On Apr 28, 5:45�pm, Muzaffer Kal <k...@dspia.com> wrote:
> On Thu, 28 Apr 2011 05:49:19 -0700 (PDT), Rune Allnor
>
> <all...@tele.ntnu.no> wrote:
> >Nope. A Kalman filter might remove a *lot* of
> >noise, but no one will ever claim that it is
> >'as good as possible' or even 'good'.
>
> I seem to remember reading that (if used properly of course) a Kalman
> filter is optimal in MMSE sense so it would be difficult to do better
> given the same constraints and the same input, no?
Define 'used properly'.
These filters require the designer to come up with a
'good' model for the underlying system. Since these
filters *require* system models, they also *depend*
on the system models.
By induction, any error / glitch / flaw / blunder /
inaccuracy in the system model will directly affect
what results are attainable.
Rune
Reply by HardySpicer●April 28, 20112011-04-28
On Apr 28, 9:15�pm, Lox <skolpojke...@yahoo.se> wrote:
> Hello
>
> Is it possible to use a Kalman filter to remove as much measurement
> noise from a sensor signal as possible?
>
> I have seen a Kalman filter used this way when filtering a thermo
> couple to estimate its temperature and temperature derivative.
>
> Would it be possible to use this kind of filtering for any signal, for
> example a rate gyro?
>
> How do I set up the linear model and measurement model for this?
You're better with an H infinity Kalman filter since you don't need to
know the noise covariances.
Reply by Tim Wescott●April 28, 20112011-04-28
On 04/28/2011 08:45 AM, Muzaffer Kal wrote:
> On Thu, 28 Apr 2011 05:49:19 -0700 (PDT), Rune Allnor
> <allnor@tele.ntnu.no> wrote:
>> Nope. A Kalman filter might remove a *lot* of
>> noise, but no one will ever claim that it is
>> 'as good as possible' or even 'good'.
>
> I seem to remember reading that (if used properly of course) a Kalman
> filter is optimal in MMSE sense so it would be difficult to do better
> given the same constraints and the same input, no?
If you know all of the relevant processes exactly, and they are linear,
then the Kalman filter is optimal in the MMSE sense. One of the notable
features of the Kalman filter is that its accuracy is really degraded
when the system doesn't match the model well, however.
If you're into taking stabs in the dark with technologies that you don't
understand, an H-infinity filter is a lot more robust. (A Kalman filter
is an H2 filter, meaning it minimizes the error^2. An H-infinity filter
minimizes the "error^infinity", or more accurately it minimizes the
worst excursion from 'truth'. For whatever reason, this gives a better
filter when the models don't match).
Dan Simon's "Optimal State Estimation" is a good book to get.
--
Tim Wescott
Wescott Design Services
http://www.wescottdesign.com
Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html
>Nope. A Kalman filter might remove a *lot* of
>noise, but no one will ever claim that it is
>'as good as possible' or even 'good'.
I seem to remember reading that (if used properly of course) a Kalman
filter is optimal in MMSE sense so it would be difficult to do better
given the same constraints and the same input, no?
--
Muzaffer Kal
DSPIA INC.
ASIC/FPGA Design Services
http://www.dspia.com
Reply by Rune Allnor●April 28, 20112011-04-28
On Apr 28, 11:15�am, Lox <skolpojke...@yahoo.se> wrote:
> Hello
>
> Is it possible to use a Kalman filter to remove as much measurement
> noise from a sensor signal as possible?
No. The most effective way of removing noise,
is to multiply the recording by 0. That way
all the noise goes away, which is 'as good as
possible'. The downside is, of course, that
any signal of interest also goes away.
> I have seen a Kalman filter used this way when filtering a thermo
> couple to estimate its temperature and temperature derivative.
Nope. A Kalman filter might remove a *lot* of
noise, but no one will ever claim that it is
'as good as possible' or even 'good'. The Kalman
filter is one of many possible approaches for
solving this type of task.
> Would it be possible to use this kind of filtering for any signal, for
> example a rate gyro?
It depends. The Kalman filter is quite good
for the problems where it can be used; the
problem is to decide if it suits the problem
at hand.
The best hammer in the world is no good if your
task is to cut a piece of wood in half.
> How do I set up the linear model and measurement model for this?
The Kalman filter is a rather complicated tool.
You will have to know a lot of linear algenbra,
physics and programming.
Rune
Reply by Lox●April 28, 20112011-04-28
Hello
Is it possible to use a Kalman filter to remove as much measurement
noise from a sensor signal as possible?
I have seen a Kalman filter used this way when filtering a thermo
couple to estimate its temperature and temperature derivative.
Would it be possible to use this kind of filtering for any signal, for
example a rate gyro?
How do I set up the linear model and measurement model for this?