Simulating non linear magnetics

[Kevin Aylward]

Mike Engelhardt wrote:

Kevin writes:

Maxwell's equations are wrong. This was conclusively proved
by the photo electric effect around the early 1920's.

Best Regards,

Your previous difficulty with inductance, electricity and
magnetism

Ho hum...

could let one to fear

Oh, please don't be frightened on my account.

that ferro-magnetism
with all that remnant flux was false, so I was thinking I
should tape on my refrigerator magnets in place so that
don't pop off lest they get word of your work. Now I
guess I'll have to keep the sunlight off of them too.

Indeed you will.

Best regards to you too,

Well, after a couple of days wading through all those bloody equations I
seem to have come up with a very simple solution to the air gap problem.
As I prior noted, xspice implements non-linear inductors by having a
non-linear resistance core, driven by an l_couple. It seems just adding
a series resister from the l_couple to the core, appropriate to the
reluctance of the air gap, does the job nice and dandy. Its quite weird
really. Increasing the resister, makes the current go up, and linearises
the transfer function, right up to the now sharp saturation point
occurring at a much higher current, as expected.

I'll let you know how it compares to your LTSpice core model once I have
done some more tests on it.

err...best wishes...

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.

 

[Mike Engelhardt]

Kevin writes:

Maxwell's equations are wrong. This was conclusively proved
by the photo electric effect around the early 1920's.

Best Regards,

Your previous difficulty with inductance, electricity and
magnetism could let one to fear that ferro-magnetism
with all that remnant flux was false, so I was thinking I
should tape on my refrigerator magnets in place so that
don't pop off lest they get word of your work. Now I
guess I'll have to keep the sunlight off of them too.

Best regards to you too,

--Mike

 

[Mike Engelhardt]

I'll let you know how it compares to your LTSpice core
model once I have done some more tests on it.

Good luck with it. The LTspice implementation is an exact
analytical implementation of Chan et al's equations and the
gap is also solved exactly as a magnetic circuit as outlined
in LTspice's help documentation. It does not use iteration to
get H from the ampere-turns. This allowed simple analytical
verification.

BTW, the inductance really is discontinuous when H reverses,
so you might want to come to terms with that fact instead of
wondering why the model can't be right. It has nothing to do
with the Chan model's particular minor hysteresis loop shape.
It's just that the slope of B with H *does* change under
reversal of H. It's the way it is. The problem with your
interpretation of that discontinuity is that it doesn't case a
voltage spike, but a voltage jump. And it's a finite jump at that
even when working into an infinite impedance current source. But
having a problem with that would be sort of analogous to saying
the the inductive kick one gets when you open one of the switches
in a SMPS means that there can't be SMPS's. The fact is that
kick is real and of extreme practical utility in that it flips
voltage on the other switch so that it can usually just be a
diode, automatically conducting when it's supposed to in the
SMPS conversion algorithm.

--Mike

 

[Mike Engelhardt]

Analog,

No, that certainly is not a flaw in the Chan et al model.
When a inductor with a magnetic core is transversing a
minor B-H loop, the inductance changes with a discontinuity
then H reverses. This is true for any minor loop, both
symmetric and asymmetric. The corners of the football
shape are as pointy as I can measure. BTW, if you read the
article, the Chan model was proposed in order to remove
discontinuities of the previous Jiles-Atherton model used
in PSpice.

Chan's modeling approach seems to be fundamentally different
than most of the other methods. Chan's approach more or less
starts with positive and negative saturation curves and that
are piecewise summed and/or offset to map out a B-H trajectory.
This seems to lead to smooth curves with a minimum of
computational overhead. Most of the other methods seem to
start with an anhysteretic B-H curve and add in hysteresis as
a correction term.

BTW, I believe the Chan et al. model is fantastically better
than the prior art and that shouldn't get missed here. You've
seen the B-H curves in LTspice's help documentation. Compare
those to that the Jiles-Atherton model does:

http://www.concentric.net/~Pmte/jilesatherton.bmp

This was the hspice deck used to generate it. It's
based on their help documentation.

