Beast Jr. with 42 JFETs

I wanted to write about this amp for a long time. Build it in 2021, finally feel like writing it now. Ideas are flowing in like a flood recently after I recovered from covid last month.

Here is the amp sitting on top of my Gungnir Multibit. Sitting next to it are Cinemag transformers and Noise Nuke. Mooncake box as enclosure is kinda a heritage meme in the ChiFi DIY niche.

Pop the hoods off:

 

Extremely accurately drawn schematic for each channel:

Basically it’s like Nelson Pass’s Beast with a Thousand JFETs, but with inverted gain, cap output, single ended, and for use with headphones. This single stage design pretty much fits the Shortest Way design philosophy except for the output cap. I used J112, which is the most bang for the buck JFET series in production. Parallel 21 of them makes it as quiet as 10x 2SK170.

With single stage design, I can’t achieve gain and uninverted phase without transformers. Thus I buckle a pair of phase inverters on my DAC, or use phase inverted headphones like HE6SE.

Note the output is taken from the 24V rail to reduce power supply noise. Combined with the Noise Nuke, this amp is so quiet that I can plug in sensitive IEMs, max the volume, and still hear no hiss. The caveat is that the output sits at 24V above ground, so it could cause problems when plugging in other devices into it. Thus I didn’t measure it directly.

 

Here's a pull out of my ass measurement I did with a single J112 in a similar setup on a breadboard. Can't remember the exact values of the setup and measurement, but the harmonic characters should be pretty similar.

There's only 2nd and 3rd harmonics, and pretty much nothing beyond that. I found it pretty interesting that I could drive my HE6SE with it, max the volume, and there's still no nasty distortions, albeit a lot of compression.

The actual amp will have much lower noise floor than this, due to the poor testing conditions and paralleling 21 JFETs.

 

I found this amp to have tube amp qualities. It sounds warm and natural, has some compressions depending on the load, and makes a lot of heat too. J112 actually measures closer to an ideal pentode than an ideal JFET. It has enough driving power for most of the headphones, but still needs some more for the hungry orthos like HE6SE. I don't find the output caps to be hazy, but I don't have any endgame amps to compare to. Panasonic FR and MKT1820 did a pretty good job.

To my ears, it pretty much sounds like a wire with gain…… plus much compressions. Its quietness is unmatched by the more complicated designs like ICs based ones, complicated feedback/feedforward, and Jotunheim 2. If I could address its sound stage compression, I believe it'll be a perfect amp.

 

There are many improvements I could make, diode bias, upgrade pot, or put it in a different box, etc. But I've learned so much from making it that if I had extra energy I would just build a different amp. I've been looking into a lot about I/V curves of transistors for this amp, and wanted to write about them for a long time too. It might happen after I finish my summary in the digital filter world.

 

Are you just using a generic wall wart power supply?

 

Yeah I'm using a Mean Well 24V wall adapter. Holding up very well. Had a lightning last month that killed serveral SMPS and LED bulbs, and it survivde and still sounded very good. Noise Nuke completly nuked its noise.

 

This is a really cool project. SE JFETs sound great, and hats off for not using up 42 rare and expensive JFETs when the J112 is probably the better part here.

IME the compression effect can have a lot to do with the op point. What's the drain voltage? There's some great info about tuning the distortion of the J112 in the Firstwatt H2 Injector paper.

One particular nugget of truth:

The source degeneration should afford you much greater linearity, but there might be merit to tweaking the resistor values if you're chasing lower compression.

 

Haha 2SK170 would probably feel very sad in that jancky box.

My operation point is around 14V on the drain, and 1.2V on the source. It's probably on the low side. I plan to up the rail voltage into 30V to give it a little more room to breath, but that's getting very close to the power limit, have to increase the source resistor or something.

I had some thoughts about biasing J112, but their power dissipation really sucks. I wish some manufacture could put 20x of them on a TO220 package to give more design freedom. Otherwise, I had to pick between output impedance, heat, distortion, and usable gain for headphones. Higher gain + higher rail voltage gives more linear output swing.

I did some curve tracing with J112, their I/V curve actually follows very well with power of 1.5, due to internal resistance. Tubier than tubes I guess Howerver, their drain impedacne really sucks, not really predictable and kind of low. That's actually the major benefits to use more expensive JFETs. BTW, J11x series actually has a pretty high smoke point Vds around 60-90V, but YMMV.

If I were to do it again, I would:
1. Keep the source resistor low, or even diode bias it, which would give the best H2 to H3 ratio.
2. Run it in balanced to annihilate H2. It should still leave some due to mismatch.
3. Use J113 to reduce current draw, and run it at higher rail voltage to linearize drain impedance.
4. Parallel more and a lot more of them to suffice power handlingand out put impedance, and keep me warm in winter.

