During the last few years, valves have enjoyed a return to favour with audio enthusiasts and now professionals are seeking their characteristics and perceived advantages within studio recording equipment.

This article attempts to analyse the differences between valve and solid-state, both objectively and subjectively.

Five years ago, whilst browsing in a newsagency, my attention was drawn to an English Hi-Fi magazine. What attracted me was a cover illustration of a very ornate copper-plated chassis, bearing four large output valves instantly recognised as type 2A3, which had made its debut in the early 1930's. Even though these valves were obsolete long before my passion for music and quality reproduction began in the late 1950's, I was aware of how successful and how popular these valves had become as audio output types. Yet here they were being used in an amplifier manufactured in 1992! My curiosity aroused, I bought the magazine with the intention of discovering why an amplifier with such obviously modest specifications would interest today's enthusiast.

The amplifier produced 7.5 watts of power, had zero negative feedback, (NFB) and used paralleled output valves in single-ended configuration, once associated with less expensive amplifiers. Push-pull operation had long been favoured over single-ended for its lowered total harmonic distortion (THD) content. Why then was this design selected for an amplifier competing with sophisticated solid-state designs with superlative specifications orders of magnitude better than it's own?

The reviewer was however, very impressed with the performance, according it very high honours indeed. It was about this time I wondered about the cost of such a beast; being hand-made, it surely wasn't cheap. I was right about this at least, it carrying a retail price of 6,900 pounds sterling or 8,900 pounds for the deluxe model with silver-wound transformers and silver-foil capacitors! I was directed to an article in the same magazine (more in the form of an advertisement) relating to other zero NFB amplifiers of considerably more power, including one with push-pull 845 transmitting triodes in class-A yielding 50W. RMS that sold for-wait for it- 90,000 pounds sterling- no, this is not a misprint, ninety-thousand pounds! I closed the magazine in somewhat of a shocked state, muttering something about fools and their money!

Now I must confess, along with my solid-state Quad 405 current-dumping amplifier, which has given satisfactory listening for many years, I do own a pair of Quad QII valve power amplifiers which sound very nice, but I guess the lure of all the power margin afforded by the 405 caused me to only use the valve QII's in cases of extreme nostalgia. Yes, I like valve amplifiers, they look very attractive, they sound different in some respects to solid-state models, on some recordings even dare I say better, but on others the sheer dynamics of powerful solid-state models couldn't be ignored. You must understand that during my formative years, I developed a close bond with things thermionic and perhaps I may be a little over-biased (no pun intended). However, when solid-state amplifiers came of age in the latter part of the 60's, I embraced the new technology with all the enthusiasm expected of an audio enthusiast in a quest for louder reproduction of such classics as Credence Clearwater Revival, Elton John, Wings and similar. As my tastes matured, I found a liking for female vocals, but observed that the female, and to a lesser extent the male voice, always seemed to lack clarity on the sibilant or "ess" sounds. Years later after acquiring the vintage QII valve amplifiers, I found that the reproduction was noticeably better than the solid-state amplifier that I was using at the time. However, following the acquisition of the Quad 405 100-watt per channel unit, the performance seemed to rival the valve QII's in this respect and so with the higher power reserve this became my amplifier of choice for many years. Shortly after followed the introduction of digital in the form of Compact Disc. This represented a huge step forward in my mind- no surface noise, superb dynamic range, no apparent distortion to name just a few advantages. This was the perfect recording medium- or was it? Arguments soon arose questioning sampling rates: were they inadequate? Did CD introduce artefacts more unpleasant than the ones we were accustomed to on vinyl? This even spawned a new generation of "vinyl buffs" who adamantly and fervently defend vinyl as true fidelity. Not just "music by numbers" resulting in a harsh, clinical, sound as evident on CD. I wouldn't and indeed do not have the right to disagree with their findings. They may well be right, but I'm not convinced. As a pure trade-off, just the absence of surface noise, scratch, plops and rumble tips me well in favour of digital.

