|Elliott Sound Products||Class A Amplifiers - A Brief Explanation|
Recently there has been a resurgence of two 'ancient' technologies - vacuum tube (valve) amplifiers and Class-A systems. The big question is ... is there a difference? This discussion centres on the Class-A amplifier, and explains (or attempts to) how it is different from a conventional power amplifier.
Why would someone want to build or buy an amplifier which is sooo inefficient? A Class-A power amp will typically draw anything from 1/2 to about 1½ times the peak speaker current in its quiescent state (i.e. while it is just sitting there doing nothing).
To put this into perspective, for a measly 8 Watts into 8 Ohms, the RMS current is 1 Amp. The peak current is just over 1.4 Amps, so a typical 8 Watt Class-A amp will draw anything from 700mA to 2 Amps continuous. This equates to a quiescent (no signal) power dissipation of between 17 Watts and 48 Watts, based on a 24 Volt supply (+/- 12 Volts ). At very best, such an amplifier will have an efficiency of less than 35% at full power - at worst, this will be perhaps 15% or less.
The basic premise of a Class-A amp is that the output device(s) shall conduct all the time (through 360 degrees of the signal waveform). This means that in the simplest form, the power devices must conduct a continuous current which exceeds the maximum peak load (loudspeaker) current. If we use a power level of 20 Watts (hardly a powerhouse) for all further calculations, we can see the whole picture.
In contrast, a typical Class-AB power amplifier's output devices only conduct for about 182 degrees (at full power), which means that for much of the signal's duration, only one or the other device is conducting. The other is turned off. The 'crossover distortion' so often referred to is nothing to do with the frequency divider in the speaker system, but is created as the signal 'crosses over' the 0 Volt point (see Figure 3).
Figure 1 - The Sinewave Cycle
Let's have a quick look at some of the power amp 'classes', so we have all the info:
There are many amplifier topologies which I have not mentioned above, mainly because most of them are either too bizarre, not worth commenting on, or are too complex to explain simply. Of these, Class-G and Class-H use power supply switching and modulation (respectively). This provides greater than normal efficiency and lower dissipation, but both are essentially Class-AB designs.
Although many audio amps may be called Class-B, generally they are not. Virtually without exception they are Class-AB, although most will be at the bottom end (conduction for perhaps 181° for each device). Most power amps operate in Class-A up to about 5 to 10mW, after which they become Class-B. Many run Class-A up to higher power, 500mW of more.
In the device department - For the remainder of this paper, I shall use bipolar transistors for the power devices, since they exhibit highly desirable characteristics for this application. They are also far more linear than MOSFETs, and some of the newer bipolar devices are outstanding in this regard. Note that there are two types of MOSFET in common use - Lateral devices are designed for audio, and although less linear than bipolar transistors can make a very good amp indeed (see Project 101). Power switching MOSFETs are (IMHO) not suitable for use in audio except where very high power is needed and extreme linearity is not required.
|Load Voltage (at Speaker)||12.65 Volts RMS (17.9 Volts Peak)|
|Load Current (through Speaker)||1.58 Amps RMS (2.23 Amps Peak)|
|Supply Voltage||+/- 20 Volts (constant)|
|Supply Current||+/- 2.25 Amps (peak)|
In amplifier design, we are interested in the peak voltage and current, since if these are not met, then the required RMS values cannot be achieved. The ratio of RMS to peak (for a sine wave) is the square root of 2 (1.414), so RMS values must be multiplied by this constant to derive the peak values of voltage and current. (Refer to Figure 1 to see the relationship between peak and RMS voltages.)
This is how the values in the table were determined. The supply voltage needs to be slightly higher than the actual speaker peak voltage because the output devices (transistors) are not perfect, and some voltage will be lost even when they are turned on fully. (If MOSFETs were to be used, the losses may be much greater, unless an additional power supply is employed.)
Ok. We have determined that the peak speaker current is 2.25 Amps, so in the simplest of Class-A designs this will require a quiescent current of 2.25 Amps. Given that the voltage is ±20 Volts, this means that the power output stage will have to dissipate 40 × 2.25 = 90 Watts (45 Watts per output device).
