Elevating the miniGainBrick Zip LM675 into the elite
Class A operation
by Andre Jute
Part 2 of a two-part article which starts here
Virtually all preamps operate strictly in Class A1. There is a very good reason for it: quality.
Very few silicon power amps operate in Class A. There is a very bad reason for it: cost.
To repeat, the residual trouble with the 675 to my sort of audiophile (I like the sound of Class A1 triodes and to the devil with the wasted energy) is the switching and crossover noise and odd harmonics which are the price of efficient Class B operation.
Creating the Class A bias
We will kill several crows with a single stone. The stone in our sling is called a constant current source or CCS. (I much prefer constant current load, actually, because that is more precise, or failing that the generic term constant current device.) By the judicious arrangement of our CCS we will force the 675 opamp into Class A operation, which takes care of the switching noise and general rubbish of Class B, and into Single Ended operation, which takes care of the crossover distortion and eliminates the odd harmonics as well. Note that while single-ended operation is by definition Class A operation, the converse is not true. Push-pull operation can also take place under Class A conditions.
It will be helpful to have these illustrations from
open on your screen:
The simplest and least stable CCS is a simple resistor shunted from the output of the opamp at pin 4 to the negative rail on pin 3. The bias resistor is shunted to the negative rail to force the opamp to operate only on the NPN transistors inside its body, because the NPN's are more predictable in their behaviour than the PNPs. You can try this too: shunt the CCS from the output to the +V on pin 5 and watch the distortion spectrum turn nasty; you may even be able to hear the deterioration instantly, depending on the resolution of your chain and the tolerance of your family and neighbours for loud music.
The CCS puts a current bias onto the transfer function of the opamp which forces it to operate on one side of the switchover only. Bingo, switch noise gone. The CCS also forces the opamp to operate in Class A, because single-ended operation is by definition in Class A. Bingo, third and odd harmonics gone (if all the other elements are correctly proportioned). Class A has another bingo: it is an inherently less distorted method of operating an amplification device, so not only are the third and higher harmonics gone but second harmonic is usually reduced. (So much for the 'added euphonics' argument we keep hearing from people who spend so much time grinding their axe they don't have time to do their homework.)
Bing! Bing! Bing! One single action which gives us three advantages and a very substantial further control handle on shaping the sound of our opamp to please even more.
The theory and the reality
I'm a Calvinist so I am steeled to discover that so much promised ecstasy has such a high price that it may not be realized or may be only partially affordable. You'll understand why I'm skeptical even of what in the flush of enthusiasm I write myself when I put some numbers on the CCS.
For the tube-hobbyists still with us, a transistor (of which an opamp is a unified collation) has a transfer function precisely like a tube but with much uglier curves. The other important difference is that while a tube is a voltage controlled device, a transistor is a current controlled device. So instead of setting a negative bias voltage with our cathode resistor on a tube, on an opamp we set a bias current and for exactly the same reason, to ensure that the amp operates in a particular class we find desirable. Keep in mind that these classes are not exclusive: it is common in PP tube poweramps to arrange the negative grid bias so that they operate in Class A for while and then move into Class B when more power is demanded by a higher level of signal swing, which we call a Class A/B amp.
The quiescent operating point of a transistor is Iq and it is found as the square root of the theoretical power divided by twice the load. The load in our design is the 8 ohm speaker.
Iq = SQRT(P/2RL)
Take the square root of 30/16 and discover that the biasing current should be set to 1.37A. The theoretical highest signal is the available input times the voltage gain or 22V, which is also the voltage we expect from the power supply, so the bias resistor must be 16 ohm and it will dissipate 30W so we should use a 100W component, which itself will require a substantial heatsink.
It isn't even worth calculating how big a heatsink would be required to keep the 675 from shutting down for the day or for good. There is no such thing as a free lunch. Full power Class A operation burns three-quarters of dissipation as heat. That's just the theoretical minimum. In practice Class A operation blows off more like four-fifths of dissipation as space heating.
That, in a nutshell, is the argument of proper engineers and their associated parasit-- er, I mean accountants against Class A. It is also the entire argument against running transistors in Class A.
If at this stage you still hanker after a silicon amp that runs perfectly in Class A, you don't really want a compromised 675 opamp. (Remember, I chose opamps and then serendipitously the 675 simply for convenience.) If you are willing to spend the money for the big caps and the big heatsinks, don't walk, run to Nelson Pass and his Zen amps. The rest of us will continue with a relatively inexpensive experiment.
Learning to live with Class A/B
Class A/B sound is not necessarily better than Class B. The result depends on the arrangements. The arrangement it depends on most is in fact the speakers, and we will also use the psychological interplay between your ears and sound. Unless your speakers are grossly insensitive, most of the time your amp, however powerful it may be, idles along at a fraction of a watt. That is why it is often said that the first watt is all that counts.
