|
| Elliott Sound Products | Single Supply Amplifiers |
 Main Index Articles Index |
The era of single supply (typically +60V to +80V) amplifiers has long gone, as everyone expects to see the amplifier's output direct coupled to the speaker. However, in many respects, a single supply conveys some distinct advantages, particularly when it comes to loudspeaker protection.
By definition, a capacitor-coupled output can never apply DC to a speaker unless the output coupling cap fails short-circuit. This is a possible failure mode, but it's so rare that I've never seen it happen. A loss of capacitance, increased ESR (equivalent series resistance) or open circuit are more likely, but if a quality capacitor is used, it should normally last the life of the amplifier. There are quite a few revered 'vintage' amps that use an output capacitor, and they can sound as good today as they did when first manufactured. The direct coupled amp we expect now almost always uses a long-tailed pair as the input stage, and this invariably makes compensation more difficult than it should be.
The Quad 303 was one of the best amplifiers available in 1967 when it was released, and it can still hold its own against almost anything made today. It's not fashionable of course, but 50-odd years ago it (like almost all Quad amplifiers) had excellent reviews, and they remain in use by many audiophiles to this day. I have one, and while it's part of my workshop system, it doesn't get a great deal of use. An original 303 can still be refurbished to be as good as new, and while the power output is rather meagre by today's standards (45W/Channel into 8Ω) it's generally still enough for most 'normal' listening. The amp used a regulated power supply (+67V), and a 'triple' output stage. This has NPN, PNP, NPN for the upper output stage, and PNP, NPN, NPN for the lower half. Decent PNP power transistors didn't exist at the time, making this compromise essential.
Only the output stages are provided with all values (the designators are mine). The negative version has been converted to a 'true' triple, with fully complementary devices. The original used NPN transistors for both output devices, coded as '38494' on the schematic. The remainder of the circuit is for reference, and is not the arrangement used in the Quad 303. AC feedback was not included in the original, so the output cap will have some interaction with the load. As far as I'm aware, no-one ever complained that it wasn't 'quite right' in any way.
Note R115 (2.2Ω). This is used to separate the power output stages from the signal ground to minimise the risk of a ground loop. The ground connection is 'hostile', because it carries a half-wave version of the output signal that can induce distortion into the input and VAS stages. This usually isn't needed with a dual supply amp because all 'nasty' currents are in the supply rails.
The output capacitor was 2,000μF, which is rather small by today's standards. If that's increased to 4,700μF, response is -3dB at 8.5Hz with a 4Ω load. This is not a costly exercise at all, and the opportunity exists to develop a 'new' design that would utilise current feedback instead of the almost universal use of voltage feedback we see today. One of the greatest advantages of the current feedback topology is that stability (against oscillation) is generally a lot easier, and the bandwidth of the amplifier can be controlled by varying the value of the feedback resistor.
They aren't a panacea though, as the DC stability is not very good, and a DC servo is the easiest way to ensure low DC offset for a direct-coupled design. This isn't needed if the output is capacitor-coupled, because there is no DC offset other than a very small amount caused by the output capacitor's leakage current. This is generally very low, and rarely causes any problems.
Of course, there are issues with capacitor coupled amps - in particular a 'thump' when power is applied or removed. This is most commonly solved with a 'slow-start' circuit, but it can also be accomplished with a relay. The 'thump' is caused by the output capacitor charging when power is applied, and discharging when power is removed.
The output (before the output cap) is set for half the supply voltage, but is often tweaked just a little to ensure symmetrical clipping. The upper and lower output stages have slightly different saturation voltages, and that can be compensated quite easily. A regulated supply ensures that there's no drift as the mains voltage changes during the day (it can vary by ±10%, sometimes more).
Many other capacitor coupled amplifiers of the same era (mid 1960s up to about 1980 or so) used a regulated supply, but some did not. Provided the 'half supply' voltage remained stable, an unregulated supply caused no real problems, but eliminating supply hum/ buzz was/ is always more difficult.
Quad wasn't the only 'high-end' manufacturer of cap output amps - they were also found from NAD, Sugden and many others (including Dynaco, Heathkit, etc.), along with many DIY versions. One of the latter is shown as Project 12A ('El-Cheapo'), which was published in 1964. It could easily match many of the valve (vacuum tube) amps of the day for fidelity, but had far more power. Several Sinclair amps also used an output capacitor, as did amps intended for 'more discerning' listeners.
