[rjm003]

Original
 
The LTSpice results match my measurements on a similar amplifier, so I am confident that the distortion predicted for the Szekeres circuit is realistic.
 
So here you go. The not-quite-an-answer to the question probably few people other than myself was interested in: "Is the Szekeres amp (and the derivatives) really hi-fi, or does the source follower just pleasantly warm up the sound by adding plenty of 2nd harmonic distortion?"
 
It's not an answer, but at least now we have some numbers to argue over. :-D
 
 
 

 

 

[nikongod]

Have you run a simulation for a CCS loaded source follower? 
 
I'm curious how it compares. 

 

[rjm003]

 
Like this? Hang on a tick, I'll work it out.
 
The short answer, however, is "way less THD".

 

[dsavitsk]

What happens if you use a good CCS? Impedance of that one is something like 1K Ohms.

I suppose the load swamps it here so it probably doesn't matter much.

 

[rjm003]

If you suggest a "good" CSS and I'll by happy to run it up...
 
For this "bad" CSS,
 
 

 

 

[Avro_Arrow]

Here are a few to choose from:
 
2 and 6 are simple and effective.
 
The preview is terrible, but it looks fine when you click on it.
 
 

 

[rjm003]

Thanks, nice resource. For the present, however, I have a better one:
 
 

 
And as noted by dsavitsk in his post earlier, it makes no difference. The same distortion data as for my simple CCS. The headphone load is 30 ohms and is in parallel with the CCS impedance. The CCS impedance only has to be  4-5 times larger than that (say 100 ohms) before it becomes insignificantly high.

 

[dsavitsk]

If you suggest a "good" CSS and I'll by happy to run it up...

The Z is basically R2 * Hfe of the BJT. To make it better, you want to increase R2 and cascode the BJTs. You can also just replace the BJT with a MOSFET, or add a FET cascode to the BJT. Additionally, in a real circuit, you would want to bypass R8, or replace it with an LED or something. Otherwise, you get a bunch of noise at the base/source of the CCS transistor which is amplified at the circuit output.

While into 32 Ohms the CCS can be relaxed for distortion simulation, in real terms a very high CCS isolates the PS caps from the signal current loop and sounds better. And if someone wants to use this as a preamp the high CCS will be a better load.

 

[rjm003]

Your advice is sound, as always. Perhaps someone else will be interested in taking this up.
 
I see all too clearly that I would wind up with something very close to the Pass Zen circuit. It's been done already.
 
Also, this just seems to me to be a bad place to start if the end goal is a low distortion headphone driver. A diamond buffer topology has so many relative advantages.

 

[G.Trenchev]

You can use LM317 for CCS,it will somewhat improve performance and efficiency...150-200 mA should be fine.Overall you can expect 0.5-1% THD in real life,it sounds warm,and it requires some driving.The MOSFET input capacitance is not to be neglected if you don't want high frequency roll-off.You can also improve PSRR by dividing R1 into 2x47k,and connecting their middle point through 100uF cap to ground.

 

[rjm003]

An LM337-based CCS seems like a really bad idea from a noise perspective though.