Having worked in the broadcast profession for some years had given me the opportunity to listen to live material prior to all the equalisation and compression necessary to squeeze it all onto a vinyl disc, and to me, CD had the same so-called "hard" quality to my ears; it was more realistic in the sense that it sounded similar to live reproduction of music via the control room monitoring which was very good indeed.

Amplifier specifications have continued to improve, today in the solid-state arena there are better offerings than my Quad 405, but the actual sonic qualities, to my aging ears at least, did not seem to change. I had to come to terms with the fact that approaching 50 years of age, I would never again hear any improvement on what I already had.

My interests having turned to computers led me into new and exciting areas of design using them as a tool. In recent times this led me to the Internet with its astonishing ability to research any topic imaginable. My instincts drew me automatically back to my electronics origins; my original passion for things acoustic. I typed in my search string, "audio amplifiers" and was rewarded with a very large number of "hits". The thing that became quickly apparent was that in the year 1996 the phrase "tube audio" and related topics was present a surprising number of times. Unable to resist, I focused the search onto valve related audio topics and was amazed at the amount of information, interest and enthusiasm shown to this area. I was aware that musicians have always regarded tube (sorry, valve) sound as superior, but this was, I felt, due to the addition of harmonics to their instruments by "tweaking" their amplifiers. In short, the amplifier was part of their instrument. The high accolades now bestowed upon valve amplifiers were not coming just from musicians but from audio enthusiasts as well. Curiously enough, single-ended class-A triode, zero negative feedback amplifiers were high amongst the favoured designs. This caused me to recall the magazine articles I had read several years earlier and stirred considerable interest in me. To my recollection, I had never heard one such amplifier. How much trouble would it be to build one? Why not I thought. Shortly after this train of events, I found a medium quality but generously rated single-ended output transformer at an auction and purchased it. Several weeks later, after finding sufficient parts in my junk box, I constructed an amplifier of this type. The design, I reasoned would need to be carefully thought out to produce minimum distortion and good damping without the use of NFB. I selected a very low plate impedance twin-triode designed primarily for regulated power supply and servo amplifier service (type 6080 from RCA). Unfortunately in order to gain low plate impedance, inter-electrode spacings within the valve must be small, which in turn limits the mu to only two, with a consequent available voltage gain of even less!

To avoid a step-up driver transformer, a voltage amplifier stage running on no less than 700 volts was necessary to get adequate linear drive for the output valve. I was beginning to understand the move from valve to solid-state! Finally my efforts were rewarded with a power output of around 8 watts but the distortion could not be coaxed to much less than 2% at around 6 watts. This was going to sound very ordinary or so I thought.

Fortunately, my loudspeakers are very efficient and I reasoned that if I kept the volume down it might give me a "listenable" result. After all, I had come this far; might as well try it. The unit was connected to just a CD player and volume control and all was ready for the moment of truth.

I played a familiar CD and was surprised at the sound quality. It was actually very good. I advanced the volume; It was still good, quite a useable amplifier. I advanced the volume further. The amount of sound level before obvious overload was nothing short of astounding for eight watts! Remember, I had been accustomed to 100 watts per channel previously.

Sure it wasn't the powerhouse that the Quad 405 was but it was unbelievably loud. Eight watts! Something's wrong; maybe it somehow routed through the Quad. Double check; no it's the 8 watter I'm listening to. There was something else as well. The sound had more "punch". Percussion had a really tight (referred to many as "fast") sound as it should. Instruments could be heard individually and were readily indentifiable. I wonder what vocals sound like? I quickly reached for a popular favourite which I felt was a particularly good recording of a powerful female voice. The result was breathtaking. I had never heard such clear sibilant sounds; no "fat" sounding esses. During quieter passages vocalising ending with the tee sound could be heard clearly enunciated. I reverted to solid-state Quad 405. Yes, if I listened hard I could hear the tee sounds but with the SE triode class-A amplifier they could be heard effortlessly. That night I tried to listen to every CD I owned but since they number close to 100 this was not possible. Every disc I listened to delighted me. The transient response was significantly better than anything I can recall listening to ever. It should be remembered that I performed this test with poor expectations. Why was this amplifier with marginal specifications giving such magnificent reproduction?

