Analogue Equipment Design (Part 2)
A Driver-Stage Dissection
More from that maestro of the multimeter, Ben Duncan.
Too many professional power amplifiers omit gain controls outright and of those with gain pots, the range is usually unnecessarily wide. For real world Rock n' Roll work, there are just two essentials: a fast and quiet mute, and enough gain variation to cope with differing drive levels - for example when associated amps in a rig want to see +10dBu; or for a measured attenuation... or interchannel balancing. None of these, either singly or combined, demands more than -20dB. We'd liked to have used an Alps stepped attenuator, but such finesse is pointless if the added cost puts the design out of reach of the people who need it most, or distract funds from the real issues. So we settled for a clever hybrid arrangement. Firstly, the DVT series are muted at switch-on and switch-off. Secondly, power switching is via a chunky, 30 Amp contactor block. This means the panel power switch isn't in imminent danger of burning out. Therefore, it's perfectly satisfactory to use the power switch as a mute; shut-down is instantaneous, and involves no bangs or bursts. As for the gain control itself, it still shuts down the signal to nil, but the main range covers the first —20dB of attenuation in an expanded format. This means that a slight brushing against the gain controls won't swing the gain out by 5 or 10dB...
Now it's time to delve into the driver stage.
Though principally a high voltage amplifier, with +36dBu output (=47V RMS), it has to feed the MOSFET devices with an appreciable current at high frequencies, where they get quite hungry! The driver's topology, outlined in Figure 1, is based on the transistor design techniques developed for linear ICs, but realised here with discrete components. Elements of linear IC circuitry have appeared from time to time in bi-polar designs over the past 15 years, but it took the advent of POWERFETs before the full and beautiful symmetry of the Widlar architecture was harnessed. That happened in 1980, thanks to some US designers working with Hitachi, who hooked up the National Semi-conductor LH0003 hybrid op-amp to POWERFETs. Then in 1981, Dr Anton Rauch analysed the LH0003 topology in discrete format. Surprisingly, it was found to outperform all the more complex, competing circuits. But more work was needed for inline with the theme of Rauch Precision. It wasn't blandly assumed that the subtleties of a monolithic IC can be successfully cribbed from the schematic alone.
Let's deviate tangentially for a moment, whilst I tell you about linear design. In the 12 months before a new IC comes on stream, the manufacturer pays out millions in producing designs, artwork, masks and tooling. Even when a chip costs as much as £5, a million+ will need to be sold, to amortise costs alone. So the design had better be a good one, and one many equipment designers will want to use. For this reason, IC innards are apt to incorporate the most elegant examples of configurating transistors around.
As an example of evaluating the parts of the IC innards which aren't explicit, Figure 2 homes in on COB. Also known as Miller capacitance. This is a parasitic capacitance between the transistor's base and collector terminals. Because the value of COB varies erratically with the drive voltage, it's a source of distortion: the so called Early Effect, in fact. Taking a typical discrete transistor, found in PA driver stages, the MPSA-93 has a COB of 8pF, whereas a typical high voltage monolithic device's COB is typically 0.3pF. Thus in the DVT series, driver transistors have been selected for their COB parameters, so the parasitic capacitance is both small, and matched across the drivers and pre-drivers.
