Ferdi says "next installment" - fine, here goes.
Power amp power supplies
Before anyone asks, this is the second and last part of the power amp issue. Going beyond would incur using pictures, and ultimately, writing a monography if not a book. For chossing power devices, see appropriate text on
http://www.zero-distortion.com .
In typical modern amplifiers, we have separate plus and minus power supply lines. We could use just one, the plus, but this would require us to use a decoupling capacitor between the amp's output and the speaker, and this is always a bad idea, because capacitors are junk in general audio terms, and should be avoided whenever possible.
The most common arrangement is to have a power transformer, which takes the power from the grid and your wall outlet, say 120 VAC, and converts this to say 30-0-30 volts. Mind you, the voltage after the transformer is still alternating, so we need to rectify it, meaning to turn it from alternating current (AC) to direct current (DC), which does not have the 50 Hz alternating component.
This is achieved by using full wave bridge rectifiers. There are half wave rectifiers, but in audio, full wave rectifiers are used, and I can't even remember when did I see anything less last. A full wave bridge rectifier is essentially four diodes aranged so that the AC signal is cut by halves and then joined at the output, resulting in two supply lines. In this process, our incoming AC voltage from the transformer will be rectified and multiplied by the square root of 2, or by 1.41. Therefore, our incoming 30-0-30 VAC will become +/- 42.3 volts DC.
But this rectification is incomplete, it hasn't removed much of what we want removed, so we must additionally filter it. To that end, we use large filter capacitors, which in simplified terms pick up the junk riding along with our voltages and shunt it to the ground. Make no mistake, they do a good job of it, but the question is how good a job? More of that anon.
By default, large capacitors are parallelled by small capacitors of different material type, typically in the 100...330 nF range. They do what the big caps don't do well at all, they filter the very high frequencies well into the megahertz range, ridding us of unwanted RF intereference.
And that would be our typical power supply in commerecial units.
A step above this is to use exactly double of the above, one for each channel. This would give us a dual mono concept, its main benefit being that channel crosstalk is greatly reduced and there is independence between channels. When one is called upon to deliver large power, the other should ideally be totally unaware of it altogether. This helps keep a stable stereo image.
Another step above this is to use not one, but two full wave rectifiers per channel, i.e. two for each channel. One rectifies only the plus side, the other only the minus side. This has many benefits - better rectification, faster response due to offloaded rectifiers (work is split in half with two), far greater obtainable currents (double, in fact), better cooling of the rectifiers (offloaded and heating split up in two), etc.
Next, we can split up the supplies for the voltage gain stages from the current gain stages. This is a VERY logical decision for many reasons. First and foremost, voltage gain stages work with small currents, so full regulation is not a problem, nor is it prohibitively expensive to make. Second, since each transistor in the signal path will naturally produce a voltage drop of 0.65V, we can increase the small power rails to compensate for these drops and to ensure that even if clipping should occur, it will happen in the current stages first, due to their lower supply rails (lesser of two evils, though clipping of any kind should be avoided at all costs). Next, this would allow us to run the current gain section at lower voltages, which in turn means we would be keeping our transistors more within their safe operating area and could draw more current from them.
This greatly improves every amplifier I have ever heard. In general terms, it produces a more stable sound stage, more depth, more definition and better overall balance. I combine this with dual rectifiers per channel for best effect.
Lastly, we can do a full regulation of the whole amp. This means electronic regulation of both the voltage gain stages, and of the current stages. This will put our entire amp in a most stable environment, which will improve its performance if properly done, but at a price roughly twice that of the initial amplifier. This is because you effectively need another amplifier to supply the original amplifier, and if you want to be safe, the power supply amp has to be faster than the audio amp if it is not to start strangling the audio amp. All this takes much time, much development work and much money, and is consequently rather expensive.
I have tried it a few times, but was never really happy with this arrangement. I don't use it, but I don't knock it either. To each his own.
OK, so that how it works in general, now let's see how to dimension it all.
The first question we have to answer is how much power do we want, with what load tolerance, and under which conditions?
There are many compromises routinely made in this, because power supplies are not seen from the outside, but the customers do see shiny LEDs, and in the end, money ends up in LEDs, courtesy of the marketing departments, while engineers tend towards hara-kiri.
But the only rational way is to work backwards - start with what you want to leave the amp, and then design the power supplies for the juice you need. To illustrate this, let's take two examples, let's assume we want to design power supplies for a 40W/8 ohm amp and for a 100W/8 ohm amp. Follow this, and that 40W per side amp will subjectively sound like a commercial 100W per side amp.
We are all audiophiles here, a nice way of saying we're by and large mad as hatters, perfectionists in audio to the marrow of our bones. Thus, we would want our amplifiers to behave as ideal voltage sources, meaning to deliver an output voltage no matter what the load is, because we want to be free to buy the speakers we like, not those we must.
Let's also assume we want our amps to deliver a little more than their nominal power rating under dynamic conditions, in short term boosts. This is called dynamic power, and while distortion may rise, your amp will not compress and will not clip.
Let's take the small guy first. Let's say we'll be happy with a 40% extra margin over our nominal rating, or just over 55W under dynamic conditions. Let's say we will tolerate a 5% power loss every time our load impedance halvs, BUT in terms of our dynamic power. Lastly, let's say we will be happy with 2 ohms as our lowest impedance for normal operation.
