I found the problem. The regular expressions weren't in Perl format. I seem to have succeeded in fixing them. It needs testing though, especially to make sure the units conversions work. New script attached.
I don't know if the expression for xin is correct. In the scripts it is
xin=xti-ne*xtb
Whereas in the VBIC PDF it is
xin=xti-xtb
On page 13 here, we see why we should use level=9 instead of level=4.
https://www.iee.et.tu-dresden.de/iee/eb/forsch/AK-Bipo/2005/AKB2005_Hammer_VBIC12_benchmarking.pdf
One last thing that might matter, in LTspice there is the option to use "3k3" syntax for 3.3k. The script might need to support this.
I don't know if the expression for xin is correct. In the scripts it is
xin=xti-ne*xtb
Whereas in the VBIC PDF it is
xin=xti-xtb
On page 13 here, we see why we should use level=9 instead of level=4.
https://www.iee.et.tu-dresden.de/iee/eb/forsch/AK-Bipo/2005/AKB2005_Hammer_VBIC12_benchmarking.pdf
One last thing that might matter, in LTspice there is the option to use "3k3" syntax for 3.3k. The script might need to support this.
Dear keantoken,
In the code we have these 2 expressions:
Code:
$r->{'xii'} = ($val->{'xti'} - $val->{'xtb'});
$r->{'xin'} = ($val->{'xti'} - $val->{'ne'} * $val->{'xtb'});To support "3k3" syntax needs some time. Maybe on weekend.
A few regex more ...
Keantoken posted already about the relevance of the QS effect, to add further i think it is enlightning what the vbic web page stated as reasons to implement the VBIC model as an open source model implementation in comparison to the STGP:
Improved Early effect modeling
Quasi-saturation modeling
Parasitic substrate transistor modeling
Parasitic fixed (oxide) capacitance modeling
Includes an avalanche multiplication model
Improved temperature modeling
Base current is decoupled from collector current
Electrothermal modeling
Smooth, continuous model
While i agree that obviously two or three entries are mainly important for ICs, the others help to get better results for audio circuit simulations.
This is a good regular expressions test site. I used PCRE, which I think mean Perl.
Online regex tester and debugger: PHP, PCRE, Python, Golang and JavaScript
Look at the conversion for xii here:
http://www.designers-guide.org/VBIC/documents/VbicText.pdf
I noticed the difference when I discovered xii was a negative value when I converted my KSC1845 model.
Online regex tester and debugger: PHP, PCRE, Python, Golang and JavaScript
Look at the conversion for xii here:
http://www.designers-guide.org/VBIC/documents/VbicText.pdf
I noticed the difference when I discovered xii was a negative value when I converted my KSC1845 model.
Code:
#------------------------------------------------------------------------------
sub normalize_units
#------------------------------------------------------------------------------
{
my $val=shift;
if ($debug) {
print STDERR "normalize units:\t$val\t=>\t";
}
if ($val=~/([\+\-\.0-9]+)f$/i) { $val=$1 * 1e-15; }
elsif ($val=~/([\+\-\.0-9]+)p$/i) { $val=$1 * 1e-12; }
elsif ($val=~/([\+\-\.0-9]+)n$/i) { $val=$1 * 1e-9; }
elsif ($val=~/([\+\-\.0-9]+)u$/i) { $val=$1 * 1e-6; }
elsif ($val=~/([\+\-\.0-9]+)m$/i) { $val=$1 * 1e-3; }
elsif ($val=~/([\+\-\.0-9]+)k$/i) { $val=$1 * 1e+3; }