The only possible explanation to me was that our time-honoured methods of quantifying amplifier performance, that is to say harmonic analysis clearly missed some vitally important parameter(s). It was clear some further investigation was required.

Part 2.

I was now faced with a perplexing question. For decades harmonic analysis was used for analysing audio amplifier performance and there were valid reasons for this. Before the advent of electronic reproduction, musicians had used their knowledge of harmonics to aid and assist their design and crafting of musical instruments. Discontinuous functions such as those that represent the complex waveforms as found in music are virtually impossible to analyse by mathematical means, so harmonic analysis of subjectively evaluated distortion was the logical solution.

But the point is this- making a reproducing system to meet the harmonic distortion criteria alone does not ensure accurate or even pleasant reproduction, or from a different perspective it is possible to create two amplifiers with similar frequency response and very low THD that sound completely different. This led me to the conclusion that THD measurements alone were not enough, so I needed to understand more about the human hearing process. I sought texts and articles relating to the physiology of the ear and our interpretation of sound. These were informative but some of the most relevant material was found in the "Radiotron Designer's Handbook", written by F. Langford-Smith, (4th. edition). Those familiar with this amazing reference work will need no introduction, but those who are not are urged to find a copy if possible. Recently this book, last printed around 1967, has just been released on CD ROM, such is its considered value.

I also collected as many issues as possible of the American "Audio" and "Audio Engineering" spanning

the period from the late forties to the early sixties. These contained many classic articles from eminent authorities on the subject, outlining discoveries and work which has since been adopted in our present-day designs. Many writers had been involved with leading recording studios and research organisations such as Bell Laboratories. What I discovered was very interesting. After almost forty years in the pursuit of high-quality reproduction, I had overlooked some very important factors, and dare I say, others appear to have done the same. The questions of transient distortion and concerns for the inadequacies of THD analysis alone in quantifying performance were almost universal. Many writers expressed their concerns of the possible overuse of negative feedback and its destructive effects on transients and particularly the blurring of subtle phase information essential for two ears to locate a sound source's position. And yet we still occasionally read of the ear's inability to hear phase distortion!

This is partially true as it does not manifest itself in the same unpleasant way as dissonant harmonics but could the lack of clear phase information due in part to NFB lead to "listener fatigue", a subtle but well-known effect of listening to recorded music. So perhaps this new found clarity with the SE triode amplifier was due to the absence of NFB but why was the sound so clean and undistorted when we are continually being encouraged to use amplifiers with orders of magnitude less THD than the 2% that this amplifier was producing?

In due course I will refer to specific authors and their articles in the context of the present discussion, but if you have access to Radiotron Designer's Handbook, (or the Radio Designer's Handbook as published in the UK), I suggest reading chapter 14 titled "Fidelity and Distortion".

Now, returning to the question of THD. When I measured around 2% THD, this as the name suggests, included all harmonics, and having tested my own hearing I found that I could detect approximately 1% THD when listening to a 1kHz sine-wave source. I summarised therefore, that 2% would be objectionable. However, I completely overlooked the fact that the 1% THD I could hear was coincident with the onset of clipping in the test set-up, which Fourier analysis shows to comprise odd-order harmonics. So just as "Oils ain't oils", THD ain't THD.