Figure 2 - Basic Class-A Amplifiers
Figure 2 shows what a simple Class-A amp may look like. The current source (left circuit) is a simple circuit, which provides a current which remains constant regardless of the load placed at its output. The output transistor 'dumps' any current which is not needed by the load (speaker), so when it is completely turned off, all the current source output flows through the speaker. Conversely, when the transistor is turned on, the speaker current flows through the output transistor (as well as the current from the current source!), so its current will vary from almost 0 Amps, to a maximum of 4.5 Amps for our example. When there is no input signal, the output transistor's current must exactly equal the output of the current source. If it does not, then the difference will cause a DC offset that causes asymmetrical clipping. It is allowable (generally speaking) for an absolute maximum of 100 mV DC to be present across the speaker terminals - this equates to 1.67 mW of DC for an 8 Ohm system, assuming a 6 Ohm DC resistance for the voice coil. (Power = V² / Impedance). The capacitor in the output circuit reduces this to near zero.
When an inductor is used, the overall efficiency of the circuit is improved. An inductor is a reactive component, so it will 'release' stored energy when the transistor turns (partially) off. The other advantage of an inductor is that peak-to-peak output voltage swing is doubled for the same supply voltage. The disadvantage (of course) is that the inductor is large, heavy, and must have an air-gap to ensure the core doesn't saturate because of the DC component. The quiescent current through the transistor and inductor needs to be the same as the peak load current. For example, with a 20V supply and an 8 ohm load, the quiescent current needs to be 2.5A to ensure linear operation. This circuit was common in early hybrid (valve and transistor) car radios.
These simple models are not really appropriate for general use, since they waste far too much power, although many Class-A amps still use the inductor principle (SET (single ended triode) valve amps use a transformer instead of an inductor, but the principle is unchanged). Many other Class-A amps use the current source version, but efficiency can only reach a maximum of 25%.
The next step is to operate the current source at about 1/2 the speaker's peak current, and modulate its current output to ensure that both current source and power amplifier output device conduct during the entire signal cycle, but are able to vary their current in an appropriate manner. This improves efficiency (which remains dreadful, but slightly less so), and lowers the quiescent dissipation to more manageable levels.
The simple Class-A amplifier described by John L Linsley-Hood and the very similar looking Death of Zen (DoZ) amp on these pages use this latter approach, and it is a sensible variant of the various Class-A designs. As an example, the amplifier will only (?) need to dissipate about 50 Watts when idle, since the quiescent current is reduced to around 1.2 Amps.
Another version of the Class-A amp looks exactly the same as a standard Class-AB (Class-B) power amp, except the quiescent current is increased to just over 1/2 of the peak speaker current. This is thought by some that this is not a 'real' Class-A amplifier. It is real Class-A, and is best described as push-pull (as opposed to single ended) operation. If the bias current is not high enough for the actual reactive speaker load (not some quoted nominal resistive load), it is still possible that one transistor or the other will switch off at some part of the signal cycle. This will happen at a much higher power level than is normally the case, but if this happens, then the amplifier ceases to be true Class-A.
As an extension of the above, it is possible to design an amp that looks remarkably like a conventional Class-AB amp, but with additional circuitry is biased in such a way that the output transistors do not turn off - ever. This technique can also be used with Class-AB, and supposedly reduces crossover distortion. I have not used this method, since in my experience the crossover distortion in a well designed output stage should be sufficiently low that the additional complexity is not warranted. Project 3B is almost identical to Project 3A, except the quiescent current is increased so the amp runs in Class-A.
The last three 'variants' cause the current to be modulated in each supply rail, so there is not the steady state current one expects from a Class-A amp, but a waveform that varies with the signal. When properly designed and biased, the output devices conduct at all times, but the power supply has to contend with a varying load. I have not investigated this fully, but it can make the design of the supply a little more difficult because of the varying load current. Tests I have performed with the DoZ amp do not show any audible effect on the sound quality - provided the supply is designed to handle the variations without any problems.