elsif ($val=~/([\+\-\.0-9]+)meg$/i) { $val=$1 * 1e+6; }
elsif ($val=~/([\+\-\.0-9]+)g$/i) { $val=$1 * 1e+9; }
elsif ($val=~/([\+\-\.0-9]+)t$/i) { $val=$1 * 1e+12; }
if ($debug) {
print STDERR "$val\n";
}
return $val;
}
Code:
0.76f => 7.6e-16
0.8569n => 8.569e-10
100f => 1e-13
100m => 0.1
101.1p => 1.011e-10
105m => 0.105
1.073u => 1.073e-06
10n => 1e-08
10p => 1e-11
113.1p => 1.131e-10
115p => 1.15e-10
118.5p => 1.185e-10
1.2f => 1.2e-15
1.2n => 1.2e-09
1.345p => 1.345e-12
13.6p => 1.36e-11
13p => 1.3e-11
141.1p => 1.411e-10
14.65n => 1.465e-08
1.4f => 1.4e-15
150f => 1.5e-13
150m => 0.15
1.588n => 1.588e-09
15f => 1.5e-14
1.5n => 1.5e-09
166.7f => 1.667e-13
188.8n => 1.888e-07
18p => 1.8e-11
1.9n => 1.9e-09
1k => 1000
1n => 1e-09
200m => 0.2
200n => 2e-07
200p => 2e-10
20m => 0.02
22.03m => 0.02203
227.6p => 2.276e-10
22n => 2.2e-08
230f => 2.3e-13
250n => 2.5e-07
255p => 2.55e-10
2.65p => 2.65e-12
2.77p => 2.77e-12
2.83p => 2.83e-12
28p => 2.8e-11
293.5f => 2.935e-13
295.4p => 2.954e-10
2u => 2e-06
300f => 3e-13
300p => 3e-10
30f => 3e-14
31p => 3.1e-11
330n => 3.3e-07
33f => 3.3e-14
3.5p => 3.5e-12
380p => 3.8e-10
4.033n => 4.033e-09
420p => 4.2e-10
4.2p => 4.2e-12
4.5p => 4.5e-12
4.948p => 4.948e-12
4n => 4e-09
500f => 5e-13
500m => 0.5
500p => 5e-10
50m => 0.05
5.26n => 5.26e-09
550p => 5.5e-10
55m => 0.055
5n => 5e-09
600f => 6e-13
600m => 0.6
6.1n => 6.1e-09
620p => 6.2e-10
62f => 6.2e-14
6.2p => 6.2e-12
6.3p => 6.3e-12
6.4u => 6.4e-06
650f => 6.5e-13
72p => 7.2e-11
750p => 7.5e-10
800.3p => 8.003e-10
800m => 0.8
80p => 8e-11
830p => 8.3e-10
85n => 8.5e-08
86.19m => 0.08619
875p => 8.75e-10
893.1f => 8.931e-13
8n => 8e-09
8p => 8e-12
90m => 0.09
939.8f => 9.398e-13
970p => 9.7e-10Attached the mapping table you hopefully mean. I have inserted the corrected calculation in the script - see this changes here:
Code:
- $val->{$key}=$value;
- if ($key eq 'rbm') {$done->{'rbm'}='yes';}
if (defined $keep->{$key}) {
- $r->{$key}=$value;
+ # will become a comment only!
+ $info->{$key}=$value;
+ } else {
+ $val->{$key}=$value;
}
+ if ($key eq 'rbm') {$done->{'rbm'}='yes';}
+ if ($key eq 'ibe') {$done->{'ibe'}='yes';}
+ if ($key eq 'ibc') {$done->{'ibc'}='yes';}
}
if (!defined($done->{'rbm'})) {
$val->{'rbm'}=$val->{'rb'};
}
- if (!defined($val->{'ibe'})) { # Prefer ibe to is, if it is available
+ if (!defined($done->{'ibe'})) { # Prefer ibe to is, if it is available
$val->{'ibe'}=$val->{'is'};
}
- if (!defined($val->{'ibc'})) { # Prefer ibc to is, if it is available
+ if (!defined($done->{'ibc'})) { # Prefer ibc to is, if it is available
$val->{'ibc'}=$val->{'is'};
}
if ($val->{'ise'} > 1.0) {
@@ -517,13 +558,15 @@
$r->{'eane'} = $val->{'eg'};
$r->{'eanc'} = $val->{'eg'};
$r->{'xis'} = $val->{'xti'};
+
$r->{'xii'} = ($val->{'xti'} - $val->{'xtb'});
- $r->{'xin'} = ($val->{'xti'} - $val->{'ne'} * $val->{'xtb'});
+ # $r->{'xin'} = ($val->{'xti'} - $val->{'ne'} * $val->{'xtb'});
+ $r->{'xin'} = ($val->{'xti'} - $val->{'xtb'}); # corrected according to the url: [URL]http://www.designers-guide.org/VBIC/documents/VbicText.pdf[/URL]Now option '-o outfile' allows to output the results directly into file, not only on stdout. With option '-d' you get many debug lines to check.
