Baxandall Super Pair

[WaltJ]

Thanks to bcarso and Bob Cordell for the welcome words. Hope the listed references are useful, and folks here can/will add more.

wj

 

[unclejed613]

Re: Re: Baxandall Super Pair

This kind of circuit needs some compensating capacitor. Try 470 pF between base and collector of the driver or output device .

the more i look at this, the more i suspect that Miller capacitance is the culprit here, and with the unique way that there are two miller capacitances interacting here, there is a lot of phase shift from input to output, which coupled back to the input through the interacting miller capacitances makes for a nice oscillator. the best way to reduce the interaction is to swamp one of the transistors with a rather large compensating cap (one that will not constantly change capacitance with Vbc). because of the virtual varactors that exist, trying to stabilize this circuit without swamping the miller capacitance is like shooting at a moving target (as was obvious from the graph of bode plots with several different base stopper resistors).

 

[Sigurd Ruschkow]

Re: More papers re. the Baxandall/Swallow CS

Walt - good to see you here again!

Most people seem to reference the cascode voltage to GND,
but the best thing, IMO, is to reference it to the source/emitter.
Is is easier to reference it to GND. Even Hiraga did that in his (and his buddies') "le Monstre" class A power amp.

Attached is Borbely's way of using a high Vp JFET:s own voltage for cascode DC voltage. Very elegant!

Not sure that only 2-4V is optimal as cascode voltage, though. I often use 10V (for 2SK170/SJ74).



Sigurd

Addl' reading that will be helpful to this topic:

A second point which falls in this thread is the driven cascode, as termed "enhanced cascode" by Hawksford. This was seen prior to Hawksford's JAES paper, one example was within Erno Borbely's "High Power High Quality Amplifier Using MOSFETs", Wireless World, March 1983, Pages 69-75. Erno provides some pictures which show the reduction in distortion in curve tracer plots.

A similar type of driver stage to Erno's Fig. 7 was used in the Adcom series 565 and 585 power amplifiers. The distinction here is that for full effectiveness, the cascoding transistor base should be driven from the emitter of the bottom device, so that the collector and base see the same dynamic voltage, and consequently, greatest reduction in distortion.

Walt Jung

 

[WaltJ]

Walt - good to see you here again! Most people seem to reference the cascode voltage to GND,
but the best thing, IMO, is to reference it to the source/emitter.
Is is easier to reference it to GND. Even Hiraga did that in his (and his buddies') "le Monstre" class A power amp.

Attached is Borbely's way of using a high Vp JFET:s own voltage for cascode DC voltage. Very elegant!

Not sure that only 2-4V is optimal as cascode voltage, though. I often use 10V (for 2SK170/SJ74).



Sigurd

Thanks, Sigurd. I think this is a bit of an apples vs. oranges comparison (power amps and preamps). Yes, JFETs and bipolars both benefit from this sort of cascoding, I'll certainly agree to that. And your point about driving the base of the cascode with the emitter, that's just what I was saying. I call this a driven cascode, as the cascode xstr bootstraps the bipolar with the collector, swinging at the same dynamic voltage as the emitter, plus an appropriate offset. Of course, with bipolars there isn't the necessity to have several volts of headroom in the cascode, as there is with FETs.

Thanks for the comments!

Walt Jung

 

[Sigurd Ruschkow]

Walt - driven cascode is a nice way of putting it. One might want to add telescopic for even more detail. Ie,
driven telescopic cascode (DTC).

Attached is a simulation I did on different telescopic cascode circuits. Both cascodes with the cascode transistor referenced to GND and also DTC.

Rather interesting to see how much better it is to use driven cascodes when high BW is wanted. Almost 10X more BW with a driven cascode compared to a cascode that is referenced to GND



Sigurd

Thanks, Sigurd. I think this is a bit of an apples vs. oranges comparison (power amps and preamps). Yes, JFETs and bipolars both benefit from this sort of cascoding, I'll certainly agree to that. And your point about driving the base of the cascode with the emitter, that's just what I was saying. I call this a driven cascode, as the cascode xstr bootstraps the bipolar with the collector, swinging at the same dynamic voltage as the emitter, plus an appropriate offset. Of course, with bipolars there isn't the necessity to have several volts of headroom in the cascode, as there is with FETs.

Thanks for the comments!

Walt Jung

 

[WaltJ]

Walt - driven cascode is a nice way of putting it. One might want to add telescopic for even more detail. Ie, driven telescopic cascode (DTC)

Yes, and thanks for sharing the sims. As you say, the right cascode makes a whale of a difference.

What models did you use for the 2sk170?

wj

 

[Sigurd Ruschkow]

Here are the 2SK170 JFET models I use. Each model has a specific Vp and corresponding Idss.

