determination and control
3.3 Characteristics of the GRACE star cameras
3.3.4 Star camera measurement accuracy
The SCA measurement is characterized by its anisotropic accuracy, which is a consequence of the sensor construction geometry. The pointing of the boresight axis, i.e. the rotation about thexSCF andySCF axes, is determined with a factorκ= 8 better than the rotation about the boresight (zSCF). The nominal measurement accuracy of the GRACE star cameras is assumed to be ˆx,y= 30µrad for the rotation about the xSCF and ySCF and ˆz= 240µrad for roll around the boresight axis (Stanton, 2000). The characteristic anisotropic noise distribution is obvious from Figure 3.12, where the attitude angles and angular rates about the SCF axes relative to the inertial frame are shown in both time and frequency domain. Between 0.01 Hz and 0.5 Hz the angular rates are dominated by noise which is characterized byf-behavior, i.e.
in terms of attitude angles the PSD is flat.
For the GRACE star cameras, no error model is available which characterizes the sensor performance over the whole spectrum. The in-flight performance is not possible to be obtained precisely because of the lack of a reference and the many unpredictable factors influencing the performance (J.L.Jørgensen, DTU - private communication, March 2nd, 2015).
(a) (b)
(c) (d)
Figure 3.12: The attitude angles (a,b) and angular rates (c,d) about the star camera frame axes relative to the inertial frame demonstrating the anisotropic noise distribution of the SCA measurement in time (a,c) and frequency domain (b,d). Derived from the GRACE-A SCA Level-1A data of SCA head#1 on 2008-12-01 with
1 Hz sampling
The star camera measurement accuracy is limited by many different factors, which are discussed in detail e.g. by Eisenman and Liebe (1998); Jørgensen (2000); Liebe (2002).
Absolute attitude error with respect to the mounting plane consists of:
I boresight error (small mechanical excursions in the star tracker which cannot be calibrated out)
– thermal drift
– ground calibration residuals – launch effects
– gravity release effects
I Relative error (the relative error is a measure of how accurately the star tracker can detect changes in attitude)
– optics error
. ground calibration errors . thermal distortion . chromaticity
. optical and point spread function distortion – centroiding error
. pixel non-uniformity . quantization error
. centroid algorithm uncertainty . CCD charge transfer efficiency effects – noise equivalent angle
. readout noise . dark current noise . stray light noise . photon noise – algorithmic errors
. time stamp uncertainty . erroneous star matches . algorithmic approximations . star catalog uncertainty
The in-flight performance of the four star cameras varies a lot despite of their nominally presumed identical performance. Although the star cameras on both spacecraft were calibrated in the early stage of the mission and parameters such as focal length, threshold for minimum number of stars in field-of-view and camera boresight were tuned in order to bring the performance of all 4 sensor heads to the level of the best one, there still remain certain differences (Hermanet al., 2004). On both satellites, the attitude data delivered by star camera head#2 are less noisy than the data from head#1. In other words, the good camera is head#2 and the bad camera is head#1 for both GRACE-A and GRACE-B according to Hermanet al.
(2004). However, the performance of the cameras changes with time and it depends on many factors which are listed above and which might also change with time. In 2014, the worst performing camera was head#1 on GRACE-B and the next worst performing camera was head#2 on GRACE-A (Witkowski and Massmann, 2014).
We have compared the measurement accuracy of the four star camera heads in 2008. The measurement accuracy was estimated as the mean noise level within 0.01 Hz and 0.5 Hz for the pointing of the boresight,xSCF andySCF, and for roll about the boresight,zSCF. Figure 3.13 shows the measurement accuracy for the primary camera of each satellite for the whole year 2008. The primary camera head switch was performed on DOY 135 and 305. The accuracy of the rotation about each SCF axis is quantified in Table 3.1.