BR, Toni
Attachments
Last edited:
There are two slashes in front:
#!//usr/bin/perl
and has not worked.
Tried under msys:
$ ./convert_sgp_modelfile_to_vbic.pl bfs17.mod -o bfs17_vbic.mod
=== WARNING - device BFS17 NPN: IRB parameter is not supported in vbic
=== WARNING - device BFS17 NPN: parameter irb is not supported in vbic
* ------------------------------------------------------------------------------
* BFS17:
* ------------------------------------------------------------------------------
.model BFS17 NPN ( 1
+ rcx=2.68 rci=0 gamm=1.0e-11 vo=0 qco=0 rbx=1.2
+ rbi=3.0e-01 re=5.0e-01 is=4.8e-16 nf=1.008 nr=1.01 fc=8.9e-01
+ cje=1.325e-12 pe=7.0e-01 me=2.2e-01 cjc=4.2e-13 cjep=6.3e-13 pc=6.1e-01
+ mc=2.4e-01 cjcp=0 ps=7.5e-01 ms=0 ibei=4.816617e-18 nei=1.008
+ iben=7.49e-15 nen=1.762 ibci=1.25e-17 nci=1.01 ibcn=2.0e-16 ncn=1.042
+ vef=5.668458e+01 ver=8.2482664 ikf=1.9e-01 nkf=5.0e-01 ikr=9.32e-02 tf=5.694e-11
+ xtf=6.8398e+01 vtf=6.0e-01 itf=7.0e-01 tr=1.0e-09 td=2.08696e-11 ea=1.11
+ eaie=1.11 eaic=1.11 eane=1.11 eanc=1.11 xis=3 xii=1.4
+ xin=1.4 kfn=1.0e-15 afn=1
+ )
* ------------------------------------------------------------------------------
But the -o option has no file generated.
Same with
$ perl convert_sgp_modelfile_to_vbic.pl bfs17.mod -o bfs17_vbic.mod
Kind Regards,
Dietmar
#!//usr/bin/perl
and has not worked.
Tried under msys:
$ ./convert_sgp_modelfile_to_vbic.pl bfs17.mod -o bfs17_vbic.mod
=== WARNING - device BFS17 NPN: IRB parameter is not supported in vbic
=== WARNING - device BFS17 NPN: parameter irb is not supported in vbic
* ------------------------------------------------------------------------------
* BFS17:
* ------------------------------------------------------------------------------
.model BFS17 NPN ( 1
+ rcx=2.68 rci=0 gamm=1.0e-11 vo=0 qco=0 rbx=1.2
+ rbi=3.0e-01 re=5.0e-01 is=4.8e-16 nf=1.008 nr=1.01 fc=8.9e-01
+ cje=1.325e-12 pe=7.0e-01 me=2.2e-01 cjc=4.2e-13 cjep=6.3e-13 pc=6.1e-01
+ mc=2.4e-01 cjcp=0 ps=7.5e-01 ms=0 ibei=4.816617e-18 nei=1.008
+ iben=7.49e-15 nen=1.762 ibci=1.25e-17 nci=1.01 ibcn=2.0e-16 ncn=1.042
+ vef=5.668458e+01 ver=8.2482664 ikf=1.9e-01 nkf=5.0e-01 ikr=9.32e-02 tf=5.694e-11
+ xtf=6.8398e+01 vtf=6.0e-01 itf=7.0e-01 tr=1.0e-09 td=2.08696e-11 ea=1.11
+ eaie=1.11 eaic=1.11 eane=1.11 eanc=1.11 xis=3 xii=1.4
+ xin=1.4 kfn=1.0e-15 afn=1
+ )
* ------------------------------------------------------------------------------
But the -o option has no file generated.
