JFet matching and measured distortion - Part1

[h_a]

...or the death of jfet-matching?

Jfets are very popular here on DIYAudio and regularly they're matched in the hope to get better performance - some even develop a true art to get closest pairs.

Interestingly there's no data on what this work might bring, in fact it seems a given that the closer automatically the better. No doubt about that, but what to expect? 20% less THD in comparison to unmatched parts? 10%?

So I scribbled a little circuit together, nothing exciting, see attached, and hooked it up. (Don't be surprised by the odd resistor values, I found them in my parts bin). I used Toshiba 2SK369GR parts as I currently do not have wide spread 2SK170; however 2x 2SK170 equal in parameters a single 2SK369.

You see the THD+N plots in the next post with a big surprise.

A discussion of results and maybe suggestions for additional measurements are highly welcome!

At last, a big thank you to MrG for allowing me the generous use of the AP, a nice little beast.

Have fun, Hannes

[h_a]

This is the THD+N for 3 different jfet-pairs and the really interesting point is that matching seems to be completely unimportant THD-wise. It seems that only current has strong influence on THD.

As you can see the jfet-pair A+D, although the worst matched pair, performs best. The only explanation I can find is that its higher current causes a more linear operating point.

Have fun, Hannes

[jackinnj]

the noise is going to be different between the high current and low current devices.

i've been experimenting with a number of other P-channel devices...

[h_a]

the noise is going to be different between the high current and low current devices.

C'mon! Noise differences due to that small gm-variations?

Nevertheless the gentleman is served the noise-less THD-plot.

Please note colours are different.

Have fun, Hannes

[syn08]

Originally posted by h_a
...or the death of jfet-matching?

Jfets are very popular here on DIYAudio and regularly they're matched in the hope to get better performance - some even develop a true art to get closest pairs.

Interestingly there's no data on what this work might bring, in fact it seems a given that the closer automatically the better. No doubt about that, but what to expect? 20% less THD in comparison to unmatched parts? 10%?

So I scribbled a little circuit together, nothing exciting, see attached, and hooked it up. (Don't be surprised by the odd resistor values, I found them in my parts bin). I used Toshiba 2SK369GR parts as I currently do not have wide spread 2SK170; however 2x 2SK170 equal in parameters a single 2SK369.

You see the THD+N plots in the next post with a big surprise.

A discussion of results and maybe suggestions for additional measurements are highly welcome!

At last, a big thank you to MrG for allowing me the generous use of the AP, a nice little beast.

Have fun, Hannes

Pretty much an expected result, I think Scott mentioned this here before, and it also matches my late experience. The whole thing is to match N and P devices (or their parallel incarnations thereof) in complementary designs. This will provide some degree of distortion cancellation.

Regarding noise, I'm not so sure. While in a very simplified model the equivalent noise of paralleled JFETs is scaling down with gm increase and the equivalent source resistance, I'm not so sure this also happens for short channel JFETs like 2SK170 or BF861. My late experiments shows that by paralleling 6x2SK100GR the equivalent noise is actually slightly (5-10%) worse than a single 2SK170V, with the same transconductance. I can think of a few device physics reasons for this behaviour, but it's beyond the scope of this forum.

[EUVL]

Try it with a LTP and measure differential output voltage across the two legs, driven by a differential signal, especially even and odd harmonics distributions.

THD does not tell anything, IMHO.


Gruss,
Patrick

[AndrewT]

I thought the matching of pairs was for LTP duty.
The circuit you show does not simulate LTP operation.
By using source resistors you are burying some of the difference with the local feedback.
It Looks like EUVL's comment is along similar lines.

[mjf]

hello.
some people match them to get lower offset at the output........
greetings.............

[EUVL]

> It Looks like EUVL's comment is along similar lines.

Exactly.

The circuit used has IMHO no meaning (forgive me for being rude, or honest, depending on how you want to phrase it). It only illustrates the case for paralleling JFETs, where matching is almost a non-issue. Where matching becomes critical is in a distortion cancellation application, of which the LTP is one (of course there are others).

> ...or the death of jfet-matching?

I really couldn't help laughing when I saw this. I'm so sorry !!


Patrick

[mjf]

hello.
what happens when using two different amplfiers for stereo application?
greetings............