EDAboard.com | EDAboard.de | EDAboard.co.uk | WTWH Media

weird LT Spice

Ask a question - edaboard.com

elektroda.net NewsGroups Forum Index - Electronics Design - weird LT Spice

Goto page Previous  1, 2


Guest

Fri Feb 08, 2019 8:45 pm   



On Friday, February 8, 2019 at 11:53:27 AM UTC-5, John Larkin wrote:
Quote:
On Fri, 8 Feb 2019 08:34:54 -0800 (PST),
bloggs.fredbloggs.fred_at_gmail.com wrote:

On Thursday, February 7, 2019 at 12:01:21 PM UTC-5, John Larkin wrote:
On Thu, 7 Feb 2019 08:36:55 -0800 (PST),
bloggs.fredbloggs.fred_at_gmail.com wrote:

On Wednesday, February 6, 2019 at 6:34:30 PM UTC-5, John Larkin wrote:
I entered a pretty simple pulse driver circuit. It starts with a
5-volt pulse generator V1, has three fets and some passives, and a 12
volt power supply for the output.

When I run the transient sim, I can plot the waveform of V1, but no
other node. None. If I try to plot any other node, it shows the
waveform grid from -10 to +10 mV but no trace.

That's weird.

Just to test probing things, I added a +1 volt supply all alone off to
the side. Then everything worked.

You have GND missing somewhere whether it's used in the actual circuit or not, you need to rig it. Probe needs a reference.



The circuit is fairly simple and visually correct. It does use some
EPC GaN fets, with an included library, but I've used that many times
before and the .lib file looks OK.

Spice doesn't complain about convergence or take long to run the sim.
It's just as though all nodes except the single pulse generator simply
don't exist, and don't show a plot line if probed. Not even the +12
volt power supply.

And it's erratic; adding a part might fix it or break it.

Does it even work at DC? That's usually easiest to debug. It's probably your fet models.


A grounded +1V DC power supply, connected to nothing in the circuit,
probes as "no line."

I should look at the .RAW file. Probably mostly empty.


Something's wrong with that power supply. They have provisions for a label and then another provision for naming a net. Either you or LTS has the net unspecified.

Quote:


--

John Larkin Highland Technology, Inc

lunatic fringe electronics


John Larkin
Guest

Fri Feb 08, 2019 10:45 pm   



On Fri, 8 Feb 2019 10:46:25 -0800 (PST),
bloggs.fredbloggs.fred_at_gmail.com wrote:

Quote:
On Friday, February 8, 2019 at 11:53:27 AM UTC-5, John Larkin wrote:
On Fri, 8 Feb 2019 08:34:54 -0800 (PST),
bloggs.fredbloggs.fred_at_gmail.com wrote:

On Thursday, February 7, 2019 at 12:01:21 PM UTC-5, John Larkin wrote:
On Thu, 7 Feb 2019 08:36:55 -0800 (PST),
bloggs.fredbloggs.fred_at_gmail.com wrote:

On Wednesday, February 6, 2019 at 6:34:30 PM UTC-5, John Larkin wrote:
I entered a pretty simple pulse driver circuit. It starts with a
5-volt pulse generator V1, has three fets and some passives, and a 12
volt power supply for the output.

When I run the transient sim, I can plot the waveform of V1, but no
other node. None. If I try to plot any other node, it shows the
waveform grid from -10 to +10 mV but no trace.

That's weird.

Just to test probing things, I added a +1 volt supply all alone off to
the side. Then everything worked.

You have GND missing somewhere whether it's used in the actual circuit or not, you need to rig it. Probe needs a reference.



The circuit is fairly simple and visually correct. It does use some
EPC GaN fets, with an included library, but I've used that many times
before and the .lib file looks OK.

Spice doesn't complain about convergence or take long to run the sim.
It's just as though all nodes except the single pulse generator simply
don't exist, and don't show a plot line if probed. Not even the +12
volt power supply.

And it's erratic; adding a part might fix it or break it.

Does it even work at DC? That's usually easiest to debug. It's probably your fet models.