******************* 2SK170 *********************************************************

*** N-Channel JFET
.MODEL 2SK170GR_4m1 NJF (AF=492.527m BETA=17.3669M CGD=19.8997p CGS=24p IS=10f
+ KF=.0007956085f LAMBDA=1m RS=5.51589 VTO=-0.50)

*** N-Channel JFET
.MODEL 2SK170V_10m NJF (AF=499.843m BETA=29.5953M CGD=19.8997p CGS=24p IS=10f
+ KF=.001681153f LAMBDA=1m RS=7.19294 VTO=-0.65)

*** N-Channel JFET
.MODEL 2SK170Y_2m NJF (AF=499.84582m BETA=13.6348M CGD=19.8997p CGS=24p IS=10f
+ KF=.002599428f LAMBDA=1m RS=1.74483 VTO=-0.39)

*** N-Channel JFET
.MODEL 2SK170BL_8m NJF (AF=500.504m BETA=27.7612M CGD=19.8997p CGS=24p IS=10f
+ KF=.002229965f LAMBDA=1m RS=8.03465 VTO=-0.60)

*** N-Channel JFET
.MODEL 2SK170V_12m NJF (AF=499.843m BETA=32.0953M CGD=19.8997p CGS=24p IS=10f
+ KF=.001681153f LAMBDA=1m RS=7.19294 VTO=-0.70)

*** N-Channel JFET
.MODEL 2SK170V_18m NJF (AF=499.843m BETA=40.0953M CGD=19.8997p CGS=24p IS=10f
+ KF=.001681153f LAMBDA=1m RS=7.19294 VTO=-0.80)


************************************************************************************
************************************************************************************



Sigurd

Yes, and thanks for sharing the sims. As you say, the right cascode makes a whale of a difference.

What models did you use for the 2sk170?

wj

 

[bcarso]

Here are the 2SK170 JFET models I use. Each model has a specific Vp and corresponding Idss.
...
Sigurd

Those are interesting---could you tell us their origin? I assume that they are still being plugged into a spice model with the basic square law and the abrupt transition from the low-drain-voltage regime to the pinchoff regime?

I pasted a model into my simulator I got from someone that has a substantially higher value for the "BETA" parameter, and I think it is probably wrong for most any SK170 part I've seen. Yours look more likely to conform to the Toshiba databook curves.

Not to get too off-topic here, but years ago I looked at using short-channel devices like these in a quarter-square multiplier, and found that they were quite unsuitable. The ones tested had a very poor fit to the typical JFET equation relating Id to Vgs. Later I noticed a remark in an article by Barrie Gilbert, which was basically a piece about the continued relevance of bipolars in the world of MOS dominance, saying that many FETs hadn't followed the square law we're taught for years. When I corresponded with him about some backup for this he just said Oh look in Tsividis or something


I know that a great deal of effort is spent on accurate models of extremely small geometry MOS parts for low voltage digital simulations, and these are far beyond the humble tools that most of us use. It would be a great service though if someone could generate some more accurate models for parts like the 2SK170, incorporating a more detailed approximation to the actual behavior. Since Linear Integrated Systems seems to be managing to make something similar to the discontinued Toshiba part (LSK170), it would be nice to see how those really work as well, even though there is still a fair amount of NOS of the original.

 

[Sigurd Ruschkow]

First, the 2SK170 is NOT discontinued.

The 2SJ74 is. But there are still loads of them available to buy at good prices.

The exact origin of the model is not known to me. I found it, and modified it. Each of the above models has a suffix, and that suffix is the Idss value at 10VDC Vds. I also changed Vp to fit that Idss.
They also show good noise behaviour.

Yes, they are used in a SPICE simulator.

Try to run some simulations with my models, and you will see that they are fairly accurate.



Sigurd

Those are interesting---could you tell us their origin? I assume that they are still being plugged into a spice model with the basic square law and the abrupt transition from the low-drain-voltage regime to the pinchoff regime?

I pasted a model into my simulator I got from someone that has a substantially higher value for the "BETA" parameter, and I think it is probably wrong for most any SK170 part I've seen. Yours look more likely to conform to the Toshiba databook curves.

Not to get too off-topic here, but years ago I looked at using short-channel devices like these in a quarter-square multiplier, and found that they were quite unsuitable. The ones tested had a very poor fit to the typical JFET equation relating Id to Vgs. Later I noticed a remark in an article by Barrie Gilbert, which was basically a piece about the continued relevance of bipolars in the world of MOS dominance, saying that many FETs hadn't followed the square law we're taught for years. When I corresponded with him about some backup for this he just said Oh look in Tsividis or something


I know that a great deal of effort is spent on accurate models of extremely small geometry MOS parts for low voltage digital simulations, and these are far beyond the humble tools that most of us use. It would be a great service though if someone could generate some more accurate models for parts like the 2SK170, incorporating a more detailed approximation to the actual behavior. Since Linear Integrated Systems seems to be managing to make something similar to the discontinued Toshiba part (LSK170), it would be nice to see how those really work as well, even though there is still a fair amount of NOS of the original.

 

[bcarso]

Thanks! That's great news that at least the SK170 is not discontinued. I know the 2SK389 dual was, and got about 1k of them a while back from a grey market source. Erno B. suggested that more were on their way out, and someone else thought the SK170 had already fallen.

