rapidly turn on and off. All these phenomena can and do affect the quality of the sound we hear because they find their way into the audio signal chain by corrupting the DC power rails.
Defining the Problem
An amplifier is often described as a modulator for a power supply. That is, a clean power supply is connected to a speaker and the current through the speaker is made to rise and fall in exact synchrony with the music signal. It is easy to see that any impurity in the supply, like AC ripple, will superimpose itself on the speaker and create sonic effects. We have all dealt with hum problems, usually created by earth loops, which are tiny AC currents in the input circuits amplified to bothersome levels.

This leads us to the characteristics of a good power supply. There are three essential functions in an amplifier power supply. They are:

1. Freedom from AC impurity. This shows up sonically as 100Hz hum (50Hz buzz originates in the transformer). In solid state circuits this is usually achieved with a large capacitor wired directly to the output of the rectifier, but further filtering is invariably evident in the use of current sources within the amplifier. Solid state amps generally use a bipolar supply, with a positive, a negative and a ground rail. This sort of supply allows hum to cancel in the output stage, simplifying design. Valve amps use high voltage supplies, often with chokes as well as capacitors, and once again the topology of the output stage largely cancels any hum. Single ended amps are not so lucky, however, and all, by definition, use single ended circuitry right to the speaker. Current sources are extremely effective in isolating the low level circuitry of all amplifiers from any power supply hum, and are normally used in solid state amplifiers. They are remarkably useful building blocks, and feature in a novel power supply design offered in this article.
 
2. Low output impedance. This prevents voltage 'sag' when the amp draws heavy current from its power supply. A high output impedance, sometimes called a 'floppy' supply, creates a short term clip of a music signal which can sound ugly. Much effort is expended to achieve low output impedance with large transformers and generous filter capacitors and such measures definitely improve the quality of the sound in a solid state amp. Valve amps benefit greatly from chokes, with smaller capacitors, because the signal circumstances are somewhat different. A major difficulty of this aspect is the fact that output impedance varies with frequency, conferring some very strange sonic effects which change with pitch.
 
3. Low impedance signal path to ground. In a solid state amp, output stage (AC) current is fed from the supply through the amplifier to the speaker, then from the ground connection on the speaker through the filter capacitor back to the supply rail. Remember, current flow is always circular. The return path from the speaker to rail via the filter capacitor is the vital aspect often ignored; clearly the filter cap(s) is in the signal path. This explains why the sound quality depends to some extent on the filter capacitors, and highlights the importance of the power supply for good sonics. In the valve amp, the very same arguments apply, and the supply must be able to carry the signal currents to earth, so the filter capacitors are equally important.
For these three reasons the power supply is very important to audio amplifiers. Generally, power supplies in commercial solid state amps and preamps exhibit low ripple and passable output impedance. But many fall down badly in the area of signal path, leading to noticeable distortion of the audio signal in its return path from the speaker. Both zero feedback and conventional amplifiers are very sensitive to this last distortion mode, and thus their power supplies, solid state and valve, are very critical.
Nonetheless, the problems of solid state and valve amplifiers and preamps are different; solid state supplies are usually pass heavy current, while valve supplies concentrate on high voltage. Preamps are low current amplifiers. But all supplies seem to 'sound' better if their voltage is fixed, rather than floppy, and if a serious attempt is made to offer a low impedance, low distortion signal path to ground.

The crucial last capacitor

The signal path to ground is largely set by the very last capacitor in the supply, right at the amplifier. Because this cap must pass AC music signals, it should be of high quality, with metallised polypropylene favoured. It is not just a power supply cap, but also a 'coupling' cap. Since capacitors are renowned for colouring the music, electrolytics, with the exception of Black Gates and Cerafines, are not trustworthy in this important application, because they confer high distortion on any signal passing through. Equally, the output impedance of the supply is largely determined by the filter capacitors, though the transformer also has a profound effect. To some extent the requirements conflict, and so, like most engineering problems, the realisation becomes a careful balance of compromises.

