Positive feedback loop strangely converge.

[Fred Bartoli]

It's not the first time I encounter this problem and I don't understand how
the solver can converge.

I've made an LTspice file enclosed at the end.

This is part of a CM reduction feedback loop. The pb is that the loop should
and do converge with E_pb with a -1E6 gain and *shouldn't* for +1E6 gain but
it strangely does, whatever I try.

Anyone have a clue about this ?



--
Thanks,
Fred.


********************
Version 4
SHEET 1 1260 1144
WIRE 48 384 48 448
WIRE 208 384 208 416
WIRE 208 384 48 384
WIRE 208 496 208 624
WIRE 208 624 416 624
WIRE 624 608 624 528
WIRE 624 528 560 528
WIRE 480 528 416 528
WIRE 416 528 416 624
WIRE 416 624 576 624
WIRE 768 944 768 528
WIRE 768 528 624 528
WIRE 768 944 912 944
WIRE 416 672 48 672
WIRE 208 960 208 1056
WIRE 208 736 208 624
WIRE 256 944 768 944
WIRE 416 672 416 896
WIRE 416 896 256 896
WIRE 48 704 48 672
WIRE 48 784 48 800
WIRE 416 672 576 672
WIRE 208 880 208 816
WIRE 48 576 48 528
FLAG 624 688 0
FLAG 208 1056 0
FLAG 48 800 0
FLAG 912 944 out
IOPIN 912 944 Out
FLAG 48 576 0
SYMBOL lib\sym\voltage 48 432 R0
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName V1
SYMATTR Value 10
SYMBOL lib\sym\res 224 512 R180
WINDOW 0 36 76 Left 0
WINDOW 3 36 40 Left 0
SYMATTR InstName R2
SYMATTR Value 100K
SYMBOL lib\sym\e2 624 592 R0
SYMATTR InstName E2
SYMATTR Value 1e6
SYMBOL lib\sym\res 576 512 R90
WINDOW 0 0 56 VBottom 0
WINDOW 3 32 56 VTop 0
SYMATTR InstName R5
SYMATTR Value 100K
SYMBOL lib\sym\voltage 48 688 R0
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName V2
SYMATTR Value 5V
SYMBOL lib\sym\res 192 720 R0
SYMATTR InstName R9
SYMATTR Value 10K
SYMBOL lib\sym\e 208 864 M0
SYMATTR InstName E_pb
SYMATTR Value 1E6
TEXT 14 1128 Left 0 !;op
TEXT 760 1040 Left 0 !.nodeset v(out)=0
TEXT 368 1032 Left 0 !.dc V1 0 10 0.1

 

[Fred Bartoli]

"Fred Bartoli"
<fred._canxxxel_this_bartoli@RemoveThatAlso_free.fr_AndThisToo> a écrit dans
le message de news:416cf528$0$6495$626a14ce@news.free.fr...

It's not the first time I encounter this problem and I don't understand
how
the solver can converge.

I've made an LTspice file enclosed at the end.

This is part of a CM reduction feedback loop. The pb is that the loop
should
and do converge with E_pb with a -1E6 gain and *shouldn't* for +1E6 gain
but
it strangely does, whatever I try.

Anyone have a clue about this ?

Just forgot to mention that this is not an LTspice pb and should be
reproducible with other spices.
At least it is with the other one I have and that's why I tried LTspice as a
doucble check.


--
Thanks,
Fred.

 

[Kevin Aylward]

Fred Bartoli wrote:

It's not the first time I encounter this problem and I don't
understand how the solver can converge.

I've made an LTspice file enclosed at the end.

This is part of a CM reduction feedback loop. The pb is that the loop
should and do converge with E_pb with a -1E6 gain and *shouldn't* for
+1E6 gain but it strangely does, whatever I try.

Anyone have a clue about this ?

You haven't supplied enough info. There is no general reason why
positive feedback loops don't converge. Indeed, they usually do. Why do
you think they shouldn't?

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Helmut Sennewald]

"Fred Bartoli"
<fred._canxxxel_this_bartoli@RemoveThatAlso_free.fr_AndThisToo> schrieb im
Newsbeitrag news:416cf528$0$6495$626a14ce@news.free.fr...

It's not the first time I encounter this problem and I don't understand
how
the solver can converge.

I've made an LTspice file enclosed at the end.

This is part of a CM reduction feedback loop. The pb is that the loop
should
and do converge with E_pb with a -1E6 gain and *shouldn't* for +1E6 gain
but
it strangely does, whatever I try.

