Using the THAT1646 As A Headphone Amp - deleted

That sounds like pretty low distortion. From observation those output devices look like they are outside any corrective negative feedback loop. 10V p-p into 68 ohms is not exactly idle current.

What is the class A current (V drop across 3.3 ohm) ?

I typically keep the output devices inside the feedback loop for such circuits. If you are getting -100dB without NFB, could it be better?


JR

I don't like to think about the inside of the THAT1646.

And this looks like reckless disregard of fine engineering.

But if I grok what you done.... change the diode to 5V, change the 3R3 to 54R, lose the 27R build-out, put heatsinks on. Idle current will be 5V/54R= 109mA, Z(out) will be 27 ohms, it will dump nearly 6V peak in 32 ohms or 0.5 Watts which is LOUD, and 12V peak in 300 ohms or 0.25W which is also plenty. Pure class A. The intrinsic emitter resistance wanders from 0.3 ohms down to 0.15 ohms and up high, but push-pull cancels most of this, and nonlinearity could be 1 in 1,000 or lower.

This also changes the risk that some fool (or you and your 'scope probe) will get ahead of a build-out resistor, short the main output node to ground, and put up to 3 Amps and 9 Watts in those transistors.

the 1646 "phase splitter" has way more output capability than you need here. why not use that juice to drive an output stage that really needs it? like some big 'ol mosfets with max output of say, 100W or so.

mike

[quote author="mediatechnology"]

And this looks like reckless disregard of fine engineering.

Click to expand...

Would you care to try again Oh Great One?[/quote]
Hmmm... I may be mistaken, but I read that as humour from PRR. -I don't think that it was real criticism, though you do have as much idea as I do.

Certainly, if you're wondering why you were pulled over by the design police, your left-rear bias voltage looks a little low, and your output was weaving a little. -Have you been drinking, sir???

I'll let it slip this time sir, but in future, kindly stick to more conventional appraches for this sort of thing: -otherwise, next thing you know, you'll be facing charges of designing M/S matrixes with no resistors, and the like... (tut!).

By the way, reckless disregard of fine engineering carries a maximum penalty of fabricating your own transistors using a bucket of sand and some bits of plastic... I should be more careful in future if I were you...

Mind how you go now, won't you sir?

Evenin' all.

Keef

Skins be way too thin around here lately :roll:

Actually, I'm perversely delighted by the unconventionality of it all! :thumb:

...but you'd better get that intermittent long-tail pair taken care of, before you get a ticket. :wink:

Tsk. That's not being reckless... That's just plain anarchic! :wink:

Call it punditry if you like. I've been scratching my head for this circuit's raison d’être.

One thing that is nice I guess, and no doubt helps it's performance, is driving both bases from low impedances, with a fixed delta V. I imagine you could lower the distortion even more (for a known resistive load), by adding a full wave rectified signal term into that base delta to first order correct for load current, but with the distortion numbers you report, I see little reason to bother with that kind of complexity.

On the topic of unneeded complexity, using two opamps with decent drive capability, you could just about combine to two opamp outputs together, using appropriate degeneration/build out resistors, and lose the transistors entirely. There was some discussion elsewhere of a 27 ohm source impedance being desirable (easily accomplished with two 56 ohms build outs), but it isn't clear (to me) what the intent of that circuit is.

You could probably trick that THAT chip to do dat (too many "that"s in that sentence), if it's indeed intended for modest impedance loads.

JR

Reckless does not immediately come to mind as a fair description of a headphone amp that throws an opamp at DC voltage management, but intentionally avoids NFB in the audio band.

I'd call that an effects unit... :roll:

JR

> if the punditry hates something I'm usually on to something.

A good guide to life.

What I meant was: THAT's designer worked hard doing one thing, well, and you saw a way to do something VERY different with the same part. Many of the details that THAT slaved over don't matter to you; one thing you might like and THAT cheated you on was external access to feedback (which in their intended market would just cause trouble).

Recklessness can lead to serendipity.

