JFet matching and measured distortion - Part 2: LTP

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
...or the death of jfet-matching? :D

2nd part that looks at matching jfets and distortion performance in simple circuits (1st part was http://www.diyaudio.com/forums/showthread.php?s=&threadid=138852 ).


It is my pleasure to bring you something more interesting, the differential pair (aka LTP).

And yes there's is benefit. Distortion is reduced by using matched jfet-pairs.

And no it's not much.

The details:

(See schematic below) A simple LTP consisting of 2x Toshiba 2SK369GR jfets (same as last time), no source degeneration, a simple tail current source using a ZTX450, setup for 1mA. Load resistors of 100 Ohm were chosen to keep Drain-Source voltage variation low as the used rails (0-13V) do not permit the use of a cascode.
 

Attachments

  • jfet0002_s.gif
    jfet0002_s.gif
    41.1 KB · Views: 2,122
THD+N plot

For starters the THD+N plot.

Jfets A+B represent the worst case where 2 randomly chosen jfets have largely different IDSS; in this case THD+N is constantly higher by about 1.5% absolute.

Jfets B+C is so-so matched (to 0.3 mA) and is nearly indistinguishable from the best matched pair in the comparison, jfets C+D (0.068mA difference in IDSS). If you look closely, you see that this closest pair actually performs worse for larger input signals than the mediocre matched B+C pair (look at 180-200mVrms).

To summarize the plot, avoid largely unmatched jfets for LTP-use, but even mediocre matched pairs can work better than tightly matched ones. So to reliably get better performance by matching, one need to use a distortion analyzer for verification.

One does not get automatically better performance.
 

Attachments

  • thdn_100-200mvrms_1k.gif
    thdn_100-200mvrms_1k.gif
    30.4 KB · Views: 1,921
Distortion spectra

As there's often the discussion of distortion spectra, I might contribute some data as well.

Here the overlay of the mediocre matched pair B+C (blue) and the tightly matched pair C+D (red). No notable differences. Small differences for 4th and 6th order; odd orders largely unchanged.


Finally the spectra of the worst case pair: large 2nd order distortion compenent,distortion cancellation no longer effective.
 

Attachments

  • fft_ov1_128_150mvmrs_1k_s2.gif
    fft_ov1_128_150mvmrs_1k_s2.gif
    33.5 KB · Views: 1,858
Part 3 would be the complementary case, however I do not have an appropriate power supply in the moment. I can build one, but as I will be busy the next weeks this has to wait a bit.

Summarizing the results, distortion is easily held under control by using even only medicre matched pairs (within about 0.5mA, just like Borbely sells them). Closer matching is not guaranteed to give better results. Of course that only applies to distortion. How important that is to 'good sound' is to everybody's taste.

Of course discussion is welcome ;)

Again a big thank you to Mr G, who made the fun possible.

Have fun, Hannes
 
Hi,
you have compared similar and dissimilar Idss.
What about transconductance comparison?

I have found it very easy and very quick to match Idss to any tolerance.

When comparing well matched pairs at any current different from Idss, the Idss matching turned out to be nearly useless. It simply grouped the devices into batches.

Getting a loose match of transconductance AND Idss was difficult.
Getting a close match of transconductance over a range of Id is very laborious. I don't believe it can be done by comparing graphs/testpoints taken with individually measured devices.
It needs a pair matching jig and careful attention to matching currents and dissipations and temperatures.
 
What about transconductance comparison?

Sure, it's doable and of course it would also be interesting. There are lots of things one could have a look at.

I'm however especially curious about the effect of simple things, these that often give most benefit. More sophisticated methods start often the diminishing returns.

Anyway, the data does not suggest that curve matching would significantly improve distortion performance.

I think that one can do that to enjoy the symmetry, but otherwise there seems nothing to be gained.

Have fun, Hannes
 
That just depends on your definition of 'matched'.

Generally IDSS is meant, some :D mean "neighbour on the same wafer" with at least 100 identical points on a transconductance map, identical Pinch-off voltage, identical capacitances at the precisely identical temperature with the same aging.