One of the articles that gave an excellent insight into amplifiers and their differing THD contents gave, I believe, the best clue yet why SE class-A triodes can sound so good without the "benefit" of NFB. According to Langford-Smith quoting 2A3 triode class-A distortion figures, distortion present was second harmonic, 3.16%, third harmonic, 0.03%, anything above less than 0.01%! Bear in mind that musically related harmonics of which the second certainly is, are not unpleasant as are dissonant harmonics. It has been stated that second order harmonics are inaudible at levels less than 5%. An interesting observation is in the case of push-pull due to even-order cancellation the third becomes the dominant harmonic and if the drive is increased to produce the same distortion percentage, the more unpleasant and dissonant higher orders now dominate. This could explain why SE amplifiers are considered by many to sound better when approaching the inevitable short duration clipped transients than some other classes. Incidentally, I believe that the presence of third harmonic causes a reduction in loudness of a particular note, an effect known to musicians as "blanketing". From Figure 1, it is apparent that severe (20%) third harmonic causes a flattening of the waveform, which most likely causes a reduction in perceived volume. Possibly this could account for the "punchier" sound attributed by some to SE triode amplifiers? To summarise, some systems with THD as high as 5% may be quite clean sounding, whereas others with 5% THD composed of different harmonics may sound terrible. My feeling is that designers erred on the side of caution when reducing THD, but with some transient artefacts still present, felt they needed to go even lower and as an additional measure, provide more power reserve. It is my conjecture that I will attempt to substantiate, that the use of lower impedance power supplies and output devices in these higher powered amplifiers probably contributed to the major part of the improvement. At this point I would like to reiterate that these are largely speculations and conclusions on my part but indeed a significant number of authors over a period of three decades have lamented that insufficient research has been conducted in many psycho-acoustic areas. Therefore I believe if our ears are telling us it sounds better we should not solely rely on harmonic analysis and incomplete science to contradict them. Rather this suggests that more research is needed and this point is made by many prominent authorities. Before moving on, I feel it appropriate to mention damping factor (DF), another constantly diminishing parameter with the advance of amplifier performance. My experience with the design of phase-locked loops (PLL) as used in frequency synthesisers taught me the danger of over-damping which was almost as disastrous as under-damping in terms of stability. Whether we are discussing PLL's, servo-loops or negative feedback loops that incorporate the loudspeaker, it is all based upon control loop theory. With the design of loudspeaker enclosures, we critically damp the speaker cones, and yet do not make allowance for the damping conferred upon the speaker by the amplifier. It seems the highest DF is most desirable in current amplifier design philosophy. One of the unexpected bonuses with my prototype was the significant improvement in bass performance. Considering the concerns about output transformer core saturation due to the DC unbalance flowing in the primary winding, I thought the bass (LF) response would be poorer than that to which I was accustomed. This fear proved to be unfounded. Far from being boomy or excessive as one may expect from insufficient damping, it was solid, with very low note extension. Kick-drums and stringed-bass instruments sounded very natural. This is possibly due to having a more realistic damping factor of around 6 to 8 rather than the common 50-100 found in more recent designs. It should be stressed at this point that the lowering of output impedance by the use of NFB will not assist an amplifier with insufficient drive capability to provide the energy required to reproduce the rapid or "fast" transient demanded by the loudspeaker. (Transient response is also important for good LF performance as well.) To emphasise this point, output impedance is occasionally calculated by noting the reduction in output voltage when the amplifier is loaded with its nominal load impedance (say, 4-ohm) and extrapolating the value of resistance which (theoretically) when placed across the output of the amplifier would reduce its voltage to one-half of the no-load value. To clarify this point, consider an amplifier designed to drive a 4-ohm load with a power of 100 watts and having a damping factor of 100. (Fairly common) Loading the amplifier with a 4-ohm load would result in a 1% drop of the no-load value. Extrapolating this result, a load of 0.04-ohm would cause the output to drop to one-half of its no-load value. A quick calculation reveals that this would result in a power dissipated in the load of 2,500-watts! Clearly the amplifier could not provide this amount of drive. The situation with our triode amplifier not employing NFB to reduce output impedance is however, quite different. My 8-watt prototype can quite happily be loaded up until the output voltage approaches one-half of the no-load value, where the power reduces to a predictable 3-watts. Admittedly the DF of this amplifier is less than 10 and the distortion is degraded excessively, but clearly the current capability is "real", resulting in a very noticeably "fast" amplifier. Therefore I reiterate my earlier comment regarding the need for very low source impedance when it is clear the current demand suggested by such a value is simply not available.

