Determining engineering.design tolerances/margins

[kean]

Hi,
can anyone point out what are the usual methods in calculating the
design tolerances of a critical section of a circuit. For example, i
have a maximum limit for a voltage for a particular signal, so how do
i determine what are the margins i need to design in order to be
reasonably sure that the signal will always be below this level?

What is the usual industry practice? Blind rule of thumb or is there
standard estimation procedures?

Thanks.

 

[Joerg]

Hi Kean,

can anyone point out what are the usual methods in calculating the
design tolerances of a critical section of a circuit. For example, i
have a maximum limit for a voltage for a particular signal, so how do
i determine what are the margins i need to design in order to be
reasonably sure that the signal will always be below this level?

What is the usual industry practice? Blind rule of thumb or is there
standard estimation procedures?

You'd have to be more specific. There usually is not only a maximum but
a minimum as well. Then you have to calculate all the tolerances in your
circuits, 1% resistors, capacitive loads and their tolerances and so on.

The way most engineers tackle this is to design a circuit with common
parts. With some experience their tolerance will be almost where it
needs to be, or better. Then they go back and calculate worst cases, all
parts in the direction of lowest gain and then all parts in the other
direction.

Then there is the Monte Carlo method which is often used in chip design
(yield calcs etc.). But in discrete designs that can be, well, what the
name says. There are large casinos in the city state of Monte Carlo .

Regards, Joerg

 

[John Larkin]

Hi,
can anyone point out what are the usual methods in calculating the
design tolerances of a critical section of a circuit. For example, i
have a maximum limit for a voltage for a particular signal, so how do
i determine what are the margins i need to design in order to be
reasonably sure that the signal will always be below this level?

What is the usual industry practice? Blind rule of thumb or is there
standard estimation procedures?

Thanks.

Two methods:

1. Add up all component tolerances worst-case, and ensure that they
can't hit your overall target.

or

2. Assume all components are worst-case, but that their RMS sum, or
some scaled version of same, doesn't hit the limit. Monte Carlo is
essentially a numerical simulation of this situation. *Some*
production units will miss the spec, but not many; fix them, throw
them away, or just sell them anyhow.



2) is more realistic for comercial products where large numbers of
parts are involved. 1) is hyper-conservative, for critical systems or
pickey customers.

"Blind rule of thumb" is nobody's engineering practice.

John

 

[Ted Wilson]

Hi,
can anyone point out what are the usual methods in calculating the
design tolerances of a critical section of a circuit. For example, i
have a maximum limit for a voltage for a particular signal, so how do
i determine what are the margins i need to design in order to be
reasonably sure that the signal will always be below this level?

What is the usual industry practice? Blind rule of thumb or is there
standard estimation procedures?

Thanks.

It's not just a case of the component tolerances - you need to work
out the coefficients for the contribution of each of the tolerances if
you want a rigourous analysis. A 1% resistor does not necessarily
generate a 1% error in the signal of interest.

I've always used partial differentiation down the years and found it
to be an extremely usefull tool

Regards



There is a tide in the affairs of men, which, taken at the flood, . . . suggests we should not inhabit certain low-lying coastal regions.

 

[Paul Hovnanian P.E.]

It's not just a case of the component tolerances - you need to work
out the coefficients for the contribution of each of the tolerances if
you want a rigourous analysis. A 1% resistor does not necessarily
generate a 1% error in the signal of interest.

I've always used partial differentiation down the years and found it
to be an extremely usefull tool

I think this is what is called a sensitivity analysis. It tells you how
the output of a system varies for changes in one or a group of
parameters.

 

[Terry Given]

I think this is what is called a sensitivity analysis. It tells you how
the output of a system varies for changes in one or a group of
parameters.

Bingo. and bloody useful, too. it allows you to concentrate effort where
best results will be achieved.....

Cheers
Terry

 

[Bill Sloman]

It's not just a case of the component tolerances - you need to work
out the coefficients for the contribution of each of the tolerances if
you want a rigourous analysis. A 1% resistor does not necessarily
generate a 1% error in the signal of interest.

I've always used partial differentiation down the years and found it
to be an extremely useful tool

Me too. I was using it for a couple of years before I realised that I
had become an electronic engineer ...