Power-Supply Technologies There are three principal ways to power an amplifier:

1. a simple unregulated power supply consisting of transformer, rectifiers, and reservoir capacitors;

2. a linear regulated power supply;

It is immediately obvious that the first and simplest option will be the most cost-effective, but at first glance it seems likely to compromise noise and ripple performance, and possibly interchannel crosstalk. It is therefore worthwhile to examine the pros and cons of each technology in a little more detail.

I am here dealing only with the main supply for the actual power amplifier rails. Many amplifiers now have some form of microcontroller to handle on/off switching by mains relays and other housekeeping functions; this is usually powered by a separate small standby transformer, which remains powered when the amplifier supply is switched off.

The design of this is straightforward – or at least it was until the introduction of new initiatives to limit the amount of standby power that a piece of equipment is allowed to consume. The International Energy Agency is urging a 1 W standby power limit for all energy-using products.

Partnered Content: NGK contributing to the speard of IoT devices through new ceramic Li-ion rechargeable batteries

To learn more about simplifying power designs, register for this  free webinar , “Simplify Power Designs with Micromodules Products” sponsored by Analog Devices.

On perusing the above list, it seems clear to me that regulated supplies for power amplifiers are a bad thing. Not everyone agrees – see, for example, Linsley-Hood [2]. Unfortunately he did not adduce any evidence to support his case.

The usual claim – in fact it is probably the closest thing to a subjectivist consensus there is – is that linear regulated supplies give 'tighter bass' or 'firmer bass'; advocates of this position are always careful not to define 'tighter bass' too closely, so no one can disprove the notion. If the phrase means anything, it presumably refers to changes in the low-frequency transient response; however, since no such changes can be objectively detected, this appears to be simply untrue.

If properly designed, all three approaches can give excellent sound, so it makes sense to go for the easiest solution; with the unregulated supply the main challenge is to keep the ripple out of the audio, which will be seen to be straightforward if tackled logically. The linear regulated approach presents instead the challenge of designing not one but two complex negative-feedback systems, close coupled in what can easily become a deadly embrace if one of the partners shows any HF instability.

Before everyone runs off with the idea that I am irrevocably prejudiced against supply regulation, I will mention here that the first power amplifier system I ever designed did indeed have regulated power supplies, because at the time I was prepared to believe that it was the only way to achieve a good hum performance. Remarkably, considering that the only test gear I had was an old moving-coil test-meter, it all worked first time and without any misbehavior I could detect. I still have it in the cellar. However, I did take away from the experience the conviction that if the power supplies were more complex than the amplifier, something was wrong with my design philosophy.

The generic amplifier designs examined in this book have excellent supply-rail rejection, and so a simple unregulated supply is perfectly adequate. The use of regulated supplies is definitely unnecessary, and I would recommend strongly against their use. At best, you have doubled the amount of high-power circuitry to be bought, built, and tested. At worst, you could have intractable HF stability problems, peculiar slew-limiting, and some expensive device failures.

In the list of the advantages of linear regulated supplies set out above, the one that seems to have most appeal to people is the first. It allows an amplifier to approximate more closely to a perfect voltage source, which would give exactly twice the power into 4 O than it gives into 8 O. In the not always rational world of hi-fi, this kind of amplifier behavior is often considered a mark of solid merit, implying that there are huge output stages and heavyweight power supplies that can gracefully handle any kind of loudspeaker demand. I disagree, for the reasons set out above, but let's follow the train of thought for a bit, until it derails.

A regulated supply clearly gives a closer approach to this ideal than an unregulated supply whose voltage will droop when driving the 4 O load. However, even if the regulated supply is as stiff as a girder of pure unbendium, there will still be load-dependent losses in the output stage that will make the 4 O output less than twice that into 8 O.

Assume for the moment that we have an amplifier which gives 100 W into 8 O. There will be emitter resistors in the output stage, and the lowest value they are likely to have is 0.1 O. (There are good reasons why these resistors should be as low as practicable, because this improves linearity as well as efficiency – see Chapter 6.) These resistors are in series with the output and so form a potential divider with the load. Their presence alone, without considering other losses such as increased output device Vbe values at higher currents, and the wiring resistance, will cause the 4 O output to be 195.1 W rather than 200 W. That perfect voltage source is not so easy to make after all.