Behind this detail lies serendipity: keeping COB low means wide-banding the driver stage. This has the advantage that the amplifier is unusually stable on reactive loads, when at home, and yet, at the same time, rudely intolerant of any imperfections in the immediate environment. This is to say that unlike the majority of bi-polar amp designs, there's no chance of hiding short-cuts in the power supply or wiring looms. This means following high-frequency wiring routines by the book, from the very start. The difference between amplifiers with and without attention to wiring detail is startlingly audible, but only when the basic amplifier guts attain a certain, minimum standard. Below this, there are two many masking effects at work — rather like struggling to pick out a turning, whilst driving in the rain, with steamed up glasses, which are also covered in raindrops, a dead de-mister, and a windscreen that's smeared with dead flies and other unsavoury substances, all dissolved in diesel and spread across the glass... (We get the point Ben —Ed)
In resistors, we found a component which could be greatly uprated at next to no cost. With voltage swings up to ±80 volts across some parts, high level linearity came in for scrutiny. There are 100's of brands of resistors on the market, all costing about 0.6p in manufacturing quantities. But their actual specs vary widely, and by checking out parameters like load-life stability (ie. the relationship between temperature and loading, versus longevity and stability) and temperature coefficient, the sonic benefits of good resistors can come hand-in-hand with greater consistency, if not enhanced reliability. In most bi-polar designs, an amp lasts as long as its output devices, but the lower suicide rate amongst the MOSFET population throws the reliability of the surrounding components into the furnace for the first time. The importance of this comes out when you've experienced amps which have become so hot that roadies couldn't undo the speaker connector on front of the rack: they had burns to prove it.
Circuit time. TR1 and TR2 comprise the input stage. Without the usual surfeit of devices to hide behind, they really have to put their oar in. In this instance, working at peak efficiency means applying current sources. These are illustrated in the diagram as twin, overlapping circles. Because the constant current level comes hand-in-hand with full voltage swing, each source looks like a bullworker to the transistors: something gutsy for it to work linearly against. Without a current source, the devices in this configuration would saturate at a low drive level, meaning a drastic rise in distortion at high levels, even with feedback applied.
Along with 99% of transistor power amps, the input devices are configured as a differential pair. Thus we can think of the two terminals as the + and - inputs on an IC op-amp. In view of the unique, inherently non-saturating qualities of this configuration, it makes good sense to use it on every possible occasion. The driver stages of common bi-polar amps use asymmetric, half-wave pairs (or triples) to build up the current. In the DVT meaning, we have employed a modified version of the basic differential stage, which is symmetrical, to boost the current up to the tens of mA's needed to feed POWERFETs. Whilst high, this current is still two orders of magnitude less than a bi-polar power device needs, to be turned on hard at HF. Put another way, we don't need to build a 40 watt driver stage; just a net 2W of class A will do. The outcome is that POWERFET driving is simplified to TR3 and TR4, and their associated current source and current-mirror. This, along with COB 'tuning', simplifies phase compensation, which can be sufficiently tricky on wideband circuits to make stability virtually unrealisable, short of dunking the lot in thick treacle! Somewhat ironically, compensation is achieved by small phase-lead capacitors between the base and collectors of TRs 3 and 4. But wait: these are real, linear capacitors, and besides not creating any distortion, they serve to swamp-out the vacillations of the deadly inter-electrode capacitance.
C4 in Figure 1 is in the lower arm of the feedback network ('NFB'). The purpose of this part is to force the DC gain of the whole circuit to zero (ie. x1), regardless of the AC signal gain, which is actually set at x23. But in every practical amplifier circuit, CB needs to be a large value — about 100μF, and thus an electrolytic, to maintain a sensible roll-off at LF. That's because, to make it usefully smaller, we'd also need to increase the feedback resistor values about 20 fold. This in turn would magnify the effects of noise, and stray capacitance, making hum, hiss and instability likely.
Any sort of polarised capacitor is bad news in this position, because it's apt to take its time to recover from DC level shifts, and high level asymmetric LF signals. Even with 100μF, there's plenty of scope for this, and in the wake of temperature changes, it will only serve to exacerbate DC level shift at the amp's output, and associated envelope anomalies. Added to this, we have the complication of distortion brought on by the fact that the capacitor appears within the error-correction (NFB) loop.