* plot B vs H for major loop
..tran 1m 25m 0 1u
..probe H=Lx1(k1) B=Lx2(k1)
K1 l1 l2 mag
L1 1 0 nt=20
L2 2 0 nt=20
R11 1 11 1
V11 11 0 sin (0 2 60)
*V11 11 0 PWL(0,0 1,2.5 2,-.07 3,2.5 $ wrong major loop
R22 2 22 1
C22 22 0 1
..model mag l bs=6k br=3k hs=1 hcr=.1 hc=.8 ac=1 lc=16
..end

If you use the commented version of V11, the thing even
returns along the wrong major loop. As far as I know,
the the Chan model, and in particular the implementation
in LTspice, is the state of the art in modelling
ferromagnetic cored inductors in a SPICE program.

--Mike

 

[Kevin Aylward]

Mike Engelhardt wrote:

I'll let you know how it compares to your LTSpice core
model once I have done some more tests on it.

Good luck with it. The LTspice implementation is an exact
analytical implementation of Chan et al's equations

As is my implementation. However, that transient discrepancy, prior
noted, might well need to be accounted for. One of the implementations
must be in error.

and the
gap is also solved exactly as a magnetic circuit as outlined
in LTspice's help documentation. It does not use iteration to
get H from the ampere-turns. This allowed simple analytical
verification.

This is interesting, as all the ways I arranged the equations I ended up
with a messy cubic for the average case. Essentially, replacing H by
(H-alpha.flux) giving flux on both sides of the equation. This is
further complicated by those abs(). So certainly, impressive that you
done the sums, but my resister method saved me much Guinness drinking
time. Or...did you cheat and use a math package?

BTW, the inductance really is discontinuous when H reverses,
so you might want to come to terms with that fact instead of
wondering why the model can't be right.

I agree, that the Chan model has a discontinuous inductance. This has
never been at debate.


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.

 

[Mike Engelhardt]

Kevin,

BTW, the inductance really is discontinuous when H reverses,
so you might want to come to terms with that fact instead of
wondering why the model can't be right.

I agree, that the Chan model has a discontinuous inductance.
This has never been at debate.

Ah, debating with the crown village idiot. That has nothing
to do with the Chan model. It's the physical inductance that
is discontinuous as H reverses. You probably should resolve
for yourself that that doesn't cause a voltage spike but at
worse a step in voltage across the inductor when working into
an infinite impedance current source.

* L1 is a table-driven inductance that jumps from 1H
* to 2H at x=1A V(1) is a theta function as the
* current runs through the discontinuity.
I1 0 1 PWL(0 0 10 10)
L1 1 0 Flux=TBL(x,0,0,1,1,2,3)
*.options maxord=1 ; removes trap ringing from last data pnt
..tran 2
..end

The above deck puts a discontinuity where there is
no reversal of I, to make its effect plainly visible.

--Mike

 

[Chuck Harris]

Hi Mike,

I presume that the discontinuity of inductance
you are discussing occurs at points (1) and (3)
on the B-H curve:

B
^
|
|
(4) -------<-------- (3)
/ | /
/ | /
/ | ^
--------------+------------------>H
v | /
/ | /
/ | /
(1)---------->------ (2)
|
|


L is proportional to mu, and mu = dB/dH

So, if we magnify the area around point (3)



(4)------------<---------------- (3)
-
-
-

(2) -

we find that there is a point at (3) where:


lim mu(H+E) <> lim mu(H-E)
E->0 E->0

or

lim L(H+E) <> lim L(H-E)
E->0 E->0

So, in other words, if you traverse (2)->(3),
when you reach (3), you cannot get back to (2)
by just reducing the current. You must travel
the path from (3)->(4)->(1)->(2).

Hence, inductance is discontinuous at (1) and (3).

Is this basically what you are talking about?

Thanks,

-Chuck Harris


Mike Engelhardt wrote:

Kevin,


BTW, the inductance really is discontinuous when H reverses,
so you might want to come to terms with that fact instead of
wondering why the model can't be right.

I agree, that the Chan model has a discontinuous inductance.
This has never been at debate.