Table 3.1:The measurement accuracy of the two star camera heads onboard GRACE-A and GRACE-B. The measurement accuracy is estimated as the mean noise level of the rotation about the SCF axes within the frequency band of 0.01-0.5 Hz. The results are based on the data from the whole year 2008, cf. Figure 3.13
GRACE-A GRACE-B
head#1 head#2 head#1 head#2
xSCF [µrad] 25 25 32 20
ySCF [µrad] 18 22 32 14
zSCF [µrad] 235 170 240 140
For GRACE-A, the accuracy of the boresight pointing is practically the same and is slightly better than the estimated nominal measurement accuracy ˆx,y = 30µrad. Moreover, the rotation about the ySCF axis can be determined even slightly better than the rotation about xSCF, cf. Figure 3.13(a). Figure 3.13(b) reveals that the accuracy of roll about the boresight, zSCF, strongly differs for the two camera heads. Compared to the nominal accuracy of ˆz= 240µrad, the SCA head#1 performs as expected, whereas head#2 performs much better than expected. Consequently, the factor κ, which represents the noise level ratio between boresight pointing and the roll about boresight, is then κ = 11 and κ = 7 for head#1 and head#2, respectively. The measurement accuracy of SCA head#1 on GRACE-B is exactly as expected and corresponds well with the nominally estimated values. From Figures 3.13(c) and 3.13(d) is obvious that the measurement accuracy of SCA head#2 is much better than of SCA head#1. The noise values are very close to those estimated during the on-ground tests. The corresponding noise level ratio factor is for both SCA heads κ= 8. Figure 3.13 also clearly shows the sensitivity of the SCA measurement accuracy to the moonlight intrusions which occur every 27 d.
(a) (b)
(c) (d)
Figure 3.13:The measurement accuracy of the two star camera heads onboard GRACE-A (a,b) and GRACE-B (c,d). The measurement accuracy is estimated as the mean noise level of the rotation about the SCF axes within the frequency band of 0.01-0.5 Hz. Figures on the left represent the zoom in of the figures on the right, with focus onxSCF andySCF. The results are shown for the whole year 2008. The primary camera head switch
was performed on DOY 135 and 305
As mentioned above, the measurement accuracy is influenced by many factors which might change over time. The most significant ones are the light intrusion into the FoV, number of stars in the FoV and thermo-elastic effects in the satellite. Some aspects of these factors are presented below.
The high sensitivity of the attitude measurement to the stray light was already discussed by Prestiet al. (2004). Obviously, the cameras are not only sensitive to the light inside the FoV but also to the light outside the theoretical FoV which is limited by the baffles. As Prestiet al.
(2004) claim, it turned out that the effective FoV is larger than expected. The possible reason for the unexpected stray light are reflections.
The other factor affecting the measurement accuracy is the number of stars in the FoV. The more bright stars are identified in the digital star image, the better instantaneous attitude can be estimated. For precise attitude determination, a minimum number of stars in FoV is defined. The threshold for minimum number of stars can be adapted individually for each camera. The stars in the sky are not evenly distributed. There is a huge density of stars in the area of Milky Way compared to galactic poles where there are not so many stars. Therefore it does matter where the camera is pointing. Moreover, the ascending node of the GRACE orbit changes very slowly, it takes almost 8 years to finish the 360◦ circle. Hence the lack of stars in specific regions might influence the camera performance for long periods of time. As the two star cameras are pointing with their boresights to different parts of the sky, the number of stars in FoV is different for each camera, cf. Figure 3.14. This is one of the factors contributing to the different performance of the two cameras.
Interestingly, Figure 3.14 also shows the sensitivity of the number of stars in FoV to orbital configuration. Between DOY 43-74 and 212-233 the satellites were flying in the full sun orbit, which means the satellites did not pass through the Earth shadow. This means that the primary camera is less disturbed by the stray light as it is pointing more or less parallel to the Sun vector, but in the opposite direction. The white areas in Figure 3.14 are caused by the sunshine and moonlight intrusions into the FoV, cf. with Figure 3.11.
(a) GRACE-A head#1 (b) GRACE-A head#2
(c) GRACE-B head#1 (d) GRACE-B head#2
Figure 3.14:Number of stars in the star camera field-of-view for head#1 and head#2 on both satellites in 2008
The thermo-elastic effects of the star cameras and their neighborhood affect the star camera performance as well and cause possible biases and systematic effects. The satellite vehicle is
exposed to extreme temperature differences which vary according to the satellite’s position in orbit and its orientation towards the Sun. While one of the satellite panels is directly illuminated by the sunlight, the opposite panel is pointing into the outer space and hence is in shadow. The outside temperature difference reach up to 200◦C, cf. Figure 3.3. Inside the spacecraft, thermistors and heaters are mounted to sense and to change the inside temperature, in order to compensate the effect of the outside temperature differences and to ensure thermally homogeneous environment inside the vehicle. The effect of the temperature variations on the SCA measurement is shown in the following section.
The analysis of the impact of all these factors on the SCA measurement is beyond the scope of this thesis. The SCA performance analysis is a task for JPL, GSOC and DTU who have the required expertise and the access to the necessary housekeeping data, to the information about all satellites components and their properties and also to the onboard algorithms, which are confidential.