Quaternion State-Transition Model¶
All state propagation equations used in this paper are based on the following Taylor-series expansion:
where terms higher than first-order are neglected. For attitude, the quaternion is propagated according to the expression:
where dt is the integration time-step (sampling interval) and \(\vec{q}_{k-1}\) is the current estimate of system attitude.
The kinematical equation that describes the rate-of-change of the attitude quaternion, \(\dot{\vec{q}}_{k-1}\), is a function of true angular velocity, \(\vec{\omega}_{true}\), as follows:
where \(\Omega_{true,k-1}\) is formed from the components of the angular rate vector, \({\begin{pmatrix}{^{N}{\vec{\omega}_{true}}^{B}}\end{pmatrix}}^{B}\) and specifies the angular-rate of the body relative to an inertially-fixed frame, measured in the body-frame. As all angular-rate measurements made with MEMS sensors are relative to the inertial-frame, the notation is simplified to \({\vec{\omega}_{true}}^{B}\).
The quaternion propagation matrix, \(\Omega_{k-1}\), at time-step k-1 is:
where (as noted above) all the rate components are estimates of the “true” rate measurements.
From the above expressions, the full state-transition model for system-attitude is:
To find the noise term in the state-transition model, \(\vec{w}_{q,k-1}\), expand the expression for \(\Omega_{true,k-1}\) using the rate-sensor model described earlier to explicitly show the constituent terms:
Substitute this result into the expression for the attitude state-transition model:
\(\Phi_{k-1}\) is the state-transition matrix, defined as:
and \(\vec{w}_{q,k-1}\) is the quaternion process-noise vector:
Note
In this expression, the components of \(\Omega_{noise}\) are the noise components of the angular-rate signal, \(\sigma_{\omega}^{2}\). This can be expressed in terms of the sensor’s Angular Random Walk (ARW).
Recasting \(\vec{w}_{q,k-1}\), so the rate-sensor noise (\(\omega_{noise}^{B}\)) forms the input vector, results in the final expression for the quaternion process-noise resulting from rate-sensor noise:
with the variable \(\Xi_{k-1}\) relating the change in process noise to system attitude
and \(\begin{bmatrix} {\vec{q}_{v} \times} \end{bmatrix}\) is the cross-product matrix.
The quaternion process noise vector is used to form the elements of the process covariance matrix (Q) related to the attitude state. The covariance is computed according to the following equation:
As mentioned previously, all processes considered in this paper assume white (zero mean) sensor noise that is uncorrelated across sensor channels. This simplifies the expression for the covariance to:
In addition to the assumption that the noise terms are white and independent, all axes are assumed to have the same noise characteristics (\(\sigma_{\omega}\)). Resulting in the final expression for \(\Sigma_{q}\):