Ah, debating with the crown village idiot. That has nothing
to do with the Chan model. It's the physical inductance that
is discontinuous as H reverses. You probably should resolve
for yourself that that doesn't cause a voltage spike but at
worse a step in voltage across the inductor when working into
an infinite impedance current source.

* L1 is a table-driven inductance that jumps from 1H
* to 2H at x=1A V(1) is a theta function as the
* current runs through the discontinuity.
I1 0 1 PWL(0 0 10 10)
L1 1 0 Flux=TBL(x,0,0,1,1,2,3)
*.options maxord=1 ; removes trap ringing from last data pnt
.tran 2
.end

The above deck puts a discontinuity where there is
no reversal of I, to make its effect plainly visible.

--Mike

 

[Mike Engelhardt]

Chuck,

Kevin writes:

I agree, that the Chan model has a discontinuous inductance.
This has never been at debate.

Ah, debating with the crown village idiot. That has nothing
to do with the Chan model. It's the physical inductance that
is discontinuous as H reverses. You probably should resolve
for yourself that that doesn't cause a voltage spike but at
worse a step in voltage across the inductor when working into
an infinite impedance current source.

* L1 is a table-driven inductance that jumps from 1H
* to 2H at x=1A V(1) is a theta function as the
* current runs through the discontinuity.
I1 0 1 PWL(0 0 10 10)
L1 1 0 Flux=TBL(x,0,0,1,1,2,3)
*.options maxord=1 ; removes trap ringing from last data pnt
.tran 2
.end

The above deck puts a discontinuity where there is
no reversal of I, to make its effect plainly visible.

I presume that the discontinuity of inductance
you are discussing occurs at points (1) and (3)
on the B-H curve:

B
^
|
|
(4) -------<-------- (3)
/ | /
/ | /
/ | ^
--------------+------------------>H
v | /
/ | /
/ | /
(1)---------->------ (2)
|
|

L is proportional to mu, and mu = dB/dH

So, if we magnify the area around point (3)

(4)------------<---------------- (3)
-
-
-

(2) -

we find that there is a point at (3) where:


lim mu(H+E) <> lim mu(H-E)
E->0 E->0

or

lim L(H+E) <> lim L(H-E)
E->0 E->0

So, in other words, if you traverse (2)->(3),
when you reach (3), you cannot get back to (2)
by just reducing the current. You must travel
the path from (3)->(4)->(1)->(2).

Hence, inductance is discontinuous at (1) and (3).

Is this basically what you are talking about?

Yep, exactly. Nice drawing. Ferromagnetics
have a physical discontinuity in ľ when H reverses --
basically as a consequence of remnant flux/permanent
magnetism.

Also, not that you're asking, this kink in the BH
curve, which gives a discontinuity in the ľ and so
also the inductance, doesn't cause a spike in voltage
anyway -- even if it would occur when H didn't
reverse. You'd get a spike if you had to, e.g.,
lift the pencil when you draw the BH curve.

--Mike

 

[Kevin Aylward]

Mike Engelhardt wrote:

Kevin,

Decrepancy:

So, I have now been doing some AC tests on the aforementioned circuit.
The Br issue is still there. Using the series R set to one ohm, I
compared AC response at values of current in each region of the graph
sowed by the transient response in my prior posted circuit, that is, at
40ma, 72 ma and 100ma (vdc=40mv, 72 mv and 100mv). I also manually
calculated the slopes of the transient graphs at these points to
calculate the small signal inductance. Everything matched up. That is
for LT Br=0.283, SS Br=0.250. Both gave AC run 3db points that agreed
with the manual slope calculations at each bias current, and with each
other, such that it is highly unlikely that there is an issue with the
basic structure of the equations being used. Setting LT Br to 0.25 gave
a significant difference in 3db points.

This seems to rule out the integration method as a source of the error
since it is not used in AC analysis.

Have you compared LT's results, in this way, with any other spices that
implement this model?

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.

 

[Mike Engelhardt]

Kevin,

BTW, the inductance really is discontinuous when H reverses,
so you might want to come to terms with that fact instead of
wondering why the model can't be right.

I agree, that the Chan model has a discontinuous inductance.
This has never been at debate.

Ah, debating with the crown village idiot. That has nothing
to do with the Chan model. It's the physical inductance that
is discontinuous as H reverses.