So returning to the subject of NFB. What of all those line amplifiers, equalisers etc. as used in recording studios, the dozens of NE5534 or similar low-noise operational amplifiers, all with very large amount of NFB applied. Well, the fact is they behave very transparently, seemingly not degrading transient or phase information at all and it would likely be undesirable to use valves in their place. They exhibit flat frequency response, low noise and distortion and can be cascaded to a large degree without problems. Why? Because they have low output impedances feeding successive stage high-impedance loads.

The problem arises when a reactive element such as a loudspeaker is introduced into the global negative feedback loop. Its complex load characteristics which result from its generated "back-EMF" are responsible for many of the problems encountered. I will use the term "back-EMF" to describe the voltage generated by the loudspeaker in response to its being put in motion by the amplifier signal. I am more comfortable when using the term to describe the voltage generated by the collapse of a magnetic field in say, a transformer, rather than being generated as in this case, by dynamo action, namely by a conductor cutting across a static magnetic field. Remember, the loud-speaker is just as effective a generator as it is a motor.

At this point, I feel it is probably appropriate to discuss the critical relationship between the output stage and the loudspeaker, as this is where the electrical signal is converted back to acoustical energy that is once again audible. Ideally we would like the process to be as efficient as possible, but there are impediments to our goal. At first glance, it would seem logical to power match the load (say 8 ohm) to the source such as it is done with radio transmitters and the like, but power-matching should not be confused with highest efficiency. Power-matching simply refers to the condition where the maximum power possible can be extracted from the source and does not mean the highest efficiency energy transfer. Remember, efficiency as a percentage is (audio power out / DC power in) x 100%. Typically, a class-A amplifier is generally stated as having a maximum efficiency of 33%. Fairly obviously, this varies from maximum efficiency at rated output down to zero-percent efficiency at zero signal level while the nominal impedance ratio of source to load impedance has not changed! By definition, the conditions for a maximum transfer of power from source to load occur when the source impedance is equal to the load impedance. Clearly from our damping factor example, it is far from power matched. To further illustrate this point consider our 240volt AC mains power supply. We are clearly interested in efficiency, and yet no attempt to power match is sought. It is far more important to maintain a constant voltage of 240 volts at the consumer's outlet irrespective of load. Power-matching would mean a voltage variation of almost 2:1 between low-load periods versus full-load periods. This would obviously result in an intolerable situation. Reverting to our radio transmitter example where the load characteristics are well defined in addition with the fact that we wish to get the maximum power into the antenna, power-matching is desirable. Let us now consider a hypothetical loudspeaker with zero cone-mass and with an infinitely powerful magnetic field. This would result in a back-EMF equal in amplitude and phase-angle to the drive voltage applied to the loudspeaker, resulting in cancellation of the current flowing in the voice-coil. In other words, our constant-voltage source amplifier would be driving an infinite load impedance. In the real-world, the back-EMF is considerably less than the applied voltage and the phase-angle of the resultant current is consequently lagging due of the energy required to accelerate the cone and the surrounding air-mass. The load's reactance as a result changes from being inductive to capacitive, depending upon the degree of phase shift. From our vector diagram (Figure 3) showing the relationship of impedance with reactance and resistance we see that as the phase-lag approaches 90 degrees, the load appears inductive (+j), becoming resistive again at 180 degrees, then capacitive (-j) at 270 degrees until finally, at 360 degrees or one whole cycle, resistive again. Examination of a typical impedance v. frequency graph of a loudspeaker reveals how much variation in phase-angle is experienced over the audio spectrum. Ideally, we would aim for a non-reactive load where the impedance is constant, ie. equal to the resistive or DC resistance value of the voice-coil (VC). This would ensure a more constant current through the VC with frequency variation, thereby ensuring a correspondingly proportional magnetic field to be generated. Remember, it is the generated magnetic flux that performs the work of moving the cone, and so in striving to maintain constant impedance, we ensure that the loudspeaker responds more faithfully to the applied signal. As we increase the resistance of the VC, ie. tend towards a constant-current drive, this tends to minimise the effect of VC reactance upon the impedance. However, it is obvious that more voltage-drive must be applied to the VC to provide the same current. This of course, results in higher applied power, and this is simply lost in the form of dissipated heat. Another approach would be to phase-equalise the load at the output of the amplifier, and many amplifiers incorporate a "Zobel" or step-network connected across the output. This is very limited in its effect, as it only compensates for the general trend of rising loudspeaker impedance with frequency. To be really effective, such a network should provide a conjugate load to correct for every reactive or impedance variation in the impedance graph. This is clearly not possible, and even if it were, its characteristic would need to be changed for loudspeaker loading changes! Therefore, this power-wasting resistance is perhaps the only practical approach to minimising impedance variation. Let us now look at how the amplifier's source impedance is related to the load impedance in proportion to the damping factor. For example, with a DF of 10, the source impedance would be 0.8 ohm. Apart from the risk of overdamping the loudspeaker, it is obvious that an increase of damping factor will not be significant when considering the comparatively large dc resistance of the voice-coil. What is the point of this criticism of over-damping? Well, in order to achieve this damping factor, usually large amounts of NFB are employed, and it is the effect on transient response and phase detail that is of concern.

