Thoughts on new simulator - QSpice

I started using Berkley spice on a main frame and then work paid for PSpice - neither was 'free' in the context of needing computational resources. Decades roll on and new variants of Spice software becomes available in waves, and anyone picking holes in free software that is in the process of being developed is either arrogant or obstinate or naive imho. Obviously the designer has the credibility and experience and insight to propose and develop a better widget - so I can happily wait for it to mature, and dive in at some stage.

Perhaps an analogy is REW, which has continued to develop over recent years. I had other software options going back a decade, but was only too happy to find bugs and propose enhancements for it along the way. I can only hope that the QSpice developer maintains a similar direct input into his software over the coming years, as that is a great recipe for leap-frogging other decades-old platforms.
 
Engelhardt's JFET curve tracer:
 

Attachments

  • QSpice JFET Curve Tracer.png
    QSpice JFET Curve Tracer.png
    899.2 KB · Views: 80
  • Like
Reactions: basreflex
I agree that Mike made nice improvements to the UI and overall ease of use. And, the video I just watched that shows how to use the internal tool to make diode, JFET or MOSFET models from the datasheet in just a few minutes is worth the price of admission as far as I'm concerned 👍 : link
 
  • Like
Reactions: jackinnj
I agree that Mike made nice improvements to the UI and overall ease of use. And, the video I just watched that shows how to use the internal tool to make diode, JFET or MOSFET models from the datasheet in just a few minutes is worth the price of admission as far as I'm concerned 👍 : link
I ported the QSPICE created JFET model of an MMBF5461b to LTSpice. There are a few parameters which LTSPICE doesn't recognize :

Error on line 155 : .model mmbf5461b pjf rs=63.9 rd=63.9 vto=-1.32 beta=920µ lambda=38m ronx=1.07 cgs=12.5p cgd=12.5p eta=75m vtotc=-2m nlev=3 gdsnoi=2 vds=-40 ron=511
* Unrecognized parameter "ronx" -- ignored
* Unrecognized parameter "eta" -- ignored
* Unrecognized parameter "vds" -- ignored
* Unrecognized parameter "ron" -- ignored

Compare to the LTLWIKI Standard JFT model of J2N5461:
.MODEL J2N5461 PJF(Beta=1.699m Betatce=-500m Rd=1 Rs=1 Lambda=23m Vto=-1.883 Vtotc=-2.5m Is=222.4f Isr=2.177p N=1 Nr=2 Xti=3 Alpha=29.8u Vk=400.1 Cgd=2.34p M=482.2m Pb=1 Fc=500m Cgs=2.92p Kf=1.055f Af=1)
 
You mean I have to actually push keys on the keyboard?
And I have to remember that R stands for resistor? Waaaaahhhhh!
And that F2 is delete. F3 is wire. Or whatever 1980s Solaris UI BS. Imagine if as much work had been put into making the LTspice UI usable and intuitive as went into loading that completely useless background image or changing the mouse cursor graphics.

QSpice does have some of the legacy feel of LTspice. No surprise. But at least from my quick poking around in it, it seems pretty usable. I would have preferred a dialog box to set up voltage sources rather than having to remember the SPICE syntax for, say, a piece-wise linear, but QSpice does give you the syntax, so I won't cry myself to sleep over it.
I'm not sure I'll use the ability to include behavioural models, but having that feature could be very cool for mixed-signal circuits.

The drag-n-drop model import blew my mind! I hope that works on the Mac once QSpice is ported through Wineskin.

I think it'll be very much worth learning. I'd also keep an eye on the simulation abilities of KiCAD. Last I checked (1-2 years ago) it wasn't super useful. It had no ability to simulate distortion, for example. Not that I could immediately find anyway. But KiCAD has come a very long way the past few years, so it's worth having another look at the simulator in it.

Tom
 
I have very little in LTspice. Maybe 2-3 circuits that I haven't touched in quite a while. I have vastly more in TINA-TI, which is no longer TI's preferred tool. So my choices are to go back to LTspice or to learn something new. For me, it makes more sense to learn QSpice or the simulator in KiCAD (NGspice, I think). But each to their own.

