I've done several other versions of headphone amps and wanted to do a simpler version. In summary it is a differential tube input driving a Mosfet follower output section in class A with the output pull down current derived from the tube heater current. One could argue that the current is non-linear which would be true for large voltage swings. For the minimal output voltage swing it is not a concern.
The output is feedback to the neg input of the diff pair. The heater current of the 12AU7 is around 160mA and if you use a 12BH7 it is raised to about 300mA. With a 24V wallwart PS, there is 12 to 18 volts across the tubes for a little more class A swing, and the tubes enjoy a little more potential.
I completed stuffing it a couple nights ago and have been enjoying it every free minute. It can drive 24 ohm cans with ease and black hole quiet with no input. ( almost DC on the heaters with a little audio ripple)
Only 2 adjustments, current through the diff pair, set at 6mA, 3 each tube section and an adjustment for adjusting the heater voltage on the tubes, 12.7 volts.
It turned out to be a simple toner transfer PCB, easy to stuff and adjust. Very pleased by the dynamic range, lower end should be about -.36db at 20 hz with 24 ohm cans. I had 3,900u electo's and .47u mylar caps to use on the output coupling.
Its design is a combination from the days at National Semi in the linear group and work done by Millet on his starving student Head Phone Amp. For those purists who whine about why a hybrid, try it and then make your comments! A lot of the components can be had at the local Rat-Shack, you'll have to scrounge a few components, but they are easy to find.

ps: Last pic is a Compactron Mosfet Hybrid amp with tone section, increased complexity especially in power supply.


  • head_amp_10.jpg
    340.7 KB · Views: 1,008
  • headamp-12bh7A.jpg
    148.4 KB · Views: 899
  • P3140191.jpg
    881.6 KB · Views: 886
  • glass-sand.jpg
    458.9 KB · Views: 941
  • plan2b-4.jpg
    177.6 KB · Views: 896
You can use any small signal NPN and PNP's like 2N3904, or 2N2222 and 2N3906 or 2907. At low current and low voltage levels most will work, just pick a decent Beta. I did no matching with them to try to duplicate what everyone else could do. The PNP mirror would be the most important to match if one were to take the time. Emitter resistors could be used to desensitize mismatch, say 100 ohms. The MosFet is the IRF510, 5amp and 100 V with about a 3volt Vgs. Any N-Mosfet with low Vgs should work, again low current and low voltage. I wanted to use cheap readily available product.
I've been happy with Baldwin branded 12AU7's, not sure who really made them.
Tubes, sockets, heatsink and PCB should be the only thing needed out side of a rat shack visit, They do not have the same pots, but they do cary some. A darlington NPN can be used for the output, but they are prone to be unstable, may require a little cap to tame them if used. I chose components out of what I had on hand and a trip to the shack....


  • head_amp_5b.jpg
    175 KB · Views: 335

JPG of the PCB attached....
I use ground stitching to minimize ground loop potential and use insulated wire on any non-ground jumpers. I'm using a 24V wall-wart power supply so no power supply on the PCB except for 1k voltage dividers across the power supply caps. This creates the V/2 ground used as signal ground and the ground plane. Another option is to add 12V zeners on PCB across the caps for a more solid reference, or go external PS.
The indicator LED's are bright on first fire-up as the heaters are still at their low impedance. as they warm up the LED's almost go out. With the 12BH7 tubes they stay on with the sense resistor chosen.
I cannot say that this amp has "tube coloration", it is pleasant to listen to without any fatigue, you just don't want to put it down.
If others choose to try it, be sure to add your experience to the link. The design concept is open to modification to tone section etc.... I just chose this as a simple low cost concept project. From my searching I found Millet's Starving Student Hybrid that is similar in double using the heaters in the signal path.


  • glass-sand_PCB.jpg
    177.2 KB · Views: 235
I would guess 20V on the heater string until V1B's plate begins to conduct?
Just eyeballin' the schematic, not doing any valid actual math or anything...

Does that 100K cold pull up also want for a 220K cold pull down?
Once V1A side of mirror sees current, cold bias resistors won't matter.
DC point set by the loop takes over as both plate currents increase...
Last edited:
I havn't fully thought this through. Just noodling.

Elevated heaters.
Filament bias same, warm or cold.
Little extra (sandy 4x) open loop gain.
Greatly increased output swing.

Don't take these claims too seriously.
Its all very half-baked, maybe less.


  • LameIdea.gif
    21.1 KB · Views: 347
Last edited:
I took the start up 100k and just ran it to ground instead of V+. It takes a little longer to cold start but that is the desired effect. That should be a little kinder to the tubes.
Keep the ideas coming! No apparent change to the sound when up and running...


  • glass-sand_PCB_2.jpg
    211.1 KB · Views: 316
Last edited:
I can believe an output swing all the way up to 12V.

