CUBETH SENSOR CHARACTERIZATION: SENSOR ANALYSIS REQUIRED FOR A CUBESAT MISSION

Transcription

1 IAA-AAS-DyCoSS CUBETH SENSOR CHARACTERIZATION: SENSOR ANALYSIS REQUIRED FOR A CUBESAT MISSION Stefano Rossi *, Anton Ivanov, Gaetan Burri, Volker Gass, Christine Hollenstein **, Markus Rothacher Swiss Federal Universities of Lausanne and Zurich have initiated a new 1U CubeSat project. The main objective of the CubETH is to demonstrate use of commercial GNSS receivers in space and test Precision Orbit Determination algorithms. One of the key driving requirements is to provide zenith pointing with a 20 degrees of precision and rotation rate less than 2 degree / second, in order to track GNSS constellation satellites. This paper will describe details of Attitude Determination and Control Subsystem (ADCS) based on magnetotorquers, sun sensors, gyros and magnetometers in order to satisfy the science requirements. We will discuss the lessons learned from the ADCS implementation on the first Swiss nanosatellite SwissCube (in operation since 2009) and we present the characterization of the selected sensors in the laboratories of the Ecole Polytechinque Federale de Lausanne (EPFL) in Switzerland. The main lesson learned from SwissCube is to establish vigorous testing procedures for all sensors. This paper will present tests set-up and procedures used for the present and the future tests in the laboratories of the Swiss Space Center that take experience from that heritage. We describe the main issues encountered in the preparation of tests set-up, interfaces and procedures to establish the trade-off study for the sensors: static and dynamic characterizations, temperature behaviors and others. We address, as conclusions, the main performances obtained after the three test campaigns done. The paper gives to the CubeSat community a validated and simple process to characterize the sensors for the attitude determination. 1. INTRODUCTION CubETH is a project to evaluate low-cost GNSS sensors on a nano-satellite by following the Cubesat standard. GNSS sensors will be used for precise orbit determination and validation of * Aerospace Engineer, stefano.rossi@epfl.ch, Swiss Space Center EPFL, Lausanne, Switzerland Scientist, anton.ivanov@epfl.ch, Swiss Space Center EPFL, Lausanne, Switzerland Microelectronic Eng. Student, gaetan.burri@epfl.ch, EPFL, Lausanne, Switzerland Professor, volker.gass@epfl.ch, Swiss Space Center - EPFL, Lausanne, Switzerland ** Scientist, christine.hollenstein@geod.baug.ethz.ch, ETHZ, Zurich, Switzerland Professor, markus.rothacher@ethz.ch, ETHZ, Zurich, Switzerland 1

2 attitude determination of the cube. The project shall verify in-space use of COTS GNSS detectors and novel algorithms for on-board data processing. The main scientific goal of CubETH satellite is precision orbit determination using COTS GNSS sensors. Additionally, we aim at characterizing attitude determination using GNSS sensors. Programmatic goal is to implement this project in cooperation between ETHZ and EPFL schools, involving engineers and students from federal schools as well from HES / FH domain. This project will serve towards education of new generation of highly qualified engineers. CubETH shall have a nadir pointing with a relaxed requirement on the precision of 20deg from the nadir axis maintaining an angular speed below 2deg/s (stability requirement). An Extended Kalman Filter (EKF) algorithm has been developed in two versions to guarantee the nadir pointing during daylight and eclipse [1], controlling through a PWM interface the three magnetotorquers. Since the precision and the stability are not stringent, the results obtained by preliminary simulations showed that an ADCS based on magnetotorquers meets the requirements. This paper describes the sensor characterization process for the CubETH mission, based on the SwissCube project and its results after more than four years of successful operations. Swiss- Cube heritage allowed to identify improvements for our next project. This paper shows the main problems that can be encountered on orbit starting from the SwissCube data and addresses the milestones for necessary characterizations in order to avoid on-board determination problems. With the current set of COTS (Commercial of the shelf) sensors selected for CubETH, these tests were performed in the laboratories of the Swiss Space Center and the paper details test setup and analysis, highlighting information that can help the CubeSat community to understand where the main effort must be focused in the ADCS validation and tests. The main goal is to underline that vigorous testing procedures for the sensors are essential for building Cubesats capable of performing scientific or technological goals. 2. CUBESAT MISSION: CUBETH CubETH follows the CubeSat design specification with the 10x10x10cm volume (1U) and 1kg mass. Precision Orbit Determination functionality is implemented via reception and analysis of GNSS signals from either GPS or GLONASS constellations in order to estimate the attitude and location on orbit of the spacecraft. Six GNSS Antennas are mounted: four on the top (zenith) and two on the sides. The payload design is lead by the ETH Zurich while the Swiss Space Center is in charge of the satellite s bus with the collaboration of other engineering school (HSR, HSLU, HES-SO) Figure 1: CubETH CAD. On the top and on two sides six GNSS receivers are placed. 2

