monte carlo/ worst case analysis in pspice

[kristo]

Hi,
I am trying to study the effect of parameter variations of various
components in a circuit. I read i can most likely use monte carlo or
the worst case analysis in pspice. My aim is to find the worst ouput
condition for the various combination of the circuit parmaeter
variations. Please suggest how i can do it. I was thinking to use
worst case analysis. But, i read in the OrCad spice's manunal that
says the worst case analysis does not search for the set of parameter
values that result in worst result. I am confused here. My aim is to
find the set of parameter values that result in the worst result.
Kindly suggest how i can do it. Thanks a lot.
Regards,
kristo

 

[Heinz Schmitz]

kristo wrote:

I am trying to study the effect of parameter variations of various
components in a circuit. I read i can most likely use monte carlo or
the worst case analysis in pspice. My aim is to find the worst ouput
condition for the various combination of the circuit parmaeter
variations.

Imho you feed the Monte Carlo program with a range of possible
variation for every circuit component. Monte Carlo then "throws dice",
determines an accidental value for every component within its given
range of variability and calculates the outcome for that set. And
again and again, as often as you command it to repeat.

Thus you simulate the outcome of a production line, where in 100
produced sets may be 3, which fail to work (worst case). One of the
three, however, may be the only one you build (at home ).

Eg if you choose resistors with +-5%, caps with +-10%, then there may
be a combination of parts that cause your circuit to not to work as
designed (worst case). Thus you may have to buy some of the caps
with +-5% to avoid that.

Regards,
H.

 

[Rich Webb]

On Mon, 25 May 2009 18:14:57 -0700 (PDT), kristo
<krishmaniac@hotmail.com> wrote:

Hi,
I am trying to study the effect of parameter variations of various
components in a circuit. I read i can most likely use monte carlo or
the worst case analysis in pspice. My aim is to find the worst ouput
condition for the various combination of the circuit parmaeter
variations. Please suggest how i can do it.

Monte Carlo analysis simulates operational variation that resulting from
specified component tolerances.

Worst case analysis aims to determine the single "worst" operating point
for your *particular* definition of worst. For example, in an adjustable
linear regulator (LM317) circuit is there a combination of setting
resistors, Iadj, etc. that results in an out of band value for Vout?

I was thinking to use
worst case analysis. But, i read in the OrCad spice's manunal that
says the worst case analysis does not search for the set of parameter
values that result in worst result. I am confused here. My aim is to
find the set of parameter values that result in the worst result.
Kindly suggest how i can do it.

1. Determine the specified guaranteed min/max values or 1-sigma variance
for applicable components.
2. Determine which extreme value will result in your particular "worst
case." In the case of components with a range of variation, determine
how much risk you're willing to accept.
3. Run the numbers.

--
Rich Webb Norfolk, VA

 

[Charlie E.]

On Mon, 25 May 2009 18:14:57 -0700 (PDT), kristo
<krishmaniac@hotmail.com> wrote:

Hi,
I am trying to study the effect of parameter variations of various
components in a circuit. I read i can most likely use monte carlo or
the worst case analysis in pspice. My aim is to find the worst ouput
condition for the various combination of the circuit parmaeter
variations. Please suggest how i can do it. I was thinking to use
worst case analysis. But, i read in the OrCad spice's manunal that
says the worst case analysis does not search for the set of parameter
values that result in worst result. I am confused here. My aim is to
find the set of parameter values that result in the worst result.
Kindly suggest how i can do it. Thanks a lot.
Regards,
kristo

Kristo,
The reason that worst case analysis does not necessarily give you the
actual worst case is the way it is done. During your design, you need
to be sure and specify the tolerances on all your critical parts, such
as resistors, capacitors, inductors, and even on your transistors.
This means that each parts actual value will be in a range of values
around the nominial value that you usually use in simulation.

When you do a Monte Carlo analysis, the software literaly goes to each
of these parts, and throws the dice, and gives you a random value
somewhere inside this range. You normally do many runs in this type
of simulation, anywhere from 30 to 100, so that you get many possible
combinations of your component values. Everytime you run a
simulation, you get a trace of your critical output variable. After
all the runs, you look at that output, and you measure the range of
those responses, and see if they are within your acceptable
performance. It can take quite a bit of time to run the necessary
simulations, especially if you have a large and complex circuit where
each individual simulation takes a lot of time, but the results can be
very informative.

Now, on a worst case analysis, PSpice simplifies things a bit. It
starts with a quick simulation at the nominal value. It then goes to
the first toleranced part, tweaks the value a bit, and then runs
another simulation, and sees if your output value goes up, or down. It
repeats this process on each part, getting a direction of change for
each part. Once it has done this for all the toleranced parts in your
design, it does one final simulation where it takes each part and pegs
its value at the extreme of its tolerance values either up, or down,
in the direction indicated by those earlier simulatons. You get a
choice as to whether you get the max, or the min of the output
variable.

Now, the obvious question is "Why isn't this the guaranteed worst
case?" for your circuit? Well, in real circuits, there are
interactions between all these parts, especially in filters and other
resonances, where the actuall worst possible performance is somewhere
in the midst of all the possible values. The worst case analysis will
miss this combination, but a Monte Carlo analysis, with enough runs,
will probably get some simlulations with values near this worst case.

So, if you have the time and budget, you do MC analysis, especially on
a final circuit design. However, worst case is very informative,
especially in the early design stages.

Oh, and one final note. When you are setting up tolerances, you have
two major choices for specifying the variation of your tolerances -
uniform and guassian. If you chose uniform, then your specified
tolerance is the top and bottom of the expected range. If you choose
guassian, though, then be aware that your tolerance value specifies
the one sigma point. When you do a worst case, it sets the part to
the three sigma point, which is then three times the variation that
many engineers expect! So, be warned... ;-)

Charlie

 

[kristo]

Thanks to Heinz, Webb and Charlie a lot. My special thanks to Charlie
for long and detailed explanation.
kristo.

 

[kristo]

hi,
Any other simulation tool that provides the features of advance
analysis of Pspice .(e.g sensitivity, smoke) etc.
kristo