However, to make a rather ambitious generalization (and all generalizations are of course dangerous) it can be said that the power deficit from this cause is rather less than that due to unregulated supply rails drooping, which can cause twice the loss in terms of watts. This factor depends very much on how big the mains transformer is, how big the reservoir capacitors are (because that affects the depth of the ripple troughs, which is where clipping occurs first) and so on – I said it was a generalization. It is therefore perhaps worthwhile to look a little closer at the regulated supply issue.

I was once faced with this situation: the managing director wanted exact power doubling in a high-power design, but I was less than enthusiastic about trying to make heavy-current regulated power supplies work dependably. Time for some thought.

If you accept that there is no problem in making a hum-free amplifier that runs from unregulated and ripply rails – which is emphatically true, as demonstrated in the second half of this chapter – then the function of the regulators is simply to keep part of the supply voltage away from the amplifiers. In effect, the output stage is a giant clipping circuit. So why not do the clipping at the input of the amplifier, where it can be done with a couple of diodes, and go back to an unregulated power supply? The idea is shown in Figure 9.1.

Figure 9.1: Putting a small-signal clipping circuit at the amplifier input to emulate a regulated power supply

The electrical power previously wasted in the regulators is now absorbed by the output devices, perhaps necessitating a bit more heat-sinking, but all the complications of regulators disappear. As with a regulated supply, the clipping will be clean and uncontaminated by ripple – in fact probably cleaner because a small-signal clipping circuit will have no time-constants that may gather unwanted charges during overload.

Now you may think that this is cheating – the managing director certainly did, but even he was forced to admit that what I proposed was functionally identical to an amplifier with regulated supplies, and much cheaper. However, the idea of deliberately restricting amplifier output – and this new approach simply makes it obvious that that is what regulated supplies do – did not appeal to him any more than it does to me, and the project went forward with unregulated supplies. And no hum.

In the foregoing argument there is one point that has been oversimplified a little. Making a small-signal clipping circuit is straightforward. Making a clipping circuit that is wholly distortion-free below the clipping point is anything but straightforward. As I described in Chapter 2, it can be done, with some non-obvious circuitry. You will, I hope, forgive me for not revealing it at the moment, but I rather hope that someone might buy the idea off me.

A typical unregulated power supply is shown in Figure 9.2. This is wholly conventional in concept, though for optimal hum performance the wiring topology and physical layout need close attention, and this point is rarely made.

Figure 9.2: A simple unregulated power supply, including rectifier-snubbing and X-capacitor

In a multichannel amplifier, the power supply will fall into one of three types. In order of increasing cost, and allegedly decreased interaction between channels, these are:

In reality the only interaction experienced with (1) and (2) is a variation in maximum power output depending on how the other channels are loaded. With competent design signal crosstalk via the power supply should simply not happen.

For amplifiers of moderate power the total reservoir capacitance per rail usually ranges from 4700 to 20,000 µF, though some designs have much more. Ripple current ratings must be taken seriously, for excessive ripple current heats up the capacitors and reduces their lifetime. It is often claimed that large amounts of reservoir capacitance give 'firmer bass', presumably following the same sort of vague thinking that credits regulated power supplies with giving 'firmer bass', but it is untrue for all normal amplifier designs below clipping.

I do not propose to go through the details of designing a simple PSU at this point, because such information can be found in standard textbooks, but I instead offer below some hints and warnings that are either rarely published or are especially relevant to audio amplifier design.

Mains Transformers The mains transformer will normally be either the traditional E-and-I frame type, or a toroid. The frame type is used where price is more important than compactness or external field, and vice versa. There are various other types of transformer, such as C-core or R-core, but they do not seem to be able to match the low external field of the toroid, while being significantly more expensive than the frame type.

The procurement of the mains transformer for a given voltage at a given current is simple in principle, but the field of audio power amplifiers always seems to involve a degree of trial and error. This is because when transformers are used in unregulated power supplies for audio power amplifiers, the on-load voltage has to be accurate; the power output in watts depends on the square of the rail voltage. Watts do not have a direct relation to subjective volume, but are psychologically an important part of the written spec.

