Adaptive state noise compensation (SNC) in the different local frames

[ChristopherRabotin]

High level description

In the real world, SNC is typically configured once at the start of a pass, and left alone. With the Cr estimation of #94 merged in #350, the SNC is very finicky: too little noise and the Cr oscillates a lot, and too much causes the Cr to not converged onto the correct solution fast enough.

At the moment, there are three SNC options:

1. Fixed
2. Exponential decaying
3. Set thereof with different activation times

The purpose of this issue is to add a self-tuning SNC option, and probably remove the set of SNC configurations.

Requirements

• Provide a first self-tuning SNC option using the trace of the outer product of the innovation and the trace of the innovation covariance
• Provide a second self-tuning SNC option using the maximum likelihood SNC
• Allow for each dynamical model to specify its own state noise compensation: SRP should add noise in the direction of the Sun-line vector, and the gravity field should add noise in the R direction of the RIC frame.

Test plans

Rerun the Cr estimation OD with all three SNCs and assess their performance. Document the findings are recommendations in the selection of the SNCs.

Design

• Add a LocalFrame parameter to the SNC to allow for different configurations of the SNC
• Consider using the TypedBuilder trait for SNC building.

Algorithm demonstration

Refer to the attachments:

• https://repository.library.carleton.ca/concern/etds/t435gd948
• Self-tuning-SNC.pdf