John Curl's Blowtorch preamplifier part II

[Bonsai]

quote
Any off-centering (which the servo would do) would place the pair in a less linear region.
unquote

Jan, surely wha t you would do here is simply move it along the bias line (I am not talking about a gross mismatch here).?

Anatech,
I am using 4 bipolar input devices on my amp and my source and speakers are quite good - I don't pick up the brittle sound you mention. Betas were matched on a meter to within 10%.

[jan.didden]

Originally posted by Bonsai
quote
Any off-centering (which the servo would do) would place the pair in a less linear region.
unquote

Jan, surely wha t you would do here is simply move it along the bias line (I am not talking about a gross mismatch here).?
[snip]

Bonsai, as you know, that bias line is that familiar S-curve where the center is the most linear. Move it off-center and you go to a less linear part. I have the same opinion as you, that the difference would be very small, but don't have any hard data, just opinion.
I wouldn't think it would make such a big difference because it is a relatively smooth non-linearity and feedback is very good at making that all but disappear anyway. But it is measureable in the even harmonics.

jd

[AndrewT]

Originally posted by anatech
Hi Andrew,
Have you ever tried to measure beta to compare how close Vbe measures? Just curious. I have, as mentioned above. I would say that matches in beta area subset of matches in Vbe.

yes, I have.
I find little correlation, although others have posted the opposite view. Maybe we test differently.

I think Vbe and Vgs are far more important, then I select for a close match of gain (hFE) or transconductance (gfs). BUT ALWAYS at operational quiescent currents.
I am not usually able to test at operational Vce nor Vds, due to dissipation limitations, but I doubt our disagreement is due to this lack of ability.


As an aside I used to measure hFE first and then try to select similar Vbe from the close batches. It was a complete waste of my time.
The Vbe matches could come from any hFE batch.

I now test Vbe first. Batch them according to this parameter.
The slight problem with this is that the next amplifier may run at a different current and that initial selection must be rechecked at the new current.

[AndrewT]

Originally posted by anatech
In tested incoming parts for grading, I have seen that when beta matches, so does Vbe. Of course, Vbe is always within a few mV anyway, so I don't see this as a valid measure to grade parts on. AndrewT and I disagree on this basic point. The current sharing is better when matched for beta in my experience, and so is the balance of a differential pair. I use a steady state current measurement method, not the "transistor test" function on a multimeter. I have also made jigs to encompass higher voltage and current levels, as well as a jig to balance transistors in long tail pairs. The circuit is simple a long tail pair running open, no feedback of any kind. The parts are in thermal contact and pairs matched in this way do react to the matching of emitter degeneration resistors. That seems to be an extremely close match, and it works well in practice.

This spells out the difference in testing techniques that we use.
I accept that I cannot hold a device to a set temperature.
Instead I select a reference device and thermally couple a DUT to it.
I set up a long tail pair with the bases/gates directly connected.
I also directly connect the emitters/sources.
Now add a pair of accurately matched collector/drain resistors.
Apply a Vbe/Vgs and measure the difference in collector current.

This measurement method with Zero value for the External emitter/source resistor gives a very different result compared to Anatech's method.
The zero Re is very intolerant of Vbe differences.
It shows them up whereas using a nominal Re hides the parameter mismatch very effectively.

[AndrewT]

Originally posted by Joshua_G
Hi Chris,

What would you match for JFETs in complimentary differential pairs, other than IDSS?

Originally posted by PMA
Transfer function (dId/dVgs), if possible.

I test Jfets exactly as BJTs.
If the Vds is matched at the operational quiescent current then the drain current will be the same.
Now I adjust the Vds to monitor how far the drain current mismatch becomes at other Vds. This shows up transconductance slope variations very effectively.
The Idss selection is not matching by my definition. It is merely selecting for Idss.
I have matched k170 at various <=Idss and find that similar or even identical Idss is worthless when the current (Id) is reduced slightly or a lot below Idss.
The slope of the transconductance curve must be matched at all reasonable currents. I try for 100% of Idss down to ~ 25%Idss.

