See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/269254512 The characterization and testing of MEMS gyros for GIOVE-A Conference Paper · August 2006 DOI: 10.2514/6.2006-6044 CITATIONS READS 0 199 3 authors, including: Alexander Cropp Stephane Dussy European Space Agency European Space Agency 25 PUBLICATIONS 114 CITATIONS 42 PUBLICATIONS 164 CITATIONS SEE PROFILE Some of the authors of this publication are also working on these related projects: Intermediate eXperimental Vehicle (IXV) View project All content following this page was uploaded by Stephane Dussy on 25 May 2015. The user has requested enhancement of the downloaded file. SEE PROFILE AIAA 2006-6044 AIAA Guidance, Navigation, and Control Conference and Exhibit 21 - 24 August 2006, Keystone, Colorado The Characterization and Testing of MEMS Gyros for GIOVE-A A Cropp* and C Collingwood.† Surrey Satellite Technology Ltd, Guildford, UK Stephane Dussy‡ ESA/ESTEC, Noordwijk, The Netherlands This paper describes the characterization and testing of the Systron Donner QRS11 MEMS rate sensors for the GIOVE-A mission. The rate sensing requirements for this mission exceeded the QRS11 data sheet specifications; hence testing was required to prove the unit's enhanced performance. This paper describes the test methods used, as well as the resulting bias and bias instability values measured. Effects such as the bias sensitivity to temperature, initial bias transient on switch-on, and vacuum effects on bias are measured and compensated for. The end results are bias and bias instability values that are well within the values stated in the QRS11 data sheet, and are verified by initial in-orbit results. I. Introduction T he QRS11 MEMS (Micro Electrical Mechanical System) analogue rate sensor is an inexpensive robust unit with previous space flight heritage. It has flown on the NASA SAFER EVA backpack experiment, on the ESA INTEGRAL mission, as part of the NASA Mars Sojourner Rover mission, as well as on a previous Surrey Satellite Technology Ltd. (SSTL) mission UoSAT-12. It was chosen for GIOVE-A due to cost constraints, and because of the relatively relaxed mission rate-sensor requirements. The unit was required in the following mission modes: • DTM (Detumbling): The gyro rate output measures the tip-off rates after launcher separation. In this mode, the gyros are mission-critical sensors. • SAM (Sun Acquisition Mode): Before acquiring the Sun, 3-axis gyro rate information is required. After acquisition, the gyro rates are required to control rate about the Sun vector. • SAM-C (Sun Acquisition Mode with Coning): Earth searching coning maneuver about un vector, the gyro rates are required to control rate about the Sun vector. However, the unit specifications were significantly outside of requirements for maximum bias and bias instability parameters. Table 1 below gives the GIOVE-A gyro requirements, along with the relevant Systron Donner specifications, taken from the QRS11 data sheet (correct at time of writing). Table 1: A comparison of GIOVE-A gyro requirements with QRS11 specifications Parameter Rate measurement noise (assumed white when sampled at 1Hz), also known as Angle Random Walk (ARW) Maximum bias at any time GIOVE-A Requirement < 0.010 deg/s (1 ) QRS-11 Specifications < 0.010 deg/s/ Hz (1 ) = ARW of 0.6 deg/ hr < 0.25 deg/s (hard limit) Steady-state bias instability (random walk of gyros) < 20 deg/hr/ hr (3 ) 0.5 deg/s (at 22°C) + 0.35 deg/s deviation from 22°C 0.01 deg/s in 100 sec = 216 deg/hr/ hr * Alex Cropp, email a.cropp@sstl.co.uk., tel. +44 1483 803870 Cheryl Collingwood, email c.colligwood@sstl.co.uk, tel. +44 1483 803859 ‡ Stephane Dussy, email stephane.dussy@esa.int, tel. +31 715 654018 † 1 American Institute of Aeronautics and Astronautics Copyright © 2006 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved. This paper briefly describes the tests carried out to Temp Temp Temp Temp Temp Temp demonstrate that the chosen QRS11 units can meet Gyro Gyro Gyro Pack Gyro Gyro Gyro Pack Temp Temp X Y Z X Y Z these requirements. The primary tests in Section II are based on the characterization of bias with temperature, transient effects, and vacuum effects. Bias instability AIM 1 AIM 2 tests are included in Section