Just the other day the fantastic work by Sijosae caught my attention. There are some threads here on Headwize (search for "multi-hybrid" but I'm not sure if his latest is posted here. I found it on the Korean site www.headphoneamp.co.kr,
here. Long story short he has a similar circuit running on a 24V supply. Transistor CCS on both the plate load of the tube and source of the FET. Runs the heater from the MOSFET source, so no additional supply required. Very efficient, simple, and impressive.

In his circuit, the tube plate voltage is about 16-17 V. The cathodes are about half a volt above the grids.

Simply put, is this a good idea? Is there a problem running the grids so low that the input signal could easily drive the grids positive? Have people come to any conclusions on what the real lower pracical limits on the plate voltage are before things go from hi-fi to lo-fi? Specialized tubes aside, just keeping the discussion to the common 6922 type triodes for now...

How low would you go?

/rjm

---

[Quoting rjm]

Simply put, is this a good idea? Is there a problem running the grids so low that the input signal could easily drive the grids positive? Have people come to any conclusions on what the real lower pracical limits on the plate voltage are before things go from hi-fi to lo-fi? Specialized tubes aside, just keeping the discussion to the common 6922 type triodes for now...

---

mu on these is high enough, and since gain = mu with a ccs plate load, that you rarely need to turn them up very high, so the grid is generally not driven positive. That said, if you did drive the grid positve it would sound bad.

As for low plate voltage, it increases second harmonic distortion. This works well to produce a "tube" sound, but I would think that this is inherently lo-fi.

---

[Quote]

frankly the power supply was going to be a pain

---

The PS need not be a pain. The CCS should have quite good ripple rejection and effectively keeps the PS out of the signal path, so a basic CRCRC or CLCRC will be sufficient.

How about leaving off the mosfet output and instead using some cheapo line matching 70V transformers in a parafeed circuit? You'll need the high voltage, and probably will want to parallel the tube for enough power, but it will make a quite nice circuit.

-d

F

I guess if you have a plate voltage of 23.7V on a 24V power supply, then you battery biased the anodes as Sijosae has it in some of the versions. Another neat trick I forgot to mention earlier.

Anyhow thanks for the comments. Plenty to think over.

/rjm

What, 130V is "high"?

> a similar circuit running on a 24V supply

Old news. Look on this forum for YAHA, maybe SOHA also.

> the tube plate voltage is about 16-17 V. The cathodes are about half a volt above the grids. ...is this a good idea? Is there a problem running the grids so low that the input signal could easily drive the grids positive?

YOU choose to "drive the grid positive", or not, with your hand on the volume control.

The only question is: can you get useful output from a common tube? (I would not call 6922 "common"; but 12AU7 will do much the same lo-volt work.)

ALL common triode lines converge to zero. They all "work" at low voltage.

A triode may be approximated as a voltage variable resistance, with the minimum resistance rather high, several K ohms. This means the power available out of the tube declines as square of voltage. 1/2 the voltage, 1/4 the power.

Even a "voltage amplifier" must deliver some power to the load.

The MOSFET needs essentially zero power to hold a DC voltage, but has substantial Gate capacitance. Your IRF610 is good: as a follower, with strays, you may be under 100pFd. At the top of the audio band, that's maybe 100K impedance. I think you underestimate the power which might be wanted in headphones, but 2V peak in 100K is 0.02mA peak current, 0.04mW peak audio power.

A further kink: most tubes double in plate resistance below ~50V. That 6922 is 2.2K at high voltage and current but 5K at low voltage.

Still we can probably be flowing 50mW in tube and DC load, so a 0.04W peak audio output should not be impossible.

I've just seen that the Twilight is bootstrapped, and does not use the MOSFET as a follower (although it may appear that way). Interesting variant; I'd just wire the triode as a straight voltage amplifier, with DC plate resistor considerably higher than the (low-volt) plate resistance but considerably smaller than the load impedance (essentially gate capacitance).

> How low would you go?

Until you can't make enough power.

Rule O Thumb: triode voltage amplifier can make a peak signal about 20% of supply voltage. THD will be near 5%; you usually don't want to go this far. You can do somewhat better, but not easily (and you are already overwhelmed).

Your number for HD600 implies ~2.5V peak. Assume follower has gain so close to unity that we can write "1". You need a supply voltage of 2.5V/20%= 2.5/0.2= 12.5V.