Same with
$ perl convert_sgp_modelfile_to_vbic.pl bfs17.mod -o bfs17_vbic.mod
Kind Regards,
Dietmar
I don't think we can decide which xin expression is correct without consulting an expert. Colin McAndrews wrote both the original script AND that PDF you used to correct the original script. Perhaps Colin knows which one is correct? That said, the default value is 3, so the PDF expression gives a closer value to the default.
Bugfixes!
Revision now:$Id: convert_sgp_modelfile_to_vbic.pl 1603 2019-02-23 11:07:48Z sta $
P.S.: example output:
Revision now:$Id: convert_sgp_modelfile_to_vbic.pl 1603 2019-02-23 11:07:48Z sta $
- fix shebang
- fix missing level parameter
P.S.: example output:
Code:
* ------------------------------------------------------------------------------
* KSC1845E_kq: icrating=0.05 mfg=Fairchild vceo=120
* ------------------------------------------------------------------------------
.model KSC1845E_kq npn ( level=9
+ rcx=1 rci=250 gamm=2.0e-06 vo=500 qco=0 rbx=1.2092e+01
+ rbi=1.44908e+02 re=1.5 is=1.075431e-13 nf=1 nr=1 fc=5.0e-01
+ cje=1.8e-11 pe=7.300286e-01 me=3.619943e-01 cjc=4.525739e-12 cjep=0 pc=5.0e-01
+ mc=3.659045e-01 cjcp=0 ps=7.5e-01 ms=0 ibei=1.790296e-16 nei=1
+ iben=1.98107e-13 nen=2 ibci=7.927984e-15 nci=1 ibcn=1.8378e-11 ncn=1.5
+ vef=4.547994e+01 ver=1.527354e+01 ikf=5.5e-02 nkf=6.6e-01 ikr=1.90546e-02 tf=4.2e-10
+ xtf=1 vtf=0 itf=2.0e-02 tr=0 td=0 ea=1.1809
+ eaie=1.1809 eaic=1.1809 eane=1.1809 eanc=1.1809 xis=3 xii=1.2719
+ xin=1.2719 kfn=0 afn=1
+ )
* ------------------------------------------------------------------------------
* ------------------------------------------------------------------------------
* KSA992F_kq: icrating=0.05 mfg=Fairchild vceo=120
* ------------------------------------------------------------------------------
.model KSA992F_kq pnp ( level=9
+ rcx=1.0e-01 rci=100 gamm=2.0e-07 vo=500 qco=0 rbx=1.6084e+01
+ rbi=1.23916e+02 re=3.0e-01 is=5.7544e-14 nf=1 nr=9.5e-01 fc=5.0e-01
+ cje=1.3e-11 pe=8.55e-01 me=4.104e-01 cjc=2.732866e-12 cjep=6.192234e-12 pc=5.0e-01
+ mc=3.497e-01 cjcp=0 ps=8.0e-01 ms=3.3e-01 ibei=1.653088e-16 nei=1
+ iben=5.7544e-15 nen=1.5 ibci=1.589613e-14 nci=9.5e-01 ibcn=1.8378e-14 ncn=1
+ vef=2.862935e+01 ver=1.250379e+01 ikf=2.98e-01 nkf=5.0e-01 ikr=5.25e-02 tf=4.2e-10
+ xtf=5.0e-01 vtf=0 itf=5 tr=0 td=0 ea=1.1603
+ eaie=1.1603 eaic=1.1603 eane=1.1603 eanc=1.1603 xis=3 xii=1.7151
+ xin=1.7151 kfn=0 afn=1
+ )
* ------------------------------------------------------------------------------Dear keantoken,
have converted your kean9.lib.txt to VBIC models and tried to run a curve trace using all your npn/pnp devices included.
