Codename "Tapyrus", my ongoing DIY headphone amp

[fbx]

Hi,

dedicating my spare time to this project since quite some months, now I believe that it got to a very good point, and I'd like to share it with others.

My trigger: I never was a tube guy, even if with a strong historical technical background on the electronics subjects, and I had a big surprise (a revelation maybe) when I bought a Yaqin amp for my little studio, to substitute the Sony and Pioneer amps that I had, and I thought that they were good. My loudspeakers are a pair of small Infinity, with their subwoofer. Nothing spectacular, except for the surprising leap in sound quality.

Hence I decided that, since I love listening with headphones, I wanted a headphone amp that could possibly give me the right feeling with my music. So, after having read a lot, and having decided that I did not like the amps that I was reading about, or whose schematics I was looking at, I started with my project, putting there my ideas.

My goal: a tube-based OTL headamp that could drive headphones from 32 to 300 ohms, at a decent volume level, with as much headroom as I could, and good characteristics.
Moreover, I wanted to avoid all the capacitors that I could avoid, except for the one on the output. I consider this one as a necessary evil. Hence, DC coupling is the way I've chosen.

In the tube world it's difficult to invent things, it's quite a consolidated technology, so I cannot claim too much originality. However, I designed this from scratch, just getting inspiration from many articles and forums, like DIYAudio, HeadWize, and John Broskie's ones.

The final schematic (if "final" is a word that a DIYer may use) comes from many many simulations using ngspice and the gEDA tools. Learning the tools, making the symbols and the models was not really easy, but I appreciate a lot this suite.

This schematic can be described as an ultra linear cascode with parallel cathode follower. The bias of the CF is around 60mA, 15ma per triode section.

The tubes: I've tried to choose relatively normal ones, like the ECC82 for the input stage and ECC88 for the CF. Originally I was using ECC83, but then I realized that I did not need so much gain.

So, here's the schematic. If anybody is interested I can post also some results from the simulations. They look really very good to me. I am also building it, since I'm quite happy with the results on paper.

Have fun!
fbx

[ruffrecords]

Very interesting. I would love to see a simulation showing the distortion at 3V rms out into a 32 ohm load.

Cheers

Ian

[fbx]

Hi,

3V rms into 32 ohm is 280mW, there's no way in which a single ended stage biased at
60mA could give all that power. This is around 20 times the power of an mp3 player, quite useless I'd say.

I did not design it with this in mind, hence I'd expect a soft square wave. Moreover, normally the 32 ohm headphones are quite efficient, and play loud also with the 12-15mW of an mp3 player, if we don't take into account the silly EU power limit.

The theoretical max power of this stage should be around 50mW into 32 ohm. Still 4 times the power of an mp3 player, more than sufficient to blast my head. Moreover,
32 ohm for a tube amp is the evil case. On 300 ohm it is much more relaxed. I can run later some simulations to confirm this value.

fbx

[DF96]

A cascode is useful when Miller effect capacitance would otherwise be a problem. Sending negative feedback to the cascode upper grid stops it from doing the only thing which makes it worth using. It will raise the impedance seen at the anode of the lower valve, as this would normally be quite low in a cascode so can lead to distortion, but not as much as getting rid of the upper valve. So my conclusion is that you would get better results by dropping the upper valve of the cascode. Either get rid of it altogether, or use it as an active load.

[fbx]

I see your point, df96, and technically agree from a certain point of view. I am not able to tell what's perfect and what's not, but I may comment on why I did in that way. In some time I will also be able to verify all this in a prototype.

I had done many preliminary simulations using just one triode in the gain stage, but was never satisfied nor by the characteristics nor by the fact that it became very difficult to have a stable DC coupling with the output stage. Just slightly changing the characteristics of one tube or the power supply voltage needed to tweak the value of the resistors in order to find again the best operating point with the largest headroom. Not acceptable for me, a real world prototype would just be asymmetrically clipping 90% of the time.

My idea was to have a DC coupling that does not need tweaking with trimmers, hence the solution of the DC feedback was just perfect. It both gives a relatively low feedback for the signal and stabilizes the operating point. Moreover, the feedback path is very fast, hopefully minimizing the danger of phase shifts that could lead to ringing and oscillations. Hence, the top triode serves this purpose. As I had written, I did not need too much gain there, so the purpose of the cascode was not to multiply the gain, but have a topology that could lead to good and stable results with as few components as possible and 'easy' power supply requirements.

Just for the credits, I got the inspiration for the ultra linear cascode in one article by John Broskie, and then followed my way. I find that it gives very good results by now.

fbx

[DF96]

So-called ultralinear cascode may be OK when you need a cascode, but want more linearity. If you don't need a cascode then other circuits may be better. You could provide DC feedback without AC.

[fireworks]

Originally Posted by fbx

Have fun!
fbx

Did you use the open source GEda/gschem software to draw this schematics ?

[fbx]

Originally Posted by fireworks

Did you use the open source GEda/gschem software to draw this schematics ?

Yes, and then gnetlist and ngspice for the analysis on a Macbook.

[fbx]

As promised, I post a few simulation results:

- frequency response
- output impedance

the max output power seems to be around 130mW@300 ohms, and 40mW@32 ohms.

And here I paste some THD+harmonics. their content does not seem to vary wildly as the volume increases, except for when clipping is approached:

300 ohms, 6mw
Fourier analysis for vout:
No. Harmonics: 10, THD: 0.0949378 %, Gridsize: 200, Interpolation Degree: 1

Harmonic Frequency Magnitude Phase Norm. Mag Norm. Phase
-------- --------- --------- ----- --------- -----------
0 0 -0.0033404 0 0 0
1 1000 2.00086 179.87 1 0
2 2000 0.00140556 22.989 0.000702475 -156.88
3 3000 0.000658174 3.95188 0.000328945 -175.92
4 4000 0.000647599 4.1352 0.00032366 -175.74
5 5000 0.000518078 4.57156 0.000258928 -175.3
6 6000 0.000431779 5.45924 0.000215796 -174.41
7 7000 0.000370289 6.35176 0.000185065 -173.52
8 8000 0.0003242 7.24567 0.00016203 -172.62
9 9000 0.000288377 8.14113 0.000144126 -171.73


32 ohm, 8mW

Fourier analysis for vout:
No. Harmonics: 10, THD: 0.808642 %, Gridsize: 200, Interpolation Degree: 1

Harmonic Frequency Magnitude Phase Norm. Mag Norm. Phase
-------- --------- --------- ----- --------- -----------
0 0 -0.010557 0 0 0
1 1000 0.740426 -179.8 1 0
2 2000 0.00592933 -80.685 0.008008 99.1188
3 3000 0.000298457 174.939 0.000403088 354.742
4 4000 0.000470852 20.182 0.00063592 199.985
5 5000 0.000374785 4.81889 0.000506175 184.622
6 6000 0.000296292 5.64662 0.000400164 185.45
7 7000 0.000253508 6.78114 0.000342381 186.585
8 8000 0.000222723 8.08596 0.000300804 187.889
9 9000 0.00019768 8.54195 0.000266981 188.345


The circuit seems to have a preference for the 2nd harmonic. Anyway, I do not trust 100% a simulation, especially with the tube models got from the net, I fear that the final answer will have to come from a working prototype (if I had decent instruments to measure...). But, at least, it seems to work well. As always, comments are welcome.

fbx

[Globulator]

Why don't you use an ECC88 on the input stage instead of the rather poorer ECC82?