AoE x-Chapters, High-Speed op-amps section, DRAFT

[John Larkin]

On Wed, 24 Apr 2019 15:56:28 -0700 (PDT), "John Miles, KE5FX"
<jmiles@gmail.com> wrote:

On Wednesday, April 24, 2019 at 3:41:06 PM UTC-7, John Larkin wrote:
This all-pass doesn't like current feedback amps.


You're not using 10K feedback resistors, are you? That could lead to
odd behavior with CFB amps.

-- john, KE5FX

No, we started with 300 ohms. We tried 1Ks to lower the BGW of the
amps but that didn't help.

https://www.dropbox.com/s/hanve6f87b8vyec/Allpass_P348.jpg?dl=0

Looks like 1Ks with the LT1818s is good.

LT Spice doesn't (yet?) have some of the fast CFB amps in the library,
like for AD8009. We sort of butchered an old model into LT Spice.




--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[Winfield Hill]

Clifford Heath wrote...

On 25/4/19 9:19 am, John Larkin wrote:
LT Spice doesn't (yet?) have some of the fast CFB amps
in the library, like for AD8009. We sort of
butchered an old model into LT Spice.

It would be good if the x-chapter had tips on how to
simulate these things, where a model is unavailable.

Arrgggh!
Yes, it would be good. Do we have a guest author?


--
Thanks,
- Win

 

[John Miles, KE5FX]

On Wednesday, April 24, 2019 at 4:04:34 AM UTC-7, Winfield Hill wrote:
> Read and comment. AoE x-Chapters,

You might add the LMH6733 to your table. 3 wideband CFB amps with
shutdown pins, somewhat similar to OPA3695.

Also could be worth mentioning that the shutdown pins on many of these
parts are referenced to the positive rail, not ground. That just
hosed me the other day. The shutdown pin on the LMH6733 and similar
parts is best thought of as an analog input, not a digital one.
Some chips like the newer THS3491 have a separate reference pin just
for that purpose, but most don't. For the LMH6733 the disable/enable
thresholds are specified at 3.2V and 3.6V for 5V rails, and you need
to add a volt to them if you max out the rails at +/- 6 volts.

That could be (and was) a problem when relying on an open-drain output
to pull the shutdown pin up to 5 volts, since the current drawn by the
shutdown pin is not negligible. (Worse, the open-drain pin in question
was on an I2C extender whose data sheet specifies 5.5V compatibility at
the GPIO pins without regard to its own supply voltage, but didn't
mention the ESD diode that limits the "open drain" voltage to 1 volt
above the Vdd rail, which was 3.3V in my case.)

-- john, KE5FX

 

[Clifford Heath]

On 25/4/19 9:19 am, John Larkin wrote:

LT Spice doesn't (yet?) have some of the fast CFB amps in the library,
like for AD8009. We sort of butchered an old model into LT Spice.

It would be good if the x-chapter had tips on how to simulate these
things, where a model is unavailable.

 

[Winfield Hill]

gnuarm.deletethisbit@gmail.com wrote...

I've been using the National (is that now TI?) LM8272
for some years now. It's 15 MHz so I guess not much
different. Works a treat for audio work.

Thanks, I'm adding that one in.


--
Thanks,
- Win

 

[Winfield Hill]

John Miles, KE5FX wrote...

On Wednesday, April 24, 2019 at 4:04:34 AM UTC-7, Winfield Hill wrote:
Read and comment. AoE x-Chapters,

You might add the LMH6733 to your table. 3 wideband CFB amps with
shutdown pins, somewhat similar to OPA3695.

Also could be worth mentioning that the shutdown pins on many of these
parts are referenced to the positive rail, not ground. That just
hosed me the other day. The shutdown pin on the LMH6733 and similar
parts is best thought of as an analog input, not a digital one.
Some chips like the newer THS3491 have a separate reference pin just
for that purpose, but most don't. For the LMH6733 the disable/enable
thresholds are specified at 3.2V and 3.6V for 5V rails, and you need
to add a volt to them if you max out the rails at +/- 6 volts.