Decrepancy:

So, I have now been doing some AC tests on the aforementioned
circuit. The Br issue is still there...

So now the crown village idiot who trys to discredit the Chan model
by pointing to the fact that it has discontinuities under reversals
of H(just like physical ferromagnetics), keeps coming back because
he wants it so bad so his junk spice product doesn't look so stupid.
Why should I continue to help you when you've been such an ass to
so many posters here? Besides, I don't know why you can't get it
right even in AC. Maybe you think there's room for interpretation
of how to linearize the Chan et al. model to a small signal AC
analysis. LTspice has that correct, too. I see it's missing from
the help documentation but that will be updated as time permits with
an explanation of the correct method -- I'd expect it there in a
month or so. In the meanwhile, forgive me that I'll forgo trying to
further explain inductance, electricity and magnetism to you since
I clearly lack the skill to get through. I don't have a 2x4 and
you're not a good donkey.

* This deck will make the kind of plots you could get out of
* those dear old Ferrox Cube databooks for inductance verses
* DC bias:
* Plot V(n001) to see small signal inductance vs DC bias
L1 N001 0 Hc=16. Bs=.44 Br=.10 A=5u Lm=2m Lg=0.11m N=75
I1 0 N001 {I}
I2 0 N001 ac {1/2/pi}
..ac list 1
..step param I 0 1 10m
..end

Have you compared LT's results, in this way, with any other
spices that implement this model?

Of course. That's why I know that the Chan model has been
implemented incorrectly at other times not including your
attempt. But it's trivial to prove LTspice has it correct.
Check out attached schematic which illustrates that. Compare the
voltage across the Chan device with voltage across an arbitrary
inductance device programmed with the Chan equations or just
compare it the analytical solution solution which is also
drafted in the schematic. All three answers precisely overlap.
To prove your implementation is in error, I step Br to show that
LTspice has the variation of Br correct, as naturally anyone
enough presence of mind to consider the source might expect.

--Mike

--- proof.asc ---

Version 4
SHEET 1 2208 680
WIRE 48 128 48 160
WIRE 48 240 48 288
WIRE -208 128 -48 128
WIRE -208 288 -208 256
WIRE -208 176 -208 128
WIRE 48 -80 48 -48
WIRE 48 32 48 80
WIRE -208 -80 -48 -80
WIRE -208 80 -208 48
WIRE -208 -32 -208 -80
WIRE 784 256 784 224
WIRE 784 144 784 112
WIRE 1040 112 1040 144
WIRE 1040 224 1040 256
WIRE -48 -80 48 -80
WIRE -48 128 48 128
FLAG 48 288 0
FLAG -208 288 0
FLAG 48 80 0
FLAG -208 80 0
FLAG 784 256 0
FLAG 784 112 X
FLAG 1040 256 0
FLAG -48 -80 A
FLAG -48 128 B
FLAG 1040 112 C
SYMBOL ind 32 144 R0
WINDOW 123 36 114 Left 0
WINDOW 39 37 136 Left 0
WINDOW 40 39 160 Left 0
WINDOW 0 38 34 Left 0
WINDOW 3 37 88 Left 0
SYMATTR Value2 A={A} Lm={Lm}
SYMATTR SpiceLine Lg={Lg} N={N}
SYMATTR InstName L1
SYMATTR Value Hc={Hc} Bs={Bs} Br={Br}
SYMBOL current -208 256 M180
WINDOW 0 24 88 Left 0
WINDOW 3 24 0 Left 0
SYMATTR InstName I1
SYMATTR Value PWL(0 0 10 10)
SYMBOL ind 32 -64 R0
WINDOW 0 38 34 Left 0
WINDOW 3 43 71 Left 0
WINDOW 123 46 102 Left 0
SYMATTR InstName L2
SYMATTR Value Flux=.5*({Bs}*(x-{Hc})/(abs({x}-{Hc})+{gamma})
SYMATTR Value2 +{Bs}*(x+{Hc})/(abs({x}+{Hc})+{gamma}))
SYMBOL current -208 48 M180
WINDOW 0 24 88 Left 0
WINDOW 3 24 0 Left 0
SYMATTR InstName I2
SYMATTR Value PWL(0 0 10 10)
SYMBOL bv 1040 128 R0
WINDOW 3 31 97 Left 0
WINDOW 123 31 125 Left 0
SYMATTR Value V=.5*({Bs}*(V(x)-{Hc})/(abs(V(x)-{Hc})+{gamma})
SYMATTR Value2 +{Bs}*(V(x)+{Hc})/(abs(V(x)+{Hc})+{gamma}))
SYMATTR InstName B1
SYMBOL voltage 784 128 R0
SYMATTR InstName V1
SYMATTR Value PWL(0 0 10 10)
TEXT -200 328 Left 0 !.tran 10
TEXT -200 352 Left 0 !.options maxstep=1m
TEXT 432 384 Left 0 !.param Hc=1 Bs=1 Br=.5 A=1 Lm=1 N=1 Lg=0
TEXT 432 360 Left 0 !.param gamma = Hc*(Bs/Br-1)
TEXT -200 376 Left 0 !.options plotwinsize=0
TEXT -200 400 Left 0 !.step param Br list .1 .2 .3
TEXT 112 176 Left 0 ;Intrinsic Chan model Inductor
TEXT 112 -40 Left 0 ;Arbitrary inductor device with the Chan equation
TEXT 720 16 Left 0 ;Analytical equ. of known solution:
TEXT 720 40 Left 0 ;d(V(c))*1s should overlay V(n001)