Or... use 4x J113 as input stage, then buckle a push pull MOSFET output stage. I have found even more theories to make them sing


Here's the plot for a J113 biased with 1n4007 loaded with 1Kohm. Either the rail or the Vds is 32V, can't remember. I think the voltage swing was at 0dBu. It's just eye pleasing to look at.

I think it'll already sound great if I were to parallel 50x of them for singled ended each channel. Make 2x of them for balanced and bang out the H2 then it might truely be the wire with gain wihtout any global feedback.

 

What power output does this have?

 

I don't have the exact measurement done so I can only pull number out of my ass. The output impedance should be around 20-25 ohms. For power it's enough to drive my LCD2, but not enough for HE6SE which sounds compressed. All my dynamic headphones sound great on it. So I would guess maybe 1 to 10 mW of really linear power. For max theroretical power, it should be able to go up to 1.8W for 22 ohm loads, around 60mw for 600 ohm loads, 700mw for 8 ohm loads.

 

Sounds like a problem for a cascode. This is how people have been getting around the dissipation limits of the LU1014D at diyAudio. I would probably use a big BJT as the cascode device working into a single power resistor, but you could also use, say, a J111 as the upper device for a J112 as you're guaranteed the IDSS of the top device will be greater. But while one BJT (or MOSFET) could cascode a bunch of JFETs, an all-JFET cascode would probably want another 42 JFETs.

A folded cascode could be a neat trick to DC-couple it from input to output, but that might be a step too far from the elegance of three resistors and a FET. No point in fixing what isn't broken.

 

Oh yes bring on the cascode! It would really solve a lotta problems in this case.

I was thinking about cascoding with SiC JFETs like UJ3N065080K3S at one point, coupling some low voltage low noise JFETs like some surface mount Onsemis. It should be able to snap the drain voltage at around 9V, and support up to a heck lot of rail voltage. Could even do a transformer load or parafeed like a real big ass pentode, while the design remains a JFET only project.

Unfortunately my curve tracer fried during FETs measurements. There are still a lot left to be characterized about UJ3N series. I also lack the tools to characterize its high frequency performance. UJ3N + little JFETs should still be an intersting option nonetheless.

 

If transformer were an option, the real equivalent to a SET circuit for JFETs is to use them as followers, and drive a step up transformer for gain. 21x J112 should be able to provide around 25-50 ohm output impedance at 2x gain, which should still be good enough for headphones. It will eliminate the need for drain resistor, source resistor, and output cap all together. Also less need to worry about the rails. IMO this is the holy grail to use JFETs to compete with SETs. I was meant to save it for future threads.

However it's best of luck to find a suitable transformer for this design. The best option is to have it customized. With dual input winding, it would also be possible to switch the input output to toggle between singed ended and balanced mode. Take that Mjolnir 3!

 

If you're not already familiar with it, it might be worth taking a look at Susan Parker's amp.

I'd give some thought to leaving the OPT on the drain. It'll be a SEP of sorts rather than a SET, but would keep the distortion profile you already have while allowing you to explore current drive. Or, you could use feedback to make a real SET equivalent.

They're not free, but these are quite nice. https://www.lundahltransformers.com/wp-content/uploads/datasheets/2754.pdf

 

I found out about that project when I was thinking about using transformers in solid state amps. Definitely a super awesome project from 20 years ago. Using JFETs would eliminate the need of the input transformer, which I think is an upgrade. She had a DN2535 version of the line driver, which I think I could eliminate the input transformer and biasing RC, if the OPT has primary DC resistance of 10-12 ohm, at least for my batch of DN2535. I've figured the resitance needed for parallel JFETs too. The spots are pretty sweet.

TBH I feel a little done for the SEP configs. The linear input voltage range is quite limited with JFETs, at least without severe source degeneration. 680 to 1k ohm would work the best for J11x, but that feels like too much, and require biasing circuit on the input. The other way out is using huge gain. It would be amazing on preamps, but too much for headphones. I'm ready to explore other setups, even if they end up in a firework party

I'm not really a fan of the trioderizer, as it seems to overcomplicates the problem. Source followers is very similar to triodes that the dampling factor modulates the distortion profile. Drain out designs like this project ideally has a nearly flat output impedance instead.

I actually had a batch of Siliconix J105 that measures like triode, but it's hard to find its best operation point due to many limits.

Thanks for the recommendation! It seems like the best option for the application. I was thinking about VTB1148 or JLM111DC, but both of their DC resistance feel a bit high. LL2754 should be the perfect candidate albeit a bit costly.