About fifteen years ago there was considerable debate in US circles about the importance of a rigid power supply voltage. Audio Amateur devoted some time in 1983 discussing the so called Mike Sulzer and James Boak circuits, and investigating the properties of the new regulator chips produced by Linear Technologies and National Semiconductor. Most of the ground Sulzer and Boak covered had been well known back in the halcyon sixties when transistors were still a novelty and boffins were imaginatively hooking them together every which way. (I have a very interesting 1966 engineering text by Alley and Attwood detailing all these circuits). The Sulzer/Boak conclusions and arguments are still highly relevant today. However, what has changed in the interim has been the availability of inexpensive, robust MOSFET's ideally suited to power supply duties.

A Regulated Supply for Tubes

Just for the exercise, let us imagine that we wish to build a regulated supply. Our design will be set up for a valve preamp, where the currents are manageable and costs likely to be low. Let us select 250 volts, and a maximum current of 10mA. These figures are arbitrary, and may be varied at will.
How do we start?
We need a fixed voltage, so we could use a Zener (which is a shunt device), or a series pass element like a transistor, MOSFET or valve. This addresses the very first quality of a good power supply both topologies are capable of very low ripple.

Series elements are very efficient but they display erratically changing impedance with frequency, and do not make a good signal return path because current through them can flow only in one direction; outward, from the supply. The commonly available regulators like the LM317/337 series are designed specifically for series regulation, and while their performance is excellent, their output impedance rises steadily with frequency, leaving a just discernible sonic signature which many listeners find irritating. The signal path issue means such regulators must rely on high quality bypass capacitors, and these must be large, making them expensive and bulky and leading to charge problems at switch-on.

Shunt elements, like Zener diodes, are wired across the amp to be powered, and drop a fixed voltage. A series resistor is usually placed upstream near the rectifier to 'drop' the voltage between the raw supply and the Zener. This resistor must be judiciously chosen so that the Zener does not dissipate too much heat, and yet can accommodate the maximum current drawn by the load and the Zener combined, so that the voltage does not drop below the Zener voltage, abruptly losing regulation. Shunt elements are inefficient, because they must pass about the same current as the load, and this doubling of current represents lost energy, which must be dissipated as troublesome heat. But the Zener, although it is a little noisy, is nonetheless an excellent element for power supplies, because it meets all our criteria listed above.

In particular the shunt regulator, including the Zener, creates a marvellous signal path to ground. The reason is that current can flow from the powered circuit down through the shunt to ground, and to the powered circuit from the power supply. Bilateral flow to and from the power supply is the hallmark of the shunt regulator, satisfying the third requirement of low AC impedance to ground. This removes undue dependency on expensive, high quality bypass capacitors, allowing the designer to use a small capacitor in this role, more the size of a coupling capacitor. A smaller capacitor usually has a less damaging sonic influence, so this is excellent design practice.
While the Zener has all the required virtues, for larger power supplies the noise, heat and cost aspects lead us to more robust, quieter devices. Such a device is the MOSFET.

After more than thirty years of development, MOSFET's are extremely fast, robust and inexpensive devices. They are made in a variety of ratings to beyond 1000V and 50A. They are commonly found in traction control systems on trams, trains and servo systems and can withstand enormous electrical and thermal abuse. A whole class of electronic consumer products has been based around MOSFET's. Certainly at low currents they are not particularly linear, but in shunt applications this shortcoming is unimportant. They make wonderful shunt regulators, principally because they are fast and exhibit very low dynamic impedance. These are ideal qualities in a shunt regulator.

Here is a very simple Boak shunt regulator I have modified for MOSFET use. It uses a current sink and a resistor to set the voltage standard, and features great speed, stability and low dynamic impedance. It will source up to 10mA at 250V in a valve preamplifier. It may easily be scaled up to handle larger currents and different voltages. With bipolar transistors, the same circuit is suited to the lower voltages of solid state preamplifiers. It is a most useful tweak for an audiophile.