Anyone have a clue about this ?

Hello Fred,
I was involved in a discussion about a similar "problem" with a simple
inverting opamp circuit a year ago.
If you use an E-device in an inverting amplifier configuartion, then you
can exchange the + and - input and LTSPICE and PSPICE still converge.

There is really a mathematical solution which fulfills Ohm's law
even when you exachange the + and - input of an opamp. It's just noise
that guarantees that you never can experience that in a real circuit.

The only safe way to discover such a wrong connection is a .TRAN
simulation after you have added e.g. 100p parallel to the 100k feedback
resistor in your circuit.

Best Regards,
Helmut




----- Original Message -----
From: "Helmut Sennewald" <HelmutSennewald@t-online.de>
Newsgroups: de.sci.electronics
Sent: Sunday, November 09, 2003 2:14 PM
Subject: Re: Spannung subtrahieren mit OP

"Helmut Sennewald" <HelmutSennewald@t-online.de> schrieb im Newsbeitrag
news:bojrse$rm8$02$1@news.t-online.com...
"Christof Voelkle" <cvoelkle@yahoo.com> schrieb im Newsbeitrag
news:bojoon$1e6e7i$1@ID-151351.news.uni-berlin.de...
Hallo,

Also bei mir haengt der 2te OpAmp (LT1013) in SwCad3 an oberer
Versorgungsspannung, wenn ich den ersten falschrum habe.

hab beim aufbauen den erstbesten (sprich LT1001) OpAmp genommen. Wenn
ich
den LT1013 nehm wird das ganze so wie ich das erwartet habe. Nur was
ist
am
LT1001 anders?

Hallo Chris,
ich wollte es ja erst nicht glauben, aber dem Modell scheint es echt
egal
zu sein wie herum man es anschließt. Damit nehme ich meine Behauptung
zurück,
daß das nicht sein kann. Da bleibt nur der Umkehrschluß, das Modell vom
LT1001
ist miserabel.


Hallo,
für alle die gerade neu hier einsteigen, es geht um die Frage wieso
LTSPICE beim LT1001 die .OP oder .DC-Analyse funktioniert, obwohl
der Plus- und Minus-Eingang in der Schaltung eines invertierenden
Vertärkers vertauscht waren.
Um der Sache auf den Grund zu gehen habe ich mir mal das SPICE-Modell
vom LT1001 vorgenommen. Ergebnis, es ist in Ordnung.

Dann habe ich einen Verstärker mit nur einer E-source(Gain=100)
und zwei Rückkopplungswoderständen gebaut.
Ergebnis, genau das gleich Problem. Man kann den Plus- und
Minus-Eingang beim invertierenden Verstärker vertauschen und erhält
fast das gleiche Ergebnis. Wenn man sich die simulierten Ausgangs-
spannungen anschaut, dann erfüllen tatsächlich beide Fälle das
Ohmsche Gesetz. SPICE konvergiert bei der falschen Beschaltung
des OPs einfach auf einen theoretischen Grenzfall. Ich habe diese
Schaltung dann noch mit PSPICE(Demo V8) simuliert und hatte genau
das gleiche Problem.
Ganz anders sieht das bei der .TRAN Simulation aus wenn man
Energiespeicher( C, L) in der Schaltung hat. Dann "schwingt"
der Simulator quasi über diese pathologischen Werte hinweg.

In der realen Welt ist das Verharren in diesem Grenzfall
(Arbeitspunkt) nicht möglich, da allein schon durch das Rauschen
die Schaltung bei falscher Beschaltung "wegkippt".


Wie kann man nun aber solche Fehler erkennen oder vermeiden?
Beim LT1001 hat es geholfen bei einer .DC Simulation die Option
"x NoOpIter" im Control Panel->SPICE zu setzen.
Bei der einfachen Schaltung mit der E-Source hilft das aber auch nicht.

Mein Tipp:
Die Simulation im Zeitbereich .TRAN hilft, da dann durch die
Verzögerungen bedingt durch L und C in der Schaltung der Simulator
nicht so leicht auf diese "pathologischen" Werte konvergiert.

Ich habe das Ganze dann mit einer Rampenfunktion(Rise time 1ms)
in einer .TRAN-Simulation getestet.
Dann stellt sich zumindest auch beim LT1001 das richtige Verhalten ein.
Beim LT1001 war hierbei kein besonderes Setting im Control Panel
notwendig. Der hat ja auch einige Cs(Tiefpass) im Modell.