 

[Kevin Aylward]

Mike Engelhardt wrote:

Kevin,

Besides, I don't know why you can't get it

right even in AC. Maybe you think there's room for interpretation
of how to linearize the Chan et al. model to a small signal AC
analysis.

Not at all. As I explained, the discrepancy is in a value change in the
equations, not the equations themselves. The equations match up
precisely, showing that the basic implementation is indeed correct.

The linearization is nothing more than taking the derivative. The point
is that the discrepancy is in two sets of equations. The basic function,
and the derivatives.

The idea that you would attempt to question my basic math ability
(http://www.anasoft.co.uk/physics/gr/index.html) again shows that you
have a chip on your shoulder. Just what with is it with you that you
must keep doing this?

Sure, its possible that I might have made a minor error, but this has no
relevance to my overall ability. Or they may be something strange in
xspice. This is what I am investigating.

LTspice has that correct, too. I see it's missing from
the help documentation but that will be updated as time permits with
an explanation of the correct method -- I'd expect it there in a
month or so. In the meanwhile, forgive me that I'll forgo trying to
further explain inductance, electricity and magnetism to you since
I clearly lack the skill to get through.

Oh dear...

Have you compared LT's results, in this way, with any other
spices that implement this model?

Of course. That's why I know that the Chan model has been
implemented incorrectly at other times not including your
attempt. But it's trivial to prove LTspice has it correct.
Check out attached schematic which illustrates that.

You obviously lack a basic understanding of "proof". I said an
*independent* proof. This supposed "proof" simply checks LTSpice with
LTSpice. Dah...

I'll certainly have a good look at this, as it does look persuasive. The
issue is that my implementation has apparently, the exact same
equations. You seem to miss the point that you cant, usually,
accidentally get completely wrong equations to match up over a range
with just one parameter change. Especially, since my derivative
calculations match up with the manual calculation from the curves. The
error is simply one of the exact value of Br, which you are making every
effort to imply is proof that I am, yet again, an idiot.

As a side note, you keep on with this inference that everything I do is
crap. This is simply not supportable by the evidence. What I do, for the
most part, works well. If you have a technical agreement, by all means
present it, but please keep your personal insults to yourself. Life is
simply way too short to for me to get involved in this area. Maybe you
should actually re-read what you write, seriously, as if it was someone
else and ask yourself is the type of person you want to come across as.
You do little more than non stop sour grape-ing.

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.

 

[Kevin Aylward]

Mike Engelhardt wrote:

Kevin,


So, I have now been doing some AC tests on the aforementioned
circuit. The Br issue is still there...