> There wasn't much difference other than twice the supply current.

Sorry. I ran away with the simplicity of changing bias this way, and the synergy of building the build-out into the bias. I already knew (but didn't think) that Class A is no cleaner than a good dry AB stage. If you don't need to plaster "Pure Class A!!!" on your sales-lit, I think you can usually/always build a better amp in near-B than in True-A.

But now we know.

> I've been scratching my head for this circuit's raison d’être.

There are darn few GOOD headphone amps as simple as this one. Before the ADSL chips, nothing with 100+mA of drive had great audio performance. You have opamp+buffer, and massively parallel, and HOT class-A SE. I myself have recklessly built too-simple too-hot amps and too-too-complicated amps. I doubt they work better, and probably worse, than the subject amp.

I do miss my old push-pull tuner-triode headphone amp. Military servo-amp iron ("servo" used to mean AC) with a lot more bandwidth than was stamped on the bomb-proof cans. 3U rack cabinet. Real reckless excess and misuse of military material.

> Hell, I'm being reckless just by not using a tube.

Yeah! Subversive chip-lover!!! You probably use LEDs for pilot lights instead of a vintage #47 filament lamp. (Says the owner/rebuilder of the world's only all-chip 1937 Allen B. DuMont oscilloscope.....)

> intentionally avoids NFB in the audio band.

The Kumisa III is a feedback amp. The input device is degenerated about 80:1. The gain-mirror is degenerated about 180:1.

It is a high local feedback VCVS dumping into an assumed perfect resistor. The resistor voltage is not sensed, but if all is well it can't deviate from I*R.

The Darlington emitter follower buffers are feedback amps of course.

It also runs super-rich. Q7 Q8 run 4mA and have peak signal near 0.01mA. That's what keeps the I*R=perfection assumption reasonably valid.

> throws an opamp at DC voltage management

I too am an old-fashioned guy who thinks that an audio amp must first find and hold a DC level. And I'm inclined to see a DC servo as a rock under a saggy porch: the designer didn't put in enough support. (Full disclosure: my front porch is all stray rock support. Please don't line-dance.) But fooling with otherwise good audio amps, sometimes a good audio plan is more or less weak on DC. Especially if the goal is "capacitorless". And the annoying ~20mV offset between NPN and PNP. So if you can't naturally get DC below a few mV, why not let it go to Volts and wipe it away with a 19 cent chip optimized for DC perfection?

I bet the Kumisa is really clean. And the local NFB is not prone to recirculate IM as bad as other plans.

> That's just plain anarchic!

A union wants to represent us tech-folk where I work. CWA, appropriate but 'yawn'. If I'm gonna give 1.15%, I'd rather have someone "fun". Like the Wobblies (IWW). They are still active, goodness knows why or how. They are not true Anarchists, but their survival proves that anarchy rules the universe.

> powdered Boric acid.

I apologize for recklessly putting Borax up your shorts. It was uncalled for.

Once again I wasn't adequately precise. I meant overall NF. I am a fan of lots of local NF but advocate closing the complete loop. Customers have a bad habit of connecting less than well behaved loads to gear, and I prefer a consistent output sound, despite their attempts to make it "special" with funny wire and such.

I have yet to hear the horrors of well executed NF, but have heard much benefit.

Headphone amps, especially for 32-600 ohm cans, is not exactly heavy lifting. Is someone suggesting an improvement in accuracy from eliminating overall NF? I am aware of a small amount of phase shift caused by that NF (due to integrator compensation pole), but this phase error is tiny when adequate loop gain margin is provided.

YMMV

JR

PS:

Well if a 741 opamp is only comfortable driving 2k, 44 would only handle something like 45 ohms... Perhaps they should have used 1458s.

I have actually doubled up on sections of TL074 before to get a little more drive capability, rather than bring in a special part for one circuit node. (design was for a small company who only had one opamp type in the product)

--------

I used a pretty unusual discrete opamp buffer in an old phono preamp I did back in '80s. I used a NPN and PNP common collector, but instead of biasing the bases above and below zero, I tied both bases together and to the opamp output, then used a resistor divider from the two emitters to get a final output down around zero volts.

