Ver. 1.0 Headphone Amplifier
I don’t know about you, but I purely detest people who say stupid things like, “If life hands you a lemon, make lemonade!” What if, like me, you’re not particularly fond of lemonade (or platitudes)?
So what do I do if life hands me a lemon?
I wish you hadn’t asked me that.
All right, I’ll go ahead and make the confounded lemonade... then hammer together a lemonade stand and sell it off at a dollar a glass. Which, oddly enough, leads me to the circuit at hand. The entire story is longer than I can justify and not much fun, anyway. Suffice it to say that I ended up with a large quantity of P-ch JFETs and began experimenting with recipes to make lemonade out of the blasted things. My intention was to use all P-ch devices, but then I remembered that lemonade requires sugar to balance the flavor. I relented and allowed myself a single “N” device. Okay, okay, so it’s an NPN... picky, picky, picky.....
The JFETs in question are J271s. While not explicit complements to the J310, they’re pretty similar, and I may show something that uses both in the future. They’re pretty hefty in the current department, which got me to thinking about power applications, which got me to thinking about headphones, which got me to thinking about my Grados. Not to put too fine a point on it, but Grados are a booger to drive. Those of you who’ve been at this a while might remember the Apogee speakers. Beastly, low impedance things that ate amplifiers for breakfast, lunch, and dinner. Current hogs. Think of the Grado headphones as the Apogees of their kind and you’ll be headed in the right direction. Whereas the Stax electrostatic headphones have as near infinite resistance as you’ll find in the real world and the Sennheisers come in at 300Ω, the Grados clock in at a punishing 32Ω. Ouch!
But what about my old Yamaha HP-1s? I still use them on occasion. At 150Ω, they’re half-way to Sennheiser territory; they need less current, but more voltage swing. Could I build a single circuit that would accommodate both? And the Sennheisers... what about them? Still more voltage swing. And the Stax? What about... Whoa! Put on the brakes, fella! A truly universal headphone amp was out of the question, though the Grados and the Sennheisers might just be doable within the working parameters of the J271.
So I got busy.
The SSTART circuit had one stage. The Difference Engine had two. This one has three. In fact, it’s a classic opamp topology, albeit with a quirk—you don’t usually see JFETs as power output devices. But wait, aren’t opamps bad? Hmmm... many are the words I could spill on this topic, but in the interest of brevity, I’ll confine myself to pointing out two things:
1) When people say “opamps” are bad, they’re generally speaking of integrated circuit opamps. “Chip” circuits, in the vernacular. An operational amplifier (frequently shortened to opamp in daily use) is a topology, not a manufacturing technique for building miniaturized circuitry. Quite a few people — and I’m guilty of it, myself — use the term opamp to refer to chip circuits. Let’s agree to draw a distinction between the operational amplifier topology and integrated circuits, at least for the duration of this project. We can all go back to sloppy use of the terminology tomorrow.
2) If “opamps” were necessarily bad, you’d find that you’d disqualified the vast majority of audio power amplifiers and not a few preamps. It’s a very popular topology.
The P-amp is not a difficult circuit to understand. It starts with a differential (see the Difference Engine for more on differentials), followed by a voltage amplification stage, and ends with a class A power output stage. (Class A outputs are not welcome in integrated circuits because they put out an inconvenient amount of heat for the small package. Since we’re working with discrete components, we have more latitude.)
The differential accepts the signal at the Gate of Q1. To keep the output phase correct, we choose Q1’s Drain, rather than Q2’s, and use that to drive the sole non-P-ch, non-JFET active device, an MPSA18, against a current source (Q4), which provides the voltage gain of the circuit. If the Grados had a high enough impedance, we could stop right there, but alas, t’was not to be. Current gain is needed. For that, I used J271s as followers, biased by more J271s operating as current sources. In fact, the entire upper half of the circuit is comprised of J271 current sources. Q2 is rigged as a current source to bias the input differential. As mentioned above, Q4 is a current source load for the voltage gain stage. And Q6, Q8, Q10, Q12, and Q14 are all J271 current sources for the output stage.
And the fella in the front row says, “Er... right. So what’s a current source?”
Glad you asked.
You are familiar with the idea of a voltage regulator, right? An ideal voltage regulator will deliver any arbitrary amount of current from 0 amps to infinite amps while maintaining an absolutely steady voltage. That's why it's a voltage regulator—the voltage remains constant, no matter what, but the current varies. Now, a current source has another name: current regulator. What's a current regulator do? (You can already see where I'm going, I'll bet.) It locks the current...and lets the voltage vary. Kind of like an upside-down voltage regulator.