III. A brief note on rate integration is given in Section IV. For GIOVE-A, each single-axis gyro provides an CAN Bus (Primary) AOCS Interface Module (AIM) with an analogue rate CAN Bus (Secondary) and internal temperature signal. A single AIM unit has Figure 1. AOCS Interface Module Block Diagram three gyros sensing each axis, with an additional Each AIM unit has 3 gyros supplying rate and internal external pack temperature sensor, as shown in temperature information. In addition, an external Figure 1. The AIM provides filtering for each signal, temperature sensor is mounted next each set of three with a first-order time constant of approximately gyros. The secondary AIM unit is fully redundant. 100ms. Each analogue signal is converted to raw telemetry via a 12-bit ADC. Typically, each gyro is sampled at 10Hz, then averaged each second to provide 1 rate estimate and temperature estimate per gyro per second. II. Bias Characterization Gyro Bias (TLM) A. Bias Thermal Calibration Raw (i.e. uncalibrated) temperature telemetry is compared against raw gyro rate telemetry at fixed temperatures, while the sensor is at rest. In each case, three gyros and an AIM unit are placed in a thermal chamber, and are allowed to achieve steady-state temperature. Although the calibrated temperature is unavailable, steady state temperature can be observed when the average rate of change of temperature telemetry becomes less than the average telemetry noise. For these test cases, the time required to achieve steady-state temperature was less than 2.5 hours. The initial temperature range was -10°C to +40°C in 5°C steps. For each fixed temperature, an average gyro rate telemetry value and an average temperature 1980.0 telemetry value are measured. Over the full 1975.0 temperature range, the various gyro and 1970.0 X measured temperature telemetry 1965.0 X estimate points can be fitted to a Y measured 3rd order polynomial 1960.0 using least-squares Y estimate techniques. This Z measured 1955.0 polynomial is then used Z estimate 1950.0 to estimate the zero-rate gyro bias telemetry value 1945.0 for any given temperature telemetry measurement. 1940.0 The use of raw telemetry 0.50 0.70 0.90 1.10 1.30 1.50 values avoids the Temperature (Internal TLM / 2000) intermediate and unnecessary step of Figure 2. Example of measured gyro telemetry with estimated polynomial fitting having to calibrate and Points represent individual measurements of gyro rate telemetry and temperature convert temperature telemetry. The far left set of points was added after system assembly & integration thermal telemetry into °C. In tests (AIT) took temperatures outside design range down to approximately –25°C. addition, internal heating However, as only raw measurements are used, this data point could be added to the model of the gyros would mean fitting. the sensors would be at different temperatures 2 American Institute of Aeronautics and Astronautics from the thermal chamber, making accuarte calibration of temperature difficult. During these tests, hysteresis effects were observed, and hence the units were taken from 10°C up to +40°C and back down again to -10°C in 5°C steps. All the resulting measurement points were then used in the polynomial fitting. 1980 Raw Gyro Telemetry 1975 1970 1965 1960 1955 1950 0 20 40 60 80 Time (minutes) 100 120 140 Figure 3(a). Example of raw gyro rate initial transients. These measurements were taken from 3 gyros immediately after switch-on. As can be seen, in the first hour, the rates change noticeably, despite the fact that throughout this test the sensors were stationary. It is believed this is due to internal heating. Without correction, these transients can lead to bias errors on the order of 0.05 deg/s. 2250 Raw Internal Temperature Telemetry 2200 2150 2100 2050 2000 1950 1900 1850 1800 1750 0 20 40 60 80 Time (minutes) 100 120 140 B. Initial Transients Immediately after the sensors had been powered, a change in the gyro rate output was observed. Internal gyro temperature telemetry seemed to indicate the unit was self-heating. Figure 3(a) shows a typical example of gyro rate telemetry taken from 3 gyros. These gyros had been left in a thermal chamber set at constant temperature for several hours before