Actually, the idea that a Class-A amp draws a continuous steady current from the supply is true in one case only. A single ended amp using a current source as the collector load will draw a continuous steady current - but only if it uses a single supply. In the case of a dual supply, the same amp will draw a continuous current from one supply, and a varying current from the other. (My thanks to Geoff Moss for pointing this out - a detail that few published designs have ever mentioned !)
An amp that uses a fixed current source of (say) 2.5A from the positive supply will draw 2.5A regardless of load or signal level, but only from the positive supply. The negative supply current will vary from 2.5A at no signal, but will be almost zero at maximum positive swing, when the lower transistor is turned off, and the current flows from the current source to the load. At maximum negative signal swing, the negative supply current will be close to double the quiescent current, since the lower transistor now carries the current from both the load and the current source.
This 'small' detail seems to have received scant reference in any of the articles I have read, but it will make a very big difference to the power supply. In this respect, I do not feel that the single ended version should be operated from a dual supply. If it is so important to you to eliminate the coupling capacitor, then I suggest that either a push-pull Class-A design be used, or build separate power supplies for each polarity.
There is some evidence (I refer again to Doug Self) to indicate that the distortion of a 'true' Class-AB amp will often be worse than that of a Class-B design, since the switching transients are larger due to the output devices' higher gain at moderate (0.5A to 1.5A) currents. I have not been able to verify this, and the tests I have done indicate that there are definite benefits in the higher quiescent currents, provided the current is chosen reasonably carefully.
One of the biggest problems with Class-A amps is that the simple power supply used with conventional Class-AB amps is usually no good to us. The reason is that the AC ripple on the DC power rails is injected into the amp, and emerges as hum (at 120 or 100Hz, depending on location - US or elsewhere, respectively). The magnitude of this ripple is far greater than with a Class-AB amp, because a considerable amount of current is being drawn at all times, rather than during signal peaks (etc). A power supply which provides a no-load ripple of perhaps 50 mV for a Class-AB amp may have 1 Volt (or more) of ripple at a current of 1.2 Amps. This will be audible at low signal levels.
Adding capacitance helps, but by the time the ripple is reduced to a reasonable level, you have sold the car to pay for the capacitors, and no longer have a vehicle to carry them home in. You will need a ridiculous amount of capacitance to obtain reasonable hum levels (> 70dB signal to noise ratio) unless a regulated supply is used. The fact is that many Class-A power amps do not have particularly good power supply rejection (Ok, it is not generally too bad, but cannot compete with the likes of an operational amplifier), and a regulated power supply is recommended for all such amps. In case you were wondering, that does indeed mean that you need more transistors, more heatsinks, and it will cost more money. Such is the price we pay for perfection.
There is an alternative (which I have not tried for this application, but have carried out numerous spice simulations) called a capacitance multiplier, which is simpler and cheaper than a regulated supply, but should be capable of reducing the ripple to very low levels. I have had a few e-mails from readers who have built the capacitance multiplier project (see the Projects page), and the results have been very positive, so this makes the Class-A idea far more attractive from a cost and heat perspective. (Capacitance multipliers are not required to regulate, so operate with a much lower input to output differential voltage - therefore, less heat!) Indeed, the design by John Linsley-Hood referenced on these pages uses a capacitance multiplier, although its performance can be improved dramatically.
The question now is - is this really what I want to do? The answer might be a resounding yes (after all, there is no good reason that a Class-AB amp cannot be just as good) - but to be sensible, we should apply the Class-A amp for the tweeters in the system, and use conventional Class-AB amps for the low and mid frequencies. To obtain adequate sound pressure levels, most modern speakers need lots of power, since they are not very efficient (i.e. electrical power in versus acoustical power out).