I would actually consider booting up windows to run PSpice for TI, but last I tried I could not import a 3rd party model. I think I tried various ways for 2-3 hours before deciding to move on. It's supposed to be able to. I used PSpice in the late 1990s and it was a pretty solid tool even back then.

Tom
 
I ported the QSPICE created JFET model of an MMBF5461b to LTSpice. There are a few parameters which LTSPICE doesn't recognize :

Error on line 155 : .model mmbf5461b pjf rs=63.9 rd=63.9 vto=-1.32 beta=920µ lambda=38m ronx=1.07 cgs=12.5p cgd=12.5p eta=75m vtotc=-2m nlev=3 gdsnoi=2 vds=-40 ron=511
* Unrecognized parameter "ronx" -- ignored
* Unrecognized parameter "eta" -- ignored
* Unrecognized parameter "vds" -- ignored
* Unrecognized parameter "ron" -- ignored

Compare to the LTLWIKI Standard JFT model of J2N5461:
.MODEL J2N5461 PJF(Beta=1.699m Betatce=-500m Rd=1 Rs=1 Lambda=23m Vto=-1.883 Vtotc=-2.5m Is=222.4f Isr=2.177p N=1 Nr=2 Xti=3 Alpha=29.8u Vk=400.1 Cgd=2.34p M=482.2m Pb=1 Fc=500m Cgs=2.92p Kf=1.055f Af=1)
Unless I'm reading it wrong, there's quite some difference between the QSPICE created model parameters and those from LTWIKI, which I presume came from some manufacturer.

I'm no expert, but I don't think that both can be accurate. Maybe neither one is. Ideas??
 
I'm no expert, but I don't think that both can be accurate. Maybe neither one is. Ideas??
In one of Mike's YT videos he explains that previous versions of spice (I take it any spice including his own LTspice) didn't model JFETs properly because the spread in parameters from or in actual manufacturing was so large that no one cared. But now he works for Qorvo, who makes JFETs, and therefore he is addressing better JFET models, only now, in qspice.

My issue is trying to find a way to get qspice running in Linux as I am moving there. Not under a VM, for speed, but bare metal via Wine, or at least containerized under Bottles. Perusing the qspice forum there was a thread about Linux circa 2023 where Mike said "no" regarding Linux because he needs DirectX 12 graphics support and that wasn't available under Wine back then. But things have changed. There have been great strides in DirectX 12 for gaming. I'm hoping that, thanks to gamers, qspice will be unlocked in Linux.

P.S.
I know that LTSpice can run under Wine but qspice seems to converge much better than LTspice and many of my simulations take forever to converge, or don't converge at all under LTspice. Also, Micro-Cap might also run under Wine and possibly converges fast enough to run under a VM. So I will try that too. But Micro-Cap produces unexpected results with transformers and besides, eventually it will break due t an OS update as it's not being updated. So it seems that qspice is the likely future for me.
 
Actually, playing with the qspice demo files is fairly impressive. The Tian example allows for the stepping of many stability runs to yield a plot of the phase margin away from zero volts. Plotting -10v through 0v and continuing through +10v was very revealing. It ran 200 steps in 24 seconds. It reveals that the 164093-100w-ultimate-fidelity-amplifier has as little as 19 degrees of phase margin away from zero volts, whereas a Tian run at zero volts shows an apparent 73 degrees of phase margin for the amplifier. This is the circuit found at this link here at diyaudio
https://www.diyaudio.com/forums/solid-state/164093-100w-ultimate-fidelity-amplifier.html
See the attached plot of the phase margin at voltages from -10 to +10 volts as the bias point.

The demo duplicates the circuit. I wouldn't want to do it that way as I'd rather run the Tian / Weideman runs using a probe subcircuit which is stepped. I believe I can still do that as with LTspice.
 

Attachments

  • 164093-100w-ultimate-fidelity-amplifier - Tian -10 to 10 volts.jpg
    164093-100w-ultimate-fidelity-amplifier - Tian -10 to 10 volts.jpg
    72.5 KB · Views: 8