That could be (and was) a problem when relying on an open-drain output
to pull the shutdown pin up to 5 volts, since the current drawn by the
shutdown pin is not negligible. (Worse, the open-drain pin in question
was on an I2C extender whose data sheet specifies 5.5V compatibility at
the GPIO pins without regard to its own supply voltage, but didn't
mention the ESD diode that limits the "open drain" voltage to 1 volt
above the Vdd rail, which was 3.3V in my case.)

-- john, KE5FX

Excellent note John, we'll try to do the story justice.


--
Thanks,
- Win

 

[John Larkin]

On Thu, 25 Apr 2019 10:10:29 +1000, Clifford Heath
<no.spam@please.net> wrote:

On 25/4/19 9:19 am, John Larkin wrote:
LT Spice doesn't (yet?) have some of the fast CFB amps in the library,
like for AD8009. We sort of butchered an old model into LT Spice.

It would be good if the x-chapter had tips on how to simulate these
things, where a model is unavailable.

One of my guys got an old Pspice model and hacked it into LT Spice. It
was ugly.

https://www.dropbox.com/s/6dt43us3v6rzq8y/Wideband_90_shifter.PNG?dl=0



--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[Allan Herriman]

On Wed, 24 Apr 2019 17:19:53 -0700, John Miles, KE5FX wrote:

On Wednesday, April 24, 2019 at 4:04:34 AM UTC-7, Winfield Hill wrote:
Read and comment. AoE x-Chapters,

You might add the LMH6733 to your table. 3 wideband CFB amps with
shutdown pins, somewhat similar to OPA3695.

Also could be worth mentioning that the shutdown pins on many of these
parts are referenced to the positive rail, not ground. That just hosed
me the other day. The shutdown pin on the LMH6733 and similar parts is
best thought of as an analog input, not a digital one.
Some chips like the newer THS3491 have a separate reference pin just for
that purpose, but most don't. For the LMH6733 the disable/enable
thresholds are specified at 3.2V and 3.6V for 5V rails, and you need to
add a volt to them if you max out the rails at +/- 6 volts.

That could be (and was) a problem when relying on an open-drain output
to pull the shutdown pin up to 5 volts, since the current drawn by the
shutdown pin is not negligible. (Worse, the open-drain pin in question
was on an I2C extender whose data sheet specifies 5.5V compatibility at
the GPIO pins without regard to its own supply voltage, but didn't
mention the ESD diode that limits the "open drain" voltage to 1 volt
above the Vdd rail, which was 3.3V in my case.)

-- john, KE5FX

LTC6228 has a multifunction shutdown pin. Different voltages (wrt the
posive rail) will shutdown the whole opamp, or enable / disable the input
bias current cancellation circuit (allowing one to trade input current
noise for input bias current).

Allan

 

[Winfield Hill]

Allan Herriman wrote...

On Wed, 24 Apr 2019 17:19:53 -0700, John Miles, KE5FX wrote:

On Wednesday, April 24, 2019 at 4:04:34 AM UTC-7, Winfield Hill wrote:
Read and comment. AoE x-Chapters,

You might add the LMH6733 to your table. 3 wideband CFB amps with
shutdown pins, somewhat similar to OPA3695.

Also could be worth mentioning that the shutdown pins on many of these
parts are referenced to the positive rail, not ground. That just hosed
me the other day. The shutdown pin on the LMH6733 and similar parts is
best thought of as an analog input, not a digital one.
Some chips like the newer THS3491 have a separate reference pin just for
that purpose, but most don't. For the LMH6733 the disable/enable
thresholds are specified at 3.2V and 3.6V for 5V rails, and you need to
add a volt to them if you max out the rails at +/- 6 volts.

That could be (and was) a problem when relying on an open-drain output
to pull the shutdown pin up to 5 volts, since the current drawn by the
shutdown pin is not negligible. (Worse, the open-drain pin in question
was on an I2C extender whose data sheet specifies 5.5V compatibility at
the GPIO pins without regard to its own supply voltage, but didn't
mention the ESD diode that limits the "open drain" voltage to 1 volt
above the Vdd rail, which was 3.3V in my case.)