Let us now consider the sequence of events that occur during the passage of signal as it travels around the global feedback loop. The signal passes through to the loudspeaker, which, as we have discussed, produces a voltage of its own. (back-EMF.) This voltage is fed back to the input of the amplifier in such a sense as to subtract (hence the term "negative" feedback) from it. This in turn causes a reduction of the output fed to the loudspeaker, therefore, in the case of say a cone resonance which would result in a larger voltage generated by the freer moving cone to be subtracted from the input, thus reducing the drive from the output and compensate (hopefully) by damping out the excess movement. This works very well as we are aware, especially when demonstrated using a sine-wave signal of a relatively low-frequency but what happens to fast rise-time transients? Many designers believe it is sufficient to simply compensate by using a phase-lead capacitor. I deplore the term, phase-lead in this context, the implication being that somehow a future signal can be anticipated. As any video engineer can tell you, this is only possible when the past history of the signal is known such as when using a repetitive signal source, eg. a square-wave.

Consider such a square-wave signal fed through a NFB amplifier without such compensation into a resistive load and observed on a cathode-ray oscilloscope (CRO). The above amplifier has some inevitable phase-shift at the point of response roll-off. Using our harmonic analysis theory, the higher odd-order phase-shifted harmonics are causing time-shifted additions and subtractions resulting in the effect we know as ringing. By the addition of our phase-lead capacitor across the series NFB network, we are actually further lagging the high odd-orders so that the "next available cycle" can restore the signal to an approximation of a square-wave. Can you imagine what this does to a sudden single transition as found in a typical musical waveform? Of course the effects are very subtle and we recognise the instrument by its pattern of sound, such is the wonder of our auditory sensory mechanism, but vital phase information required to pinpoint positional information has been seriously degraded in the process. Naturally, the system as a whole has degraded this quality anyway, such are the imperfections of even the finest reproducers, but the additional degradation due to NFB is very significant. Further to my comments regarding the human ear, I would emphasise that the ear is very sensitive to fast rate of change sound pressure levels or transients, rather than the slower rise-times as we would find with pure tone or sine-wave. Have you ever considered the SPL of the sound of a pin dropping (fast rate of change) which is clearly audible in a quiet room?

