SRIFB#

class SRIFB(propagator: Propagator | MissionSequence, measurements: MeasurementSpec, settings: FilterSettings, traj_name: str | None = None, traj_dir: str | None = None, overwrite_traj: bool = True, verbose: bool = True)#

Bases: FilterOD

Square Root Information Filter — Batch (SRIFB) for spacecraft orbit determination.

The SRIFB is the batch counterpart of the sequential SRIF. All measurements across all epochs are accumulated at once in the square-root information domain without any intermediate time updates, so process noise (SNC/DMC) is not supported. It produces the same solution as the LSB normal-equation batch estimator but with superior numerical conditioning.

Algorithm

The SRIFB represents the accumulated information as the pair \((R_{xx},\, z_x)\), where \(R_{xx}\) is the upper-triangular Cholesky factor of the information matrix and \(z_x = R_{xx}\,\hat{x}_0\).

At each epoch \(t_k\), let \(\tilde{R}_k\) be the upper-triangular Cholesky factor of the measurement noise covariance \(R_k\) (so \(R_k = \tilde{R}_k^T \tilde{R}_k\)). The whitened observation equation

\[\tilde{R}_k^{-1} H_k\,\Phi_{k,0}\cdot\Delta x_0 \;=\; \tilde{R}_k^{-1} y_k\]

is appended below the running information block and the combined rectangular system is triangularized via a Householder QR decomposition:

\[\begin{split}\begin{bmatrix} R_{xx} & z_x \\ \tilde{R}_k^{-1} H_k\,\Phi_{k,0} & \tilde{R}_k^{-1} y_k \end{bmatrix} \xrightarrow{\;\text{QR}\;} \begin{bmatrix} \hat{R}_{xx} & \hat{z}_x \\ 0 & e_k \end{bmatrix}\end{split}\]

where \(e_k\) collects the normalized postfit residuals for epoch \(k\). After all \(K\) epochs are processed, the state deviation at \(t_0\) and its covariance are recovered as:

\[\hat{x}_0 = \hat{R}_{xx}^{-1}\,\hat{z}_x, \qquad \hat{P}_0 = \hat{R}_{xx}^{-1}\,\hat{R}_{xx}^{-T}.\]

The estimated deviation is then propagated forward to every measurement epoch via the stored STMs.

Numerical advantage over LSB

The LSB forms the normal matrix \(M = P_0^{-1} + \sum_k A_k^T R_k^{-1} A_k\) explicitly and inverts it, squaring the condition number of the problem. The SRIFB avoids this by working directly with the square root: each QR step is an orthogonal transformation that preserves the numerical rank of the information block throughout accumulation.

Schmidt consider parameters

The SRIFB supports the Schmidt consider-parameter extension. The consider Jacobian \(H_{c,k}\) is appended to each whitened measurement block before the QR step, and the resulting off-diagonal information block \(\hat{R}_{xc}\) encodes the cross-correlation between estimated and consider uncertainties. The posterior covariance is:

\[\hat{P}_0 = \hat{R}_{xx}^{-1}\,\hat{R}_{xx}^{-T} + S_{xc}\,P_{cc}\,S_{xc}^T\]

where \(S_{xc} = -\hat{R}_{xx}^{-1}\hat{R}_{xc}\) is the consider sensitivity matrix.

Process noise and stochastic parameters

Sequential process noise models such as SNC and DMC require intermediate time updates between measurement epochs and are therefore not supported in the purely batch formulation. Use SRIF or LKF when unmodelled accelerations must be compensated sequentially epoch by epoch.

A batch analogue can still be constructed through so-called “stochastics” or pseudo-epoch states, where empirical accelerations are represented as estimated solve-for parameters within the batch state vector. In this approach, first-order Gauss–Markov or piecewise-constant stochastic accelerations are parameterized over predefined intervals and estimated simultaneously with the orbital state.

Parameters:
propagatorPropagator or MissionSequence

Defines the reference trajectory and STMs. A MissionSequence can be used for multi-leg trajectories with maneuver events.

measurementsMeasurementSpec

Measurement specification object.

settingsFilterSettings

Filter configuration. settings.process_noise must be None.

traj_namestr, optional

Name for the trajectory BSP file. Auto-generated if None.

traj_dirstr, optional

Directory for the trajectory BSP file.

overwrite_trajbool, optional

If True, overwrite any existing trajectory file.

verbosebool, optional

If False, silence the printout of the propagation run at construction. Defaults to True.

Raises:
ValueError

If the initial covariance matrix is not positive definite (required for Cholesky factorisation).

NotImplementedError

If settings.process_noise is not None (process noise is not supported in the batch formulation; use SRIF or LKF).

See also

scarabaeus.LSB

Normal-equation batch estimator; same solution, less stable.

scarabaeus.SRIF

Sequential square-root information filter; supports process noise.

scarabaeus.FilterOD

Base class providing trajectory and measurement handling.

Notes

Process noise (SNC/DMC) requires intermediate time updates and is therefore not supported in the batch formulation. Use SRIF or LKF when unmodelled accelerations must be compensated sequentially. For process-noise-free problems, SRIFB is the numerically preferred alternative to LSB.

References

[1]

Tapley, B. D., Schutz, B. E., & Born, G. H. (2004). Statistical Orbit Determination. Elsevier Academic Press. ISBN 978-0-12-683630-1.

[2]

Bierman, G. J. (1977). Factorization Methods for Discrete Sequential Estimation. Academic Press. ISBN 978-0-12-097650-8.