An amplifier that on review does not quite meet its published power output gives a poor impression. The subjective difference between 199 and 200 W is utterly negligible but the two figures look quite different laying there on the paper. It is therefore normal practice to err on the side of higher rather than lower output power; this should not be taken too far as the amplifier will be running hotter than necessary.

The main reason for output power error is that the voltage actually developed on the reservoir capacitors depends on losses that are not easily predicted, and this is inherent in any rectifier circuit where the current flows only in short sharp peaks at the crest of the AC waveform.

Firstly the voltage developed depends on the transformer regulation, i.e. the amount the voltage drops as more current is drawn. (The word 'regulation' in this context has nothing to do with negative-feedback voltage control – unfortunate and confusing, but there it is.) Transformer manufacturers are usually reluctant to predict anything more than a very approximate figure for this.

Voltage losses also depend strongly on the peak amplitude of the charging pulses from the rectifier to the reservoir; these peaks cause voltage drops in the AC wiring, transformer winding resistances, and rectifiers that are rather larger than might be expected from just considering the mean DC current. Unfortunately the magnitude of the peak current is poorly defined, being affected by wiring resistance and transformer leakage reactance (a parameter that transformer manufacturers are even more reluctant to predict), and calculations of the extra peak losses are so rough that they are of doubtful value. There may also be uncertainties in the voltage efficiency of the amplifier itself, and there are so many variables that it is only realistic to expect to try two or even three transformer designs before the exact output power required is obtained. I have run projects where the transformer was exactly right the first time, but that was maybe 10% of cases, and I might as well be honest and put them down to good luck.

The power output of an amplifier depends on when it starts clipping – a common criterion is that the rated power is given when the THD due to clipping reaches 1%. Given the usual unregulated power supply, clipping is controlled by the troughs of the ripple waveform rather than its peaks, and the depth of these troughs is a function of the size of the total reservoir capacity. Since large electrolytics have relatively wide tolerances, this introduces another uncertainty into the calculations.

Secondly, the voltage losses in the power amplifier itself are not that easy to predict, some of the clipping mechanisms being quite complicated in detail. The inevitable conclusion is that the fastest way to reach a satisfactory transformer design is to make only approximate calculations, order a prototype as soon as possible, and fine-tune the required voltage from there.

Since most amplifiers are intended to reproduce music and speech, with high peak-to-average power ratios, they will operate satisfactorily with transformers rated to supply only 70% of the current required for extended sine-wave operation, and in a competitive market the cost savings are significant. Trouble comes when the amplifiers are subjected to sine-wave testing, and a transformer so rated will probably fail from internal overheating, though it may take an hour or more for the temperatures to climb high enough. The usual symptom is breakdown of the winding insulation, the resultant shorted turns causing the primary mains fuse to blow. This process is usually undramatic, without visible transformer damage or the evolution of smoke, but it does of course ruin an expensive component.

To prevent such failures when a mains transformer is deliberately underrated, some form of thermal cut-out is essential. Self-resetting cut-outs based on snap-action bimetal disks are physically small enough to be buried in the outer winding layers and work very well. They are usually chosen to open the primary circuit at 100 or 110° C, as transformer materials are usually rated to 120° C unless special construction is required. Once-only thermal cut-outs can also be specified, but their operation renders the transformer almost as useless as shorted turns do – it is rarely economic to rewind transformers. The point is that they are required for safety reasons; the transformer will fail in a controlled fashion rather than relying on internal shorting and consequent fuse-blowing, and they are significantly cheaper than self-resetting cut-outs.

If the primary side of the mains transformer has multiple taps for multi-country operation, remember that some of the primary wiring will carry much greater currents at low-voltage tappings; the mains current drawn on 90 V input will be nearly three times that at 240 V, for the same power out.

Mounting frame transformers is straightforward; bolts go through holes in the frame and into the chassis. There may be an orientation that minimizes the hum induced into the electronics, and this needs to be considered at the mechanical design stage. These transformers are usually, but not always, mounted with their sides parallel to the sides of the chassis for aesthetic reasons, and rotating them to minimize hum is not common practice.