-- john, KE5FX

LTC6228 has a multifunction shutdown pin. Different voltages
(wrt the positive rail) will shutdown the whole opamp, or
enable / disable the input bias current cancellation circuit
(allowing one to trade input current noise for input bias current).

Wow, Allan, that's one fascinating op-amp. It's new and
I had missed it, thanks!


--
Thanks,
- Win

 

[Phil Hobbs]

On 4/24/19 7:04 AM, Winfield Hill wrote:

Read and comment. AoE x-Chapters,
High-Speed op-amps section, DRAFT.

Recall, the AoE x-Chapters are advanced material that
was meant to come after each relevant chapter, can skip
on a quick first read, go back later for detailed info.
But as main book was growing over 2000 pages, we opted to
bring out main book first, including x-Chapter cross refs,
follow with x-Chapter book. As explained in the preface.

26 pages of good stuff, from new sections 4x.5 and 4x.6
Chapter 4x is to supplement H&H AoE III, chapters 4,5,8.
(Full Chapter 4x is now 146 pages long, still growing.)

DRAFT, but getting close to being complete.

Are explanations excellent, good, OK, or confusing?
Examine tables, any of your favorite parts missing?**
Want a separate low-power table? Can we skip that?
Is the CFB op-amp scene explained well enough?
Did you like VFB scatterplots? Need some for CFB?
Have a favorite trick that should be included?
Make comments here or send to winfieldhill@yahoo.com

https://www.dropbox.com/s/aemtyly16mtj87n/Op-amps_High-speed_ch4x_DRAFT.pdf?dl=0

** For op-amps that come in multiple variants, i.e.,
single, dual, quad, we list only one part, and show
# per pkg variants in next column, so you may not see
your part, e.g., we show OPA192, not OPA2192 or OPA4192.
As you can see, we were forced to economize on space.

Awesome.

One interesting part you don't list (except dissing its offset voltage
in an aside) is the LT1228, which is a 75-MHz OTA with a 100-MHz CFA
attached. It's a real OTA, not like that three-terminal OPA861 thing.

The LT1228 is kind of noisy--about 20 dB above shot noise, and worse at
higher current--but you can do a lot with a fast OTA.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Phil Hobbs]

On 4/24/19 6:40 PM, John Larkin wrote:

On Wed, 24 Apr 2019 11:51:42 -0700, John Larkin
jjlarkin@highland_snip_technology.com> wrote:

On 24 Apr 2019 11:34:28 -0700, Winfield Hill
hill@rowland.harvard.edu> wrote:

John Larkin wrote...

On a quick read, I didn't see any mention of c-load
stable amps. That could be an interesting section.

I could scribble up a half page or so on the subject.

Hey, go for it!

In the VFB bold-letter Comments notes, we have AB.
"AB. stable with unlim cap-load." Four types are
marked with the AB comment.

Oops, missed that on a quick read.

Some opamps are unstable over a fairly narrow range of cap loads, but
are happy with big caps. Use the Jim Williams technique: a big enough
capacitor will stabilize anything.


There aren't very many C-load opamps, and my memory
is most of them have fT under 10MHz, which would be
too low for these tables. But a separate section
labelled C-Load, would not have that restriction.

With a big c load, one might not expect much bandwidth!

We developed a fancy I/Q modulator box to simulate a blade-tip sensor
in a jet engine. It looked great in Spice but it oscillates big-time
in real life. We used some really fast current-mode amps, AD8009s. I
sent my guy your opamp tables and he'll use that to pick some slower
(roughly 250-300 MHz) voltage mode amps, which we hope will tame
things.

The misbehaving circuit is an all-pass phase shifter

https://www.dropbox.com/s/nhevg979w7gtxib/90_degree.JPG?dl=0

(from the Williams book) which tops out about 30 MHz. The AD8009s
simulate fine here but oscillate wildly in real life. We just tried
some LT1818s, 400 MHz VFB amps, and they behave.

