Interstage Coupled Amps
Support Information from Andy Grove
Interstage coupling was widely used in the early days of radio and audio, in fact it is still used in R.F. work extensively. I once had, now long lost, some paperwork written by Western Electric engineers, and they spoke of using transformers to minimise energy loss and indeed interstage coupling, if applied correctly, results in the most efficient coupling of one stage to the next, and the best utilisation of the available HT voltage. It was for this reason, in the early days, when valves were expensive, and didn't have the high performance of later types that transformer coupling was used; the gain and voltage swing of each valve is maximised so, essentially, you need less of them.
Transformers gradually disappeared from audio circuits, and really this wasn't due so much to issues with performance, but more to do with cost and convenience, and, sadly, a lack of understanding of their design and application. Circuit designers became lazier as the physicists and engineers designing the valves improved them, transconductance and gain increased over the years and if you bogged down each stage with RC coupling who cared? You just add another stage to make up of the loss of gain if necessary. When circuits became enclosed in feedback loops, the possibly unpredictable or, to use a messier but more accurate phrase; the “not easily modelled” high and low frequency characteristics made them unpopular, and there was a time when engineers wished that they could just design them out. Of course the likes of Philips (the manufacturer) and Futterman (the guy) did.
The want to increase feedback levels in order to decrease distortion during the THD figure arms race made transformers especially unpopular, and probably it was that which eventually killed them in most products. What is important to bear in mind is the fact that transformers do not limit the performance of these amplifiers directly. What does limit the performance is the cost of a good component, and most importantly, the fact that there aren't many amplifier designers who also design transformers. The situation was, and still is, predominantly that of a transformer being seen by the majority of engineers as an off the shelf component like a resistor or a capacitor, the problem here is that a transformer can be an incredibly complex component, and if you just plonk one in a circuit and try to wrap 60dB of feedback around it you'll get problems. If you understand what's going on then, really, you can build a valve amp with transistor amp looking figures, if you wanted to that is.
What's good about an interstage transformer?
The transformer used in the monoblock amps is different to the usual type; it's bifilar wound. Some years ago I introduced this type of coupling transformer in single ended form for driving power triodes; it has since been copied by others, well they do say plagiary is a form of flattery. Bifilar winding is a technique where two (hence the bi) are wound together, literally side by side along their entire length. In the past interstage transformers usually had separate primary and secondary windings, the two separated in the transformer by insulation and interwinding screens if necessary. This resulted in imperfect coupling between the two windings, not always a bad thing, but in this case (between driver and output valves) it's usually better that the windings are very closely coupled. There are several reasons for this, the two main ones being that the imperfect coupling between the two windings shows up as a leakage inductance, which can be imagined as being in series with one, the other, or both the primary and secondary winding, depending upon how you look at it. The upshot is that this inductance will create a resonance with any capacitance connected to it. Capacitance like the input of a big power triode, or the internal stray capacitances of the transformer which, sadly, are increased in the non-bifilar type due to the interfaces between the two separate windings. To get a flat response and to lessen overshoot on fast edges such a transformer usually needs some kind of damping resistor, or even a Zobel type network on its secondary, and not only does that affect the sound of the thing, but it partly negates one of the most important reasons for using an interstage transformer in the first place (I'll get to it soon).
When using a bifilar transformer the coupling between primary and secondary is so good, that really you can put on as many turns as you like, the main limitations are that the DC resistance of the two windings will get higher and higher, until it's becomes equal to the anode impedance of the valve you're using to drive it, and the wire gets so thin that insulation between the two closely coupled windings can be a problem. Even so, you can have big primary inductance without the penalty of high leakage inductance, you can swing lots of volts before the core saturates at any given frequency, and because the flux density is low and, due to the high turns, the perturbations in the material characteristic make less difference to the signal.
Now we have a transformer which has lots of turns on it, but still the high frequency response is smooth and extended (dependant mainly on winding technique). This means that you can do something very different, and that is run the interstage with no load on the secondary. You have a driver valve, the interstage, and then the output valve, nothing else apart from your bias components and of course the power supply. This means that the driver valve is running essentially into an AC infinite impedance (well very, very high anyway) at audio frequencies, which in modern parlance translates as constant current operation; the loadline is horizontal. To a driver triode, that means lower distortion, and maximum voltage swing, swing that can exceed the HT supply rail. This cannot happen with one of the funky, complex stacked systems often seen as clever constant current loads; things like SRPP or the mu follower.
I look at it this way; you can take bog standard parts, and connect them together in complex ways, or you design a component which does what you want. Each has its merits, but the transformer is such a neat, purist solution.
