Vacuum Tubes Part
time (in part 4) I promised a current source that would improve the performance of the tube
differential we were building. But first, let's define what a current source
actually is, because I know that I had fits trying to wrap my head around the
idea of a current source the first time I saw one... and I'm pretty sure I'm not
the only one.
It goes like this — a current source is everything a
voltage source isn't.
There... feel better?
I don't blame you. That's pretty much the same sort of
non-explanation I got when I asked about current sources. Eventually it dawned
on me that the reason I was getting such vague answers was that they hadn't the
faintest clue, either. Ahem. So
much for self-styled “experts.” And yet, that answer is correct in so far as
it goes. A current source really is the mirror image of a voltage source and
lots of people have a fairly good intuitive understanding of voltage sources,
even if they aren't accustomed to calling them by that name; they're more
commonly known as voltage regulators.
So let us consider a voltage regulator that's designed to
deliver, say, 50V. If you give that voltage regulator a 1k load, it will supply
50mA while holding a constant 50V. (50V / 1000 Ohm = 0.05A, or 50mA) If you give
it a 10k load, the voltage will remain constant at 50V, but the current will
drop to 5mA. In fact, if you stick a variable resistor in as the load and whip
the knob back and forth, a decent voltage regulator will maintain a rock-steady
50V, but vary the current up and down to match the changing resistance, no
matter how fast you change it. There is something to note about voltage sources
— they have low output impedances. A mathematically perfect one would have a 0
Ohm output impedance. You're already familiar with that, too, in terms of the
output stage of power amplifiers; the damping factor spec is an indication of
how low the output impedance of the amplifier is. (Yes, the voltage of a power
amplifier varies. That doesn't mean that it's not a voltage source, it's just
that we're programming it — via the input signal — to have a varying output
So to summarize, a voltage source (aka voltage regulator)
holds the output voltage steady and allows the output current to vary. It has a
low output impedance. Now, given that a current source—which you could also
call a current regulator if you wanted to — is the opposite of a voltage
source, you can predict its characteristics. It will hold the current steady
while allowing the voltage to vary, and it will have a high output impedance. Of
note is that high impedance is exactly the ticket for biasing our differential,
because it will force the signal coming from the input side of the differential
to go to the other tube, rather than down to ground. Sounds good, but how do you
build a current source?
They can be as simple as a single JFET or as complicated as you want them to be. Let's start with the single part options. We've already seen that a low value resistor doesn't work too well. Broadly speaking, you could say that it's a current source, but it's a pretty poor one, performance-wise. However, if we replace it with an N-channel JFET with its Gate connected to ground, then we've got a pretty good current source. If you think that sounds too good to be true, your suspicions are well-founded. The problem is that you're running the JFET at IDSS, in other words it is wide open, which isn't really a problem, per se, but you don't have a good way to set the current to the value you want. You have two options here: You can sort through a large number of JFETs to find one that has the right IDSS value for your design, or you can pay someone else to do pretty much the same thing. There are commercial versions of exactly this current source and you can buy them pre-sorted according to your needs.
However, by adding a resistor you can gain a little control over your current output and improve the performance in the bargain. The resistor goes under the Source and it will reduce the current output of the JFET, so you will need to plan ahead. You'll need a JFET with higher IDSS than you intend to use in the circuit. The resistor gives you two benefits. The first is that you can set the current output of the JFET. The other is that it tightens up the response of the current source. Let's begin by assuming that the JFET is delivering exactly the right amount of current. Then along comes a bit of a breeze and the output of the current drifts away from the planned value because the breeze changed the temperature of the JFET. If the output increases, then the voltage drop across the resistor also increases and this decreases the current because the Source is lifted farther above the Gate, which is held at ground potential. (JFETs work pretty much the same way as tubes in this regard.) If the output drops, then the voltage drop across the resistor also drops and this brings the Source closer to ground, thus increasing the current. In other words, it acts as a feedback loop, correcting errors in the current source's behavior.