This all-pass doesn't like current feedback amps.

Interesting. Usually they're OK unless you use too small a feedback
impedance.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Winfield Hill]

Phil Hobbs wrote...

Awesome.

One interesting part you don't list (except dissing
its offset voltage in an aside) is the LT1228, which
is a 75-MHz OTA with a 100-MHz CFA attached. It's a
real OTA, not like that three-terminal OPA861 thing.

The LT1228 is kind of noisy--about 20 dB above shot
noise, and worse at higher current--but you can do
a lot with a fast OTA.

Yes, thanks Phil for point that out. I'm a longtime
fan of the LT1228, and have found it super useful in
several projects. We'll carve out a spot for it.


--
Thanks,
- Win

 

[Phil Hobbs]

On 4/24/19 6:55 PM, John Larkin wrote:

On Wed, 24 Apr 2019 23:17:57 +0200, Gerhard Hoffmann <dk4xp@arcor.de
wrote:

Am 24.04.19 um 21:35 schrieb Joerg:

1. EMI behavior of opamps. This is generally not understood at all by
engineers and (still!) not taught at universities from what young EEs
told me. A bipolar input stage will rectify RF at the first BE junction,
even stuff at cell phone frequencies. This rectification or demodulation
is very inefficient but since that is inside the loop any resultying
baseband AM will hit at full tilt because it happens at "open loop".

It's not that FETs are any worse at demodulating than BJTs, it's just
that they need a higher source impedance for the same dBms. Say, a
different cable transformation.
At the same time their higher Cs make it harder for that match to exist.
The loop gain is the same.

It's much like in HP's app note "Every Schottky diode is a zero bias
detector".



2. Back-to-back input protection diodes between IN+ and IN-. Very often
overlooked. In datasheets they are sometimes only mentioned in a
footnote under the abs max table but often there is only a +/-0.3V diff
limit. Aside from pouring gasoline on the above mentioned EMI issue
these diodes can really throw people a curve when using opamps in an
unorthodox way or as a comparator. Thou shalt not do that but ...

These diodes are anti-parallel, not an efficient rectifier. It's a very
good thing that they are there. A low noise op amp with zenered BE
junctions at the input is no longer a low noise op amp. And an op amp
with 100 dB open loop gain and a Volt between its inputs is a design
error.
Using an op amp as a comparator is not unorthodox, it's wrong.
The diodes are there to protect other stuff from dying.

Some opamps are pretty good comparators, just the right speed for some
applications. AD8033, OPA171.

LM324.

National used to make a combo LM358/LM393 amp/comparator. (LM611 and
614 maybe?)

IIRC JT said that the two were the same silicon with different
metallization.

But for slow stuff that has to be very cheap, using a section of a 324
as a comparator works fine.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Winfield Hill]

Phil Hobbs wrote...

National used to make a combo LM358/LM393 amp/comparator.
(LM611 and 614 maybe?)

LM392, alive and well, multiple manuf.


--
Thanks,
- Win

 

[Winfield Hill]

Phil Hobbs wrote...

This all-pass doesn't like current feedback amps.

Interesting. Usually they're OK unless you use
too small a feedback impedance.

Or too large, insufficient speed.


--
Thanks,
- Win

 

[John Larkin]

On Fri, 26 Apr 2019 13:11:46 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 4/24/19 6:55 PM, John Larkin wrote:
On Wed, 24 Apr 2019 23:17:57 +0200, Gerhard Hoffmann <dk4xp@arcor.de
wrote:

Am 24.04.19 um 21:35 schrieb Joerg:

1. EMI behavior of opamps. This is generally not understood at all by
engineers and (still!) not taught at universities from what young EEs
told me. A bipolar input stage will rectify RF at the first BE junction,
even stuff at cell phone frequencies. This rectification or demodulation
is very inefficient but since that is inside the loop any resultying
baseband AM will hit at full tilt because it happens at "open loop".