Toroidal transformers introduce some extra considerations. It is well known that toroids can be rotated to minimize induced hum, and it is a very good idea to allow for this by making the transformer lead-out wires long enough.

Toroidal transformers are typically mounted by sandwiching them between two dished plates, or one dished plate and a dished area pressed into a chassis plate. The plates are then held in place by a single large bolt passing through the center, as shown in Figure 9.3. Neoprene washers are used at top and bottom to prevent the pressure from the plates putting undue pressure on the outer windings. In some cases a large flat washer is used underneath the chassis to spread the loading from the central bolt.

Figure 9.3: Toroidal transformer mounting. For clarity the fixing bolt is partly withdrawn

The fixing bolt must be secured with some kind of locking nut or locking washer. The toroid will be the heaviest part of the amplifier, and you really do not want it bouncing around inside the equipment because vibration in transit has loosened the nut. It is important not to over-tighten the bolt and put undue stress on the windings; in a production situation a torque-wrench setting is usually specified.

Very small toroids can be mounted simply by putting a fixing bolt through a central filling of epoxy potting compound. This would not be safe for larger sizes as the potting compound is only adhering to the tape on the inside of the toroid, and any serious vertical force will either tear the tape or rupture the bond between tape and potting. Nevertheless, large toroids very often do have their center filled with potting compound; this is to deal with side-forces, at which it is good because one side is in compression, and not vertical forces. These side-forces are typically produced by the 1-meter drop-test.

It is well known that when a toroidal transformer is mounted, it is essential to avoid creating a shorted turn through the central bolt. However, the mistake shown in Figure 9.4a does still occur and the result will inevitably be blown primary fuses and profuse profanity. Slightly more subtle is the dangerous situation shown in Figure 9.4b, where a shorted turn is created when the equipment lid is slightly deformed by placing a heavy item on top of it.

Figure 9.4: How not to mount a toroidal transformer

The clearance between the top of a toroid mounting bolt and the lid must always be checked. If you've got it wrong and you are surrounded by 1000 sets of metalwork, a thin layer of tough insulation on the inside of the lid will get you out of trouble.

A transformer specification needs to be a formal document. There are many factors to nail down and the usual result is an electrical schematic showing the primaries, secondaries, screen, etc., and a mechanical drawing showing maximum dimensions, supplemented by quite a lot of text.

The specification process usually starts with an informal discussion with the manufacturer to determine the approximate physical size of the transformer for the VA required. This must be done before you freeze the mechanical size of your product.

The list below gives an idea of what you need to specify when ordering a toroidal transformer.

Electrical specifications Having done the basic calculations, you have (you hope) a pretty good idea of what DC voltage you will need on your main supply rails to give the desired power into a given load impedance, and have done your best to translate that into a required AC voltage from the transformer secondary.

Manufacturers are not normally prepared to give exact figures how transformer regulation will affect the DC voltage available after rectification, and so when specifying the secondary AC voltage, it is realistic to aim for getting this exactly right under only one loading condition. This is usually the 'rated load', which is almost always 8 O. (The choice of what 'rated load' you put on the rear panel can have implications for safety testing – see Chapter 19 on testing and safety.)

Safety issues The internal safety requirements, such as the thickness of insulation between windings, are usually left to the manufacturer. It is common, however, to specify the safety requirements for the lead-out wires, with phrases such as 'must be UL-approved'.

When a sample transformer is ordered there are several aspects of its performance that need to be looked at. Most are straightforward, e.g. is it the right physical size, does it have the right lead-out colors and so on, but some aspects need a little more thought.

Unless your transformer manufacturer is hopelessly incompetent, the secondary voltages should be roughly what you asked for but, for the reasons detailed above, are unlikely to be exactly right. This is of no importance in secondaries for powering regulated supplies to run op-amps, etc., but because of its strong effect on the power output figure in watts, the main HT rail secondary voltage has to be correct.

When checking power output, it is obviously important to have the incoming mains at exactly the right voltage, as errors here will feed directly into erroneous measurements. Once again, this is particularly important since the output in watts varies as the square of the voltage. The usual practice is to use a variable autotransformer to fine-tune the mains voltage, its output being monitored by a DVM with the usual measure-average-but-call-it-rms calibration. Another option is to use a ferroresonant constant-voltage transformer, but these have several disadvantages: the output waveform is usually more of a square wave than a sine, and there is a fixed output voltage. They are also heavy and expensive.

