Buffer distortion measurements

[Voodoochile]

Thanks for the excellent informational study, Kurt!

Also, thanks, Nisbeth, for sheeting it. I should point out that the first section header does read 100 Ohm load, however, and should be updated to 30 Ohm. I changed mine, but you may want to fix the .zip.

 

[Nisbeth]

Quote:

I should point out that the first section header does read 100 Ohm load, however, and should be updated to 30 Ohm. I changed mine, but you may want to fix the .zip.

Oops! Done.

/U.

 

[aos]

Kurt, thanks for the info. Did you measure noise in audible range (to 20kHz) or did you measure it broadband?

By the way, from my own limited measurements, value of potentiometer has a big impact on amount of noise. Not just because thermal noise is proportional to resistance but because high impedance lines are more prone to acting as antenas for whatever EMI is around. 10k is a very good value to use in that regard.

 

[morsel]

Agreed, aos. 100k is too big, however, I like to recommend 50k as that is close to the 47k standardized audio input impedance, and 10k can sometimes be too low for very wimpy substandard outputs. How do you feel about that?

 

[Jan Meier]

Dear Kurt,

Thank you for your measurement. Very interesting indeed!

One suggestion I would like to make: Why not do some measures with a 500~1000 pF capacitive load in parallel with the Ohmic load? This will be closer to every day practice as it mimics the capacity of the headphone cable.

Just a suggestion though.

Thanks for the work done

Jan

 

[KurtW]

aos, the noise measurement was a 20-20kHz reading.

Jan, your idea of more accurately simulating a real headphone load seems like a good one. I took it a step further and used a real headphone, a Philips HP550. I backed off considerably on the level and used 100mv which was quite loud with these phones, way too loud to put them on. Here's the THD measurement with a few different buffers:

1x 2001, 0.48%
2x 2001, 0.16%
1x 2002, 0.07%
4x 2002, 0.07%
2x BUF634, 0.07%
1x HA5002, 0.06%

The measurement is at 1kHz, but I also checked several other frequencies for each buffer combo including 20 and 20kHz and it was very similar. The first measurement is not a typo, and I tried 3 different units with similar results. In fact two of those were later soldered together for the 2x 2001 measurement. I then repeated the test with the same level with the Senn. HD580 with the blue Clou cable for the first three buffers and this time they all measured the same 0.07%. I also compared a 1KHz square wave with the single 2001 and a 2002 (back with the Philips headphone) and they looked similar with some ringing which had a slightly higher amplitude on the 2002. I'm not sure what to make of these measurement, but here they are for what its worth.

 

[morsel]

Fascinating, Kurt.

 

[Jan Meier]

Dear Kurt,

Yep, that's what I expected. The distortion of a single EL2001 rose more rapidly than that of the other configurations. The problem is, that the rise time and bandwidth of this buffer decreases rapidly under capacitive load and that it becomes slower than the controlling opamp (AD8610). I feel with a few minor modifications the META can be improved on this one, but that's upon the designers to handle and decide

Thanks for the interesting info!

Cheers,

Jan

 

[morsel]

Well, short term we have buffer stacking as a reasonable option (one that I have always considered vital for optimum performance). Long term (late summer?), the Elantec buffers will be going the way of the dodo and thus will be replaced with some other buffer, probably the HA5002. It's unfortunate, because the Elantec buffers are the lowest current option available, which is nice for battery operation. Ah well, all good things must come to an end. Get them while you can.

 

[Voodoochile]

Nice! Thanks once again for sharing this valuable information. That's what it's all about, but I have to say it again. I appreciate it, and I'm sure many others do, also.

 

[Wodgy]

I agree! Thanks for the effort you've put into this.

 

[aos]

You can always slow down AD8620 with a feedback capacitor. This further decreases the noise by a large factor (several times in my case) as you reduce the bandwidth. I can't say I notice a difference in sound. Choosing a fast opamp only to slow it down does seem kind of self-defeating though.

I also use ferrite beads in series with buffer. This will put some resistance in series with load capacitance at high frequencies while not compromising output impedance at audio frequencies.

I remember seeing some Elantec application note about snubbers...?

 

[morsel]

Those suggestions are good ideas, and there are other ways to deal with the issue as well. We probably won't worry about it until the Elantec buffers run out (whenever that happens), since we will have to redesign at that point anyway.

 

[aos]

What other ways? The only other thing I can think about it would be some kind of compensation R/C network in the overall feedback loop... or maybe something between the opamp and the buffer... my mind is drawing blank. I remember ppl also discussed this issue.

 

[morsel]

Either changing the output buffer (obviously something that has to happen when the Elantecs eventually poof) or the output configuration, which is not a consideration until the buffer changeover.

As an aside, I'm faced with the strong likelyhood of having to move to a new place in a month since I am losing my housemate and have not found a replacement. The move will eat at least 2 months of my life, maybe more, so I won't be thinking about revising the META42 any time soon, certainly not until the Elantec buffers are drying up.

My apologies for sidetracking the distortion measurement topic.