But downward, the plate can't go any lower than the
cathodes. And I hesitate to believe even with a VGS
drop, that source could ever dip much below ground.

Take you a look at the topology of suggestion #10.
You can see a little sand gain here means the plate
swing stays above the cathodes and within reason.
Yet the output can now travel the full span of both

No change to your current source and mirror, except
to flip the mirror around. Was just drawing shorthand...
Last edited:
Half the plate resistance would be seen by the gate (6K5/2).
So maybe I've added more sandy gain than I was thinking.
Perhaps reduce the 270K and 100K, to 27K and 10K???

Now 3K25 combined plate resistance to 27K sets the ratio.
About 8.3x gain added. The plate is never asked to dive
below the cathode at full swing would be the advantage.

10K pull down reduces feeds forward and back equally,
so has no effect upon gain ratio, only DC filament bias.

A 3K3 gate stopper (not shown in #10) would cut the
sandy gain boost in half ( to the original intended 4x).
And the plates would be able to swing some voltage,
which I think is necessary for Mu to work its magic.

Without this stopper, the triodes plates might be held
too closely to gate threshold, and in the less desirable
gm gain mode. I didn't see that when I drew #10...
Anyways, easily fixed...

Have I done anything stOOpid that would make the loop
unstable? If anything, thats where I usually go wrong...
Last edited:
12.6 volts across the heater:
The cathodes are sitting at -9.3 volts
the +gate at -9V, -gate is at -7.9V
Plate V to Mosfet is at 4.6 volts, output at .75 so ~4volts Vgs
Taking the input +, the max output goes to 7.8 volts limit of the Mosfet Vgs
Taking the input -, the max output goes to -11.7volts, heater goes cold if DC and it stays there
AC coupled --So the max output swing is 19.5 volts PP on the wall-wart supply , just short of 24V.... 6.9Vrms into 20ohm cans=2.4 watts rms best case?? Again , just a headphone amp.........
With the heaters being the output current sink, it is not designed to swing rail to rail and be linear. The heaters can be replaced by a current sink if that is the desire, and the output current could be cranked up to drive an 8 ohm speaker. ( I did that with my other design)
I have to admit, this is a very unconventional design. I was pleasantly surprised that the idea worked at all!
How can those ground ref'd grids measure so negative?
And yet still slightly positive vs the cathodes...

Are forward grid currents pumping the grid bias down?
I guess with such low plate voltage, stray electrons have
to go someplace...

However you done it, if you measured as you say, its fine.
Last edited:
I searched and found a graph on grid voltage vs plate current. At low voltages, the grid can be positive 4 to 6 volts which is so strange. I've never tried using a tube at such low voltage before. It is like you have to go positive to coax the electrons to leave the cathode where at high voltage you have to do everything to contain them, they want to leave real bad.
I read something about the real old car radio low voltage space charge tubes that had 2 grids, one was positive to coax the electrons to leave and then the 2nd grid controlled the flow.
I thought of another mod, a pot in the cathodes for offset adjustment, a simple change to try.
Sometimes reality is stranger than fiction.........


  • graph 12AU7_low voltage.png
    graph 12AU7_low voltage.png
    242.4 KB · Views: 263
Last edited:
If we accept big leaky grid currents, whats the point of a rigid
current source below? Would a simple resistor do just as well?

If you add a little resistor under the filaments, such that it sees
a voltage swing similar to (but slightly less than) the input.
Then let the lower end of the tail resistor connect to this active
node, instead of directly to the negative rail.

Its kinda bootstappish. I dunno if this makes the tail to be a
genuine higher impedance, or just a feedback problem.

If so, then 6 components could maybe be replaced by just 2.
Last edited:
Like R10 in this old drawing of mine, same principal could be applied?

R10 above my bootstrap, but in your case would be below filaments.
Size to be slightly less than than 1/Gain, to stiffen the tail resistor
without going overboard and becoming a positive feedback. I drew
mine at exactly 1/Gain which is the threshold of becoming a problem.
If you try, I wouldn't advise cutting it quite so close.

Then maybe you don't need 6 parts active current source thats just
going to leak out the grids anyway...

You do seem one to forgive an imperfection in pursuit of minimalism.


  • SRJLH_Simple.gif
    42.2 KB · Views: 264
Last edited:
The current sink was used to check the sensitivity to plate current to sound quality. I backed it down to 100uA and it still sounded good and then raised it up to 6mA and still sounded good. I have it set a 3mA. Original thought was 2mA/side 4ma total tail current.
If I could set the plate current by cathode to grid voltage your idea cold work. What I'm concerned about is the variation found in the tubes when subjected to this low voltage operating point. They are less predictable that when used at high voltage.