This was not very efficient, as it needed quite a bit of class a current (resistors from emitters to +/- supplies) to support a low output Z, but this shared your circuits benefit of driving the bases from a very low impedance (but only one opamp). Since I used global feedback (and a DC servo in that particular design) I wasn't much concerned about npn vs pnp voltage drops.

This was only intended for light, line level loads (50 ohm source impedance, 10k nominal load). For heavier loads perhaps another pair of devices (PNP followed by NPN, etc), which would get back to nominal 0v out. (This has probably been done before, while a little parts intensive. Essentially replacing the bias diodes, with emitter followers in typical buffer designs)

JR

[quote author="mediatechnology"] I don't have a convenient way to bring it back into the 1646 [/quote]

the 1606 version has the -input, seems to me like it is BEGGING for feedback.

It is only available in smd, I know, but so what? its pointless to fight it at this point. these days you can get a hot-air rework station for $150. IMO prototyping SMD is easy. you dont have to drill as many holes.

mikep

> use two diodes as the Vbe reference and divide it down to one Vbe plus a smidgen. ....It just burns my biscuits to see 18 mA or so being burned off...

I wuz wondering where all that current was going.

One "righter way" is to grab a third output device and a couple of resistors, a Vbe multiplier. The 3rd device WILL be the same Vbe as the outputs, +/- temperature and currrent. Unlike most diodes, it has a hole to nail it to the heatsink. I forget what your output current is, but if you want 60mA there and 1mA in the reference, at 60mV/decade, then you want almost 120 more mV on top of the 1mA Vbe. Take the base-emitter resistor much larger than 600mV/1mA or perhaps 5K. Put a 1K trimmer collector-base. You can get from 1.0 to 1.2 times the 1mA Vbe, which should bring the output devices to good heat without huge heat in the reference source.

Or you could stack three #47 lamps down from your +18V to a diode and get a large Vbe reference plus enuff light to read by.

> I'm not exactly sure why it takes 44! op amps either to make a stereo HP driver

I've used 8, plus a quad of 3055/2955 power devices. Pure Class A to 250mW and heavy NFB.

Figure a 5534 will pull 600 ohms well. 600/32= 18.75 chips, 37.5 for stereo, which rounds-up to 44 easy. (They may like me have a complex gain-control structure before a fix-gain power array; or balanced inputs or other frills.)

> something to the stiff high current voltage-source drive on the 1646 output that works to our advantage

There's a simple "proof" that a good BJT near-B stage should drop 28mV in emitter resistors to get first-order compensation for crossover distortion. Going low is nasty and going high brings up 3rd but reduces 5th, 7th, etc. Indeed stage-amps are often biased 20mV (cool is good) and hi-fi at 30mV-50mV.

(And THD numbers this way are lower than in Class A. The two devices are not fighting each other and driving-up the THD.)

That 30mV-50mV in Re is correct only if the drive impedance is ZERO.

For non-zero Z drive, I believe you divide source Z by Hfe and subtract that much from Re.

If you could have true hi-Z drive, Re seems to be zero. Indeed true current-drive would not invoke voltage-threshold crossover distortion. But what current driver has infinite slew?

But most such amps (and IIRC that 990 you were making fun of) take capacitive NFB from the drive point. In the top of the audio band, the source Z to the output devices may fall quite small. Since abusive overall NFB will kill bass crossover, you pick idle current and Re for the high end of the band, which is where drive impedance is typically low.

> I used a NPN and PNP common collector, .., I tied both bases together and to the opamp output, then used a resistor divider from the two emitters

A similar ploy was once popular as a video line driver. Aside from lo-Z base drive, I guess it was Class A (AB cutoff recovery time is a problem at 5MHz on 1970s parts), short-proof, had benign clipping, and a defined output impedance through the resistor string. This was +12V power and coupling caps, so DC offset was not an issue. The gross waste of power was orders of magnitude less than the tube amps of a generation before.

