This book is a collection of the articles I wrote for the journal Wireless World (now Electronics World) between the years 1979 and 1999. The vast majority of these deal with hi-fi preamplifiers and power amplifiers, and this book concentrates on this field.

In the last twenty-five years the scope of technology available to the audio designer has greatly widened. At the beginning of this period the only choice in preamp design was between discrete transistor stages and the relatively new op-amps. The latter had dubious characteristics as regards noise and distortion—particularly crossover distortion. This most unloved of audio defects was tolerated in power amplifiers because it had to be, but there was considerable resistance to incorporating it in preamplifiers. At this time no-one would have considered using valve circuitry in a new design.

My association with this influential and much-loved magazine actually began with a design for a compressor/limiter in 1975, which started life as my third-year project at university. This design is not reproduced here as it is unarguable that it has been overtaken by advances in technology.

This debut was followed by what I called ‘An Advanced Preamplifier’ in 1976. The Advanced Preamplifier certainly gave (and gives—I still have the prototype) exemplary performance, obtained by making each stage a discrete-component operational amplifier. This made necessary the use of dual IC regulators to produce +/− 15 V, and at the time the cost of this power-supply scheme was significant.

As a reaction to this complexity, I decided to try my hand at what might be called ‘traditional’ discrete circuitry in a preamplifier, and this became the ‘High Performance Preamplifier’ published in 1979, though actually designed nearly two years earlier. It was conceived in an era when opamps were still regarded with considerable suspicion by designers seeking the best possible audio performance. In the search for simplicity a single supply rail was used, without regulation, but with a simple RC filter after the reservoir capacitor to reduce ripple to a manageable 50 mV or so. Experiment had proved that this minimal-cost arrangement could give hum and noise results that were as good as those yielded by the dual-IC-regulator approach.

In contrast, the Precision Preamplifier of 1983 was designed at a time when the remarkable 5534/5532 op-amps had become available at reasonable prices. Since they delivered very low noise with almost unmeasurable distortion, it was clearly time to try a ‘third way’ as regards preamp design. Having explored discrete op-amps, and conventional discrete circuitry, an IC solution was an obvious next step. The return to op-amps meant a return to dual power supplies, but this was a small price to pay for the convenience of dual rails. This design later gained a moving-coil head amplifier. I had designed several of these stages before, at least two of which made their way into commercial production, but this was the first version that got both noise and distortion down to what I considered to be acceptably low levels. The salient features are the discrete transistor input devices which then, and indeed now, provide the best possible noise figure. At the time many head amps were outboard units, often relying on battery power, presumably to sidestep intractable ground-loop problems. However, no difficulties were found in grafting this design onto existing preamplifiers.

Some years later, having devoted much time in between to power amplifier design, I felt the call to take another look at preamplifiers. My last design was twelve years old, and it seemed likely that some significant improvements could be made. Much thought and a lot of calculation and simulation led to the ‘Precision Preamp 96’ articles, including in-depth mathematical modelling of the noise generated by the RIAA stage. This allowed each noise contribution to be studied independently, and permitted comparison between the actual noise and the theoretical minimum. The latter is rather sensitive to the exact assumptions made. It was also possible to discover why op-amps that appeared to be quieter than the 5534 in theory, were actually slightly noisier in practice. The answer was that op-amp bias-cancellation networks maybe great for DC precision, but the extra common-mode noise they generate in audio circuitry is just an embarrassment. The 96 preamp is in fact not so much an updated version as a thorough re-design, with only the moving-coil input amp remaining essentially unaltered. It demonstrated, amongst other things, that obtaining an interchannel separation of 100 dB on a stereo PCB is perfectly possible with careful component and track layout.

In 1990 I had again turned my attention to power amplifiers. For many years I had felt that the output stages of power amplifiers presented very great possibilities for creative design, and so I explored some of them. One of the first difficulties I met with was the problem of determining how much of the overall distortion was produced in the small-signal sections, and how much was generated by the output stage. Traditionally the latter was regarded as the major source of distortion, but there was very little published research to back this up, and so I attacked the problem myself. When I began it was not clear if there were two, twenty, or two hundred significant distortion mechanisms, but after a good deal of study it suddenly became clear that seven or eight were sufficient to explain all the observable distortion. This is not to say that there are not other distortion mechanisms—there almost certainly are—but the non-linearities they produce are currently below the level of practical measurement with THD analysers. When output stage distortion has been banished forever (and it has to be said there is no sign of this happening in the immediate future) then it may be time to dig into the deeper levels of non-linearity. It quickly became clear that by taking a few simple circuit precautions, it was possible to design amplifiers with very much lower distortion than the norm. Such amplifiers, with their very low THD figures, are rather distinct from average designs, and so I looked around for a suitable name. Once the critical factors are identified, designing an low-distortion amplifier becomes more a matter of avoiding mistakes rather than being brilliant, so I decided to call them ‘blameless’ amplifiers, rather than ‘hyper-linear’ or something similar, to emphasise this. The results and conclusions of this major investigation were published in eight parts as ‘Distortion in Power Amplifiers’. ‘Distortion Residuals’ followed this up, providing a visual guide to the appearance of the various distortion mechanisms on the oscilloscope screen.