A grounded +1V DC power supply, connected to nothing in the circuit,
probes as "no line."

I should look at the .RAW file. Probably mostly empty.

Something's wrong with that power supply. They have provisions for a label and then another provision for naming a net. Either you or LTS has the net unspecified.



No, it's an initial conditions convergence problem. With no warning.


--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

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

George Herold
Guest

Sat Feb 09, 2019 3:45 am   



On Friday, February 8, 2019 at 9:37:24 PM UTC-5, John Larkin wrote:
Quote:
On Fri, 08 Feb 2019 08:06:50 -0800, John Larkin
jjlarkin_at_highlandtechnology.com> wrote:

On Fri, 8 Feb 2019 07:15:36 -0600, John S <Sophi.2_at_invalid.org> wrote:

On 2/6/2019 5:34 PM, John Larkin wrote:


I entered a pretty simple pulse driver circuit. It starts with a
5-volt pulse generator V1, has three fets and some passives, and a 12
volt power supply for the output.

When I run the transient sim, I can plot the waveform of V1, but no
other node. None. If I try to plot any other node, it shows the
waveform grid from -10 to +10 mV but no trace.

That's weird.

Just to test probing things, I added a +1 volt supply all alone off to
the side. Then everything worked.


Which version? If it is XVII, that may be the problem. Try it with
version IV.

XVII, which is otherwise pretty good. IV isn't supported any more.

I think the problem is that I have some optional peaking inductors in
series with some fet gates. If I set the values small, around expected
circuit parasitics, the equivalent time constants or resonant
frequencies get crazy and the initial conditions solution doesn't
converge. But there is no warning, the sim runs, and the symptom
becomes this invisible node thing.

Removing the inductors, or skipping the initial condition solution,
either one fixes it.

No, wrong again. It now seems like the EPC2038 model breaks LT Spice.

But skipping the initial operating point solution does work.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics


I know nothing, but there is a yahoo LTspice group/thing I've found useful.

George H.

John Larkin
Guest

Sat Feb 09, 2019 3:45 am   



On Fri, 08 Feb 2019 08:06:50 -0800, John Larkin
<jjlarkin_at_highlandtechnology.com> wrote:

Quote:
On Fri, 8 Feb 2019 07:15:36 -0600, John S <Sophi.2_at_invalid.org> wrote:

On 2/6/2019 5:34 PM, John Larkin wrote:


I entered a pretty simple pulse driver circuit. It starts with a
5-volt pulse generator V1, has three fets and some passives, and a 12
volt power supply for the output.

When I run the transient sim, I can plot the waveform of V1, but no
other node. None. If I try to plot any other node, it shows the
waveform grid from -10 to +10 mV but no trace.

That's weird.

Just to test probing things, I added a +1 volt supply all alone off to
the side. Then everything worked.


Which version? If it is XVII, that may be the problem. Try it with
version IV.

XVII, which is otherwise pretty good. IV isn't supported any more.

I think the problem is that I have some optional peaking inductors in
series with some fet gates. If I set the values small, around expected
circuit parasitics, the equivalent time constants or resonant
frequencies get crazy and the initial conditions solution doesn't
converge. But there is no warning, the sim runs, and the symptom
becomes this invisible node thing.

Removing the inductors, or skipping the initial condition solution,
either one fixes it.


No, wrong again. It now seems like the EPC2038 model breaks LT Spice.

But skipping the initial operating point solution does work.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

boB
Guest

Sat Feb 09, 2019 6:45 am   



On Thu, 7 Feb 2019 18:28:58 +0100, Gerhard Hoffmann
<ghf_at_hoffmann-hochfrequenz.de> wrote:

Quote:
Am 07.02.19 um 17:31 schrieb boB:

I have seen LTspice circuits not converge until I added something
weird, for example, I got it to work once by adding just ONE end of a
resistor to the circuit. Later versions did not need it.

Also, sometimes you need to add a high value (several MegOhms)
resistor from a floating node to a grounded node to get it to
converge.

