Topics

NPN-PNP diff pairs; Was 2235 Horizontal Calibration Issue


Tom Lee
 

Hi Ozan,

Yes, it's funny how once you notice something, you suddenly see it everywhere.

In the case of the 741, that NPN-PNP mashup was done to work around the terrible performance of PNPs. By combining NPN followers with PNP common-base pairs, you could get the level shift of a PNP, with the high beta of an NPN. Widlar did it in the LM301 before the 741, but Fullagar had a much less clumsy scheme to set the bias.

I'll take a look at the Gray and Meyer reference. Thanks for finding it!

--Tom

--
Prof. Thomas H. Lee
Allen Ctr., Rm. 205
350 Jane Stanford Way
Stanford University
Stanford, CA 94305-4070
http://www-smirc.stanford.edu

On 1/17/2021 23:37, Ozan wrote:
Hi Tom and keantoken,
Thank you for the information.

I don't want to hijack this thread, maybe we should start a new one if we have more ideas.

I have seen series NPN-PNP pairs before, LM 741 input uses a similar structure
https://www.researchgate.net/figure/Circuit-diagrams-of-two-amplifiers-at-the-top-of-a-widely-used-IC-LM741-type_fig1_237625509
and
MOS class-AB input stage with NMOS-PMOS version has been around for some time:
http://www.cse.psu.edu/~chip/course/analog/lecture/01OpampTut.pdf (Gray & Meyer paper Fig 20).

However, I haven't seen a discrete implementation with differential input and differential output before. It turns out in reality I have but not noticed: My Tektronix 485's horizontal output is NPN-PNP type.

As Tom mentioned it looks like this choice has more to do with the characteristic of the sweep signal, where ramping up and down has different requirements and class-AB operation is desirable.

Ozan








keantoken
 

I see now that C780 and C770 allow the differential pair to briefly exceed their circuit limit in order to charge the output node. I was thinking as if this were a vertical amp rather than horizontal. This circuit makes use of this kind of differential pair's virtually unlimited slew limit in one direction. I wonder whether the differential pair actually turns off during the retrace. Or maybe I have it backward and the differential pair exceeds class A on the retrace rather than the sweep.

Here is an example of multiple used to provide extreme slew rate in both directions:

https://mrevil.asvachin.com/amp/topologies/rush/


Ozan
 

Considering the ramp up is the important portion of the signal, using NPN for + side and PNP for the - side is the right choice as long as PNP performance is decent which is the case for discrete components. You are right for vertical amp the response has to be symmetrical and schematics are symmetrical. In a horizontal driver IC (only one I can find the schematic for), Tek used symmetric NPN diff pair (most likely PNPs didn't perform well in their IC process, may not even exist).

For a history of CMOS class-AB input stages I came across in CMOS design space:
Gray and Meyer Paper I cited earlier
https://ieeexplore.ieee.org/document/1052449
Analysis and Design of Analog Integrated Circuits, Gray, Hurts, Lewis, Meyer, 4th edition page 689.

The circuit in GHLM book looks very much like the "Rush Cascode" but CMOS version. The reference of the book cites
https://ieeexplore.ieee.org/document/1052734 as the source.

(some of these are ieee papers but google may find "free" copies)
Ozan


Tom Lee
 

Hi Ozan,

As far as I am aware, pre-Maxim Tek's IC processes never had a good PNP -- only the standard lateral and grounded-collector vertical. Wink Gross (who teamed with John Addis on the 485 and 7104) even once wrote an article titled something like "You don't need PNPs". I have never been able to find a copy of it, but he apparently made a sufficiently convincing argument that Tek never (to my knowledge, anyway) undertook development of a complementary bipolar process.

Thanks for the CMOS AB references. I haven't looked at those yet, but I see from the dates that they are comparatively recent, relative to their bipolar ancestors. Do those papers cite any prior bipolar implementations?