But as in my other recent posts, I use ref+DUT thermally coupled. This, I believe, is important. It ensures Id and Vgs and Vds and Pq and Tj all match, while the measurements are being taken.

[forr]

Hi Jan,

---I once heard Doug Self talk about that. One point he made was that if you use a servo to correct DC offset from a non-matched diff pair, you also base the pair off of the center point where the transfer function is as linear as it gets. Any off-centering (which the servo would do) would place the pair in a less linear region. Intuitively I feel that the effect should not be very great, and there is still the global nfb to straighten things out, but the effect is there and can be measured on the ol function.---

Perrot's US Patent #5635874 adresses the problem by feeding the servo output signal after the input stage. However simulation seems to show this should'nt be an issue. The effect on harmonic distorsion of a +/-100 mV input DC offset are minute on 2 * 3 mA differential input stage with 100 Ohm degenerative resistors à la Self, even when not loaded by a current mirror.

[jan.didden]

Originally posted by forr
Hi Jan,

---I once heard Doug Self talk about that. One point he made was that if you use a servo to correct DC offset from a non-matched diff pair, you also base the pair off of the center point where the transfer function is as linear as it gets. Any off-centering (which the servo would do) would place the pair in a less linear region. Intuitively I feel that the effect should not be very great, and there is still the global nfb to straighten things out, but the effect is there and can be measured on the ol function.---

Perrot's US Patent #5635874 adresses the problem by feeding the servo output signal after the input stage. However simulation seems to show this should'nt be an issue. The effect on harmonic distorsion of a +/-100 mV input DC offset are minute on 2 * 3 mA differential input stage with 100 Ohm degenerative resistors à la Self, even when not loaded by a current mirror.

I believe that. Gut feeling.

PS This Perrot is that the guy aka Hephaistos?


jd

[hitsware]

>It's not a resistor value, it's a current,
>that for IRF whatever needs to be too
>high for them to be temperature stable.
>That means, when they are temperature
>stable they are already too hot.

Then you can't use source resistors
the way you use emmitter
resistors with bi-polars ?

>Laterals are weak mosfets, they change
>tempco direction on too low currents.

For class AB they are perfect (for my purposes).
Stable @~100ma.

[h_a]

The slope of the transconductance curve must be matched at all reasonable currents. I try for 100% of Idss down to ~ 25%Idss.

As interesting and certainly painfull your method is, I would be interested if you ever measured distortion of your matched pairs?

If you say 'must' you certainly have checked that? How could you be sure otherwise?

Is there significant improvement over my measurements?

JFet matching and measured distortion - Part1
JFet matching and measured distortion - Part 2: LTP

Please post your figures on that topic, I think it helps everyone to see how much that brings in real life. We would then also see how much better your extensive method is in comparison to simple IDSS selection.

Have fun, Hannes

[AndrewT]

Hi Ha,
"must" in the context of my definition of "matched".
If the device is deemed matched then it's amplifying ability "must" be the same. For that to be true then the hFE "must" follow it's partner as Ic (for the pair) changes, similarly gfs "must" follow it's partner as Id varies.
I called this tracking in earlier posts.
It is easy to see the mismatch or preferably lack of mismatch of a matched pair as Vds/Vbe is altered and monitor the difference in voltage across the two drains/collectors.
The time consuming part is finding the pairs in the first instance.

I cannot measure real life distortion, so I cannot comment on whether this form of matching produces either audible or measurable improvements to the overall amplifier.

What I can say is that selecting by Idss or hFE alone gets nowhere near reliable matching. I would guess that I would be lucky to get 1in in 100 pairs that gave a reasonable match by selecting by those parameters.

Your earlier thread did a comparison that I thought was fundamentally flawed. That's why we had such a long winded disagreement on the applicability of the method.

But, having last week installed an E-MU1212 sound card I will shortly be asking for assistance in this distortion measuring process, going by my previous abysmal history of understanding plain English instructions. It even took a while to realise I needed an adaptor to couple up my stereo headphones to the two balanced output channels.