On the autonomous in orbit calibration of satellite attitude sensors Yu.V.Kim, K.J.Di Filippo, A.Ng Spacecraft Engineering, Space
Technologies, Canadian Space Agency 6767 Route de l'Aéroport, St-Hubert,
Québec, Canada J3Y 8Y9 A routine task for satellite operation is the
calibration of on-board sensors. Typically, this task is performed on ground
using spacecraft telemetry data. The calibrated parameters are then uploaded to
the spacecraft. As the trend in satellite operation is towards extended
operation autonomy, there is a need to develop an autonomous on-board
spacecraft sensor calibration. This
article proposes a general approach to solving the sensor calibration problem
autonomously using an onboard processor with a sub-optimal Kalman Filter (KF). The
approach is illustrated with RADARSAT-1 magnetometer calibration as an example. In-flight satellite attitude estimation and
sensor calibration is an important part of satellite mission success. Many
publications have been devoted to this subject using a variety of different
levels of complication of sensor error models and estimation techniques. A few
of these references are cited here, noting that they consider "in-orbit
calibration with on-ground estimation" requiring operation personnel
intervention. The publications devoted to autonomous calibration mainly
consider some partial sensors with very detailed models and sophisticated
optimal algorithms; however, such an approach is inappropriate within the microsatellite
paradigm, where the driving philosophy is that of maximal performance from
minimal resources. Specifically,
current publications do not take into account the limited on-board computation
power on micro satellites. To implement a KF algorithm on a micro satellite,
the algorithm must be minimal and robust enough to handle the effects of
estimation process potentially diverges in case of including state vectors some
badly observable parameters or the effect of limited time optimal recursive procedures
for a deterministic model. This paper presents a general approach for
calibrating a wide class of sensors, namely Vector Measuring Devices (VMD). The
approach is based on using the simplest generalized sensor error models (which
take into account only main components) and the simplest way of achieving
computational economy: by pre-calculation of KF coefficients for some
predetermined set of orbits chosen for the calibration. Typical attitude determination accuracy
requirements for a modern satellite range from moderate (~1º) up to
extremely high (<0.01º). In
order to achieve the required accuracy, satellite operations often require
on-orbit sensor calibration to compensate for deterministic residual error
components, such as mutual misalignments, biases, and scale factors. The misalignments between the instrument
axes of different satellite attitude sensors are usually the main contributors
to the total error budget. In the current satellite operations paradigms,
calibrations to sensor outputs are typically determined on-ground, based on
on-board sensor telemetry data. The corrections are subsequently uploaded to
the satellite. Typically, satellite
attitude control software includes a set of algorithms, commonly known as
Attitude Determination Mode (ADM), to determine satellite attitude. The most
accurate ADM is designated as the primary ADM and is used as the reference base
(physical platform) to calibrate redundant auxiliary sensors. They are not involved in closed loop control
of satellite attitude in the primary ADM; however, they are used as part of the
control loop under special circumstances: e.g. eclipse, primary sensor failure,
recovery from Safe Hold Mode, and special attitude manoeuvres. This strategy
has been adopted by two Canadian satellite missions: RADARSAT-1 and SciSat,
both of which are operated by the Canadian Space Agency (CSA) Mission Control
Centre (MCC). This paper proposes the transfer of calibration
authority from a ground-based MCC to on-board algorithms, while preserving the
underlying calibration strategy. A recursive KF algorithm is used for real time
on-board estimation of the calibration parameters of satellite attitude
sensors. To have the method applicable to even a microsatellite with a
resource-limited processor, some efforts was spent to sub-optimize the
developed KF in order to make it more economical from a computational loading
point of view. The approach presented
in this paper avoids the computation of covariance matrices and weight
coefficients - which are the most computationally demanding aspects of KF - by
approximating these coefficients as analytical functions of time. The decision
concerning the insertion of the derived estimates into the control algorithms
is based on a set of criteria that include the evaluation of the values of the
estimates and their stability in time after some pre-determined observation
period. General Approach is based on generalized error
model for a vector measuring devise (VMD) that measures some physical vector to
determine satellite attitude and KF sub optimization by pre-calculation of KF
coefficient matrix. Examples of VMDs include Sun sensors and Star
Trackers (which determine star pointing vectors), Earth sensors (which
determine the direction of the nadir vector), and magnetometers (which
determine the local direction of the geomagnetic field). The output of a VMD is
typically used in conjunction with a target, reference vector, which is
the expected value of the corresponding physical vector when the satellite is
at the nominal attitude. It is showed that the approach presented allows
reaching sufficient level of calibration accuracy with very economical
implementation of the filter in on-board processor. The payment for such a simple implementation is
some extension of the required observation period (w.r.t optimal KF), which can
be tolerated for the considered problem. |
© 1995-2008 Kazan State University