So now the crown village idiot who trys to discredit the Chan model
by pointing to the fact that it has discontinuities under reversals
of H(just like physical ferromagnetics),

Not at all. I made no assertion that the Chan model was at fundamental
issue because of H reversals. Would you care to point out exactly where
I made such a claim?

I asserted that there was a basic issue with minor loops rejoining and
*continuing* on to the average curve. This still seems to be the case,
noting that all of my comments on H reversal were speculative, not
assertions. In the continuing on case, H is increasing monotonically and
smoothly, yet there is a supposed step change in slope at the join
point. This seems to be unlikely to me. Some of these slopes intersect
at almost right angles for the Bd>Br case. I certainly agree that I do
not have *detailed* knowledge of magnetic behaviour, as in line with the
majority of other competent EE's, but this purported behaviour seems to
be at fundamental odds with expectations.

I would expect, that starting from any initial point on the BH curve, a
smooth continuous monotonic increase in H will always result in a smooth
monotonic increase in B. This can not happen in the Chan model, and if
this expectation is correct, it means that the Chan model is indeed
flawed. Do you have any real evidence that this is not the case?

This argument leads to the suggestion that the Rugby ball shape may be
erroneous, and that the curve starts shanging shape and heading to the
average curve such that the join up has the same slope, thus more likely
that there may be a cats claw shape at that point *if* H just happens to
be reversed at that point.

Again, I admit that I don't know what the real truth of the Rugby ball
shape, what I will say here, is that it doesn't seem consistent with the
supposition that a continuous smooth increase in H will result in a
continuous smooth increase in B, independent of any initial start point
on the curve. If the real curve does have such a discontinuity in slope
like this, do you have any references that show such a discontinuity? If
it doesn't have such a discontinuous slope, than the Chan model is
flawed, as I asserted. If I am wrong, I can live with it, as this
behaviour would most definitely be rather strange.

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.

 

[Kevin Aylward]

Kevin Aylward wrote:

Mike Engelhardt wrote:
Kevin,

Besides, I don't know why you can't get it


I'll certainly have a good look at this, as it does look persuasive.

Well, what do you know. I ran your current pulse test in SS, and the
results are absolutly identical to LTSpice. That is, the voltage peaks
up and falls for each value of Br, with the exact voltage value all
along the curves agreeing with the same Br in each version. You can run
the test yourself if your so inclined.

The mystery deepens...

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.

 

[Kevin Aylward]

Mike Engelhardt wrote:

Kevin,

Have you compared LT's results, in this way, with any other
spices that implement this model?

Of course. That's why I know that the Chan model has been
implemented incorrectly at other times not including your
attempt. But it's trivial to prove LTspice has it correct.
Check out attached schematic which illustrates that.

You obviously lack a basic understanding of "proof". I
said an *independent* proof. This supposed "proof" simply
checks LTSpice with LTSpice. Dah...

Incorrect.

Ho hummm...dont be so daft Mikey.

There's three *independent* implementations
there.

Nonsense. Its 3 implementations all using LTSpice, therefore they are
all inherently dependant on each other. What part of "they all depend on
LTSpice" do you have trouble with? How you don't understand this point
is pretty amazing really.

In physics, for example, an independent verification is an *independent*
verification, i.e. one made by someone else with *different* physical
equipment. The idea is to eliminate both operator error and equipment
error, in addition to er.. fraud.

You just don't want to follow the proof.

See above..., where you show you are clueless about the scientific
method.

I have already posted that I have put your so called "proof" to a full
test, and that I *agree* with the results of that set-up. SS gives
identical results in that particular set-up.

It
verifies LTspice's intrinsic Chan model implementation with
the equations you can read yourself, also implemented twice
independently,

ROTHLMAO...read up on some physics methods sonny.

once simply plotting the analytical result
and the other by programming the arbitrary inductance with
the Chan model equations for initial saturation. These are
three *independent* implementations,

Nope. See above on the scientific method. Ohh.. I keep forgetting, your
a software engineer...so maybe you still wont understand this
fundamental point.

two of which are not
part of LTspice, but a test case presented to it. QED.