Rather than extend this page to a short text book on the subject, I shall leave you with a simplified model, which I produced for a reader who had a speaker system which was even less efficient than is common. The table shows the power needed to achieve various peak SPLs (at one metre) for a speaker with an efficiency of 85dB/m/W. Based on the sensitivity of these speakers, the following shows roughly what you can expect, based on a single amplifier for clarity (i.e. not bi-amped or tri-amped):
|dB SPL at 1 metre||Amp Power, Watts, one channel|
This is not good news for the most part, as it clearly shows that vast amounts of power are needed to achieve a realistic SPL in a typical listening environment. Remember that the figures shown are at a distance of only one metre - the SPL will fall by a further 6dB each time the distance is doubled. (Mind you this is a theoretical figure, which is generally not met in practice - perhaps 5dB would be closer to the truth?)
Realistic SPL in this context is worthy of a page (book?) in itself, but remember that for an average SPL of (say) 85dB, transients will require between 10 and 20dB of headroom. This means that the peak power needed will be between 10 and 100 times the power needed to reproduce the average of 85 dB. At a distance of 2 metres, something around 3 Watts will be needed for this example. To reproduce the transients, the actual power needed must be between 30 and 300 Watts!
In case you were wondering, 85dB SPL is not loud (although 's/he who must be obeyed' will almost certainly disagree). In fact, it is only marginally louder (by about 5dB) than the recognised optimum level for normal speech.
Since Class-A amps are inefficient, generate lots of heat, and require a far more complex power supply than conventional Class-AB amplifiers, there have to be some compelling reasons to use this arrangement. The first is circuit simplicity. In the light of the above discussion, the circuit is not simple, but for the audio signal it can be far less complex than for a conventional power amp.
The benefit of this is that the signal is subjected to comparatively little amplification, resulting in an open loop (i.e. without feedback) gain which is generally fairly low - probably less than 250 (48dB), and possibly as low as 50 or so (34dB). This means that very little overall feedback is used, so stability and phase should be excellent over the audio frequencies. A well designed Class-A amplifier should not require any frequency compensation (or very little), so the open loop gain will remain reasonably constant over the audio range. This may result in superior transient response, and dramatically reduced 'Transient Intermodulation Distortion' (or TID, aka Dynamic Intermodulation Distortion), which is thought by many designers to be caused by phase and time delays between the input and feedback signals. It may be possible that this is the cause, although the existence of TID is virtually zero in any competently designed amp.
The simple fact is that the more amplifying devices that are introduced into the chain, the more phase shift must be introduced. No amplifying device is capable of responding instantaneously to a change of input - all have some inherent delay (which usually includes different turn-on and turn-off times). With fewer devices in the audio circuit, there must be less delay between a change in the input causing a change in the output. The simplified topology used for most Class-A amps can also be used with Class-AB - often with very good results indeed.
Figure 3 - Crossover Distortion
Figure 3 shows the crossover distortion of a Class-B type amplifier. This is exaggerated for clarity, and the 'clean' signal is included for comparison. As can be seen, when the signal is reduced, the ratio of distortion to signal will become much worse, resulting in an increase in distortion as power is reduced. Indeed, this is exactly what happens in many amplifiers, but it generally is 'swamped' by so much feedback that it seems to disappear. It can be seen from the diagram that for this crossover distortion to appear, the amplifier's gain must fall as the signal level approaches 0 Volts. Indeed, the amplifier's loop gain really is reduced to zero when the transistors are turned off!
The point that distortion 'seems to disappear' is the operative term here - it does not go away at all, and worse, as the crossover point is reached, the open loop gain of the amplifier is reduced, meaning that there is not as much feedback as at higher signal levels. This will be apparent to readers with an electronics background - note that near the crossover point, the amplitude of the signal is much lower than it should be (this is what causes the problem in the first place!). Since the amplitude is reduced, it is obvious that the amplifier's gain must be lower at this level than at higher levels.