In a particularly good article from "Audio" magazine (December 1962 p34) written by George Fletcher-Cooper a regular contributor at the time simply titled "Series Feedback", he makes the following observations. A transient arriving at the loudspeaker causes a resultant voltage to be generated by the loudspeaker which after being fed back and subtracted from the input arrives (after the delay caused by passing through the amplifier again) at the loudspeaker, where the process is repeated indefinitely. This according to Mr. Cooper has the potential of doing sufficient harm to the transient wavefront shape to be of concern. He shows this diagrammatically and mathematically in the article.

The conclusion from this is simply testing an amplifier for THD at various frequencies (sine-wave) would not disclose this form of transient distortion. Clearly many high-powered solid-state (and expensive) monitor and reference amplifiers using NFB do not fall into this criticism. I believe that the explanation lies in the huge current-sourcing capabilities of these large amplifiers, many capable of hundreds of watts typically used in professional applications. Having such high current capability ensures a very low output impedance (inherent) therefore not relying on NFB to perform this task, thus significantly reducing the loudspeaker's back-EMF voltage contribution to the voltage fed back. Note that NFB only reduces output impedance for in-phase signal components at the summing node for source and load. Generally speaking, this is open-loop source impedance versus DC resistance of the loudspeaker. This allows the NFB to linearise the characteristics of the amplifier proper, a job it performs very well, while the loudspeaker damping is provided by the inherently low source impedance. The reason given for the use of high-powered amplifiers only to provide sufficient peak overload margin may be a fallacious one or at least not the whole story.

The subject of NFB and the problems involved with achieving unconditional stability and satisfying Nyquist's criteria for stability would fill a volume alone, and still we have the dilemma of the effect of the reactive or complex load when connected, so lets look at the possibility of using an amplifier that uses zero NFB, thereby side-stepping all these problems.

Unfortunately, this limits our options, as a triode valve operated under class-A conditions would appear to be the most linear and simple amplifier we have available to us. Thus we must accept the disadvantages associated with such an approach, such as high-voltage, heat, inefficiency and the need to use an output transformer to name but a few.

Perhaps is now a pertinent time to look again at our SE, zero-NFB triode amplifier and review various types of harmonic distortion and their respective audibilities. The graphs included show computer simulated waveforms of fundamental plus second harmonic only and fundamental and third harmonic only. A relatively high percentage of 20% was used in order to emphasise the differences. Recalling that as we approached the onset of overload, and thus the limit of usefulness of the amplifier, our distortion looks similar to the graph showing second harmonic. With the use of push-pull, the cancellation of even order harmonics allows us to push the usefulness to the point of odd-order harmonic distortion, and with the addition of NFB, the onset of clipping becomes more aggressive. On the surface we can conclude well, so what? At a given power the performance is still just as good and we have heavily reduced distortion allowing considerably more power reserve. Well, unfortunately, in the interests of producing more efficient and cost-effective amplifiers in conjunction with losing much of this power by the use of less efficient loudspeakers, we restored the status-quo with the result that for the same acoustic output, peaks and short duration overloads are exhibiting very different harmonic distortion characteristics. It is a well-known fact that the harmonic content of a musical instrument increases as it is played louder. This is in the case of say, a stringed instrument due to the limits of vibration excursion of the string being approached, and clearly the resultant compression of the waveform resembles that of second harmonic distortion, the same as in our SE class-A triode amplifier. Perhaps this is why these amplifiers sound apparently louder, as the resultant distortion is more familiar to us.

The explanation for the presence of second order harmonic that resembles "squashing" of one half-cycle of the signal is as follows. During the half-cycle of signal causing the single output valve's (or transistor)

current to increase, the resultant plate (or collector) impedance is lower than the opposite half-cycle when the device is approaching device cut-off point. This results in a lower output voltage in the load and the degree of compression increases with reduction of load impedance. A glance at the transfer characteristic of the output device will confirm this point. With the push-pull amplifier came the solution and the trick was to have matched output devices such that as one was approaching cut-off, the other was beginning to conduct and in the ideal case, (matched output devices) both would work to approach a more constant output impedance. This effect could be employed according to the whims of the design engineer and became known as the class of operation. These are commonly class-A, AB1, AB2 and B. Accordingly NFB was used to reduce forms of cross-over distortion as a result of less-than-ideal device matching. It is the author's opinion that this form of distortion is the most objectionable by far, so much care needs to be taken by the designer employing class-B. The types of THD present in push-pull amplifiers have already been covered.