Attached the script to generate the files and results. To automatically generate the png files you need the ngspice gnuplot patch some posts before.
If you edit the resulting e.g. tmp_curve_trace.BC550C_C.inc file
and define the steps and limits which corresponds to the datasheet, you can generate a plot which can be easily compared to the datasheet.
E.g. run the resulting inc file on console:
BR, Toni
have converted your kean9.lib.txt to VBIC models and tried to run a curve trace using all your npn/pnp devices included.
Attached the script to generate the files and results. To automatically generate the png files you need the ngspice gnuplot patch some posts before.
If you edit the resulting e.g. tmp_curve_trace.BC550C_C.inc file
Code:
*
.control
* --------- BJT specific values -------
* ngspice-30 BUG: need a non numeric character and not +/- in front of the bjt name, otherwise the bjt name will be converted to a number
set mybjt="_BC550C_C"
set ibase1=( 2u 4u 6u 8u 10u 12u 14u 16u )
set sweep1end=5
set ibase2=( 0.2u 0.4u 0.6u 0.8u 1u 1.2u 1.4u 1.6u )
set sweep2end=100
* --------- BJT specific values -------
.endc
Q1 net1 net2 0 0 BC550C_C
Vce net1 0 DC=0 AC=0
I1 0 net2 1u
.endE.g. run the resulting inc file on console:
Code:
ngspice -i tmp_curve_trace.BC550C_C.inc curve_tracer.ngspiceAttachments
So now all those models are sitting there dead until someone can go through them. And we hope that person knows what they are doing. And without data, we don't even know if the converted models are better!
Think about it. The only metric we have for how good the converted models are is how well they fit the original SGP models. Which means that without basically reworking the models entirely with strong data, the conversion can only make them worse.
Think about it. The only metric we have for how good the converted models are is how well they fit the original SGP models. Which means that without basically reworking the models entirely with strong data, the conversion can only make them worse.
Don't understand your 
Think the perl script will help you providing you a basic vbic conversion with the most parameters usable. And why not converting all models in one run?There are many versions of model library files out there using the same filename with different values, because no one uses the potential of a version control system on e.g. github. (Maybe due to the problems presenting a modified version using the original from big enterprise...)
So the example of the vbic conversion is only one more version of a file which may be incomplete/unusable or whatever.
Obviously you haven't seen the potential of ngspice + scripting + some tooling to help
BR, Toni
Think the perl script will help you providing you a basic vbic conversion with the most parameters usable. And why not converting all models in one run?There are many versions of model library files out there using the same filename with different values, because no one uses the potential of a version control system on e.g. github. (Maybe due to the problems presenting a modified version using the original from big enterprise...)
So the example of the vbic conversion is only one more version of a file which may be incomplete/unusable or whatever.
Obviously you haven't seen the potential of ngspice + scripting + some tooling to help
- checking and comparing SGP, VBIC ...
- align the output exactly to datasheet values to easier see a difference.
BR, Toni
Yes - this is what I meant before. The transformed model will not have more informations as the original sgp model.
The way can only be to identify (with the knowledge of this forum) the devices showing the effects which are covered by vbic like quasi-saturation, unsufficient accuracy in go or avalanching. Then using your script to generate a starting vbic model and manual fit the responsible parameters for this effect.
E.g. the attached heavy metal guy (and partly his pnp counterpart) show clear quasi-saturation. I am sure that is possible to tweak RCI, GAMM, VO, HRCF to get a more or less accurate mapping of this effect. Not to forget that some other parameter have to modified too.
BR
Dietmar
Attachments