There are countless well-regarded (and many quite the opposite) amps that use or used capacitor-coupling. The article Power Amp Development Over The Years shows several, and the John Linsley-Hood Class-A, the Pass 'Zen' and ESP 'DoZ' are examples of popular designs that still use an output capacitor.
An unexpected effect of the output capacitor was (sometimes) 'better' bass. With a common (at the time) 1,000μF output cap, it could form a series resonant circuit with the inductive (mass controlled) component of many bass loudspeaker drivers. The 'improved' bass was highly unlikely to be particularly accurate, but the cap could shift the resonant frequency down by up to 10Hz, but almost invariably caused a peak (sometimes very audible) at the 'new' resonant frequency. Ideally, the speaker and amplifier will be designed as a pair for this to work well.
This doesn't seem to have been understood by many designers, but there were quite a few who took feedback from after the output cap, preventing the interaction (and reducing capacitor distortion at the same time). Even a 2,000μF (2,200μF now) cap will have some effect if it's not within the feedback loop, and its influence depends on the loudspeaker itself.
Neville Thiele wrote a paper on this very topic, where it was stated that "Adding a series capacitor to a closed-box loudspeaker modestly extends the low-frequency response, enabling the use of a smaller enclosure. In addition, the capacitor offers substantial protection against excessive voice-coil excursions from subsonic input signals. The capacitor converts a second-order system to third order. A design procedure for the composite system in closed form is provided, allowing the resulting transfer function to be optimized for group delay, frequency response, and cone excursion." [ 1 ]
I don't have access to the paper, but it describes the procedure. It's not difficult to experiment with the idea, but without speaker measurement software the results will be unpredictable. You also need to be aware that we (as humans) often perceive 'different' as 'better', even when the opposite is true. More bass output will almost certainly be described as being 'better', even if it results in response that is no longer accurate.
Almost all power amps made today are direct-coupled, with no output capacitor. These amps use a dual supply, with voltages from ±25V to ±100V. With voltages above ±60V many amps use Class-G, with a secondary, lower voltage supply at half the maximum voltage. For example, with ±80V main supplies, the half-voltage rails will be at ±40V.
The real problem with direct coupled amps only shows up if (when?) a transistor fails. The output will swing to one or the other supply voltage (i.e. positive or negative), sending a significant DC voltage directly to the speaker. With passive crossover networks (within the speaker enclosure), the tweeter (and midrange if included) are protected by series capacitors in the crossover, but the bass driver will have to handle the full current provided.
If an amp has ±35V supplies, a (nominal) 4Ω driver may have a DC resistance of perhaps 3Ω. That means a current of 11.6A initially, but it will collapse somewhat due to the high current. If we look at it at the basic level (not accounting for transformer regulation), 11.6A through a 3Ω resistance is a power of over 400W - the speaker's voicecoil will overheat very quickly and serious damage is guaranteed.
Direct-coupled amps also need to have a low DC offset. While 100mV is considered by most people to be too high (and I agree), it won't harm a speaker. That's only 2.5mW into a 4Ω load. In general, you'd aim for less than 20mV. There's no reason to attempt better than 1mV though - 250nW is nothing to worry about - ever.
When amps are direct-coupled, it becomes necessary to add DC detection circuitry (e.g. Project 33) that will remove power to the speaker if there's an amplifier fault. This becomes much harder with voltages greater than ±35V or so, and the only solution may be to use relay contacts in series, or a MOSFET/ hybrid relay. These are covered in several articles and projects, and if done properly the speaker will be protected.
Even within the 'sub-genre' of direct-coupled amplifiers there are significant differences, and for reasons that continue to be deeply mystifying, someone, somewhere, decided that audio amplifiers should be DC coupled from input to output. This has made the problems even worse, and for zero audible benefit ('phase shift' is usually the cited reason, but static phase shift is inaudible). As I've noted elsewhere, DC is also inaudible - we can't hear it and speakers can't reproduce it. This is one of the many areas where audio has descended into bullshit.