Well at this point having had the audacity to question so many accepted norms, possibly people may well wish to point out flaws in my discussion (I prefer not to think in terms of an argument, as this article is simply conjecture on my part) and their comments, suggestions, corrections etc. would be most welcome as it is the acquisition of knowledge in these fascinating fields that is of prime importance to me. It seems the more I research and learn only serves to highlight my ignorance of this topic.

If I have failed to convince by objective reasoning, then that is due to my inadequacy, but one thing that I state emphatically and having enthusiastic approval from all who have heard my prototypes, SE triode, zero-NFB amplifiers are much more pleasing to listen to and any person wishing to purchase a Quad 405 amplifier can contact me and I will be pleased to negotiate a deal!

In conclusion, it is perhaps not hard to see why amplifier designs have (in my humble opinion) erred slightly from the ideal path. Firstly, more "efficient" pentodes were utilised to allow radio manufacturers to use at least one less valve, then class-AB afforded higher efficiencies, followed by many designs intended to restore pentode performance to more closely approach, but never quite reach, triode performance. Several very successful examples include the development of the beam-power or "kink-less" tetrode and partial triode operation, otherwise known as "ultra-linear". These were cost motivated in a very competitive radio and later hi-fidelity market. With the advent of solid-state and complementary NPN-PNP devices it enabled the elimination of the costly and unwieldy output transformer and without the phase-rotation effects around the band-edges of the frequency response, highly efficient class-B could be employed using huge amounts of negative feedback to bring the THD (in some instances tested at 1kHz only!) down to levels we were comfortable with for valve amplifiers. In order to satisfy the requirements for unconditional stability, considerable HF roll-off within the feedback loop had to be introduced. With many of the early solid-state amplifiers any attempt to produce the stated power output at frequencies much higher than 10 kHz resulted in the sine-wave input becoming a triangle waveform at the output. They had produced an integrator for high frequencies! With the limited energy content of HF information on the vinyl recordings of the day, it was not as apparent, but astute listeners soon denounced these products. You may recall the "birth" of transient intermodulation (TIM) or slew-induced distortion measurements, introduced by manufacturers of solid-state amplifiers that relied upon less NFB or nested loop, rather than global NFB, eager to demonstrate their amplifiers did not fall into the above category.

It would appear that some form of solid-state amplifying device that is more linear without the high dependence of NFB would be the answer, but until that time those people to whom the highest quality of reproduction of music is paramount, will gladly accept the extra difficulties associated with valve amplifiers.

The power MOSFET appears to be a likely candidate, but has some problems. There are volumes that could be written not only about power MOSFETS, but about all of the components in a high-quality reproducing amplifier, especially the wrongfully and much maligned audio output transformer. Perhaps discussion on such topics including the use of load-lines and transfer characteristics in the design of more linear amplifiers and these components could form the basis for a future article? Also new improved valves are now available from the Russian valve manufacturers who have been busy working to improving the humble thermionic valve long after the western world had abandoned them in favour of the transistor. Some of the new specifications coming from these companies are astoundingly linear and very worthy of further consideration. I would like to acknowledge the invaluable contributions made by the many very knowledgable people in the various articles and papers I have read, and include a list of authors and journals that I would strongly recommend reading. These may be a little difficult because of their age to locate, and although I have these in my possession, I am not sure about the possibility of copyright violation if I was to distribute copies. I hope in some way this discussion has at least encouraged others to question established practice, as for me I feel a great sense of relief in having at least consolidated many of my thoughts on paper. Most important of all- trust your ears above all else.