The secondary winding at DC is a length of wire, and thus has a low DC resistance to ground. This is good for power valves. There has been plenty written about it in the past, but there's two big effects which can happen at clipping, the first can be disastrous, but to be honest I've never witnessed it myself. Put simply output valves have big grids, and therefore substantial grid current. Usually this current flows (according to convention) INTO the grid, thus, if a resistor is connected there it creates a negative potential at the grid and biases it a bit more negative, when you see “grid bias” with the 10M resistor in the grid, this is what's happening. However, when the grid is driven towards zero volts the grid current approaches zero and then changes direction (actually there are two opposing currents, one usually dominates the other); current flows OUT of the grid and the voltage across any resistor placed in the grid circuit will increase, thus biasing the valve further into conduction. The larger the resistor the more pronounced this effect will be, and it's why a maximum grid resistor is usually specified with power valves. I seem to remember it was Osram who referred to the “Trigger Effect”, what can happen is that under clipping if the grid to ground resistance is too large when the grid is driven momentarily positive grid current will increase, pushing the valve further into conduction until it melts, like I said, I've never seen it, but in theory it could happen.
Under more normal operation which, with the kind of amps we're usually dealing with, means regular clipping is a fact of life. If you don't believe it, look with a scope! With a usual RC coupled stage when the grid hits zero volts it will suddenly go from megohms to being kilohms to ohms in impedance, this clips one half of the AC cycle in the output valve. If you look at it another way, what you have is a diode there, a rectifier. What happens with a RC coupled stage is that the grid rectifier and the coupling cap generate, and store a DC voltage at clipping, and the output valve is biased more negative. With SE amps you can see the effect as pumping of the bass unit. With clipping the output valve bias is suddenly reduced, and this change in current is transmitted via the output transformer to the speaker as a subsonic signal. What is worse is that the effect persists for some time, depending upon the coupling cap and output valve grid resistor time constant. With cathode biased amps the DC feedback effect of the bias resistance helps to compensate, with fixed bias the effect is more drastic, the valve can become cut off completely. I've seen it in some highly respected amplifers.
The low DCR of the transformer secondary and the lack of coupling cap mean that both the effect of “normal” grid current and that of clipping are minimised, if not eliminated completely, and recovery after a transient overload is more or less instantaneous under ideal conditions. When the output valve grid does conduct the low source impedance means that clipping is fairly soft.
Using parallel feed, and choke loading and then cap coupling are not the same thing, briefly I'll explain why just because it bugs me a little when they are compared.
The advantage claimed of parallel feed is that your output transformer can be made of a better material such as mumetal, and you use a big silicon steel choke to carry the DC. Well as in all things you get nothing for free; first you are now swinging the material through its initial permability point, which doesn't happen with an SE transformer because the DC current biases the material to the best point on the curve. You still need a shed load of turns to support the voltage, and at the end of the day you've still got a lump of silicon steel there across the primary anyway, this will have a direct affect upon the sound. And of course you need a blocking capacitor.
Using choke loading and then capacitor coupling is a waste of time, you need the resistor to ground at the other side of the cap for biasing, and so therefore you load the driver stage, or if you use a choke there then you have two chokes in parallel with uncoupled magnetic fields, why not put both chokes in the same field?
One disadvantage of the bifilar transformer is that the input must be in phase with the output, this is due to the relatively large capacitance which exists between the two closely spaced wires of the primary and secondary, when the input and output are in phase then this capacitance actually forms part of the coupling. If the phase of the output were reversed compared to the input, in other words if you were to use the transformer as a phase inverter, then the magnetic coupling would be fighting against the capacitance of between the wires and the capacitance gets reflected across the primary and secondary. This shunt capacitance will severely limit the high frequency response of the stage. When using bifilar transformers in very low impedance circuits it is possible to alter the phase of one winding compared to the other, the magnetic coupling beats the capacitive, however we're not talking about “normal” valve impedances.
The other main disadvantage is that because the wires are so close together the maximum voltage you can put between them is limited, usually to around 350V on a small transformer to be on the safe side, and it partly depends on the wire insulation and the post winding processing. The insulation on copper wire is usually excellent if it comes from a good source, the insulation on silver wire is never usually as good; it seems that the coating doesn't adhere as effectively to the silver surface. If the transformer is made larger then the wire, and therefore the insulation thickness can be made thicker, and thus you can get away with more voltage.
The seemingly low voltage isn't such a problem, because you need much less HT than with an RC coupled stage, or in fact any stage which is loaded down in some way with resistors. Effectively you could use a 5687 at 250V and swing +-200V just about. On the subject of driver valves for these transformers it's usually best to pick a lower impedance type, such as a 5687, ECC82, parallel ECC82, 417A and so on.
Andy Grove - Audio Note design engineer