Therefore, if the open loop gain is lower, then the available feedback must also be lower. This is an area that has received some study, and this is illustrated by the very 'flat' gain vs. collector current curves of many of the more desirable audio output transistors. It is certainly a cause for some concern, and indicates that the open loop behaviour of a power amp should minimise crossover distortion before any feedback is added. Simply increasing the quiescent current is not always a complete answer, because this problem is created by the inherent non-linearity of the output devices as they commence (or cease) conduction. Increasing quiescent current will move the 'kink' further away from the 0 Volt point, but it will still be there - and may actually be worse than at lower quiescent currents. A major advantage is that the distortion components will be (potentially) somewhat less audible, and will affect the signal while it is comparatively loud - this will reduce its audibility further.
|I bagged MOSFETs earlier in this article because they are actually more non-linear in this region than transistors. Since this is the most critical part of the signal, it is important that
it is treated with the utmost respect. However ...
This does not mean that MOSFETs are not capable of exemplary performance. A carefully designed lateral (not switching!) MOSFET amp will sound every bit as good as (or perhaps 'better' than) a bipolar amp, whether it is operated in Class-A or Class-AB.
In the light of this, it is a wonder that any Class-AB (conventional) power amplifiers sound any good at all. Historically, it is exactly the problems I have highlighted here which created the term 'transistor sound' (used in a derogatory sense of course) when transistor or 'solid state' amplifiers first appeared. Despite anything you may read, these problems are caused by the physical and electrical characteristics of transistors, and have never gone away. New devices are far more linear than those of the 60s and 70s, but they are not perfect. Operation at higher quiescent currents (i.e. more into the Class-AB region) will reduce the non-linearity at crossover, but it can never be eliminated altogether - at least not with any devices currently available.
It is fair to say that although the problem cannot be eliminated, the effects can be reduced to such an extent that many amplifiers have almost unmeasurably small levels of crossover distortion. It is not at all uncommon that to be able to see the distortion residual (after the fundamental has been removed with a distortion analyser), it is necessary to use a digital oscilloscope that can apply averaging. The distortion is buried below the amplifier noise floor, and is not visible without the averaging feature. In tests I have performed, listening to the residual noise + distortion reveals that the distortion component (in isolation) is barely audible over the system noise - itself normally below audibility with typical loudspeakers.
So, it is entirely possible to design an amplifier whose distortion at any level below clipping is virtually unmeasurable. Marginally higher levels are commonplace, and it is thought by many that the typical distortion level in most well designed power amplifiers is inaudible under most listening conditions. There are (of course) others who deny this - either because they have done proper comparisons under controlled conditions, because they have hearing that is far more acute than most of us, or because they have been told that they must be able to hear the difference - if they can't, they must have 'tin ears'. Nothing like a bit of peer group pressure to influence one's perceptions.
Where does this leave Class-A? There is an emotional connection with the idea of a Class-A amp, and it has to be considered that sometimes there is simply a 'feel good' aspect to this - technicalities don't even enter into it. Despite my own ambivalence, I was still a bit disappointed in my decision not to use P36 for my own tweeters - and this in spite of the fact that I could hear no difference between the P36 and the high quality power opamp which I am using for my tweeters.
Because the transistors in a Class-A amplifier are never switched off, there is obviously no crossover distortion (after all, there is no crossover - where one transistor turns off, and the other supplies the load current). There is distortion though - it is caused by all the normal non-linearities in any active device, and in particular the wide current variation in the output device (in combination with elevated temperature). It is worth noting that crossover distortion is exactly the same as clipping distortion, but with a different phase with respect to the signal. Consequently, it contributes odd harmonics (as does clipping) - 3rd, 5th, 7th, etc.
If properly designed, a Class-A amplifier should be capable of a maximum open-loop distortion of perhaps 5% at full power, reducing as the input signal (and hence output power) is lowered. This distortion is believed to be predominantly 2nd harmonic, which (in moderation) is far less intrusive than the odd-order distortion created by conventional push-pull Class-AB amplifiers, however this may not be the case. In contrast, most common Class-AB amps will have an open loop distortion of perhaps 10% to 15% at full power, although some will be much lower.
Such amps typically rely on global feedback to reduce this distortion, and usually have very high open loop gains. Another problem is that the open-loop gain is not constant with frequency, so the amount of feedback applied is reduced at the higher frequencies - not at all what is really needed. However, it does not mean that all such amplifiers are unlistenable - despite claims to the contrary.