These endeavours built up to a substantial body of information on just how to minimise amplifier distortion, and two designs that exemplify this are included in the ‘Distortion In Power Amplifiers’ articles, one working in Class-B and the other in Class-A. This foundation of knowledge simply begged to be put to further use, and so two major power amplifier projects were created; the trimodal amplifier based on Class-A, and the load-invariant amplifier in relatively conventional Class-B.

The trimodal amplifier demonstrated how to make a Class-A amplifier that coped gracefully with varying load impedances. This project had its roots in an insistent demand for a PCB for the Class-A power amplifier presented in the last part of ‘Distortion in Power Amplifiers’. I find it goes against the grain to reproduce a design without trying to improve it, and the trimodal article was the result. My first intention was to demonstrate how a Class-A amplifier could, with appropriate design, move gracefully into a relatively linear version of Class-AB when the load impedance became too low for Class-A operation to be maintained, rather than clipping horribly as some configurations do. Improvements were also made in the noise and DC offset performance of the basic amplifier. I was concerned to guard against catastrophic currents flowing if there were errors in building the quiescent-current controller, so a safety network was added to set an upper limit on the bias voltage. It was simple to make the amplifier switchable between A and B by changing the limiting value of this second bias circuit, and the trimodal was born.

The load-invariant power amplifier project was a direct development of the work done on amplifier distortion. Power amplifiers always give worse distortion into lower load impedances. For bipolar output devices, as the load value drops from 100 Ω to about 8 Ω, the crossover distortion increases steadily and predictably. However, at about 8 Ω (depending on transistor characteristics) an extra low-order distortion appears that can easily double the THD at 4 Ω. I decided to see to what extent I could thwart this extra distortion, aiming to produce the first semiconductor amplifier that gave exactly the same THD at 4 Ω as it did at 8 Ω. While it did not prove possible to quite attain this, I did get reasonably close. This design seems to have generated a lot of interest.

A few of my articles have been written in reaction to contributions to Electronics/Wireless World that suggested promising new approaches, or that I simply found intriguing. Investigating a particular amplifier topology takes a lot of time and effort, but preparing the results for publication does not add a great deal to this, so some more articles resulted. They may not have advanced the art of audio greatly, but they did explore a few paths which would otherwise have remained untrod. Two examples are given in this book: ‘Common-Emitter Amplifiers’ and ‘Two-Stage Amplifiers’. In neither case were the results sufficiently encouraging for me to proceed further with the concepts involved. When the idea that loudspeakers could, in certain circumstances, draw much more current from an amplifier than its impedance curve suggested first came to my notice in the mid-1980s, I must confess I felt a degree of scepticism. I was wrong; the effect is real, though its relevance to real-life signals rather than artificial stimulus waveforms is rather doubtful. The abnormally high currents that flow are provoked by using the stored energy in the circuit elements, such as the inertia of the speaker cone. This requires a rectangular stimulus waveform with rapid full-amplitude transitions, carefully timed to catch the speaker resonance at its worst moment, and the difficulty is that real waveforms do not have these. Eventually I got around to putting the idea to the test, using an electrical analogue of a speaker system. The article ‘Excess Speaker Currents’ describes the effect and how to produce it.

When semiconductors were first applied to audio amplification, the choice of operating mode was simple: Class-A or Class-B. It took several years before proper complementary pairs of output devices were available, but it was clear that they allowed a good deal more flexibility in the design of output stages, and variations on the standard configurations began to appear, becoming more radical as time went on and the technology of audio developed. The 1960s gave us Class-D, though the rudimentary versions available then were not much of a gift. The 1970s saw the advent of the Blomley concept, current-dumping, and, significantly, Class-G. In this situation it was inevitable that extra letters of the alphabet would be called in to describe new methods of operation, though it was clear that calling something, say, ‘Class-Y’ said nothing about how it operated, and the prospect of trying to remember what an alphabet soup of 26 (or more) class letters actually represented was not enticing. With this in mind, I produced the article ‘class distinction’ which attempts to simplify amplifier classification into simple combinations of A, B, C and D in such a way that at least some information about the mode of action is given. I will not pretend that I expect my classification system to sweep the world overnight. Should it fail to sweep the world at all, I think the article is still useful because it allows the generation of a matrix of amplifier types, some of which no-one has got around to inventing yet. Give me time.