There's no doubt that many of the later designs are better than their 1970s counterparts, but that does not mean that you will hear a difference in a blind test. It's common to see distortion figures that are so low that even test instruments have a hard time measuring the performance. However, once you get below 0.1% THD, the distortion is generally considered to be below audibility. However, this depends on the exact nature of the distortion (e.g. crossover distortion vs. 'ordinary' non-linearity) material and the resolution of the speakers or headphones used.
Certainly, below 0.01% THD the differences are such that it's unlikely that anyone will hear a difference, let alone any degradation. At that level, the distortion is at -80dB relative to the signal, or 100μV for a 1V input. Perhaps surprisingly, it's often the case that you can hear lower levels of distortion with a single sinewave tone than with music or speech. I've verified that (under the right conditions) I can hear less than 0.1% distortion of a sinewave, but it's not easy.
Single-supply amplifiers can also be direct-coupled, by using the BTL configuration. Two amps are used, with the speaker connected between the outputs. Each amp is driven from the same signal, but the signal to one amp is inverted. Each amplifier 'sees' half the total impedance, so driving 4Ω loads is difficult because each amp sees 2Ω. This technique is now very common with IC based Class-D PWM (pulse width modulated) amplifiers. The IC shown above is the TDA7297, designed specifically for low-cost applications (e.g. consumer products). The signal inversion is provided by the cross-linked inverting inputs, eliminating the need for a separate polarity inverter stage. The internal values are based on calculations - they are not provided in the datasheet.
The DC offset between the two amps has to be fairly close (ideally within 20mV or so), but this is easy to achieve with an IC. The TDA7297 quotes the 'maximum' DC offset to be 120mV. That's much higher than we're used to, but it's unlikely to cause any problems. A 'typical' figure isn't given. Maintaining low offset can be a little harder with discrete designs, but it's not especially difficult. In some cases, IC Class-D amps (in particular) are rated for around 6Ω minimum impedance, but 4Ω loads are often provided for by using paralleled amps on each side, known as PBTL, so four amplifiers are needed for each channel. This connection requires excellent DC offset control to prevent circulating current between paralleled amps.
Everyone (including me) tells you that the equivalent load impedance seen by each amp is half the actual load, but why? Consider Fig. 2.1, where the speaker load is 8Ω. The voltage at the mid-point of the load is zero, because each end is driven from an equal amplitude but opposite polarity signal. If it helps, substitute the 8Ω load with two 4Ω loads in series. If the voltages are equal but of opposite polarity, the voltage at the mid-point of the two resistors is zero. It's now immediately obvious that each amp does in fact 'see' half the total load, the voltage across the total load is doubled, and power is quadrupled (compared to one amplifier).
This approach was uncommon with 'vintage' hi-fi amps, and few examples exist that I could find. However, car radio manufacturers adopted this scheme fairly early, with many ICs developed for the purpose. A disadvantage was (and still is) that there is a significant DC voltage on each speaker terminal, and an accidental short to ground can be fatal (for the amplifier). If the supply voltage is high enough (e.g. a high-power Class-D amp, the DC voltage can be above the allowable limits for human safety as well. Almost certainly unlikely to cause serious injury, unless contact should cause one to fall from a ladder or scaffolding in a large-scale installation. The use of Speakon connectors is highly recommended.
Distortion is a topic that's been done pretty much to death, by a multitude of different people, and with endless opinions. Some forms of distortion are far more audible than others, with crossover distortion being the most objectionable. As the level is reduced, the distortion increases, and it's generally high-order harmonics that are generated. While there are some who maintain that it's still a problem with modern amplifiers, for the most part this is simply not true.
We see an apparent increase in distortion at low levels with many circuits, opamps in particular. This is the 'N' part of THD+N - noise. No circuit is noiseless, and in many cases the source resistance can be a major contributor to the overall noise figure. There's a comprehensive article on the ESP site - see Noise In Audio Amplifiers. In most cases you can't do much about noise from the source, as once it's mixed into the wanted signal it cannot be separated (at least not in 'real-time').
If you can't hear any amplifier noise from the listening position, then reducing it further makes no difference. Apparent silence (other than normal background noise within the listening space) is silence. Other than active noise cancellation (very difficult in a room) or room treatment, there's not much more you can do. Choosing quieter electronics won't help.