> 4V P-P driving some old Koss PRO-1As.

???? 10mW? You the guy who likes "big loud manly headphones driven really loud"????

You musta missed Hot Tuna in their over-loud days. 10mW is kiddie-level.

OK, I do work near 1mW-2mW most of the time, but my 100mW amp wasn't enough for live recording monitoing, why I built the 250mW. I can't clip it with Pro-4 phones and live, but some modern phones (the kind that don't sound like wider-range telephone earpieces) will clip 250mW before my eyeballs bulge. (Kids, don't try this at home. You'll end up like some of our recent students: obviously half-deafened by iPod addiction. And these are musicians!)

[quote author="PRR"]>

Take the base-emitter resistor much larger than 600mV/1mA or perhaps 5K. Put a 1K trimmer collector-base. You can get from 1.0 to 1.2 times the 1mA Vbe, which should bring the output devices to good heat without huge heat in the reference source.

[/quote]

Amp designers often put the trim pot, in Vb-e multipliers, in the b-e leg with a series limit resistor. That way when the trim pot fails open, the amp doesn't go to excessive bias and release it's magic smoke. I don't think I've even seen a trim-pot fail shorted, but end limit resistor makes the circuit "technician proof". This way trim-pot failures are cheaper to repair, a consideration in production design, although some large scale production approaches avoid trims entirely.

JR

I designed far more equipment that wasn't power amps, but find it a useful experience.

WRT headphones, it's been a few decades since I researched this but as I recall, there were some rat shack cans down at 3.2 ohm, and upper limit for all but real obscure Hifi stuff was 600 ohms.

The popular approach in headphone design is to use a reasonable build-out resistor so that 3.2 ohm cans will not get too much power or current limit circuitry, while 600 ohm will not drop too much voltage.

Designing a headphone amp for personal use, or a narrow market, is quite a bit easier to accommodate. If you want to get fancy for personal use, perhaps investigate what source impedance your headphone maker of choice designed for. If the driver in the headphone exhibits a non-flat load that will interact with any source impedance causing a response error, which they may have anticipated (or not). While not easy to play with open loop, using NF you could shape source impedance for perhaps flatter response in such a way that doesn't cost as much headroom or add noise like simple EQ.

JR

Like I said I haven't researched HP amps for quite a while..

I did one for a kit back in '78 (DJ Mixer) that used one of those cheapo single supply ICs (LM377?? from memory). Also from memory I think it advised a build out R. I would be nervous about making anything, including a power amp without some kind of load isolation.

Although I doubt anyone will miss 3.2 ohm cans, If I sold a HP amp today, I'd find the one customer who still has a pair of 3.2 ohm cans.

JR.

I have both the Sony's and the Beyer's, my preference is the DT-770's. They aren't as schreechy on the high end and middle.

I've always been happy with using one of my preamp cards as headphone amps, reasonable output power and voltage swing, very quiet and non-zero output Z to taste.

[quote author="mediatechnology"]
I got curious and wanted to check output stability with capacitive loading. So while driving 30 ohms through a 27 ohm build out at 4V P-P 100KHz square wave I installed a 0.1 uF from output to ground. This was ahead of the build out from the emitter resistors directly to ground. No oscillation, no ringing, just an RC exponential curve. Still darn square too. With the 0.1 uF on the right-hand side of the buildout there was obvious RC triangulation.

But stable. Very stable.

Update: I did try the BD139 and BD140 and I think I like them better. They worked identically DC and THD and with the 100 KHz/0.1uF torture test performed even smoother. The main thing is that they're isolated packages. Thanks Roger! I've got some slick heatsinks to drill with the BD1XX/LED combo.[/quote]

I'm not sure what you're testing with your capacitor torture test? If there is no overall feedback, the cap will just give you a LPF with the output impedance.

Am I missing something here?

JR

Not to demur, but you get that same benefit from a simple build-out resistor. Thus my previous advocacy for build-outs... Old school, but works.

JR