being switched on. During DTM mode of GIOVE-A, there was a 6-hour critical period when the satellite needed to reduce initial tip-off rates using rate measurements from the gyros. If the gyros had to be switched on and left for 1 or more hours before being used to allow initial transients to disappear, this would have taken a significant proportion of the available time, and reduced margin to an unacceptable amount. Errors due to transients alone have been seen to reach 0.05 deg/s. As can be seen in Figure 3(b), the internal sensor temperature increased just after switch-on. As the units had been at constant temperature before being powered, this increase was assumed to be due to internal heating. In addition, it was assumed that the internal heating was causing the transients observed in the rate output. Hence, applying the bias thermal calibration model to the initial transients should remove most of the effect. Figure 4 below shows the calibrated bias error based on the raw sequence from Figure 3(a). In addition to removing the initial transient effects, a total bias error of less than 0.015 deg/s is demonstrated, much less than the quoted 0.5 deg/s. Figure 3(b). Example of gyro internal temperature measurements during transients These measurements were taken at the same time as the measurements in Figure 3(a). These show the observed increase in internal temperature immediately after switch-on, followed by a gradual attainment of steady-state temperature. C. Vacuum Effects As part of the AIM + gyro Thermal Vacuum Tests (TVT), the sensors were placed in a vacuum at controlled temperatures. Due to gyro requirements for long-duration testing, no vacuum testing had been carried out. A vacuum environment was not expected to cause any change to the unit behavior, as the quartz sensing element is in a hermetic chamber at 1 atmosphere. However, on testing, an apparent offset in bias of up to 0.1 deg/s was observed. The units were tested after completion of TVT in ambient conditions, and it was found that this bias offset was removed, and bias values had returned to normal. From these results it was estimated that each gyro acquired a relatively fixed but unique bias offset when in vacuum. 3 American Institute of Aeronautics and Astronautics Initial in-orbit results confirm this, although longduration ground tests in vacuum are required to see if these bias offsets due to vacuum change with time. III. 15 10 Bias Instability Characterization 5 mdeg/s The bias instability specification of these devices is also considered to be overly conservative. Although not proven here, it is estimated that the quoted bias instability figures given in Table 1 cover the effects of initial transients discussed above. Bias changes on the order of 0.06 deg/s over 1 hour are equivalent to a bias instability of 216 deg/hour (per square-root hour). Hence, the uncalibrated initial transients due to expected internal heating may appear to be a large initial bias drift. However, if the initial transients are either removed through calibration or are allowed to settle, the true bias instability of the unit is much lower. The bias instability is measured using an Allan Variance technique, as described below. 0 -5 -10 -15 0 20 40 60 80 Time (minutes) 100 120 140 Figure 4. Example of calibrated bias error The results in this figure are the bias errors after calibrating the raw results shown in Figures 3(a) and (b). The bias is estimated by taking the mean of the calibrated gyro rate output every minute. As can be seen, the output bias error is within 15 millidegrees per second (mdeg/s). A. Bias Instability and Allan Variance Bias Drift can be modeled as a Random Walk, i.e. the integration of a white noise source. A random walk process is time varying and highly correlated, and as such it is impractical to try to take the mean or standard deviation of a sample set. However, the white-noise process that is integrated to form the random walk is easily characterized by a mean and standard deviation. For a signal composed of a random walk process with additional white noise added, over a short period of time the measured variance of the signal will be dependant on the added white noise. However, over long periods of 10 time, the random walk process will begin to dominate. Predominantly due to Allan variance1 is a technique used to extract the Output Noise variance of the white noise process driving the random walk, as well as the variance of the added white noise. 