If you really want to go that way, you will never be able to use SPICE ;)

As said before, one can estimate the gain with better matched devices from the current data. Remember, 1st and 2nd order effects. :)

Have fun, Hannes
 
Matched means two devices having the same Id vs Vgs curve over the entire working range.

I thought about not commenting, but I could not resist :

a) I have never thought of using JFETs at 0.5mA bias. Even 5mA is a bit on the low side.
b) I would not call -30dB second harmonics well matched, if there is any match at all.

I promise these would be my last words on this thread.


Best Regards,
Patrick
 
Hi Patrick,

don't worry, your comment is very welcome ;)

ad a.) yes you're right, it's a bit low. I'm sorry, I don't see how that affects the results at the moment. For reference, at least the XOno runs the differential at 2 mA. If it helps, I can redo the measurements at higher current?

ad b.) Please feel free to do so; if highly sophisticated matching methods are required to get THD down, I'm not really interested.

If however somebody wants to send me a matched pair that meets your stringent criteria, I will put it onto the bench.

As I have more the feeling that my little fun is not really positively appreciated here - I'm not surprised as the results are not nice - I don't expect any jfet package.

Am I wrong? :D

Have fun, Hannes
 
By the way, Patrick I want to build your DAO in the near future and put it onto the bench.

Actually I thought of comparing at least 2 versions (stock and the 2SK246 variant), however I don't have any 2SK246. Funny I thought of asking you wether you would spare some to make the comparison possible.

It would be a pity if my little measurements made me your personal persona non grata :D

Take it easy, it's just a handful of plots ;) Feel free to ignore it for whatever reason.

Have fun, Hannes
 
I guess you simply missed that the LTP runs on a single rail. So you cannot assume that the jfet source sits at ground potential. If you connect the gate to ground, the source sits much higher -> -VGS. The same as source degeneration.

As the current source takes all the voltage, the jfets would just cut off.

Have fun, Hannes
 
PMA said:
Your circuit in
http://www.diyaudio.com/forums/showthread.php?postid=1761341#post1761341
is a perfect example how NOT to design with JFETs.

Rude as usual. Somebody tried to do more than an armchair simulation and give away the results for free and instead of encouraging and showing a better/useful way you had just a knee jerk reaction.

The results on a CCS driven JFET LTP are certainly interesting, and it would be great if you could redo at higher currents, around Idss.

h_a, JFETs have a unique property: they self bias. Here are three self-biasing configurations that would be very interesting to investigate experimentally. I understand that's a huge amount of work (lot's of variables), but at least one of each cases will reveal some results. E.g.

- Two parallel JFETs (matched/mismatched) with 0.0/4.7 ohm Rs at various VIN (4 measurements/passes)
- LTP with two JFETs (matched/mismatched) with 1.0/4.7 ohm Rs and Rs2=0.0/2.2ohm at various VIN (8 measurements/passes).
- H configuration with two pairs JFETs (3 mismatches P/P, P/N, N/N) with 1.0/4.7 ohm Rs and Rs3=0.0/2.2ohm at various VIN (12 measurements/passes).
 

Attachments

  • jfet matching.jpg
    jfet matching.jpg
    40.1 KB · Views: 1,212
Hi Ovidiu,

thanks for your input!

Am I missing something, but #1 of your suggestions was part 1 of my little series.

#2 should be easy to implement, #3 needs a new power supply :D

Of course, especially the complementary case would be interesting.

It's a pity, but I don't have the time at the moment, also the Audio Precision is not free currently.

Have fun, Hannes

PS: geez, see what the archive here at DIYAudio brought up!

in the "John Curl amp"-thread

Upupa Epos
"But remember, that you must select pair for the same Vgsth ( they have very wide tolerances at this parameter ) and also transconductance must be the same..."

and what John Curl replies

"Don't worry about the fets, just get the Idss close and everything will follow. How do you think that we made the first series, before complementary jfets were even available?"
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.