There is no such thing as a floating node in Spice. Everything
has to have some connection to node 0 (gnd). The first thing
that spice does to your circuit is constructing a conductance
matrix and zero or not-completely-0-but-nearly entries mess up
the algorithms that follow. Just imagine computing a voltage
difference between 2 nodes from current and NO conductance. i/0.

To avoid that, Spice normally adds huge resistors to nodes that
are not connected "enough". The size of these resistors is hard
to determine. Make them too large and the equation system will
not converge, make them too small and the user will complain
that the results are not accurate. And resistances/conductance
may change wildly during simulation.



My particular circuit did not need a large value to any other node as
I remember. I know you can choose some of those values. But LTspice
doesn't converge well sometimes. Requires some playing around with
adding extra conductance sometimes. I believe that Mike Englehart
(and others?) have done a lot of work..

John's battery I think he said wasn't connected to anything.
Might be default be connected to ground by some high value. Not sure
why that would help anything though.


Quote:
So you don't add just one end of a resistor to the circuit.
The other end is connected also via the invisible large R.

And WRT LTspice support: If you feed your problem into the
provided channels, it will be fixed pretty soon. Last week I
heard of a case where writing sound files did not work for
resolutions other than 16 bits. That was fixed within hours.

regards,
Gerhard


John Larkin
Guest

Sat Feb 09, 2019 5:45 pm   



On Fri, 08 Feb 2019 21:02:30 -0800, boB <boB_at_K7IQ.com> wrote:

Quote:
On Thu, 7 Feb 2019 18:28:58 +0100, Gerhard Hoffmann
ghf_at_hoffmann-hochfrequenz.de> wrote:

Am 07.02.19 um 17:31 schrieb boB:

I have seen LTspice circuits not converge until I added something
weird, for example, I got it to work once by adding just ONE end of a
resistor to the circuit. Later versions did not need it.

Also, sometimes you need to add a high value (several MegOhms)
resistor from a floating node to a grounded node to get it to
converge.

There is no such thing as a floating node in Spice. Everything
has to have some connection to node 0 (gnd). The first thing
that spice does to your circuit is constructing a conductance
matrix and zero or not-completely-0-but-nearly entries mess up
the algorithms that follow. Just imagine computing a voltage
difference between 2 nodes from current and NO conductance. i/0.

To avoid that, Spice normally adds huge resistors to nodes that
are not connected "enough". The size of these resistors is hard
to determine. Make them too large and the equation system will
not converge, make them too small and the user will complain
that the results are not accurate. And resistances/conductance
may change wildly during simulation.



My particular circuit did not need a large value to any other node as
I remember. I know you can choose some of those values. But LTspice
doesn't converge well sometimes. Requires some playing around with
adding extra conductance sometimes. I believe that Mike Englehart
(and others?) have done a lot of work..

John's battery I think he said wasn't connected to anything.
Might be default be connected to ground by some high value. Not sure
why that would help anything though.


I added a 1 volt supply that was grounded on the low end but not
connected otherwise. That temporarily fixed the problem of the
un-probeable nodes. But then adding another part broke it again. The
power supply just shuffled the circuit a little and stirred up the
random initial condition problem. LT Spice is funny that way.

I didn't have floating nodes unless they were inside the EPC fet
model. I think that LT Spice doesn't like that model. The EPC library
is all text, but it's too complex for me to decode.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

legg
Guest

Sun Feb 10, 2019 4:45 pm   



On Sat, 09 Feb 2019 08:00:06 -0800, John Larkin
<jjlarkin_at_highlandtechnology.com> wrote:

<snip>
Quote:

I added a 1 volt supply that was grounded on the low end but not
connected otherwise. That temporarily fixed the problem of the
un-probeable nodes. But then adding another part broke it again. The
power supply just shuffled the circuit a little and stirred up the
random initial condition problem. LT Spice is funny that way.

I didn't have floating nodes unless they were inside the EPC fet
model. I think that LT Spice doesn't like that model. The EPC library
is all text, but it's too complex for me to decode.


If the issue is performance of GaN models, you could always consult
the LTspice group on Yahoo.

https://groups.yahoo.com/neo/groups/LTspice/conversations/messages

I don't recall anyone else having issues with them, but if you've
loaded the EPC library somewhere in the past, it may be first
reference for the model, even if it or a newer library or model
is located in your working library.

