Hello, I'm using a Kalman-filter for state estimation issues. My question is dealing with the signals that I apply to the filter. Let's think of an analog sensor with a changeable low-pass filter at its output. I could think of tuning the RC-elements to a bandwith of let's say 40 Hz or 400 Hz. When I measure the noise levels and variances I can use these values in both cases to feed my Kalman-filter with the measurement covariance-matrices. I wonder if there is a trade-off, maybe from a theoretical point-of-view, in filtering at an early point my signal with a low frequency and then apply it to the filter, or filter with a higher frequency and use the more noisy signal in the Kalman-filter with modelling the noise accordingly. I hope the question becomes clear. TIA, Kai
signals applied to a kalman filter or other estimator
Started by ●September 10, 2008
Reply by ●September 10, 20082008-09-10
On 10 Sep, 10:45, Kai <inva...@invalid.invalid> wrote:> Hello, > > I'm using a Kalman-filter for state estimation issues. My question is > dealing with the signals that I apply to the filter. > > Let's think of an analog sensor with a changeable low-pass filter at its > output.Are you talking about simulations? If so, why deal with analog models? Or are you talking about anti-alias filters?> I could think of tuning the RC-elements to a bandwith of let's say > 40 Hz or 400 Hz. When I measure the noise levels and variances I can use > these values in both cases to feed my Kalman-filter with the measurement > covariance-matrices.You can simulate this without talking about analog electronics.> I wonder if there is a trade-off, maybe from a theoretical point-of-view, in > filtering at an early point my signal with a low frequency and then apply > it to the filter, or filter with a higher frequency and use the more noisy > signal in the Kalman-filter with modelling the noise accordingly.The wider the bandwidth of the pre-filter the more noise reaches the Kalman filter. I would assume that the amount of noise has an impact on th eperformance of the *system*. A narrow-band signal will be relatively insensitive to changes at the input and slow to react, wherea the wide-band signal will give faster reactions but maybe a more 'nervous' system. So you are left with all the usual tradeoffs in system design. Rune