|
Re: Warning: Multiple definitions of model ...
开云体育Perhaps it's as well I'm a bit of a dog, then.--
Regards, Tony On 10/07/2023 19:54, John Woodgate
wrote:
|
|
Re: Warning: Multiple definitions of model ...
开云体育Yes, I did mean 'will'. My (limited) experience is that even different data books don't give the same graphical data, and models are no different in that respect.? Yes, you can tweak a model, but most people don't have your insight into the arcane parameters that are used in models. I agree about diodes being calculable, as long as the ohmic resistance component is included, which is rarely specified for small-signal diodes. Curiosity, of course, killed the cat. ? ======================================================================================
Best wishes John Woodgate OOO-Own Opinions Only Rayleigh, Essex UK I hear, and I forget. I see, and I remember. I do, and I understand. Xunzi (340 - 245 BC) On 2023-07-10 18:42, Tony Casey wrote:
I presume you meant "It's unlikely that a model "will* produce curves that closely match those in the data sheet"? Of course that's not true. In many cases, you can tweak a model to produce almost overlay what's in the datasheets. Why not make the model as good as you can? Don't give the "simulations can't be trusted brigade" more ammunition. |
|
Re: .imp file
开云体育I say 'fictitious', because no
such signal with that waveform was the input for the
simulation.? I agree that the result of the inverse FFT is the
impulse response (within limitations). There is quite a lot
about cepstrum analysis on the web, although not millions of
hits, of course. ======================================================================================
Best wishes John Woodgate OOO-Own Opinions Only Rayleigh, Essex UK I hear, and I forget. I see, and I remember. I do, and I understand. Xunzi (340 - 245 BC) On 2023-07-10 18:39, Andy I wrote:
John wrote, "...?an inverse FFT will produce a (fictitious) waveform whose spectrum is the frequency response. I can't see much use for that."? It might be useful to see the impulse response of your network.? Why do you say fictitious? |
|
Re: Warning: Multiple definitions of model ...
I presume you meant "It's unlikely that a model "will* produce curves that closely match those in the data sheet"? Of course that's not true. In many cases, you can tweak a model to almost overlay what's in the datasheets. Why not make the model as good as you can? Don't give the "simulations can't be trusted brigade" more ammunition.
toggle quoted message
Show quoted text
And there again, there are many cases where the model is nowhere "close enough", even for diodes. Basic discrete diodes are quite simple to optimise as there are relatively few model parameters. You don't even need LTspice for this, it can easily be done in a spreadsheet using its internal solver. There's nothing very mysterious about the diode equations. As far as understanding what each parameter does, LTspice gives us the tools to analyse them: just change one and see what it does. Whatever happened to curiosity? --
Regards, Tony On 10/07/2023 17:58, John Woodgate wrote:
|
|
Re: .imp file
开云体育The output from an .AC
simulation is a high-resolution spectrum, so indeed an inverse
FFT will produce a (fictitious) waveform whose spectrum is the
frequency response. I can't see much use for that. But an FFT
of an FFT is a cepstrum, and in some fields, cepstrum analysis
if found useful. ======================================================================================
Best wishes John Woodgate OOO-Own Opinions Only Rayleigh, Essex UK I hear, and I forget. I see, and I remember. I do, and I understand. Xunzi (340 - 245 BC) On 2023-07-10 17:52, Andy I wrote:
Going back to this discussion adding another tidbit: |
|
Re: Warning: Multiple definitions of model ...
开云体育No doubt you can carry out an optimization, but it requires a deep knowledge of what each of the model's parameters determines, which is far from straightforward. It's unlikely that a model with
produce curves that closely match those in the data sheet, and
determining whether a match is 'close enough' also requires
deep understanding unless the 'match' is a gross mismatch. ======================================================================================
Best wishes John Woodgate OOO-Own Opinions Only Rayleigh, Essex UK I hear, and I forget. I see, and I remember. I do, and I understand. Xunzi (340 - 245 BC) On 2023-07-10 16:51, Tony Casey wrote:
Good stuff. For accurate model assessment and optimisation, I also have a method of importing the datasheet data into LTspice for direct comparison. It's this latter bit that's onerous, because it requires digitising the datasheet graphs, which can then be overlaid on the simulated results. You can then derive an error function between the two curves which can be minimised by model parameter optimisation. |
|
Re: Warning: Multiple definitions of model ...
开云体育Good stuff. For accurate model assessment and optimisation, I also have a method of importing the datasheet data into LTspice for direct comparison. It's this latter bit that's onerous, because it requires digitising the datasheet graphs, which can then be overlaid on the simulated results. You can then derive an error function between the two curves which can be minimised by model parameter optimisation.--
Regards, Tony On 10/07/2023 17:37,
aburtonline@... wrote:
Fabulous guys. I had thought it might require me to write my own test setup but it's good to hear that these have been made available. I'm not talking about anything exotic here. I only mean BJTs, JFETs, MOSFETs, valves(?) and the like.? I'm off for a dig around in the files! |
|
Re: Warning: Multiple definitions of model ...
开云体育What you say about this is true, hFE does have significant statistical spreads, but the datasheet limits are probably set at 3 sigma (any outside this being rejected, being only a 0.3% loss). 3 sigma covers 99.7% of the total distribution, whereas 1 sigma covers 68%. 95% of the distribution is contained within 2 sigma, so you are very likely to get something not too far away from the typical (or nominal), and the chance of a random device being near the limit is vanishingly small.BTW, you should also search for "testjig", as it's more often used than "curve tracer". --
Regards, Tony On 10/07/2023 15:34, John Woodgate
wrote:
You put your device and its model into a curve-tracer .ASC and set it up to produce (if it will) the same curves as in the data sheet. Compare the results. You can find curve-tracer .ASCs by going to Files on the web site and Searching for 'curve tracer'. For a bipolar, I would start with collector current as a function of collector voltage with base current as parameter. But remember the large variation of current gain between samples, even for graded devices like BC847x. |
|
Re: Warning: Multiple definitions of model ...
开云体育You put your device and its
model into a curve-tracer .ASC and set it up to produce (if it
will) the same curves as in the data sheet. Compare the
results. You can find curve-tracer .ASCs by going to Files on
the web site and Searching for 'curve tracer'. For a bipolar,
I would start with collector current as a function of
collector voltage with base current as parameter. But remember
the large variation of current gain between samples, even for
graded devices like BC847x. ======================================================================================
Best wishes John Woodgate OOO-Own Opinions Only Rayleigh, Essex UK I hear, and I forget. I see, and I remember. I do, and I understand. Xunzi (340 - 245 BC) On 2023-07-10 13:59,
aburtonline@... wrote:
I have searched for this and can't find anything that is generic and at my level but, is there somewhere information on a sensible validation procedure for the simple classes of model?? I mean something like at best a tutorial video going through the process, or a description of the process and what needs to be validated, or maybe at least a simple step by step guide.? I do hear what you experts say on the models being flawed and it would be good to verify the ones we each use regularly.? It's an area I have al;ways meant to dip into but I've never been able to address the first step. |
|
Re: Periodic signal from PWL file
开云体育That’s what I was initially thinking of (and also posted it). What you are mentioning is BNF (Backus-Naur form of descripion of higler level programming language syntax).Only I was missing the left part of the or-sign in the Wiki. Maybe there were meant to be time/value tuples in angle brackets and they were fallen victim to the HTML conversion? — Christoph
|
Tony Casey
John Woodgate
Christoph