Gate capacitance, for IRF610 as true Follower, approximates 100K impedance at the top of the audio band. Rule O Thumb, pick previous plate resistor 3 times lower, 33K. Pick tube with plate resistance 3 times lower than that, 10K. Shop for tube.

10K will be either a very fat tube or a low-Mu tube. Low-Mu is OK: we need output of 2.5V peak, most any source has 0.3Vpk, gain of 8 or 10 is ample.

Rejecting the exotic 6922 (it'll work), I grab the 12AU7 out of my cheap VTVM. It will pass 1mA at 13V, or about 13K plate impedance. Close to our 10K goal. If we actually had a fixed load impedance, we'd then multiply plate resistance times load impedance and take the square-root: SqRt(13K*100K)= 36K plate resistOR. But the "100K" will be much higher over most of the audio band; we don't want full power up there. And I have a string of 33K resistors.

Squinting a 12AU7 graph, I round-off to 15V and 30K to avoid eye strain.

We see that a mere 12AU7, fed 15V through 30K (Red load-line), will swing nearly 10V peak-to-peak, and might only need 1Vpp to do so.

All these numbers are approximate. The graphs are crude at normal current and dubious at low current. But we know from physical action that the tube won't quit, just get weak. So while it may be a bit less than 10Vpp, and may need a bit more than 1Vpp, it will give gain and signal. And just a bit less if you care to try 12V supply.

Since the tube will be running about 1/4mA, we could run two such stages from 15V, with a 3K filter resistor, and hardly lose any voltage. A 100uFd 25V cap would clobber all supply ripple, giving tube clean DC from nasty power supply. The MOSFET has near-infinite "plate" impedance so as a Follower it has huge supply rejection, its Drain can be fed nasty power. Now your power supply reduces to a 12VAC winding, a bridge, and say 100uFd 25V cap. The heaters, if "DC heated", will actually be a worse problem: they need cleaner DC because 120Hz ripple will radiate worse than 60Hz AC, worse than ripple leaking through tube filter and MOSFET Drain.

The MOSFET will support huge current. Assuming its Gate is direct coupled to tube plate, which sits near 10V, the Source will be near 6V. Put a 60 ohm 2W resistor Drain to ground, 100mA idle, it can swing 2Vpk and 60mApk or ~50mW into 32 ohms, or 4Vpk 25mW in 300 ohms.

It is even possible to cleverly contrive things so the MOSFET source load is the tube heater. Nobody ever believes it will work, but many Fishers and GEs fed their preamp tube heater from a quad of 6L6 power output tubes.

[Edited by PRR on 02-24-2007 at 01:32 AM.]

You can't trust SPICE any further than you can throw the punch-card reader it was written for. It will give bad answers with 9-place precision. The common models for vacuum tubes are gloss approximations outside "normal" operating points. I've salted my SPICE with some experience.... I don't say this is right, but I doubt it is very wrong.

For 15V supply, 12AU7, IRF150 (a larger MOSFET), 0.5V peak input (a hair into grid current), I get 2.9Vpk out of the FET into a 100R load at 10% THD. That is 9.7% 2nd harmonic, 2% 3rd, 0.9% 4th, 0.3% 5th... a nice drop. And that is about 40mW in 100 ohms.

At half that input voltage I get half the output voltage (10mW) at about half THD, 6%.

It isn't "uncolored". It may be fun. We are not talking the nasty 0.3% distortion of an over-complicated transistor amp, but mostly sweet tooby bend.

Huh. Biasing the tube a bit cooler reduces gain to 6 but gets THD at 2Vrms output down near 4%. I don't trust the exact voltage shown; you'd get here by playing Real Loud, and fiddling the cathode resistor 2K to 10K for maximum unclipped output.

The 33K grid resistor is just to confirm low grid current. In fact grid current won't hurt anything (or not for the currents that can flow to 12AU7 grid from Hi-Fi sources). In fact if you have a low-Z source, working Class A2 is quite valid. And for the first tenth of a volt positive, 12AU7 grid will look like over 5K, no big strain for hi-fi outputs.

Cathode cap could be smaller, output cap should be larger.

It plays fairly loud into lo-Z headphones and gives
comfortable levels when fed into speakers.

I realise that as the plate voltage is lowered the tube rather than the MOSFET comes to define output power; either because the grid voltage is too low to support the input signal without going positive, or because the output voltage swing on the plate generates too much distortion.