Suppose we wanted a steady 1 Amp. If you give a 1 Amp current source a 1 Ohm load, it will develop exactly 1 Volt across the load. Simple application of Ohm's Law: I*R=E...1 Amp * 1 Ohm = 1 Volt. But what happens if you give it a 2 Ohm load? It will do whatever it has to do to force 1 Amp through the load. In this case, it will develop 2 Volts of output. 1 Amp * 2 Ohms = 2 Volts. An ideal current source could develop 1kV across a 1k resistor, simply because you told it to deliver 1A, no matter what. In the real world, of course, it’s no so easy to build a current source with that wide a range, but the principle remains constant, so to speak. In each case, the current source could be replaced by a simple resistor. In fact, resistors sometimes sound better. My impression is that this is because current sources are imperfect. More particularly, they are notably inconsistent about compliance (that’s how well a current source bounces along with the music) at different frequencies. Walt Jung did a little research on this and you can read the resulting paper on his website. He doesn’t address the frequency-related problems directly, but they’re clear to see in his graphs. That said, I elected to go ahead and use current sources because my design goal was to use up scads of J271s. Hey, they’re paid for... why not?
So let’s run through the circuit again, addressing a few of the details that got glossed over previously. The circuit includes a volume control. If your CD player, or whatever you use for a source, happens to have an output level control, feel free to replace this with a 10k fixed resistor. The differential is completely normal. The variable resistor, V2, is there to compensate for real world differences in Idss that you’ll see when you start setting up Q2 as a current source. In practice, you set V2 for 0Vdc offset at the output, let the circuit run for thirty minutes or so, repeat, and you’re done.
The voltage amplification stage could serve as a jumping-off point for any number of topics. Q4 serves as a high impedance load for Q5. This yields more gain, while at the same time allowing the stage to both push and pull current in the process of driving the output stage. C1 is there to forestall oscillations, and in this case I recommend that you leave the cap in the circuit. Without it, the circuit (in typical opamp fashion) is quite capable of generating its own RF nasties. In passing, I’ll note that the node where the collector of Q5 meets the Drain of Q4 provides an opportunity to experiment if you happen to feel that large quantities of feedback aren’t necessarily the ultimate audio answer. By placing a resistor from this node to ground (designated R* on the schematic), you can decrease the open loop gain, which in turn reduces the amount of negative feedback. I’d suggest starting with values on the order of 1k to 2.21k. It also provides an option should you find that you need to decrease the overall gain of the circuit.
From there, it’s on to the output stage. Q7, Q9, Q11, Q13, and Q15 are simple Source followers (more formally known as the common Drain configuration). Note that there are no Gate stopper resistors. If you want to put some in, feel free to do so, but the Gate capacitance of the J271 is so low that the circuit’s pretty stable the way it is. The followers work against the current sources along the top, providing push-pull class A output. Simple.
I mentioned using J310s along with J271s earlier, and this is an interesting place to do so. It’s a trivial matter to replace the J271 current sources with J310s (arranged just like the J271s, but upside-down, meaning that the J310 Sources [and Source resistors] are pointed towards the output line, and their Drains towards the positive rail) and have a complementary single-ended, push-pull class A output. Cool, eh? But I’m trying to use up J271s here, so I’ll stick to lemonade for the moment. In passing I should note that the J310 has a slightly lower voltage rating, so you might want to consider lowering the rails to +10V instead of +12V.
The feedback loop is a simple two resistor affair. R16 and R17 serve as a voltage divider that sends a small sample of the output back to the differential as negative feedback. By some peoples’ standards, the amount of feedback used is rather modest. If you want to increase the amount of feedback or reduce the gain, increase the value of R17. The capacitor in parallel with R16 limits the frequency response of the circuit as a whole. If you lead a charmed life (like me), or if you simply happen to live in an area where there’s little or no RF, feel free to delete C2 or use a smaller value.
As far as parts matching goes, you should match the Idss of Q1 to that of Q3. The pair do not have to be matched to anything else. Ideally, the output follower devices (Q7, Q9, Q11, Q13, and Q15) should be matched to each other and the output current sources (Q6, Q8, Q10, Q12, and Q14) should be matched to each other. Alternatively, you could match the followers and use different Source resistor values to trim the current sources along the top to about 8-10mA each. In any event, you will need to use higher resistor values for R3, R4, R6, R8, R10, R12, and R14 if you find that your JFETs run much above 10mA, as the J271 is only rated for 350mW, and I recommend staying below 150mW in actual use. Q2, Q4, and Q5 don’t need to be matched to anything. If you’re hazy about how to match JFETs, check the SSTART article, but reverse positive for negative since the J271 is a P-ch part instead of N-ch.
(You should be able to easily better the distortion specification — see disclaimer in the Difference Engine write-up.)
The overall cost of the project is low and the parts are not particularly hard to find. You should buy extra JFETs for matching, but they’re not particularly expensive and any leftovers can be used for other projects as they really are pretty nice JFETs. Come on... P-amper yourself.