Here, we may conjecture significant even-order (ie. 2nd and 4th harmonic) distortion at the bass-end as a result, ironically, on asymmetric signals alone! That is, CB chucks in large dollops of even-order overtones, most especially when these components are already predominant in the legitimate music signal. CB isn't, therefore, a source of ear-splitting distortion, but more like a pair of deep crimson spectacles, which leads to pleasant perceptions, so long as the wearer isn't aware of, or seeking reality. In the DVT series, advanced capacitor techniques take care of this little setback. The outcome is a bass-end that supports the theory that alleged 'transistor sound' and 'MOSFET sound' are, in reality, nothing of the sort, but simply 'bad capacitor sound'.
With few exceptions, the power supplies inside professional power amplifiers look like one big afterthought. This is a pity, because the power sourcing performance is obviously capable of holding back the performance of the amplifier as a whole, however brilliant its conception. This in turn dictates the outcome of the BIG test: the venue is a dimly lit warehouse, the audience an old Rock 'n' Roll cynic. Once the amplifier salesman is fully wound-up, the acid question is brusquely pitched into the air. 'Can it kick ass, man?'
Let's begin by reflecting on the nature of instrument sounds: every guitarist knows that a well tuned valve instrument amp can sound as loud as a 1kW PA. In part, this is down to 'narrowbanding', because with only one instrument to deal with, the spectrum is confined enough for sounds to be pushed harder, without ill-effect. This much is common knowledge but we haven't yet explained where the energy for hyper-powerful guitar chords comes from. Somewhere implicit in the gusto of tube amps, is a hidden reservoir of energy; inside the power supply's capacitors to be precise.
Energy is measured in joules: it's the product of voltage squared x the capacitance. E=(CV2/2) joules. So let's compute the joules in a 200 watt tube-amp's supply, say 1000μF charged to 440 volts. It's (0.001 x 440 x 440/2) = 196 joules. Now let's take an average 100 + 100W stereo transistor amp, by way of comparison. With 4700μF charged up to 60 volts, the energy preparedness is just 9 joules (in mono), or 4½ joules per channel.
With ten times the energy to draw on in an axe battle, it's small wonder a good tube instrument amp ends up sounding twice as loud as bi-polar transistor amps, of the same nominal rating. Drawing on this radical knowledge in the DVT, we looked beyond the simplistic provision of BIG capacitors, and instead, began by setting up the highest supply rail voltage that's safe, within the limits of today's POWERFETs. At ±80 volts, even a modest 4700μF capacitor can manage 30 joules. Having set a voltage, this leaves the designer free to fill in the remaining space with μF. For the DVT250s, that's 15,000μF per rail, per channel, or 60,000μF in all. Turned into joules, the single channel asymmetric pulse capacity is 48 joules, while in mono-bridge mode, it comfortably exceeds a Marshall stack, at 196 joules...
This, in a nutshell, explains the DVT's undoubted balls, in the face of a real loudspeaker loads. However, joules alone aren't enough: it's only amplifiers without V-I limiting (like the DVT) which can translate the store of CV2 energy into useful amps, to placate the appetite of hungry, reactive loudspeakers.
A healthy power supply is a holistically conceived one. So, having visited the reservoir capacitors, it's time to dip into the mains wiring. For the average UK power amp, the internal 240 volt wiring has a 2 or 3 Amp rating, based on the simple reasoning that because it doesn't glow red hot, it must be adequate. And even if an open-minded person was to try uprating it, it would be a waste of time, in the face of V-I limiting. But once we dispense with this output stage protection, the amp is free to draw as much juice as it needs. How much? Well, at Glastonbury, in June 85, we recorded a peak mains current greater than 30 Amps on a 250s, which was doing bass and mid-range for Echo and the Bunnymen. Under these conditions, it's evident that even if the usual 2 Amp bellwire arrangement doesn't actually burst into flames, it would lay the bass sound to waste. Accordingly, the DVT series amps are served via chunky mains cabling, with a defined maximum resistance.
Next month, Ben Duncan continues his sideways look at amplifier power supplies, and examines V-I limiting, and 'MOSFET' sound.
Feature by Ben Duncan
Previous article in this issue:
Next article in this issue:
mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.
If you value this resource, you can support this project - it really helps!