So, if 55W is what we want for 8 ohms, then our required voltage will be sq.root ((2x55)
, or 29.66V peak, call it 30V. We must now accommodate for inherent voltage drops, say 3V, and must add some reserve, say another 2 V, so we need +/- 35V supplies for the current gain stages. Now, every transformer when pushed will sag, meaning it will deliver less voltage, so what we need are transformers made and sold with nominal voltages being declared UNDER FULL LOAD, not in the usual off load conditions. The differences can reach 7-8% otherwise. In short, we need power transformers rated at 25-0, 25-0 volts (twin secondaries, twin neutral wires) under full load.
Since 30V peak into 2 ohms represents - hang on now! - 226W into 2 ohms, we obviously need at least a 200 VA transformer PER CHANNEL. We could go for less, say 120-150 VA, if we decided we weren't interested in continuous operation into 2 ohms. Remember that every transformer will deliver approximately 1.5-2 times its nominal rating in impulses, so even lower values will actually do fine. Also, don't forget the filter capacitors, which will also help ride out any nasty transients.
Speaking of which, we need to consider them as well. Now, all this jazz about power supplies is really talking about energy, not only voltage, or only current. We need them bound together, not piece by piece.
Empirical experience shows that we need 1-2 joules of energy per every 10W of dissipated power. The reason why it's a range is that it depends on just how reactive a load is; if it's mostly resistive, 1 joule will do, but if it has nasty phase shifts, we will need 2 joules for the same power. Since 226W is our maximum power, we will obviously need 22.6-45.2 joules of energy. We know what our supply voltage is, we know how much energy we want, all we have to do now is calculate how large capacitors do we need. There's a very simple formula for this, and while not terribly precise or mathematically corrct, it does the required job just wonderfully:
1/2C (V squared) = x joules,
where 1/2C is the capacitance on either one of two symmterical supply rails and assumes the same value on the other rail in Farads, and V is the supply voltage of that same rail. For a start, let's take 20,000uF as an initial value (for a whopper value of 40,000uF per channel, or 80,000uF for the whole amp - not many 40 watters around like that). We have:
0.02 (35x35) = 24.5 joules.
A bit over our lower level, but with quality capacitors, and using two 10,000uF caps in parallel for each of the four supply lines, more than enough in real life. I would therefore use this and call it a day, I know for a fact that this would be more than enough. Since music pulsates, it changes in level, "continuous sine wave" power is hardly a meritory way of measuring power, though it is accepted (just as watts are accepted, when they are a measure of heating and when we really need to talk about volts).
You could use 40V capacitors, but I would advise against it. A 40V rating is too near our actual rails, I'd go for a 50V rating, I like being safe. Since they are parallelled, you don't need ultimate quality, go for good quality, such as say Nichicon, Panasonic, etc.
Regarding the 100W amp, let's say we are happy with +/-50V, as this would give us about 144W of peak power into 8 ohms. But this is 288 watts into 4 ohms, or 576W into 2 ohms. Really rockin' here. We need 57.6-115.2 joules of energy.
Using parallelled 10,000uF caps per channel rail, or keeping what we had in the previous example, we'd get 50 joules of energy storage.
But let's get ambitious here, hey, we don't do this every day. Let's say we want to go to the hilt, so let's keep the two 10,000uF caps for good filtering, but let's also add one 4,700uF cap for speed. Our energy reserves rise to 61.75 joules. As in the previous example, while this is below what we might nominally require, in real life this is more then plenty assuming we use double rectifiers per channel and sufficiently large power transformers. Speaking of which, I would recommend no less than 400 VA per channel, and 500 VA is better yet. More would be a waste of time, I think, for everything but benchmark pure sine wave testing. That's 1,000 VA for the whole amp - how many do that? And in between, you have the third, say 100VA toroid, feeding the voltage gain stages and protection circuits.
The best place for the filter capacitors would be as near to the power devices as possible; this is a good point to remember, proper placing makes a lot of difference. Unfortunately, this assumes you make your own printed circuit board layout, which most can't.
Please remember, the above is an example only, and should not be taken as conclusive.
One last item - voltage gain stage regulation. As everything else, so this too can be done in many ways. For example, you could use our trusted LM 317/337 regulators; while they nominally go up to 38V only, you could always put their ground pin under a voltage from a simple zener diode, which would raise the adjustment level by 38 plus the zener voltage. Or you could build a discrete regulator, which takes time, but can provide excellent results. Or, you could use a single power MOSFET, such as say IRF 540/9540, to produce a "virtual battery", which is essentially a voltage regulated power MOSFET circuit followed by a high quality capacitor of some size, say 1,000uF or larger, so between the two of them, they deliver something very near a battery qiality of supply (but not quite battery).
Based on still inconclusive tests I am conducting, a line filter which actually works, in my case a DeZorel, will make your power supplies far more efficient. This is an interesting aspect, because it makes me think that instead of pouring inordinate amounts of money into massive power supplies, I would be better advised to invest into a line filter, and then downsize my power supplies simply because I will get the same effects with more reasonable components, which after filtering have a much easier job to do. But I still have quite a few things to try before I'm sure of this.
The problem is that my initial results show that I cannot duplicate the effects after the filter with any reasonable number of ultra high quality capacitors. Putting 6 4,700uF Siemens Sikorel caps per supply line on a +/-33V circuit is what I'd call already quite unreasonable, yet those six cannot do what just two in parallel with a filter preceding them can. And just one channel pays for that filter in less capacitors, so I get the second channel for free. But as I say, I have some work to do on this subject yet.
Cheers,
DVV
