IRAS Explanatory Supplement
V. Data Reduction
B. Pointing Reconstruction
Table of Contents | Index | Previous Section | Next Section
The term "pointing reconstruction" as used here refers to the process of reconstructing the pointing direction of a fiducial point in the focal plane, as well as the twist angle of the telescope about the reference point, referred to as the boresight. Since the accuracy with which the pointing could be reconstructed varied as a function of time, uncertainty histories were also specified. Use was made of all available fine attitude sensor data to maximize the accuracy of the reconstructed pointing during the various attitude modes. A smoothed estimate of the pointing was obtained using a recursive form of an extended Kalman filter. A more detailed discussion of the pointing reconstruction than is given here is provided by McCallon and Kopan (1985).
The pointing reconstruction was based on the output of the fine attitude sensors which include eight slit-type visible star sensors, a three-degree-of-freedom gyro package and a dual-axis fine Sun sensor. The star sensors were mounted on the periphery of the focal plane, four slits being normal to the scan direction, and the remaining four slits skewed at angles of approximately ± 40 degrees (see Fig. II.C.6). The slits were used in pairs, one normal and one skewed, to provide two axes of boresight information at each visual star sighting, hereafter called a fine attitude calibration, or FAC. The gyros and fine Sun sensors were mounted together on the outside of the spacecraft. There were, of course, misalignments between the gyros and fine Sun sensors due to limits on the accuracy with which they could be positioned. Unfortunately, the misalignments between the gyro-Sun sensor package and the telescope were found to vary with time (mainly in cross-scan) due to bending of the telescope mount with respect to the spacecraft due to temperature gradients.
Pointing reconstruction was accomplished by integrating the gyro outputs to provide an initial estimate of the attitude history. Concurrent processing of frequent Sun sensor measurements and relatively infrequent FACs was performed with an extended Kalman filter to estimate model parameter errors and refine the attitude history. The model parameters estimated include initial-attitude errors, spacecraft-telescope misalignment errors, gyro drift, gyro scale factor errors, and gyro alignment errors. Gyro drift was modeled as a constant rate plus a random Gauss-Markov process. Process noise was added to all model parameters with the exception of the initial attitude parameters to allow for modeling errors and slow changes with time.
The recursive Kalman filter was run first forward in time and then backward over a scan and then the two estimates were combined to provide a smoothed estimate sampled every second. The Kalman filter was reinitialized with the updated model parameters, and the process was repeated for the next scan using any available boundary conditions. The scan was chosen as the basic processing block because it normally contained at least two FACs and was of manageable length. A special file of boundary conditions was maintained for each SOP, and processing of observations with fewer than two FACs was deferred until the multi-FAC observations were finished in order to pick up final boundary conditions for the others.
This approach to the pointing reconstruction worked quite well as shown by the statistics on known star position matches (see Section VII.C). One danger which was carefully guarded against was giving bad data to the filter. Considerable effort was put into algorithms to perform automated consistency checking and to provide measurement rejection capability. There were problems associated with each type of attitude sensor.
The fine Sun sensor hardware had severe spiking problems from the beginning of the mission. This had the greatest effect on spacecraft control, causing several fallbacks to the safe attitude control mode. The spikes were thrown out in the pointing reconstruction algorithm by consistency checks. Spikes occasionally caused a brief cross-scan excursion in the actual pointing.
Another problem associated with the fine Sun sensors was less dramatic but potentially of greater consequence for the pointing reconstruction. The y-axis fine Sun sensor characteristics varied with time over the mission. Thus the transfer function used to convert from the integer output of that Sun sensor to the desired cross-scan angle should have been varied slowly over the course of the mission. This problem was only partially alleviated by the fact that the spacecraft-telescope misalignment angle about the y-axis was being continually re-estimated. The pointing reconstruction used only two transfer functions for the y-axis fine sun sensor over the course of the mission. Ideally, a separate transfer function would have been used for every 75 to 100 SOPs.
The gyros also had their difficulties. Early in the mission it was apparent that the gyros (especially one of the z-axis gyros, denoted Gyro ZA) were noisier than expected, and that their characteristics, especially those of the x-axis gyro, varied with time. Gyro ZA was used by the on-board control system for in-scan control. Unfortunately the control system was not designed for a gyro as noisy as ZA, and this resulted in the occasional occurrence of a phenomenon referred to as the limit-cycle-burst problem. During a burst, the limit cycle amplitude and frequency increased by a factor of three or four; amplitudes of 13" were observed. This problem occurred much more frequently after the star of the third hours-confirming coverage and was finally solved by switching to Gyro ZB for control. The limit cycle bursts were, however, reconstructed without apparent difficulty. The long-term changes in gyro characteristics were tracked by the Kalman filter, but process noise had to be increased to give the filter enough freedom to follow the changes. There was also some indication of even faster changes suggestive of thermally induced misalignments between the gyros and the fine Sun sensors. These problems worsened as the mission progressed. The x-axis gyro was dropped from use in reconstruction beginning with SOP 187 and the y-axis gyro was dropped beginning with SOP 316.
The gyros were far more susceptible to the Earth's magnetic field than expected. This problem was discovered early in the mission, and a software compensation algorithm was developed for pointing reconstruction.
Very early in the processing it was noted that errors in the reconstructed position of known infrared sources were sometimes larger near certain FACs rather than smaller as expected. It appeared that the FAC was degrading the solution. These cases became generally known as "biased FACs". Most biased FACs had errors on the order of 5" to 10", but some were much worse, up to 6'. It was determined that the worst biased FACs were the result of having selected stars for FACs which had other bright stars close enough to interfere with the star sensor observations. Occasionally, biased FACs were the result of catalog errors. An algorithm was developed to identify the FACs with disturbing stars, and in reprocessing SOPs 29 through 446 these FACs were either given larger uncertainties or deleted. The worst biased FACs (errors greater than 30") in the 446 to 600 SOP range were corrected in the same manner.
Another problem was the thermal misalignment about the y-axis (cross-scan) between the gyro/Sun sensor package and the telescope. This caused particular difficulties during the third sky coverage because of large and frequent changes in solar aspect angle.
Lack of a time-variant y-axis fine Sun sensor transfer function had the greatest effect on cross-scan reconstruction for observations with slews between the FAC and the survey scan. In this situation the Kalman filter was unable to compensate for the y-axis fine Sun sensor errors by adjusting the telescope misalignment about the y-axis as it would normally do. In the worst-case, cross-scan slews on the order of one to two degrees resulted in reconstruction errors as large as 25". The actual magnitude of the error varied greatly depending on the exact star and end points of the slew and was not necessarily greater for longer slews. The only survey scans affected by this problem were those with out-of-scan FACs.
Lack of the improved y-axis thermal misalignment model was significant only for observations which were both without a FAC and without a boundary condition on the same end. In the worst-case situation, cross-scan reconstruction errors as large as 30" were possible for a one-FAC observation; this could happen on full length observation following a maximum cross-scan slew with the worst possible placement of the FAC. Such large slews were executed only after the star of the third survey coverage (SOPs 426 to 600). For SOPs 29 to 425, peak errors due to this problem were less than half as large.
Table of Contents | Index | Previous Section | Next Section