10 Allan Variance is useful for analyzing signals with 2 x Power Spectral Density (PSD) of the form Sx(f)=Sx0 f(where Sx0 is a constant and f is frequency in Hz). For a white-noise source y(t), the PSD is constant for all 10 Predominantly due frequencies such that PSD{y(t)} = Sy(f) = Sy0. For a to Bias Drift random walk process x(t)= y(t), the PSD of x(t) is given as Sx(f)=Sx0 f—2. Time variance (TVAR) is a form of Allan Variance, and 10 for discrete signals is the variance of the double 10 10 10 10 10 Decimation Ratio, n difference of the mean of a block of data, of varying length: Figure 5. Example TVAR This figure shows an example TVAR plot from a N 3 synthetic signal. With low decimation ratio n, the 1 2 2 2 measured variance is dominated by added white (output) (1) xi x () = 6(N 2) i = 0 noise. As the decimation ratio increases, the mean of larger blocks of data reduce the effect of noise, reducing the measured variance. This decrease continues until the where N is the length of the total data set, and x k is the random walk process begins to dominate, where mean of a non-overlapping block of n samples (where n variance increases with time and hence block size is the decimation ratio), and = n 0 is the time duration (decimation ratio). of the block of data (where 0 is the sample interval). -2 -3 -4 -5 0 1 2 3 4 (( 4 American Institute of Aeronautics and Astronautics )) Table 2. PSD and Standard Deviation using Allan Variance (TVAR) Noise Type PSD vs. TVAR Bias Instability (Random Walk) Sw = f2 Output Noise Sv = Standard Deviation vs. TVAR 2 x 2 3 2 n 6 2 x f 2 2 2 B = 2 x x v n 0 ( w2). The relationship between the TVAR variance, PSD, B=1/2 0). 2 x w 1990 60 2 60 2 1980 Raw Gyro TLM = Sw 2 2 and 2000 and decimation ratio n, the random walk PSD SW/f2 with units (deg/s)2/Hz can be estimated. The bias instability in deg/hr/ hr is then given by: Bias Instability = 2 v 2010 Hence, given a point on the TVAR plot to the right with TVAR variance 2 0 w Figure 5 shows an example of the typical “v” pattern of a TVAR plot, with reducing effect of added noise on the left, and increasing effect of random walk on the right. Away from the central knee-point where the two effects balance, taking any point on the curve will give either the variance of the added noise ( v2), or the variance of the white noise driving the random walk process are given in Table 2 (where bandwidth (2) 1970 1960 1950 1940 1930 Estimates of output noise and bias instability made from TVAR plots tend to be overly pessimistic from experience. An alternate, more direct method of estimating the bias instability and output noise is as follows: The mean for each consecutive hour of data is found, and the standard deviation of the differences in these mean values was used as an estimate of the bias instability. The mean of the variances of each 10 10 1910 0 10 10 10 -3 -4 -5 -6 -7 -8 10 0 10 1 2 3 10 10 Decimation Ratio, n 10 4 10 20 30 40 50 Time (hours) 60 70 80 90 Figure 6. Example data set for bias instability estimation An example 90-hour data set from 3 gyros, showing a relatively stable bias. 2 X 10 1920 10 5 Figure 7. Example TVAR plot from real data The solid line is the TVAR plot, and the dotted lines are the estimated output noise and bias instability line fitting. At low decimation ratios, there are more blocks of data, resulting in a smoother curve. Conversely, at larger decimation ratios necessary for estimating bias instability, fewer blocks result in a plot with more pronounced jitter. hour of data was taken as an estimate of the output noise. Such an estimate for output noise would also include effects of random walk, and would therefore be conservative. Figure 6 shows an example 90-hour data set from 3 gyros, taken at constant temperature after initial transients had settled. As can be seen qualitatively from the figure, the gyros demonstrate a very stable output with low bias instability. Figure 7 shows the TVAR plot from a single gyro, along with the estimated