Provided the output power is sufficient to drive the headphones there's nothing wrong with this. From a design perspective though I consider it bad practice. In good design the amplifier output power should be limited by the output stage not the voltage amplifier stage. Actually I would prefer to have 3 dB of headroom on the tube stage compared to the MOSFET output.

Design example:

With this design "rule", and working with a 300 ohm headphone load, I first arbitrarily set the max power output at 100 mW. That's 5.5 V, or 15.5 V p-p into the load and 18 mA. I can configure the MOSFET stage to give those specifications.

Turning to the tube stage, 3 dB headroom of 5.5 V is 7.75 V. With the "5x" rule of thumb for 5% distortion, the minimum plate voltage is about 40 V. The grid voltage will depend on the tube and the plate current, but at 40 V it is going to be approximately 1.7 V. Thus as long as the tube gain is above 3x the grids will not be driven positive.

40 V on the plate also puts 40 V on the source resistor. This is still manageable: for 100 mW R_source should be 1.2 kohms, dissipating 1.25 Watts. Its not the most efficient operating point, but that is hardly a concern.

For low impedance headphones or if less output power is desired a lower plate voltage is appropriate. For 300 ohms and 100 mW, however, it looks like 40 V plate voltage is a reasonable lower limit.

/rjm

What is "good"?

I too admire elegance, where everything is exactly right and knits together well.

The problem here is that a volt-amp triode's peak output is (more or less) 20% of supply voltage, which a finely-tuned sand-state device can approach 49% of supply voltage (though maybe 33% with resistor-coupled power stages).

This means the class A output stage will dissipate 2 to 3 times the heat that it "needs to".

One "fix" is separate supply voltages. But for an amp with just two active devices, adding another supply rail is a significant complication. And you've already stated "frankly the power supply was going to be a pain. ...have yet to start building." I know the feeling. It is much easier to draw something than to put it together.

I also hate to waste power. But here the added power supply is at least $10 of transformer and caps. OTOH the cost of 4 Watts of "excess heat" is, at my electric rate, $0.0006/Hour. For the first 16,000 hours it is cheaper to waste the 4W than to buy the added power supply output. Your numbers will vary. Certainly battery power turns the equation upside down. And the planet is running out of juice, though we must note that making transformers and caps is energy intensive, and may well need more planet-energy than several thousand hours of 4W waste.

> I would prefer to have 3 dB of headroom

Look at the real problem. People listen over a 30+dB range, from 70dB SPL to >100dB SPL. That's a 1,000:1 range of electric power. There is no power spec which is exactly right. If I soberly compute "56mW", and then get drunk and put on my Neil Young at max-blast, 400mW may not be enough. Design for "overkill within budget", and don't fret a few dBs.

> I would prefer to have 3 dB of headroom on the tube stage compared to the MOSFET output.

Someone sang: You can't always get what you want.

In particular: FET power is cheap, tube power is expensive, even at this microWatt level. It is no big sin to have a "200mW" FET stage which the tube can only push to 100mW, as long as 100mW covers your needs. The FET stage is hardly any cheaper if "crippled" to 100mW. (This is the big fun of headphones: we can do things which are "uneconomic" and get away with it because the excess cost is pennies. I sure would not build a 200W output stage wiht a driver which only pushed it to 100W, that's real money.)

Don't get too hung-up on elegance. If it meets spec: power, distortion, cost, labor, weight, heat; you done good. Over-complicated is the enemy of "good".

3dB "headroom" on a simple triode volt-amp just means THD drops from maybe 4% to 2.8%. Simple triode volt-amp THD falls roughly as Voltage. If you are looking for point-oh numbers, you need very high voltage or heavy NFB. OTOH, 4% triode THD is not grossly distorted, often enjoyable. And you won't work AT 100mW all the time, probably much lower unless you are in "a mood".

> Design example: ...arbitrarily set the max power output at 100 mW.

That's ten times the power stated on your web page. I agree that 10mW may be inadequate in some situations, but moving targets make design difficult.

As an easy guess: 10X the power means 3.16X the voltage, and indeed my 15V times 3 is your 40V, within tolerable error.

My last MOSFET buffer gave 50 mW into 300 ohms, and I never felt it was remotely close to running out of juice. But this time I figures, 100 milliwatts and my eyes lit up like cathode heaters ... why not dream big!!

But I see what you're saying: Let the tube define the maximum output voltage, and overbuild the MOSFET stage cover that completely. The output characteristics are then just the distortion of the the tube.