One watt into eight ohms is 2.83V and the required bias is 0A35 or 350mA. The required resistor to arrange this is 60R and it would dissipate over 7W so a 20W or 25W wirewound would be required. You could use the easily available 56 ohm resistor or two 120R in parallel if you are obsessive. The proper engineers and accountants have now stopped screaming in rage and are rolling on the floor, tearing out their hair in frustration that we would waste so much power in heat in the resistor and the device (or more precisely, in their heatsinks). But we are still refusing to be deflected, resolutely heading for that glorious 300B sound at a very reasonable price, at least compared to the likely cost of our adventures in the section immediately above.
The bias resistor is in parallel with the output load, so you need to check that the resulting combined load has not fallen too low for the 675 to drive. 60 ohm in parallel with 8 ohm is 7 ohm so we're okay.
One more thing. When I first wrote this article, people wrote to me saying all that was required was one or two milliamp of bias. That's just silly in an opamp used to make real power; it arises from a misunderstanding of a common professional trick to ensure the opamp actually operates in Class B rather than on the margins of Class C. Opamps operated low down on their transfer functions are particularly nasty.
At this point, having bolted everything to the biggest heatsink you can afford, you can test your new pseudo-300B melted sand amp. Now it will definitely sound like a tube amp. What is more, you will be able to make it sound more or less like a tube amp by varying the bias current as well as the other elements we have already described such as the stiffness of the supply, its voltage, and the feedback resistor. If you have a spectrum analyzer you will also be able to verify my finding that what matters is the make-up of the harmonics, not just their total elevation. But even if your only tool is an oscilloscope you will be able to see and hear that the shapes of some kinds of distortion are less offensive to the ear than others. Use your tools, including your ears, to optimize the miniGainBrick Zip LM675 to your taste.
Is this black magic?
Absolutely not. It is not even sleight of hand. It is all based on science with proper repeatable experiments. The ear is more sensitive to all kinds of distortion at lower listening levels than at higher volume settings. This applies even more to third and higher harmonic distortion. We mostly listen to quite low levels, so the nasties give us a double whammy. But even on only moderately sensitive speakers, in-room 1W can sound huge. The transient peaks 10dB or 20dB higher can 'afford' much more distortion because they are so much louder. Our ear makes its judgment in the first 90dB where our one watt of Class A suffices.
Is that all there is to transistors as tubes?
It seems to me more than enough.
I have a test. I put a world class performer of classical music, a man or woman with certified golden ears, in front of a curtain. A CD player plays their latest disk. Behind the curtain, being switched by a third party, is a Class A ZNFB 300B SE, a low or zero NFB triode-linked Class A EL34 PP, and Class A/B commercial valve and SS amps linearized with oodles of NFB. These are all truly silent amps. Invariably the Class EL34 PP wins in the unsighted test, and in sighted tests the 300B wins.
What's cooking here? The winners produce Class A sound from triodes or triode coupled pentodes operated so as to sacrifice power for silence; the unavoidable residual distortion is all second harmonic. What's absent? Switching distortion and crossover distortion from Class B, higher order harmonics, any significant level of negative feedback.
The problem with negative feedback is that it may reduce noise but it doesn't shape the components of the residual. Tiny amounts of third and higher but especially odd harmonic distortion are far more objectionable than quite large amounts of second harmonic. Rather than the much-paraded 'smearing', the reason so many sophisticated listeners find large amounts of NFB, or in some cases any NFB, disturbing and objectionable may be that NFB disturbs the natural balance of the harmonics.
It may be that the ear is a natural adherent of ZNFB Class A triodes!
The resistor we have so far used as a constant current source is poor at its job. You can improve it by splitting the resistance in two with bypass caps from the junction, which splits the load on the opamp into AC and DC components, one part of which also filters high frequency noise on the power rail.
An N-channel JFET makes a better constant current source than a resistor; use it with a resistor of its own to stabilize it. A JFET cascode does better still. A current-regulating diode or CRD also does a better job than a resistor or a single JFET.
After that, if you are absolutely obsessed with making a transistor amp sound like an ultra-fi tube amp, you want to start working with discrete transistors rather than opamps, so that you can gain control over negative feedback and perhaps minimize or eliminate the last element over which you do not yet have control.
At this point you should have all the tools to mimic a ZNFB SE300B fairly well with a silicon amp. I don't imagine it will be all that much cheaper than the real thing or even smaller or lighter. That realization was the reason I gave up the quest. A 300B amp is prettier.
So, was the journey worthwhile?
For me it was. You hear a lot of waffle on the net from people who claim to be the only prophet of this niche and that, with the gaps in comprehension gouged even deeper by diplomaed quarterwits who insist their opinions are facts only heretics would question. They all have axes to grind. A great deal of so-called 'knowledge' on the net is in fact casually regurgitated gossip. The LM675T opamp has been one of my cheaper experiments in bypassing the gossip mill on the road to knowledge proven to my own satisfaction. That achievement is by itself satisfying. It has also left me with a little amp that satisfies as an amp in its own right.
If I were poor and didn't know the hams who gave me a junkbox when I started out in electronics, I would consider building a miniGainBrick Zip LM675 as my first or even only amp.
All text and illustration is Copyright © Andre Jute 1996, 1999, 2004
and may not be reproduced except in the thread KISS xxx on rec.audio.tubes