Power FETs first began to reach the market in the mid-1970s, and as so often with new technology, they were claimed to be superior to existing methods in just about every possible way. However, experience soon showed that FETs were not dramatically more linear than bipolar transistors, nor were they inherently short-circuit proof. There appeared, however, to be definite advantages in their high bandwidth and freedom from carrier-storage effects. Initially I was intrigued by the possibility that power FETs, with their much-advertised speed and bandwidth, would allow the implementation of various ingenious output stages involving local feed-back that in bipolar format had proved difficult or impossible to stabilise, apparently due to the slowness of the output devices. The results were not encouraging: any increase in stability due to the faster devices was more than outweighed by their tendency to parasitic oscillation when used in anything other than the simplest of configurations.

Most serious power amplifiers are fitted with a muting relay that disconnects the electronics from the loudspeaker at turn-on and turn-off, to reduce thumps and bangs. The same relay is used to protect the loudspeaker from incineration if the amplifier suffers a fault which puts a large DC voltage on the output. Because of the safety implications, this relay must operate promptly and reliably, and designing its control circuitry is not trivial. My article on relay control delved deeply into the control circuit design, with special emphasis on a rapid relay response to dangerous conditions or intrusive transients; this is not, as it might appear, merely a matter for the relay designer. Apparently tiny details of circuit design have a major effect.

The final two articles in this book introduce a new way of displaying amplifier efficiency and power dissipation that I call ‘power partition diagrams’. The first one dealt with many different kinds of amplifier, and studied how they disposed of the power involved in driving resistive and reactive loads with sine waves. This produced the interesting conclusion that a proper appreciation of peak transistor dissipations into real loads with phase-shift could be the most crucial factor in determining amplifier reliability. Whenever sine waves are use for testing—and there are many good reasons why they should be used sometimes—the criticism is likely to be levelled that this is unrealistic, which of course it is. The second article therefore looked at the statistics of music (which are surprisingly obscure) and ways of calculating true power dissipations from the results. This showed, amongst other things, that Class-A amplifiers are in reality not more than 1% efficient. This raises serious questions about their desirability in an energy-conscious world.

In the course of the investigations that led to these articles, I found over and over again that the conventional wisdom on power amplifiers was more conventional than wise. Some examples are given here, though they may not make much sense until you have read the relevant article. Some statements turned out to be half-true: an example is the widespread assumption that a current-source loaded Voltage Amplifier Stage (VAS) gives current drive to the output devices. The reality is much more complex; the impedance might be high at low frequencies, but the Miller capacitor around the VAS causes the impedance to fall with frequency until it is a few kilo ohms at the top of the audio band. This hardly counts as a current source. The drive point is also loaded by the non-linear input impedance of the output stage, which complicates matters further. There is much more to most questions about amplifiers than at first meets the eye.

Some long-held beliefs turned out to be completely wrong, though plausible in theory and workable in practice. The best illustration of this is the universal belief that the crucial parameter in biasing a Class-B output stage to minimise crossover distortion is the quiescent current. In actual fact the critical factor is the voltage across the emitter resistors. If the value of these are changed, then the quiescent current can be radically altered although the amplifier remains at the same optimal bias point, because the voltage drop is unchanged. However, emitter resistor values are rarely changed in this way, so setting up for a given current is the same as setting up for a given voltage. The difference is unimportant if you are simply repairing or adjusting amplifiers, but vital if you seek to understand how they work.

In some cases there was no argument about the distortion mechanism operating, but very little, if any, published information quantifying the size of the effect. This applied to most of the amplifier distortions examined, and it took a little thought to develop ways of measuring each one separately, wondering in each case why the relatively simple test had apparently never been done before. It may be, of course, that parts of this work have also been done by various audio manufacturers, who have every reason for keeping their private research to themselves.

This collection of articles is not totally exhaustive, as various topics that have already been fully expounded in The Audio Power Amplifier Design Handbook have been omitted. There is some overlap in the material relating to distortion, but that is unavoidable if each book is to stand alone. The articles here are not in chronological order as it is more useful to keep preamplifiers and power amplifier material grouped together. Writing the articles reproduced here has been a stimulating intellectual journey.

I hope that reading them will share some of the sense of discovery that I felt, and that this collection will be both useful and entertaining to all those concerning themselves with audio electronics.