output noise and bias instability lines of best fit. Table 3 gives the estimates of output noise and bias stability, based on 90-hour tests on 6 gyros, using both the Allan variance TVAR method and the alternate method described above. As discussed above, the alternate method of estimating output noise results in an overly conservative estimate, as the variance of 1-hour data sets includes both the output noise and to a lesser extent the bias instability. All direct estimates of output noise are within the requirement of 0.01 deg/s. However, the Allan 5 American Institute of Aeronautics and Astronautics Table 3: Estimates of Gyro Output Noise and Bias Stability Gyro Allan Variance Noise (mdeg/s) 13.7 13.2 11.3 13.8 12.6 11.3 1 2 3 4 5 6 Allan Variance Bias Stability (deg/hr/ hr) 1.4 1.54 1.28 2.94 0.28 3.44 Alternate Noise Alternate Bias Stability (mdeg/s 1 ) 9.76 9.39 7.99 9.72 8.91 7.99 (deg/hr/ hr 1 ) 1.42 1.51 1.32 2.42 1.47 2.47 variance estimates of output noise are all larger than these conservative estimates. The bias instability measurements from both TVAR and the more simplistic alternate method are all within 4 deg/hr/ hr. Assuming this is a 1-sigma value, the 3 bias instability of these units are all less than 12 deg/hour/ hr, which is within requirements. IV. Rate Integration Given the low values of calibrated bias and bias instability given above, these units can be considered for short-duration rate integration modes, where the gyro rate output is integrated to give an estimate of the attitude with time. For such maneuvers, it is assumed that the gyro bias is estimated to within some residual error, for example 0.001 deg/s, using some absolute attitude sensor such as a star tracker, at low satellite inertial rates. A residual bias error of 0.001 deg/s is relatively small compared with the output noise of 0.01 deg/s, but should be achievable with appropriate filtering. Such maneuvers are envisaged to be short-duration, for example when the primary attitude sensors are temporarily unavailable. As rate is integrated, errors will propagate and generally increase with time. Figure 8 gives the expectation of attitude errors due to integrating the initial bias, angle random walk (ARW) and bias instability. As can be seen, even though the initial bias of 0.001 deg/s is relatively small, this noise term dominates the estimated angle error for almost 2 hours. This would indicate that performance is determined primarily by the initial bias estimation, and factors such as gyro bias instability and ARW are secondary considerations. degrees V. Conclusion This paper highlights tests carried out with Systron Donner QRS11 gyros for the GIOVE-A mission to achieve calibrated bias and bias instability errors much 8 smaller than the values given in the official data sheet. Bias calibration errors after temperature 7 Init. Bias 1e-3 deg/s ARW 0.6 deg/rt-hr compensation was less than 0.015 deg/s, and bias Bias Instab 4 deg/hr/rt-hr 6 instability was less than 4 deg/hour/ hr. Fixed offset variations in bias was also observed in 5 vacuum. A brief analysis indicated that the bias 4 instability error is low enough to enable this unit to 3 be used as a low-cost short-term attitude sensor in rate-integration mode. 2 In addition, in-orbit results have indicated through indirect measurements that the gyros 1 performed successfully on GIOVE-A. 0 0 20 40 60 80 100 120 Acknowledgments Time (minutes) Figure 8. Comparison of angle estimation error sources for gyro rate integration This figure shows the increase in angle error standard deviation with time due to initial bias error (linearly increases with time), angle random walk (increases with the square-root of time) and bias instability (increases with time to the power 1.5). An example initial bias of 0.001 deg/s dominates almost from the start until almost 2 hours later, even though 0.001 deg/s is a relatively small residual bias. A.Cropp would like to acknowledge the help of Graham Baker of BEI technologies Inc. on his assistance with QRS11 test support and information. References 1 D. Allan et.al., “The Science of Time Keeping”, Hewlett Packard Application Note 1289 6 American Institute of Aeronautics and Astronautics View publication stats