[Top] | [Previous] | [Next]

E-6. Attitude Analysis Performance Drivers

The SSCS FRD requires that processing associated with attitude determination be completed within ten minutes of updates to the orbit ephemeris. For this benchmark exercise, it was assumed that activities related to attitude data preprocessing (i.e., filtering, thinning, and smoothing) would be sufficiently automated to minimize the need for manual editing by human operators.

Performance benchmarks for attitude determination were defined using sensor data from a GPS mission and a Research, Development, Testing, and Evaluation (RDT&E) program. Results were generated using two Aerospace-developed programs called SPNATT2 and AFG2 run on a 486 class PC. For each case investigated, an attitude sensor observation file was created based on preprocessed telemetry data containing raw observation data from sensors onboard the spacecraft. Data regarding the vehicle's spin rate was used as well. The data processing activities also included the vehicle's orbital ephemeris, as well as an initial guess on the attitude solution.

Several cases were investigated to determine the impact of statistical analyses (i.e., determination of sensor biases and assessment of data quality) on data processing and hence performance of the algorithms. The two satellite systems provided data for two different attitude determination systems and two orbit classes. Orbit ephemerides for the study were generated using a tenth order Adams orbit propagator. The attitude sensor observation files contained measurements recorded by the Earth sensors and sun sensors. Values were also time-tagged to support attitude determination algorithms. A sampling of the most stressing results is presented in Table E-1. Results were generated with and without determination of data quality, confidence factors, and statistics.

Table E-1. Performance Benchmarks for Attitude Determination

Attitude Data Processed

Data Products Generated
Execution Time for Processing (sec)
30 minutes of sensor observations; data frequency at 1/sec; total of 1800 observations in input file.
Inertial attitude state with standard deviations and residuals.
190.3
16 minutes of sensor observations; data frequency at 1/sec.
Inertial attitude state with standard deviations and residuals.
102.1
30 minutes of sensor observations; data frequency at 1/sec; total of 1800 observations in input file.
Inertial attitude state, sensor biases, residuals, spin rate errors, confidence measure and data quality assessment.
395.4

The attitude determination algorithm is based on a combination batch processor/Kalman filter. The algorithm uses a variation of the "cone intercept" method to determine orientation of two independent vectors. Measurements of these vectors can be individually deweighted to minimize effects of sensor inaccuracy or poor performance. Determination of sensor biases and associated statistics provided the most stressing conditions to the attitude determination process. Typically, the algorithm would converge on a solution after three iterations.

The attitude determination application must be capable of analyzing observation data files for sensor biases, spin-rate errors, spin-axis misalignment, and precession and nutation of the spin axis. Computation times could be improved by incorporating a second polynomial fit to smooth observation values before data processing. Note that the performance results can vary as a function of 1) initial guess (or starting point), 2) quality of the sensor data (i.e., the degree of "random noise" in the observations), and 3) sharing of system resources with other processing routines.

Table E-2. Performance Benchmarks for Attitude Ephemeris and Events Reporting

Attitude Data Processed

Data Products Generated
Processing Time (sec)
Attitude events for a 24-hour time span with a 60 second step size.
RGF look angles, spacecraft transmission angles, sensor visibility intervals. Data generated for 9 tracking stations and formatted.
10.4
Attitude events for a 24-hour time span with a 30 second step size.
Same As Above
19.1
Attitude events for a 24-hour time span with a 15 second step size.
Same As Above
38.0
Attitude events for a 24-hour time span with a 5 second step size.
Same As Above
114.4

Benchmark results were also generated for output associated with attitude events. These data items include RGF look angles and spacecraft transmission angles, as well as prediction of sensor coverage intervals. These data products were generated using software tools developed by The Aerospace Corporation to support launch and early orbit operations. Results were generated at varying step sizes to illustrate the degree of linearity in processing routines. Orbit ephemerides were generated using an tenth order Adams orbit propagator. The results of the exercise are displayed in Table E-2.

The most stressing activity in Attitude Analysis is the Determine Attitude application because determination of statistics will drive computation times. The benchmark numbers illustrate that, for an attitude determination system using two Earth sensors and one sun sensor (the typical configuration for most spin stable spacecraft), a 486 class PC is capable of solving for attitude and associated statistics. The primary factor to consider during attitude determination is the amount of human intervention during editing and filtering of observation data files. If this process is sufficiently automated, a 486-class PC will be capable of meeting the FRD requirements for Attitude Analysis. However, in anticipation of growth requirements (and requests) for higher fidelity atmospheric and thermal modeling as well as enhanced graphics, a mid- to high-powered desktop workstation is recommended for Attitude Analysis to support mission operations of current and future space systems.