For additional comment on Class-A, the 'Death of Zen' (DoZ) article may be an interesting read.
|Class-A Myth #1|
A Class-A amp maintains the same current through the transistors, therefore ensuring that they remain in their most linear region at all times.
This is not the case at all - the current varies widely in the output device in the case of a current source amplifier, and it varies widely in both output transistors for other types of Class-A amp. While it is possible to make the current reasonably constant, it is neither practical nor sensible to do so.
As often happens when writing, I suddenly decided that I just had to run a simulation on a pair of output stages. One is Class-AB (essentially the same as that used in Project 3A) and a Class-A emitter follower circuit. Both were operated with zero feedback, and the Class-AB stage was run at a quiescent current of 14mA vs. 2A for the Class-A circuit.
Rather than make this article longer than necessary, if you want to see the details see Class-A Part 2
Class-A is the most desirable of the amplifier configurations from a purist point of view, but is not suited to high power systems unless outrageous power dissipation is acceptable (like between 825 to 1500 Watts of pure heat, to get 300 Watts of audio). However, if used for the high frequency amplifier in a tri-amplified system, it is possible to obtain the SPL you desire in your listening room, but without having to install a dedicated air-conditioning system to remove the heat generated.
When used for the frequency range of 3000Hz and above, comparatively little power will be needed, and the sonic benefits should be readily apparent - crystal clean highs, without any harsh distortion components. The distortion generated may be (but is not necessarily) predominantly 2nd harmonic, and will be greatest at high power levels where it is least likely to be audible. Bear in mind though, that a great many Class-AB amplifiers will be capable of performance that is just as good, and in a lot of cases, far better.
|Class-A Myth #2|
Class-A amps give predominantly 2nd order distortion.
They might, or they might not, depending entirely on the topology. A great many Class-A amps will produce distortion components that are almost identical to those produced by a Class-AB amp. This excludes clipping distortion, which should be avoided in any class of amplifier used for high quality audio. Almost all amplifier topologies produce some third harmonic distortion along with second - it's almost impossible for it to be otherwise.
Where it is not feasible (economically or otherwise) to use a Class-A amp in the tweeter frequency range, a modified Class-AB amp could be used. The modification needed is to increase the quiescent current (to perhaps 1 Amp or so) so that the amplifier operates as Class-A for any signal below about 8 Watts - assuming a well behaved 8 ohm load such as a tweeter. Such a modification to an existing amp is quite simple for an experienced electronics engineer or service person, but will almost certainly require that the heatsinks be upgraded to prevent the destruction of the output devices. It is also probable that additional capacitors will be needed for the power supply - and possibly a regulator or capacitance multiplier circuit, too. Without these, the hum level may become intrusive, which rather negates the whole purpose of the exercise. Some basic experimentation is required for anyone thinking along these lines.
Bear in mind that you can say a fond farewell to any warranty which may exist on your amp - few manufacturers will accept that ripping their product to pieces and rebuilding it as something 'new' is a perfectly reasonable thing to do.
Despite the cost of modifying an amp in this way, it is bound to be cheaper than buying or building a Class-A amp from scratch - even more so if you have a perfectly good (but underpowered) amp just lying about waiting to be put to use. For not a lot of work and relatively few dollars, a potentially fine amplifier can be yours.
Please be aware that the above section is more in the line of 'musings' than established fact with full testing. The theory is (more or less) sound, but one cannot predict the exact behaviour of any amp once modified, and I suggest that if any such mods are to be attempted, they should be done with 'before and after' measurements to allow proper comparison. Operation at a higher than normal quiescent current may actually degrade performance with some amplifiers.
|Copyright Notice. This article, including but not limited to all text and diagrams, is the intellectual property of Rod Elliott, and is Copyright © 1999-2005. Reproduction or re-publication by any means whatsoever, whether electronic, mechanical or electro-mechanical, is strictly prohibited under International Copyright laws. The author (Rod Elliott) grants the reader the right to use this information for personal use only, and further allows that one (1) copy may be made for reference. Commercial use is prohibited without express written authorisation from Rod Elliott.|