Nope. However, *now* we *do* have an *independent* verification of
LSpice's implementation of the Chan equations for that particular
set-up. Its SuperSpice. I agree absolutely that in the set-up you
presented, the LT version is as correct as the SS implementation. All
results are exactly the same.

So, despite the fact that is demonstrable that, under certain
conditions, LTSpice and SS agree absolutely, on alternative conditions
(the voltage step case) there is a discrepancy. I find this quite
interesting. It certainly shows why the scientific method is so
valuable. So far, we only know that the ramp current test is supported
by two completely independent sets of physical equipment and operators.
What is an open question, is why the disagreement in a different set of
tests. It seems most peculiar that this can occur.

I'll certainly have a good look at this, as it does look
persuasive. The issue is that my implementation has
apparently, the exact same equations.

Actually, I even have a hypothesis where your error is,

Oh...The actual *evidence* to date, shows that where the error is, is an
open question. If you could present the same test using a *different*
simulator this would give support to your assertion here.

Lets hope you do have an idea of where "my" error is, cos if you don't,
its a 50% chance that the error is not mine.

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.

 

[Paul Burridge]

On Sat, 8 May 2004 18:27:15 +0100, "Kevin Aylward"
<kevindotaylwardEXTRACT@anasoft.co.uk> wrote:
[...]

Hello, Kev,

You got the Nobel Prize yet?
;-)
--

"What is now proved was once only imagin'd." - William Blake, 1793.

 

[Mike Engelhardt]

Kevin,

Have you compared LT's results, in this way, with any other
spices that implement this model?

Of course. That's why I know that the Chan model has been
implemented incorrectly at other times not including your
attempt. But it's trivial to prove LTspice has it correct.
Check out attached schematic which illustrates that.

You obviously lack a basic understanding of "proof". I
said an *independent* proof. This supposed "proof" simply
checks LTSpice with LTSpice. Dah...

Incorrect. There's three *independent* implementations
there. You just don't want to follow the proof. It
verifies LTspice's intrinsic Chan model implementation with
the equations you can read yourself, also implemented twice
independently, once simply plotting the analytical result
and the other by programming the arbitrary inductance with
the Chan model equations for initial saturation. These are
three *independent* implementations, two of which are not
part of LTspice, but a test case presented to it. QED.

I'll certainly have a good look at this, as it does look
persuasive. The issue is that my implementation has
apparently, the exact same equations.

Actually, I even have a hypothesis where your error is,
but you would have to read other people's writing pretty
carefully to see it.

--Mike

 

[Mike Engelhardt]

Kevin,

Have you compared LT's results, in this way, with
any other spices that implement this model?

Of course. That's why I know that the Chan model has
been implemented incorrectly at other times not
including your attempt. But it's trivial to prove
LTspice has it correct. Check out attached schematic
which illustrates that.

You obviously lack a basic understanding of "proof".
I said an *independent* proof. This supposed "proof"
simply checks LTSpice with LTSpice. Dah...

Incorrect. There's three *independent* implementations
there. You just don't want to follow the proof. It
verifies LTspice's intrinsic Chan model implementation
with the equations you can read yourself, also
implemented twice independently, once simply plotting
the analytical result and the other by programming the
arbitrary inductance with the Chan model equations for
initial saturation. These are three *independent*
implementations, two of which are not part of LTspice,
but a test case presented to it. QED.

Ho hummm...dont be so daft Mikey. Nonsense. Its 3
implementations all using LTSpice, therefore they are
all inherently dependant on each other. What part of
"they all depend on LTSpice" do you have trouble with?

That circuit certainly verifies that the Chan model
in LTspice agrees with the behavioral implementation
written in equations next to the Chan device. If you
don't believe the behavioral equations, enter different
equations to see those waveform no longer overlay
the LTspice Chan implementation. You can even export
the waveform of the Chan device as text and check it
else where. The circuit should act basically a Rosetta
Stone showing three independent ways of computing the
same thing. That ought to be persuasive enough to make
an honest man go look for his programming bug or
conceptual error. If you meant independent proof as
independent from me, when I can't by definition give
that. It doesn't happen very often in Usenet that such
a clear and concise prove something so emphatically.
But if you can't follow that simple proof I posted, I
probably can't help you anymore.