An output capacitor's contribution to measured (or audible) distortion is generally close to non-existent. Even where it's not within the feedback loop, distortion only becomes easily measurable when the reactance of the capacitor approaches the impedance of the load (the -3dB frequency). At higher frequencies, the voltage across the capacitor is negligible, and so too is its distortion contribution.
If there is no voltage across any component, it cannot contribute any distortion. We never get to the 'zero voltage' condition, but the real voltage is small. The reactance of a 1mF (1,000μF) cap is 159mΩ at 1kHz, amounting to 318mV at 2A RMS output current. This doesn't include ESR. A more-or-less typical 1mF 63V capacitor will have an ESR of about 40mΩ.
With the 1mF cap and a 4Ω load, the response is -3dB at 40Hz (close enough). At that frequency, for a 1V input, there will be 707mV across the cap and the same across the load. The capacitor will add some distortion, but it's a great deal less than that from a loudspeaker, and tends to be low-order (no high-frequency harmonics).
By the time we get to 1kHz (the point where distortion can be very audible), the voltage across the cap is 41mV (for the same 1V input), just under 28dB below the input voltage. This is reduced at higher frequencies, falling to -37dB at 10kHz. Despite claims otherwise, a small film cap (e.g. 100nF) in parallel makes exactly zero audible difference (simulated at 0.0015dB at 10kHz) at 10kHz. The simulation used an ideal cap in parallel, with zero inductance (not possible if leads are longer than a few millimetres) and zero ESR.
If the voltage across the capacitor is (say) 20dB below the applied voltage, it follows that any distortion from the capacitor is reduced accordingly. In fact, a low voltage across the cap means that it will contribute far less distortion than you might anticipate. The whole notion of caps being somehow 'bad' is just silly, and it will never stand up to scrutiny in practice. In almost all cases, the THD contribution is negligible. It's there (of course), but at such a low level with a quality part that any distortion is inaudible.
Unfortunately, this is almost impossible to simulate, but based on some tests I performed, it's also extremely difficult to measure. The contributed distortion is so low that it can generally be ignored. If you wish to test this for yourself please do so, but remember that if it's not a blind test the results are meaningless.
In the article Capacitor Coupled Output Stages, the charge and discharge processes are examined. There's no point repeating everything here, but one thing to be aware of is that current is drawn from the power supply only when the output is greater than the quiescent DC output - i.e. when the output signal to the speaker is positive.
For an amp using a +70V supply, the quiescent voltage (before the output cap) is 35V, and when the signal is positive (>35V) current is drawn from the supply. When the voltage is negative (<35V), current is provided by the output capacitor. All capacitor coupled stages perform the same way. While we tend to think of a capacitor as a component that passes AC and blocks DC, there's more to it than that.
In a dual supply amp, negative current is provided from the negative supply (no surprise there), and both supply rails carry a half-wave rectified current. In a capacitor coupled amp, the negative half-wave current is in the ground lead, so care is needed to prevent this from contributing high levels of distortion to the output.
It's quite easy to make a mistake with grounding that causes problems. The speaker should (in most cases for cap-coupled output stages) return to the negative terminal of the amplifier, usually not directly to the filter capacitor (or regulator negative terminal) as we would do with a dual supply amp. It's not uncommon for the low-level amp circuitry (input stage and VAS) to have a low-value resistor between the main (DC power) ground and the 'reference' ground used by the input stage.
This is seen in Fig 1.1 (R115). The resistor is usually around 1-3Ω - just enough to prevent interactions and/ or break a ground loop. It's generally not needed with dual-supply amps, because there is very little ground current except for the speaker return wiring. In most cases, this should go directly to centre-tap of the power supply filter capacitors, although some amplifiers are designed specifically to have the speaker return connected to the amp PCB.
As noted above, many early single-supply amps used a regulated voltage. This made operating conditions more stable, eliminating the possibility of infrasonic 'disturbances' created by the quiescent voltage point changing with signal level or mains variations. Most regulators were fairly simple, offering perhaps 40dB or so of 100/120Hz ripple rejection.