This circuit is very simple. The bridge converts AC to DC, albeit with considerable ripple (Four 15nF ceramic capacitors may be strung around the four terminals of the bridge for even lower noise). Unfiltered DC is passed to C1, which filters and stores energy, smoothing ripple to a low level around 320 mV rms. R1 is the series dropper resistor, and dissipates around 1.2W at 23mA. Bipolar NPN transistor Q2 is a current sink, which because it has around 1.3 volts at its base, presents 0.65 volts across R4, the current setting resistor. This generates a constant 2.05 mA for the values shown, which is drawn through R3, developing a voltage standard of 245 volts. Thus the gate of MOSFET Q1, which is biased on, will always lie 245 volts below its drain, the upper terminal. C2 ensures that any AC variations are passed directly to the MOSFET gate. Since the gate will always be around 3.85 volts above the source, which is at ground, then the total voltage across the shunt device is 249 volts, give or take a couple.

Should the supply voltage drop due to heavy current demand (which cannot exceed about 12mA), the voltage dropped across R3 will remain constant because of the current sink, but the potential at the gate will drop by a precisely equal amount. This dropping gate potential will partly turn the MOSFET off, so it draws less current. With less current drawn through R1, the drop will reduce, and the rail voltage will immediately rise. Conversely, if the supply voltage should rise, the MOSFET switches on a little harder, drawing more current. This in turn drops more volts across R1, thus lowering rail voltage. Z1 is included to prevent the gate voltage rising more than 12V above the source. More than 20V between gate and source will destroy MOSFET's, their one vulnerable characteristic. In all others, particularly heat, drain/source voltage and current capacity, MOSFET's are very hardy beasts indeed.

R5 is included to stop the MOSFET self-oscillating (normally around 10 MHz!), and C3 is included to stabilise the supply further, whilst also acting to damp any overshoot created as a result of fast transient demands. The capacitor also supplements the MOSFET in shunt duties to ground for any AC signals present.
Fit the MOSFET with a generous heatsink, or be prepared to lose it in the silicon crematorium. It dissipates 2.5 Watts, and this calls for at least a 12 degree/watt sink. Be sure to use an insulating washer and paste since the MOSFET body is at 250 volts.

There are more refined versions of this circuit, which already has a very low output impedance of milliohms across the audio range. Incidentally, by design it is short circuit-proof, drawing just shy of 100mA and frying the dropper resistor. The most obvious refinement is the replacement of R1 by a current source, which then forms a powerful noise/ripple barrier with the rectifier. A simple LM317 and resistor can be used in this application for voltages up to 36, but an additional MOSFET must be used for higher voltages. A second, simpler modification brings worthwhile improvements in voltage stability. Split R2 into two 68K resistors in series, with a 10uF 200VW electrolytic cap strung between the centre of these two resistors and the 250V rail. This confers additional stability on the current sink, Q2.

The Sulzer regulator circuits achieve vanishingly low output impedance by capitalising on the very high gains of modern op-amp's. They are designed to amplify tiny error signals, thus detecting and correcting infinitesimal variations in rail voltage. With rare exceptions such as instrumentation extreme rail accuracy is not required. While it is certainly possible to team an op-amp with a simple MOSFET to achieve comparable gain, stability problems and overshoot difficulties soon intervene and the net improvement over less complex circuits is arguable.

The simple circuit at Figure 1 is a useful addition to the low level circuitry of any valve preamplifier or power amplifier. It will deliver from zero to 10mA with just on 2mVrms of 100Hz ripple. Using the LM317 variant will reduce this ripple by almost two orders of magnitude, making it suitable for a phono stage. If you are concerned about the solid state corruption introduced by your alterations, investigate the Acoustic Research preamps and power amps. Quite a few of them use solid state power supplies! It is even arguable whether a full-on choke supply is better sonically, particularly as such supplies are quite 'floppy'. Interestingly, solid state devices, particularly MOSFET's, appear to be outstanding in the role of power regulators. Low level valve circuits are sensitive to supply noise and voltage stability, and the benefits of a constant voltage across a wide frequency range are evident in superior imaging and increased resolution. These are high-end features, and well worth achieving.

This paper attempted to identify the basic design aspects of audio amplifier power supplies. The circuit at Figure 1 is an extremely effective power supply for low power valve stages. Should any MAC members wish to build the circuit to different voltage and current ratings, I would be pleased to assist with the amended design. Either email or a phone call will reach me, Hugh R. Dean
A Nelson Pass' Article on Power Supplies (for Beginners) is on his website.

Hugh Dean

© Copyright Hugh R. Dean 1999
All rights reserved