Or you could do the opposite. Set the MOSFET output power and insist that the tube stage remains below a certain level of distortion up to and including the MOSFET clipping.

The former is for people who want a fairly heavy tube sound available when the amp is pushed. The latter is for pencil-necks like myself who want a clean sound up to the specified maximum power.

If I really insisted on 1% distortion at 100 mW, the plate voltage would have to be 90 V. That's doable within my self-imposed thermal/power budget of 10 W. I could back off and go with 2% distortion at 100 mW, in which case 40 V on the plate would be sufficient. Power draw / heat dissipation drops by about 2x, power supply is a little easier to source ... but there isn't much there to force a decision either way.

(Aside: for a 32 ohm load the plate voltage of 15-20 V works just fine with low distortion since the voltages are so much lower.)

I guess that must be why I find starting these headphone amp projects so difficult. The thermal and power issues are not significant enough to contrain the design. Nor really is the power supply, or, honestly, the complexity of the circuit. Anything goes, yet perversely "anything" does not an amp make.

You say it much better than I did.

> 1% distortion at 100 mW, the plate voltage would have to be 90 V.

Sounds right.

It is possible your bootstrapped variant will do better than a simple resistor-loaded voltage amp.

The plate will see about 50 volts. That's about 40 mW at 1% distortion. The MOSFET stage clips at about 80 mW.

See the attached for the updated circuit diagram.

Anyway, its almost done now. Main amplifier unit fully assembled. Power supply 90% assembled. A couple of trivial hardware elements to trac down and install tommorrow. It's 2am now though, so time to call it a day.

Some data:

B+ : 84 V simulated, 87 V measured
plate-cathode voltage : 49 V simulated, 52 V measured
cathode votage : ~2 V estimated from plate curves, 2.0 V measured
filament : the transformer winding is specified at 6.3 V at 1 A, I measured 6.77 V at the 6CG7. Its a tad high, not a big deal but I may add a little resistance in series with the heater to drop it down a shade.

output hum and noise : not audible, below measurement threshold of my DVM. (AC filaments? You wouldn't beleive it!) This was the big unknown going into the build phase, and I am very happy to report its not an issue.

gain : 10x voltage gain predicted, I havent measured it but it seems about that. The volume control is in the right position, anyway. (8-10,11 o'clock when listening.)

heat : 5 W simulated dissipation combined from the FETs and Rsource. The case/heatsink is about 10 degrees above ambient. Just warm to the touch.

Ok, that's all for now. Got it looping through music to run it in for a few hours. Stuffed with Black Gates as it is, there is no sense saying anything about how it sounds for a couple of days at least.

Yes, when I get the chance, I'll show you the insides, I promise.

and That:

Not exactly professional workmanship, but it gets the job done.

The large R core transformer, by the way, is 80VA size and custom made by Pheonix (RA transformers), 2x 60V 0.4A secondaries, one for each channel, and a center-tapped 6.3 V 1A filament winding. There is also an electrostatic screen.

For the rest: Black Gate electrolytics, Multicap RTX, Holco and Riken resistors, and RCA 6CG7.

let us know how it sounds when things have settled in. Did you change anything from the previously posted schematic and layout files? I am guessing not. congrats again..dB

Perhaps because the BG Std types I'm using are all under strong DC bias, the break in process was much faster than with the non-polarized N type, and the transition is very abrupt. One channel "popped" a few hours before the other, so for a while there it was like having cotton wool in one ear. Then the other channel settled and all was as it should be.

Bad news: I tried the Electro-Harmonix 6CG7 and it hummed something awful. Its possible I have a defective tube but I suspect instead its because the EH tube has a cheaper filament design that does not balance out the hum on a symmetrical AC waveform. Back to the RCA. (black plate clear top)

Currenly my reference source is my computer, .wav files ripped from CD, played through an Onkyo SE-90PCI sound card. The sound card has 24bit/192kHz DACs and an S/N of 110dB.

I compared the Twilight amp to my Szekeres VE.

The Szekeres VE is a powerful amp, neutral in tone with good stereo separation and strong bass. It is utterly demolished by the Twilight.

The Twilight is so far superior to anything I have ever built before I am at a loss for words to describe it.

Sorry for the tease, but I'll be fair and give it a couple of days to come to grips with. Anyone in Kyoto this weekend? Feel free to drop by I could use a second opinion.