Nur bei der simplen Schaltung unten mir der E-Source war hier ein
zusätzliches Bauteil(Speicher), z.B. C=10p vom Verbindungspunkt
der Widerstände nach Masse, notwendig.
Zuätzlich benötigte diese Schaltung noch die Option "x Noopiter"
im SPICE Control Panel damit die falsche Beschaltung auch zum
Fehler führt.

Hallo Christof,
ich möchte mich besonders bei dir bedanken, daß du auf das
Problem mit dem LT1001 hingewiesen hast. Gestern hatte ich noch
gesagt das kann nicht sein. Heute bin ich um einiges schlauer.

Gruß
Helmut



Die richtige Beschaltung vom +/- Eingang


10k 10k
___ ___
Vin ----___-----o--------___-----
1V | |
| ____ | -0.98V
| | | |
----|- |------o------ Ausgang
| E1 | Vertärkung=100
----|+ |----
| |____| |
| |
GND ------------o--------------o--------



Die falsche Beschaltung vom +/- Eingang


10k 10k
___ ___
Vin ----___-----o--------___-----
1V | |
| ____ | -1.02V
| | | |
----|+ |------o------ Ausgang
| E1 | Vertärkung=100
----|- |----
| |____| |
| |
GND ------------o--------------o--------

 

[Kevin Aylward]

Fred Bartoli wrote:

"Kevin Aylward" <salesEXTRACT@anasoft.co.uk> a écrit dans le message
de news:Icdbd.72726$BI5.15451@fe2.news.blueyonder.co.uk...
Fred Bartoli wrote:
It's not the first time I encounter this problem and I don't
understand how the solver can converge.

I've made an LTspice file enclosed at the end.

This is part of a CM reduction feedback loop. The pb is that the
loop should and do converge with E_pb with a -1E6 gain and
*shouldn't* for +1E6 gain but it strangely does, whatever I try.

Anyone have a clue about this ?

You haven't supplied enough info. There is no general reason why
positive feedback loops don't converge. Indeed, they usually do. Why
do you think they shouldn't?

Hi Kev,
Ok, I've badly worded my thought. It's not that it shouldn't converge
but that it shoud not converge to the values it does.

Of course my first reaction was that I've made a mistake somewhere
since this kind of stuff is pretty obvious, but I've carefully
checked and rechecked to not avail. I'm still convinced I overlooked
something, but what ?

What I meant is that a positive feedback loop, having an overall
positive 10^12 loop gain and some incoming signal should clamp to the
supply rails or go to PB's gigavolts values (DC operating point).

ISTR that you have LT spice installed. If so you should have a look
and try to play with E_pb gain sign. It's pretty obvious from the
schematics.

In that case, the operating point is the same with positive loop gain
and negative loop gain of the same absolute value (10^12).


Strange indeed.

This can happen. In fact I had it happen the other day. I also had a
feedback amp with its inputs swapped.

In many circuits, if the input is effectively zero, the output can be
zero irrespective of whether the feedback is positive or negative.
0*x=0, independent of x. The zero point may actually be a non-zero with
respect to ground.

I'll have a look to see if this is the case here.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Fred Bartoli]

"Kevin Aylward" <salesEXTRACT@anasoft.co.uk> a écrit dans le message de
news:Icdbd.72726$BI5.15451@fe2.news.blueyonder.co.uk...

Fred Bartoli wrote:
It's not the first time I encounter this problem and I don't
understand how the solver can converge.

I've made an LTspice file enclosed at the end.

This is part of a CM reduction feedback loop. The pb is that the loop
should and do converge with E_pb with a -1E6 gain and *shouldn't* for
+1E6 gain but it strangely does, whatever I try.

Anyone have a clue about this ?

You haven't supplied enough info. There is no general reason why
positive feedback loops don't converge. Indeed, they usually do. Why do
you think they shouldn't?

Hi Kev,
Ok, I've badly worded my thought. It's not that it shouldn't converge but
that it shoud not converge to the values it does.

Of course my first reaction was that I've made a mistake somewhere since
this kind of stuff is pretty obvious, but I've carefully checked and
rechecked to not avail. I'm still convinced I overlooked something, but what
?

What I meant is that a positive feedback loop, having an overall positive
10^12 loop gain and some incoming signal should clamp to the supply rails or
go to PB's gigavolts values (DC operating point).

ISTR that you have LT spice installed. If so you should have a look and try
to play with E_pb gain sign. It's pretty obvious from the schematics.

In that case, the operating point is the same with positive loop gain and
negative loop gain of the same absolute value (10^12).


Strange indeed.


--
Thanks,
Fred.