The only way to enforce use of local libraries is to use
..inc.\xxx.lib
rather than the basic
..inc xxx.lib

But that's only one possibility.

I ran into this issue with long forgotten non-standard libraries that
were decades+ old, resident in a new XVI 'over-install'.

RL

John Larkin
Guest

Sun Feb 10, 2019 4:45 pm   



On Sun, 10 Feb 2019 10:03:34 -0500, legg <legg_at_nospam.magma.ca> wrote:

Quote:
On Sat, 09 Feb 2019 08:00:06 -0800, John Larkin
jjlarkin_at_highlandtechnology.com> wrote:

snip

I added a 1 volt supply that was grounded on the low end but not
connected otherwise. That temporarily fixed the problem of the
un-probeable nodes. But then adding another part broke it again. The
power supply just shuffled the circuit a little and stirred up the
random initial condition problem. LT Spice is funny that way.

I didn't have floating nodes unless they were inside the EPC fet
model. I think that LT Spice doesn't like that model. The EPC library
is all text, but it's too complex for me to decode.

If the issue is performance of GaN models, you could always consult
the LTspice group on Yahoo.

https://groups.yahoo.com/neo/groups/LTspice/conversations/messages

I don't recall anyone else having issues with them, but if you've
loaded the EPC library somewhere in the past, it may be first
reference for the model, even if it or a newer library or model
is located in your working library.

The only way to enforce use of local libraries is to use
.inc.\xxx.lib
rather than the basic
.inc xxx.lib

But that's only one possibility.


What's interesting is that I don't have any .inc statement. Spice
seems to use the epc.lib file in the local folder just because it's
there, or referenced through some path not obvious to me. If I rename
it to .liz, it won't run.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

legg
Guest

Tue Feb 12, 2019 4:45 am   



On Sun, 10 Feb 2019 07:29:05 -0800, John Larkin
<jjlarkin_at_highlandtechnology.com> wrote:

Quote:
On Sun, 10 Feb 2019 10:03:34 -0500, legg <legg_at_nospam.magma.ca> wrote:

On Sat, 09 Feb 2019 08:00:06 -0800, John Larkin
jjlarkin_at_highlandtechnology.com> wrote:

snip

I added a 1 volt supply that was grounded on the low end but not
connected otherwise. That temporarily fixed the problem of the
un-probeable nodes. But then adding another part broke it again. The
power supply just shuffled the circuit a little and stirred up the
random initial condition problem. LT Spice is funny that way.

I didn't have floating nodes unless they were inside the EPC fet
model. I think that LT Spice doesn't like that model. The EPC library
is all text, but it's too complex for me to decode.

If the issue is performance of GaN models, you could always consult
the LTspice group on Yahoo.

https://groups.yahoo.com/neo/groups/LTspice/conversations/messages

I don't recall anyone else having issues with them, but if you've
loaded the EPC library somewhere in the past, it may be first
reference for the model, even if it or a newer library or model
is located in your working library.

The only way to enforce use of local libraries is to use
.inc.\xxx.lib
rather than the basic
.inc xxx.lib

But that's only one possibility.

What's interesting is that I don't have any .inc statement. Spice
seems to use the epc.lib file in the local folder just because it's
there, or referenced through some path not obvious to me. If I rename
it to .liz, it won't run.


That's a good indication that there's only one library, the local one,
involved.

RL

Goto page Previous  1, 2

elektroda.net NewsGroups Forum Index - Electronics Design - weird LT Spice

Ask a question - edaboard.com

Arabic version Bulgarian version Catalan version Czech version Danish version German version Greek version English version Spanish version Finnish version French version Hindi version Croatian version Indonesian version Italian version Hebrew version Japanese version Korean version Lithuanian version Latvian version Dutch version Norwegian version Polish version Portuguese version Romanian version Russian version Slovak version Slovenian version Serbian version Swedish version Tagalog version Ukrainian version Vietnamese version Chinese version Turkish version
EDAboard.com map