It's not that FETs are any worse at demodulating than BJTs, it's just
that they need a higher source impedance for the same dBms. Say, a
different cable transformation.
At the same time their higher Cs make it harder for that match to exist.
The loop gain is the same.

It's much like in HP's app note "Every Schottky diode is a zero bias
detector".



2. Back-to-back input protection diodes between IN+ and IN-. Very often
overlooked. In datasheets they are sometimes only mentioned in a
footnote under the abs max table but often there is only a +/-0.3V diff
limit. Aside from pouring gasoline on the above mentioned EMI issue
these diodes can really throw people a curve when using opamps in an
unorthodox way or as a comparator. Thou shalt not do that but ...

These diodes are anti-parallel, not an efficient rectifier. It's a very
good thing that they are there. A low noise op amp with zenered BE
junctions at the input is no longer a low noise op amp. And an op amp
with 100 dB open loop gain and a Volt between its inputs is a design
error.
Using an op amp as a comparator is not unorthodox, it's wrong.
The diodes are there to protect other stuff from dying.

Some opamps are pretty good comparators, just the right speed for some
applications. AD8033, OPA171.

LM324.

The horror!

National used to make a combo LM358/LM393 amp/comparator. (LM611 and
614 maybe?)

IIRC JT said that the two were the same silicon with different
metallization.

But for slow stuff that has to be very cheap, using a section of a 324
as a comparator works fine.

Cheers

Phil Hobbs

One thing about the classic National LM324 is that if you rail any
section, it wrecks the shared bias supplies for the other sections. So
one comparator switching *really* messes up whatever the other three
amps are doing.

I think some peoples' later versions didn't do that.


--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[Winfield Hill]

John Larkin wrote...

One thing about the classic National LM324 is that
if you rail any section, it wrecks the shared bias
supplies for the other sections.

I remember JT claiming that was the designer's
fault, we should never let an opamp saturate.


--
Thanks,
- Win

 

On Friday, 26 April 2019 22:02:28 UTC+1, John Larkin wrote:

On Fri, 26 Apr 2019 13:11:46 -0400, Phil Hobbs

LM324.

The horror!


National used to make a combo LM358/LM393 amp/comparator. (LM611 and
614 maybe?)

IIRC JT said that the two were the same silicon with different
metallization.

But for slow stuff that has to be very cheap, using a section of a 324
as a comparator works fine.

Cheers

Phil Hobbs

One thing about the classic National LM324 is that if you rail any
section, it wrecks the shared bias supplies for the other sections. So
one comparator switching *really* messes up whatever the other three
amps are doing.

I think some peoples' later versions didn't do that.

The 324 is my favourite opamp ever. You get so much for so little.


NT

 

[John Larkin]

On 26 Apr 2019 14:41:43 -0700, Winfield Hill
<hill@rowland.harvard.edu> wrote:

John Larkin wrote...

One thing about the classic National LM324 is that
if you rail any section, it wrecks the shared bias
supplies for the other sections.

I remember JT claiming that was the designer's
fault, we should never let an opamp saturate.

Single SOT23 opamps don't have a shared bias supply problem!

Many new rrio opamps behave beautifully as comparators. They come off
the rails fast and clean. Some older amps would wind up some internal
node and take forever to unstick. One Intersil part took *seconds*.

https://www.dropbox.com/s/0tf0qyvjphi4yog/AD8033_Rails.JPG?dl=0

I've used RRIO opamps in things like ideal rectifier circuits, just
let'em rail.


--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[Winfield Hill]

John Larkin wrote...

On 26 Apr 2019, Winfield Hill wrote:

John Larkin wrote...

One thing about the classic National LM324 is that
if you rail any section, it wrecks the shared bias
supplies for the other sections.

I remember JT claiming that was the designer's
fault, we should never let an opamp saturate.

Single SOT23 opamps don't have a shared bias supply problem!

Yes, but it's trivial to add one transistor, and
circumvent the problem. I felt we should blame Jim,
rather than have him blaming us for our standard use.


--
Thanks,
- Win