In some cases, the regulator was designed so it could be shut down with a fault condition, such as a shorted speaker lead. The methods varied widely, and with modern parts we have even more scope to provide very good protection against 'mishaps'. The Quad design is 'upside down', for reasons that remain obscure. My best guess is that it allowed TR202 to be mounted directly to the heatsink, as the collector is at the zero volt ('-Ve') point, so no insulation is needed. This makes it a bit harder to follow, but it simulates properly and obviously worked in the amplifiers where it was used. A single regulator was used for both power amplifiers (L & R). R200, R201 and MR200 are used only during the start-up phase, and without them the supply will not 'self-start'.
The above is the circuit from the Quad 303, and while the connections look very strange, it does what's expected well enough to prevent problems with hum/ buzz. The designations are as shown in the original service manual, but the transistor types are those I used for a simulation. The original devices are only provided with a number that is (presumably) an in-house part number. Unfortunately, intermediate voltages aren't shown on the schematic, so are those I used for simulation (-10V in particular). The transformer primary has multiple selections to suit the different mains voltage that were common in the 1960s and 70s (i.e. 220, 230, 240, 115 and 120).
Simple regulators generally have a poor reputation for stability and ripple rejection, but they can be surprisingly effective, as demonstrated by the next circuit. All amps of the era that used regulated supplies used circuits that were fairly basic, because aiming for excellent stability and low noise were never priorities. The idea was only to provide stability that was sufficient to prevent audible problems.
The regulator shown above is a modified version of the design used for 'El-Cheapo' (Project 12a), and the transistor (Q3) converts the zener diode into an 'active' zener. This has much better regulation than a zener diode by itself. A 68V zener has a dynamic resistance of about 86Ω as simulated - a BZX55C68 has a quoted dynamic resistance of <200Ω, vs. only 20Ω for the 'assisted' version. The transistor also takes most of the power, so the zener can operate at its optimum current. It's possible to get even lower dynamic resistance (<3Ω) with another transistor, but there's no real reason to go to that extent. Perhaps surprisingly, the Fig 5.2 regulator outperforms the Quad version quite comfortably. The output voltage is the same, but the Fig. 5.2 circuit has less than 1.3mV ripple, vs. 2.9mV for the Fig. 5.1 Quad design (both at a current of 2A and with the same filter capacitance). Ripple can be reduced a little further by using a 100μF cap for C3.
The boosted zener works best with high-voltage zener diodes, as they have a much greater dynamic resistance than low-voltage types. However, even at lower voltages this scheme still provides a benefit, reducing the ripple by around 3dB for a 24V zener diode. Whether you consider this to be enough to warrant the extra transistor and resistor is up to you. If it helps, it lets you use a lower powered zener diode, so where you may need 1W or more in a design, a transistor booster will let you use a 400mW zener instead. To get better performance than a simple zener regulator (boosted or otherwise) requires a feedback circuit, since much of the remaining output ripple is due to the series-pass transistors that allow some 100/120Hz to get through.
With a dual supply, you'd need two regulators, which increases the cost and demands more heatsinking. This is why almost no-one uses regulated supplies for direct-coupled, dual-supply amplifiers using the 'standard' long-tailed pair input stage. Their power supply rejection is usually quite good, so supply rail hum isn't a major issue. It can still cause problems, but that's rare and requires an over-simplified design before it becomes an issue.
Adding a regulator demands that the rectified transformer voltage be greater than the regulated voltage by at least a few volts, regardless of load current and mains voltage variations (at least ±10%). For the 67V supply used in the 303, that means that at maximum load and minimum mains voltage, the DC (including the ripple component) must be no less than around 75V, preferably a bit more. The series pass transistor (Q3 in the circuit shown) will also have to cope with maximum mains voltage (253V RMS for 230V mains). The design process is actually more difficult than it may seem at first.
For a regulated preamp supply we don't need to be too concerned, because the maximum load may only be around 100mA. When the peak current required is over 10A (a very conservative estimate!), this is a great deal harder. The biggest problem is power dissipation, which can be very high. For the Quad amp, the worst case current is over 16A. At minimum mains (210V AC) the average DC must be at least 75V, allowing for ±5V ripple (and this is only just enough). At maximum mains voltage (253V), the average unregulated voltage will be 90V. The peak dissipation in Q1 could be as high as 230W, allowing for a peak current of 10A. The only thing that keeps the series-pass transistor 'safe' is the low voltage differential at maximum current, because the rectified voltage collapses under load.