The ideal solution is to use a mains synthesizer, which can output a good sine wave of variable voltage and variable frequency at a serious power level; the only downside is that it is a very expensive piece of equipment that will only be used relatively infrequently. I have only ever worked with one manufacturer that had one of these to hand (and they went bust).

Particular difficulties can arise when you are in a country with 230 V mains, and testing transformers for equipment aimed at the American and Canadian markets (115 V) and Japan (100 V). Now the variable autotransformer is required to make a major change in voltage rather than a small adjustment, and the distortion of its output waveform will usually be severe. This renders the reading of the aforementioned measure-average-but-call-it-rms voltmeter inaccurate, as the waveform distortion alters the relationship between average and peak values of the mains, and it is the peak value that determines the voltage produced by the amplifier power supply. The normal result is that the measured amplifier power output is lower than expected at 115 V and 100 V, and this can lead to baffled exchanges with your transformer manufacturer, who knows quite well that what he has supplied is correct.

If you have no plans to invest in a mains synthesizer, the second-best solution is to get a large fixed autotransformer that reduces the 230 V mains to 115 V, and use that to feed the variable autotransformer. The latter now only has to make small voltage corrections and the waveform of its output will be much less distorted than if it was doing the whole voltage step-down itself.

Evaluating a transformer sample for safety is somewhat problematical. You can do the standard insulation tests, and you can check that the lead-outs are at least labeled with the right approvals. The internal construction can only be investigated by taking a sample apart, the issue here being proper insulation between windings, especially where the lead-outs from an inner winding pass through an outer winding. If you use a reputable manufacturer you are most unlikely to have trouble in this area – if you don't, you may not find out until you submit the transformer to a test house for safety approval, by which time you're usually a long way down the road to production.

Very often the most critical part of the evaluation is the amount of hum that the transformer induces into the electronic circuitry. This has its own section just below.

All transformers, even high-quality toroidal ones, have a significant hum field, and this can present some really intractable problems if not taken very seriously from the start of the design process; the expedients available for fixing a design with excessive transformer hum are limited in number. In comparison, the fields from AC wiring are much smaller, unless the cabling arrangements are really peculiar. Here are some factors to consider:

However, with suitable layout rotation can be very effective. One prototype amplifier I have built had a sizeable toroid mounted immediately adjacent to the TO-3 end of the amplifier PCB; however, complete cancelation of magnetic hum (hum and ripple output level below -90 dBu) was possible on rotation of the transformer.

Some toroids have single-strand secondary lead-outs, which are too stiff to allow rotation; for experimental use these can be cut short and connected to suitably large flexible wire such as 32/02, with carefully sleeved and insulated joints.

References [1] I. Sinclair (Ed.), Audio and Hi-fi Handbook, third ed., Newnes, 2000, p. 266. [2] J. Linsley-Hood, Evolutionary audio. Part 3, Electronics World (January 1990) p. 18.

Printed with permission from Focal Press, a division of Elsevier. Copyright 2009. “Audio Power Amplifier Design Handbook“, 5th Edition, by Douglas Self. For more information about this book and other similar titles, please visit www.focalpress.com.

Related links: Understanding Class-D amplifier power supply requirements PSRR: The Real Story about Closed- and Open-Loop Class-D Amplifiers Distortion in power amplifiers, Part VIII: Class A amplifiers Yet more on decoupling, Part 1: The regulator’s interaction with capacitors More about decoupling and bypass capacitors, power-supply rails, models, and their interactions (Part 5.1)

Thank you very much for an informative article. You mention that regulation is not necessary if the power supply is designed for an audio power amplifier, as the rejection can be designed into the amplifier. What are your thoughts on power supply design for modular synthesisers, such as Eurorack designs? In this case the power supply is to a bus driving various audio components from separate manufacturers. There is no guarantee that any particular module will implement adequate supply rail rejection. Would you still stick with an unregulated supply in this case? Kind regards, David

You must Register or Login to post a comment.

This site uses Akismet to reduce spam. Learn how your comment data is processed.