Improved Tube Headphones Amp

[ruffrecords]

For many months I have been trying to develop an all tube headphones amp that can drive a wide range of headphones to a high level with low distortion. The designs have been based on a transformer output using a Sowter 8665A to match the tube stage to headphones of different impedances. With nearly 200GBP worth of transformers in the design it was more in the audiophile price bracket than the budget one. I have tried cathode followers, mu followers and White followers, open loop and and with NFB but with none of them could I get much more than 125mW output without a spray of harmonic distortion. So I abandoned the task for a while.

Then, back in May, I came across Pete Millett's SRPP design using the ECC99. I had not tried an SRPP design because of their reputation for high distortion so it was almost in desperation that I knocked up a prototype. Sure enough it had quite high distortion but I noticed two interesting things. First it was quite capable of delivering 3V rms into a 32 ohm load (280mW) which is enough to drive just about any headphone to an almost painfully loud level. Secondly, although it produced 3% THD at this level, the harmonics fell away rapidly, most of the distortion being 2nd and 3rd harmonic. This was in stark contrast to other designs I had tried.

That's as far as Pete Millett's design goes. There is a preceding amplification stage but no NFB so although it can provide a high output, the distortion is rather high. It seemed to me it might benefit from some NFB, so I designed a single triode stage base on one half of a 12AX7 and closed the loop from the SRPP output back the the 12AX7 cathode. Unfortunately, because there is a dc blocking capacitor in the feedback loop the closed loop gain rises at very low frequencies and it is not possible to apply enough NFB to reduce the distortion significantly without instability.

The classic way to ensure unconditional stability in tube NFB circuits is to ensure there is only a single low frequency pole in the loop which as often as not means the NFB network has to operate down to dc, i.e no series caps in the NFB loop. This can often be problematic from the point of view of setting the dc conditions in the tubes and this case was no exception and a compromise had to be made in slightly unbalancing the SRPP stage to achieve it. Despite that, the results are good:

2V rms into 32 ohms (125mW)

2H = 0.18%
3H = 0.032%
4H = 0.006%

Higher harmonics were immeasurable

3V rms into 32 ohms (280 mW)

2H = 0.28%
3H = 0.063%
4H = 0.014%

Other harmonics immeasurably low.

For 3V rms output into 32 ohms an input of 0.46V rms is required.


The prototype was built on a die cast box as a chassis but I am now well on the way with a PCB layout which looks as though it will fit onto a board 3.5 inches by 5 inches (with the transformers external).

Circuit of one channel attached.

Cheers

Ian

[tomchr]

If you want to try something different, I suggest looking at local negative feedback around the first stage. One example of this can be found in the second stage of this amp. It works quite well and has low distortion as well.

Looks like good work, though. What made you go towards a parafeed (just curious)?

I think my next project will be a headphone amp, hence, I'm exploring the field a bit.

~Tom

[artosalo]

I suggest looking at local negative feedback around the first stage.

Is it wrong place to use local NFB?
Most of the distortion is generated at the output stage.

[tomchr]

Originally Posted by artosalo

Is it wrong place to use local NFB?
Most of the distortion is generated at the output stage.

Some people don't like global negative feedback. If the majority of the distortion is caused by the output stage, then local feedback around the first stage wouldn't help on the distortion, obviously. I threw it out there as a suggestion as it sounded like the original poster was happy with the distortion but wanted lower gain. My brain isn't firing on all cylinders today so I could have misread or misunderstood, though.

~Tom

[kevinkr]

I'm wondering how you might fare with a white cathode follower output stage? I've used one successfully out to several hundred mW, but can't recall much about the linearity. (except that it was reasonably good) I used 5842/417As in the output. No global feedback though..

[ruffrecords]

Originally Posted by tomchr

Looks like good work, though. What made you go towards a parafeed (just curious)?

~Tom

What makes you think it is a parafeed?

Cheers

Ian

[ruffrecords]

Originally Posted by kevinkr

I'm wondering how you might fare with a white cathode follower output stage? I've used one successfully out to several hundred mW, but can't recall much about the linearity. (except that it was reasonably good) I used 5842/417As in the output. No global feedback though..

Of all the other topologies I tried before the SRPP, the White follower was the best, but it was still short on output power and at over 2V rms into 32 ohms it produced a spray of harmonics. Incidentally, the 5842 does not really have a large enough Vhk to be used in a White follower or SRPP circuit.

Cheers

Ian

[DualTriode]

Hello Ian,
While you are messing around try a Single End Triode headphone amplifier.
I tried this transformer EDCOR Electronics Corporation. GXSE10-16-1.7K using a triode wired 6BQ6GTB (cheap) with near 180 volts on the plate. With 32 ohm headphones attached the SET sees ~ 3400 ohms load. The Triode wired 6BQ6GTB has an Rp ~ 650 ohms. The SET 6BQ6GTB will output near 2 watts. At 0.1 watts into the headphones the distortion is much reduced.
I tried this setup on the breadboard and liked it a lot.
DT
All just for fun!

[kevinkr]

Ian, I had just one other questionably useful thought and I was wondering whether you have been able to characterize the OPT independently of the stages driving it.

Your comments about distortion spectra left me wondering, and I'm assuming that the lower the driving source impedance the lower the measured distortion until the transformer starts to saturate. This is just a SWAG on my part and possibly all wet.

One other probably very ill-advised thought I am is having is to increase the value of the output coupling caps by a couple of orders of magnitude (are they really 4n7 or are they 4u7 - I can't tell, chalk it up to my laptop screen and my eye sight) and include them inside the global feedback loop. This might be a very bad idea, but I had some success with this approach in somewhat less demanding applications. (I was using 4.7uF output caps)

The numbers you are getting in any event look pretty good.

[artosalo]

One other probably very ill-advised thought I am is having is to increase the value of the output coupling caps by a couple of orders of magnitude (are they really 4n7 or are they 4u7

With 2 x 4n7 caps. the cut-off freuency would be around 5 kHz.
So 4μ7 seems obvious to me.