I believe he is referring to internal model control. It is useful for
controlling plants with time delays or a lot of noise because you can
separate the output that is due to noise in the system from the output that
is due to the controller action. There's no point in the controller trying
to correct the current error due to a controller output that happened 15
seconds ago. The only way to fix that error would be to change the
controller output 15 seconds ago, which obviously isn't possible.

A state observer is used to determine a state variable for which it isn't
possible or desirable to directly measure. For example, you can determine
the speed of a DC motor from the back-EMF, but you can't directly measure
the back-EMF. You can however measure the terminal voltage and current and
then calculate the back-EMF (if the resistance is known). In most
applications you could just use a shaft speed sensor, but this is just
meant to be an example. The success of this technique depends strongly on
the accuracy of the model.

Matt

On Fri, Jan 4, 2013 at 5:54 AM, David Sarraf  wrote:

> Randy
>
> The general concept is called a state observer.  You construct a
> mathematical model of a system that runs in parallel with the physical
> system.  It will take the same inputs as the physical system and give an
> output that doesn't have all of the noise that the physical system
> produces.
>
> If you don't know the model parameters the state observer can generate
> them for you.  You make it adaptive by looking at inputs and outputs of the
> physical system and tuning the model parameters until you reduce the error,
> or the disagreement between the model and the system, to an acceptable
> value.
>
> Dave Sarraf
>
>
> >I was reading some of the info on LIGO recently and it appears there is
> >again use for some of this.  As I understood the paper, the concept is to
> >run a simulation of the system behavior in parallel and compare the
> >simulation to the real time measured response.  The feed back loop is then
> >constructed as a smooth predicted value.  This prevents amplification of
> >noise developed in the actual system from feeding back into future
> amplified error.
I believe he is referring to internal model control. It is useful for =
controlling plants with time delays or a lot of noise because you can separ=
ate the output that is due to noise in the system from the output that is d=
ue to the controller action. There's no point in the controller trying =
to correct the current error due to a controller output that happened 15 se=
conds ago. The only way to fix that error would be to change the controller=
 output 15 seconds ago, which obviously isn't possible.

A state observer is used to determine a state variable =
for which it isn't possible or desirable to directly measure. For examp=
le, you can determine the speed of a DC motor from the back-EMF, but you ca=
n't directly measure the back-EMF. You can however measure the terminal=
 voltage and current and then calculate the back-EMF (if the resistance is =
known). In most applications you could just use a shaft speed sensor, but t=
his is just meant to be an example. The success of this technique depends s=
trongly on the accuracy of the model.


Matt

On Fri, Jan 4, 2013 a=
t 5:54 AM, David Sarraf <david.sarraf@.........> wrote:=




Randy
<=
br>The general concept is called a state observer.=A0 You construct a mathe=
matical model of a system that runs in parallel with the physical system.=
=A0 It will take the same inputs as the physical system and give an output =
that doesn't have all of the noise that the physical system produces.=
=A0 




If you don't know the model parameters the state observer can gener=
ate them for you.=A0 You make it adaptive by looking at inputs and outputs =
of the physical system and tuning the model parameters until you reduce the=
 error, or the disagreement between the model and the system, to an accepta=
ble value.=A0 




Dave Sarraf
=A0

>I was reading some of the info on LIGO re=
cently and it appears there is
>again use for some of this.=A0 As I u=
nderstood the paper, the concept is to
>run a simulation of the syste=
m
 behavior in parallel and compare the
>simulation to the real time me=
asured response.=A0 The feed back loop is then
>constructed as a smoo=
th predicted value.=A0 This prevents amplification of
>noise develope=
d in the actual system from feeding back into future amplified error.=A0