I am trying to implement the paper Adaptive Kalman Filter for INS/GPS and there are a couple of expressions where the paper says that it comes from the Standard Kalman Filter theory, but I can't see why.

In the Annex a. Equation A1 says $$ C_{v_k}^{-1}v = R^{-1}v $$ Being $ C_v $ the covariance of the innovation, $ v$ the innovation itself and $R$ the measurement model covariance. I just can't see why is that.

What I think that this only happens only two conditions. In the optimal case, $$ C_{v_k} = HP_{k|k}H^T + R $$ Being $H$ the measurement model matrix and $P_{k|k}$ the estimated covariance matrix. It seems that for the first equation to be true, the first term of this last equation should disappear.

So I am guessin that for it to be true, 1. We must be in Steady state, where $P_{k|k} = P_{k-1|k-1}$ 2. In addition $P_{k|k}$ must be zero!

Is there any other explanation that I am not seeing?

The second equation that I don't see is equation A4. $$ (C_{v_k} - R)C_{v_k}^{-1} = H P_{k|k} H^T R^{-1} $$

Again, if I just take $$ C_{v_k} = HP_{k|k}H^T + R $$ and multiply by $R^{-1}$, I get to a situation where apparently $C_{v_k} = R$ I have to do the same assumptions, then, right?

Lastly, I would like to know why in equation (23), you can discard the terms related with the covarince matrix. The equation looks like this:

$$ \hat Q = \frac{1}{N} \sum_{j=j0}^k qq^T + P_{k|k} - H P_{k-1|k-1} H^T $$

and then just consider

$$ \hat Q = \frac{1}{N} \sum_{j=j0}^k qq^T $$

In a setting where H is the identity, and we are in a steady state, I see that you can exaclty do that. If it is not the case, then the two terms con be very different, can't they?

And a final note is that I think that there is a mistake, in equatino (16). The derivative $\frac{dQ_{j-1}}{d\alpha_k}$ should be $\frac{dQ_{j}}{d\alpha_k}$, I believe. Am I wrong?

Thank you very much. I hope I can open a nice discussion about the topic, and be of interest for people looking at the same nice stuff.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.