Determining the final requirements for the regulator will generally be done under more-or-less 'typical' operation, using an empirical approach. Trying to calculate everything is possible, but very time-consuming. Even setting up a realistic simulation is difficult, but when the amp was designed, simulation was in its infancy and wasn't terribly useful. The vast majority of regulators used at the time were 'adequate' - not wonderful, but good enough. The simple circuitry was quite deliberate, because perfect regulation was never necessary.
Another reason that no-one uses regulated supplies any more (particularly for DIY) is simply because getting the required transformer is either not possible or at least difficult. If you wanted a regulated 60V supply, you'd need to use a transformer with a secondary voltage of around 56V RMS. That's easy, because dual-secondary trannies are readily available with 30V windings. The regulator dissipation will be a bit higher than you might prefer, but it's not hard. If you wanted 80V DC, this will be a special order, as you'll need a 75V RMS secondary and that's not one of the commonly available options. You could use an 80V secondary, but with the extra voltage and current the regulator's dissipation becomes more difficult to accommodate.
No-one wants to have to increase the size of the heatsink by 50% or more just to cope with a regulator, so it's not hard to see why regulated supplies have fallen from favour. If the idea is a 'cost-no-object' project amp is appealing that's another matter of course. If an amplifier is DIY, you only have the cost of the materials and your labour, without wholesale and retail profit margins (as well as shipping costs) getting in the way. Of course you pay more for parts due to low-volume purchasing, and you still have to design and build a chassis, PCBs and a front panel (a hallmark feature of many commercial amplifiers).
However, using a regulated supply can give superb results, and an example is shown in Project 221 - a regulator specifically for tweeter amplifiers. The primary goal was to reduce the available power, but it also opens the way to using a single-supply amplifier, with an output capacitor that will provide tweeter protection. A low-noise supply can't be a bad idea for a tweeter amp.
In some cases, you may find forum posts and test 'results' (from sighted [non-blind] testing of course) claiming that a regulated supply causes the sound to be affected in some mysterious way. It may be described as 'closed in', which (apart from being meaningless) is nonsense. In a well designed amplifier, the speaker is isolated from the power supply by the power amplifier, and anything the supply may do is irrelevant so long as the output signal is within the bounds of the supply voltage(s). A regulator may reduce the available power compared to what you'd get with a 'raw' (unregulated) supply, but if the amp and regulator are designed as one, the output should be cleaner up to the point of clipping.
If the amp is driven hard, a regulator prevents 100/ 120Hz amplitude modulation of the clipped waveform, and it's hard to argue that this is somehow 'bad'.
For the most part (at least for moderate power Class-AB amplifiers), the single supply connection is dead and buried. There are a few low-power (especially Class-A) amps that use a single supply and an output cap, but they are hardly mainstream designs. There is an up-and-coming project for a 'retro' hi-fi amp (based on the Sansui AU-555A) which uses a single supply, but the vast majority of modern designs use a dual supply, and eliminate the output capacitor.
It can be argued that this is a retrograde step, since speaker protection circuitry (e.g. Project 33) becomes a very desirable addition. The output cap prevents DC from getting to the speakers by default, but at some additional cost (a high quality cap will cost more than a P33 circuit). This is offset by the single supply, so only filter caps for one polarity are needed. This can be a significant saving if you insist on 100mF (100,000μF) filter caps. If you were to follow the lead of Quad, you'd simply use 4 x 2,200μF caps - two for filtering the single rail, and two for speaker coupling. By adding a regulator, the massive filtering commonly used isn't needed, because the regulator ensures clean DC with little ripple.
If you compare the cost using that arrangement, you'll probably come out ahead. Quad always had to satisfy the requirements of users, and do so at a price that people could afford. The same applies to most other manufacturers, but there are some that aim their products at the 'high end', making very expensive products aimed at a select few. This doesn't always mean that these ultra-expensive amps are necessarily 'better' than more mainstream products - they just cost more. Of course, some are about as good as it's possible to make an amplifier, others not so much.
| Main Index Articles Index |