-- Tom

--
Prof. Thomas H. Lee
Allen Ctr., Rm. 205
350 Jane Stanford Way
Stanford University
Stanford, CA 94305-4070
http://www-smirc.stanford.edu

On 1/20/2021 22:46, Ozan wrote:
Considering the ramp up is the important portion of the signal, using NPN for + side and PNP for the - side is the right choice as long as PNP performance is decent which is the case for discrete components. You are right for vertical amp the response has to be symmetrical and schematics are symmetrical. In a horizontal driver IC (only one I can find the schematic for), Tek used symmetric NPN diff pair (most likely PNPs didn't perform well in their IC process, may not even exist).

For a history of CMOS class-AB input stages I came across in CMOS design space:
Gray and Meyer Paper I cited earlier
https://ieeexplore.ieee.org/document/1052449
Analysis and Design of Analog Integrated Circuits, Gray, Hurts, Lewis, Meyer, 4th edition page 689.

The circuit in GHLM book looks very much like the "Rush Cascode" but CMOS version. The reference of the book cites
https://ieeexplore.ieee.org/document/1052734 as the source.

(some of these are ieee papers but google may find "free" copies)
Ozan




Ozan
 

Hi Tom,
All the paper references either go to Ph.D. theses done under P. Gray or his section in the ancient Analog MOS ICs book. I have the book and the section is pretty much the paper I referenced.

I looked at Grebene's Bipolar and MOS design book, he shows the same class-AB stage (if you have the book page 295) and cites Analog MOS ICs book. I can't find a BJT version in any of the sections. With his BJT design knowledge I expected him to connect the CMOS one to the BJT one.

However, as we discussed earlier LM741 design which is based on LM101/201/301 has a BJT class-AB input stage that pre-dates these papers.

Ozan


Ozan
 

I put together a representative LTSpice schematic. Sim shows the optimized ramp-up response, most of the current drive at highest ramp (5ns/div with x10 mag) is supplied by the caps at the emitters. It is also interesting that the NPN/PNP pair is good old 2n3904/2n3906, the circuit gets the last drop of performance out of them.

If there is interest I can post the LTspice schematic.
Ozan


Dave Peterson
 

Ozan,
I'm an HSpice jockey. I'd love a look at your LTSpice model and come up to speed on component simulation (vs. IC MOS simulation).
Dave

On Thursday, January 21, 2021, 04:05:35 PM PST, Ozan <ozan_g@erdogan.us> wrote:

I put together a representative LTSpice schematic. Sim shows the optimized ramp-up response, most of the current drive at highest ramp (5ns/div with x10 mag) is supplied by the caps at the emitters. It is also interesting that the NPN/PNP pair is good old 2n3904/2n3906, the circuit gets the last drop of performance out of them.

If there is interest I can post the LTspice schematic.
Ozan


Ozan
 

Hi Dave,
I am also in CMOS IC design field, I repair test equipment for hobby. It is unlikely you will see anything you already don't know in the LTspice sim.

I used hspice quite a bit in the past, now mostly spectre. The reason I picked LTspice for this sim is since it is free, and comes nicely packaged with a schematic editor+simulator+waveform viewer anyone in this group could run. As you know although professional CAD/EDA tools are amazing, they are expensive for hobby work.

I will send you the file by private e-mail.

Ozan


Gary Robert Bosworth
 

Do you know anyone in the Los Angeles area who repairs and calibrates
Tektronics equipment?

On Thu, Jan 21, 2021, 19:20 Ozan <ozan_g@erdogan.us> wrote:

Hi Dave,
I am also in CMOS IC design field, I repair test equipment for hobby. It
is unlikely you will see anything you already don't know in the LTspice sim.

I used hspice quite a bit in the past, now mostly spectre. The reason I
picked LTspice for this sim is since it is free, and comes nicely packaged
with a schematic editor+simulator+waveform viewer anyone in this group
could run. As you know although professional CAD/EDA tools are amazing,
they are expensive for hobby work.

I will send you the file by private e-mail.

Ozan






Ozan
 

Hi Gary,
I recommend opening a new thread with this subject line. There must be some members in LA area.

I especially look for non-working equipment to repair; I haven't used any repair services myself.

Professionally we always took it to a calibration lab, or Tektronix itself for traceability. If you are looking for a professional repair place google returns few:

https://www.google.com/search?q=test+equipment+calibration+los+angeles

Ozan