Before going any further, I'd like to note that although some
people will consider it heresy to put a transistor in a tube circuit, as the
bias circuit approaches infinite impedance, the current source will be asked to
supply only DC, and surely you can trust a transistor with DC, can't you? Look
at it this way. If an electron chooses to go down the 620 Ohm path in the
original circuit, then it's going to cause a variation in the voltage across the
resistor, right? As the signal varies, you'll see an AC signal across the bias
resistor, precisely because the value of the resistor is so low. If the
impedance of the bias circuit is high enough, no electrons will choose that
route because the impedance is too high. No electron flow through the bias
circuit means no AC, which means that the current source only has to deliver DC,
something any circuit should be able to manage with ease. Face it, DC is easy;
it's when you ask a circuit to deliver AC that you run into trouble.
Okay, now that we've introduced the concept of a
current-sensing resistor, let's kick it up a notch. Let's amplify the signal
derived from the resistor to make the error correction more sensitive. While
we're at it, we'll switch to bipolar transistors, which will give us a
relatively predictable voltage reference to use as a baseline for comparison.
This circuit works slightly differently, so we'll go through its operation in
For starters, we'll assume that Q1 will switch on when the
voltage between its emitter and base is about 0.6V. This is not an exact figure.
It will vary a bit depending on the transistor (because of how they're doped),
but once you've chosen a transistor type it is far more consistent than the IDSS
of a JFET. If we wanted to program this current source for something around
6.5mA, which is roughly the amount we were getting using a 620 Ohm resistor,
then we would plug the numbers we know into Ohm's Law: 0.6V / 0.0065A = 92 Ohm.
Although 92 Ohm isn't a “normal” resistor value, 91 Ohm is, so we'll use
that. If you use a different transistor for Q1 and discover that the Vbe (the
voltage difference between the base and the emitter) value is a little
different, you can fine tune the resistor value easily — lower for more
current, higher for less.
Again, let's say a vagrant breeze upsets the apple cart. If the current through Q2 (the “output” transistor of our current source) increases, then the voltage drop across the current-sensing resistor (R2) increases. This pushes the base of Q1 more positive, which increases the amount of current passing through it... What? We haven't covered JFETs and bipolar transistors yet? Yeah, I know. Be patient. I figured you'd want actual circuits as soon as possible, so I started a tube circuit as soon as I could. Didn't want everybody snoring off during all the boring background stuff. We'll get around to FETs and bipolars at some point...
Er, where was I?
Oh yeah... Q1 is now conducting more heavily (just trust me on
this one, okay?), which means that the voltage across R1 increases, driving the
base of Q2 down, which in turn means that Q2 passes less current, and all is
right with the world. For all the fact that it's bipolar instead of JFET, the
basic circuit is pretty close, conceptually, to the last one. It's just that
we've added Q1 as an amplifier to make it even more sensitive to fluctuations.
We've now explored four possibilities for biasing the
Is that it? Are those the only options? Not by a hundred. I used to make a point of collecting current source schematics. I had dozens. I got tired of it. Suffice it to say that there are a bajillion current source topologies out there. For instance, you could use an OpAmp for the error amp. You could use a MOSFET for the output device. You could compare the current sensing resistor to a precision voltage reference. And so on, and on, and on. Mix and match to your heart's content. Warning: You can drive yourself crazy obsessing over current sources. Take it from one who knows.
At some point it's going to occur to someone to ask if it's
possible to make a tube current source. The answer is yes. The basic circuit
looks pretty much like the one for the JFET with a resistor under its Source.
The downside is that we're back to the “tall pole in a deep hole” thing that
I described as the solution for a high bias resistor value in our previous
installment. You'll need to “sink” the current source into a negative rail.
Can it be done? Of course. And for true dyed-in-the-wool purists, it'll be just
the ticket to nirvana. Maybe later I'll double back and take a look at the
options I'm bypassing at this time, but given the number of possible circuits in
the world, it would take about three seconds less than forever to look at each
and every permutation. My goal is just to stimulate your thinking. If you want
to experiment on your own, you're welcome to do so. Once you've got the basics
down you can put together a test circuit and plug in dozens of current source
variations to see which one sounds best. I'll give you a teaser to goad you on
your way: Consider the compliance of the current source. The compliance is how
the circuit responds to varying demands. You'll want a circuit with wide
bandwidth. One that responds too slowly to high frequencies will have an audible
signature that will show up in the music.
Next time we'll slap the output stage(s) back on the
differential and be done with the circuit.
Done? Did I say done?
Never say done.
There's always more to play with.