Anyway, look, I'm sorry I identified you as an idiot.
That's not polite in public forum and I regret it.
However, you started off claiming that LTspice didn't
have the Chan model implemented, but that error was
yours because you didn't realize how abrupt magnetic
saturation could be. You should have thought to check
your work or even just checked out the working examples
in the help file before making such asinine public
claims. Then you try to discredit John Chan's et la.'s
work by pointing out there's a discontinuity under
reversal of H. That's what ferromagnetics do.
Discontinuity of inductance doesn't violate any
principles and only causes a theta function step in
voltage when working into a infinite impedance current
source. So, okay, neither the trouble you had with
the math, physics, or electronics made a good
impression to me. Then that were those strange
comments that my knowledge was limited to software
and that knowledge of inductance, electricity, and
magnetism isn't necessary to implement Chan's model.
I'm really at loss there, coming from a position
of being one of the best educated people in E&M.
Those comments didn't make a favorable impression.
Of course, my view was already tainted from having
seen posts from you in the post.

Anyway, if you study that proof I gave you long
enough, you might figure out your problem. Yes, it
actually is a verification, but I hid the issue I
hypothesize you're having trouble with.

Forgive that I have no interest in using SuperSpice
as check for LTspice verification. Solely as a
matter of my own personal opinion, I don't believe
you have the intellectually integrity to be a rich
source of simulation verification. That's something
people utimately have to decide for themselves.

Best Regards,

--Mike

 

[Paul Burridge]

On 08 May 2004 15:12:41 EDT, "Mike Engelhardt" <pmte@concentric.net>
wrote:

Kevin,

Anyway, look, I'm sorry I identified you as an idiot.

You're not on the Nobel Committee, by any chance, are you, Mike?
;-)
--

"What is now proved was once only imagin'd." - William Blake, 1793.

 

[Terry Given]

"Kevin Aylward" <kevindotaylwardEXTRACT@anasoft.co.uk> wrote in message
news:2c1nc.10$ND.6@newsfe1-win...

Mike Engelhardt wrote:
Kevin,


So, I have now been doing some AC tests on the aforementioned
circuit. The Br issue is still there...

So now the crown village idiot who trys to discredit the Chan model
by pointing to the fact that it has discontinuities under reversals
of H(just like physical ferromagnetics),

Not at all. I made no assertion that the Chan model was at fundamental
issue because of H reversals. Would you care to point out exactly where
I made such a claim?

I asserted that there was a basic issue with minor loops rejoining and
*continuing* on to the average curve. This still seems to be the case,
noting that all of my comments on H reversal were speculative, not
assertions. In the continuing on case, H is increasing monotonically and
smoothly, yet there is a supposed step change in slope at the join
point. This seems to be unlikely to me. Some of these slopes intersect
at almost right angles for the Bd>Br case. I certainly agree that I do
not have *detailed* knowledge of magnetic behaviour, as in line with the
majority of other competent EE's, but this purported behaviour seems to
be at fundamental odds with expectations.

depends on your expectations. To the ill-informed, for example, proximity
effect is totally counterintuitive - add more copper, increase copper loss
exponentially....

I would expect, that starting from any initial point on the BH curve, a
smooth continuous monotonic increase in H will always result in a smooth
monotonic increase in B. This can not happen in the Chan model, and if
this expectation is correct, it means that the Chan model is indeed
flawed. Do you have any real evidence that this is not the case?

This argument leads to the suggestion that the Rugby ball shape may be
erroneous, and that the curve starts shanging shape and heading to the
average curve such that the join up has the same slope, thus more likely
that there may be a cats claw shape at that point *if* H just happens to
be reversed at that point.

Again, I admit that I don't know what the real truth of the Rugby ball
shape, what I will say here, is that it doesn't seem consistent with the
supposition that a continuous smooth increase in H will result in a
continuous smooth increase in B, independent of any initial start point
on the curve. If the real curve does have such a discontinuity in slope
like this, do you have any references that show such a discontinuity? If
it doesn't have such a discontinuous slope, than the Chan model is
flawed, as I asserted. If I am wrong, I can live with it, as this
behaviour would most definitely be rather strange.

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.