 

[jan.didden]

Re: More papers re. the Baxandall/Swallow CS

Addl' reading that will be helpful to this topic:

"A high output resistance current source", Jaeger, R.C.;
Solid-State Circuits, IEEE Journal of
Volume 9, Issue 4, Aug 1974 Page(s):192 - 194

This is a ckt almost identical to Baxandall/Swallow (B/S). However, it does *not* reference them, but rather Tom Frederiksen:

"A Monolithic High-Power Series Voltage Regulator", Frederiksen, T. M.; Solid-State Circuits, IEEE Journal of
Volume 3, Issue 6, Dec 1968 Page(s):380-387

Within this voltage regulator is a discussion of a PNP output current source that functions similar to the NPN output B/S ckt (P384). It is attributed to Jim Thompson, a fellow designer with Frederiksen at Motorola during this period. From the timelines of the two bodies of work, it would appear that B/S and Thompson/Frederiksen independently developed similar configurations to minimize errors of the output device.

This type of current source became std within IC designs, as is documented by Alan Grebene, in his "Analog IC Circuit Design", Van Nostrand Reinhold, 1972, ISBN 0-442-22827-9.

If anyone has any additional references relevant to the Thompson/Frederiksen current source, please share them (thanks).

A second point which falls in this thread is the driven cascode, as termed "enhanced cascode" by Hawksford. This was seen prior to Hawksford's JAES paper, one example was within Erno Borbely's "High Power High Quality Amplifier Using MOSFETs", Wireless World, March 1983, Pages 69-75. Erno provides some pictures which show the reduction in distortion in curve tracer plots.

A similar type of driver stage to Erno's Fig. 7 was used in the Adcom series 565 and 585 power amplifiers. The distinction here is that for full effectiveness, the cascoding transistor base should be driven from the emitter of the bottom device, so that the collector and base see the same dynamic voltage, and consequently, greatest reduction in distortion.

Walt Jung

Walt,

Great to see you here again!

Erno, if you read this, any additional comments to this concept?

Jan Didden

 

[jam]

Welcome back Mr.Jung.

Jam

 

[anatech]

Hi Walt,
It's really great to see you posting here. I'm hoping that this means you have some time now.

Take it easy and enjoy the place!

Best regards, Chris

 

[WaltJ]

Jan, Jam, and Chris, many thanks for the kind welcome(s). Hope the info mentioned is useful.

Walt

 

[AndrewT]

Walt,
thanks for the superbly informative website.

 

[WaltJ]

You are welcome, AndrewT. I wish I had more time to add to it!

I hope to add some further bib. material on this thread's topic in the near future.

Walt Jung

 

[jan.didden]

Walt,

My source for documentation appears to have dried up. Did you get those you were looking for? Otherwise, a post here might have some succes.

Jan Didden

 

[Sigurd Ruschkow]

Does anyone have a copy of the article "High Power High Quality Amplifier Using MOSFETs" by Borbely?



Sigurd

"A second point which falls in this thread is the driven cascode, as termed "enhanced cascode" by Hawksford. This was seen prior to Hawksford's JAES paper, one example was within Erno Borbely's "High Power High Quality Amplifier Using MOSFETs", Wireless World, March 1983, Pages 69-75. Erno provides some pictures which show the reduction in distortion in curve tracer plots.
"

 

[jan.didden]

If it is an Audio Amateur article I have it. Do you know year, issue?

Jan Didden

 

[bcarso]

Roll off the gain after a couple of MHz - a small cap across I2 should do it.

Jan Didden

Back to the questions at the beginning of this thread:

I'm finding in sim, and supported so far in breadboard, that a single small output capacitor to common as suggested by Jan is quite effective. Things can get messy if lead/trace inductances get significant, so a tight layout is recommended. And as something to be embedded in the middle of a feedback amplifier, especially one that may locally load the output with a frequency-dependent impedance (like a compound e-follower), especially one that can present a negative impedance at some frequencies, the required compensation will necessarily change. See jcx's remarks earlier about this situation.

As a single stage, the output C loading seems to have a disproportionate effect---it is not just a "brute-force" single-pole rolloff. Undoubtedly this would emerge from a proper analytical treatment. Odd that such has not yet shown up in the literature I've seen.

These results hold qualitatively for the various superpairs and variants, whether used as voltage amplifiers (well, really gm cells driving resistive loads) or as current conveyors driven with current signals, or as static current sources in which they are used as a compound common-base-like stage to buffer a simpler current source, or when that latter circuit's output is "probed" for output Z peculiarities by a current generator applied to the output node. However, the optimal output capacitor to common varies in each specific situation.

In cases studied so far, the capacitances required with garden-variety medium-Ft devices are so small as to be likely contributed by most anything the circuit connects to. One breadboard current source for a triode plate supply used a pair with the NPN a 2SC1815 and the output device an MPSW92, with a few hundred uA in the NPN and a few mA out of the 92.

I wish I could help with more bibliography. My thanks to all for this thread and various private communications.

Brad