3 The current design is shown in [Fig.1]: the ground Antennas are placed on the nadir face, opposite to the GNSS face (zenith). The satellite shall maintain a nadir-pointing to have access to the above GLONASS and GPS satellites for at least 5min to complete the necessary step for the experiments through the GNSS receivers. This driving requirements has given (after different simulations and orbit scenarios) two main requirements on the pointing and on the stability: Req.1 The ADCS shall maintain nadir-pointing with ±20deg of precision for at least 5min. Req.2 The ADCS shall maintain the pointing requirement with maximum 2deg/s on each axis for at least 5min. Req.3 The ADCS shall provide a determination of the attitude with a 2deg of precision and a 0.5deg/s on each axis A control based on magnetotorquers has been chosen to achieve [Req.1] and [Req.2] due to the non-strict level of these. After several simulations done [1], the results showed that even if with some limitation and hard requirements on the memory processor, the actuators can reach [Req.1] and [Req.2]. [Req.1] and [Req.2] define the actuators and the algorithm control choice. The control is based on an Extended Kalman Filter. Two algorithms are under validation at the Swiss Space Center: one based on Sun Sensors, Magnetometers and Gyros measurements during the daylight, one based just on magnetometers and gyros measurements for the eclipse [1]. [Req.3] will be satisfied by a common configuration of gyros, magnetometers and Sun Sensors. These sensors are currently under characterizations and the preliminary results obtained are here described highlighting the main issues coming from COTS sensors, the test-set-up used and the procedures. The characteristics of the Mangetotmeters (MM) and of the Gyroscopes (GY) are here investigated. The MM and GY are place on the CubETH ADCS board and interfaced with an I 2 C interface and powered by a 3.3V as all the subsystems on CubETH. The following parameters of the sensors have been studied and compared with the ones given by the supplier s datasheet: Noise model (Gaussian noise, bias), Temperature dependence, Sensitivity, Linear operating range, Resolution. These characteristics have been investigated in three tests that can be identified as: Static Tests, Dynamic Tests and Thermal Tests. These three tests are the test-campaign that has been found necessary and sufficient to provide an easy and exhaustive comprehension of the sensor for an ADCS that needs to provide determined attitude for a CubeSat project. This conclusion comes from the lessons learned with SwissCube: after four years of flight data form the sensors [3] it appeared clear that for improving the CubeSat community and bringing it as future platform for scientific missions and payload, the main effort on the ADCS team must be in a vigorous testing procedures for all the sensors. Moreover, [sec.2] some unexpected behaviors on board of SwissCube gave to the Swiss Space Center the heritage to investigate the necessary characteristics to avoid drifts, bias and oscillations. The drive requirements for the CubETH MM and GY are listed in [Table.1]. These requirements come again from SwissCube and its four years of data. The Magnetometers have the range of the magnetic field at the determined altitude for CubETH, while the temperatures are the operative range coming from the housekeeping data downloaded from SwissCube. The gyros have a very large operative range deg/s: this comes again from the SwissCube attitude monitored during the An unexpected speed over 120deg/s (angular speed vector) has been detected and saturating the gyros, the further calculations ([3]) showed a possible angular speed vector of more than 600deg/s. The thermic range for the GY has been maintained (as monitored on SwissCube) in the same range of the MMs. 3

4 Table 1: Sensor s Requirements Magnetometer Gyros Range ±65µT ± deg/s Temperature -30 C to 70 C -30 C to 70 C 3. BACKGROUND AND SWISSCUBE HERITAGE Most of the ADCS for CubeSat have the same configuration for the attitude determination (Sun Sensors or Earth sensors, Mangetometers and Gyros), and most of them these sensors are commercial-off-the-shelf components if they are not payload or models to be tested. What was observed from the past Swiss Space Center s experience is that the CubeSat community is becoming more than a test platform for demonstration and technology validation, nowadays the idea of CubeSat mission with scientific payload is becoming reality. The driving requirements on CubETH are to perform science experiments with GNSS sensors, while providing an exciting space hardware experience to engineering students CubETH derives its practices from the lessons learned on SwissCube, providing a feasible platform for a payload (GNSS antennas and algorithms from ETH Zurich). After four years of flight and thousands collected data points, Swiss- Cube has given enough feedback to the Swiss Space Center in order to provide a robust and validated platform for a scientific payload. In the next section the main focus is on the ADCS lessons learned from SwissCube that has given the inputs for the CubETH s improvements. 3.1 SwissCube Attitude Determination Results With more then four years of data, SwissCube has given enough feedback and results, especially on the ADCS subsystem, where we have identified a number of improvements Angular speed: Saturated Not Saturated 140 [deg/s] [days] Figure 2: Angular speed detected by the gyros since the start of the SwissCube mission (23 September 2009 until 5 January Note that until January 2011 gyros were saturated. After the release from the pod, SwissCube rotation accelerated and the data collected until January 2011 showed the three gyros completed saturated and with three different values of saturation even if the three GY were the same. During 2010 it was not possible to understand the atti- 4

5 tude of the body since the GY complete saturated, however, for natural dumping the body slowed down until the beginning of 2011 when rotating with 98deg/s. After 15days of ADCS operations using a B-dot controller and the Mangetotorquers, SwissCube finally reached an angular rotation of 2-5deg/s. This issue demonstrated that the GY must have a bigger range of measures, in this way on CubETH will be mounted two types of GY: one with a big operative range but low performances, and one with better performances in a smaller range of angular speed. Moreover, one other problematic not taken into account was the dynamical response of the sensors, no dynamical considerations have been previewed and no drifts could be used to post process the data (or either process on board). This has been taken into account for CubETH, since no properties are clear on this side for the COTS components. One of the behavior characterized before flight on the gyros was the bias detected in function of the temperatures on the ADCS board (close to the sensors). As shown by [Fig.3], the gyros had different behavior and bias in function of the temperatures due to their mechanical interfaces and because three one-axis sensors. This has been used for post processing the data collected and the results are shown in [Fig.5]: taking the temperatures of the sensors and using the functions shown in [Fig.3], the measurements can be filtered in real measurements deleting the bias due to the temperatures. This approach has been done just for the Gyros and not for the Magnetotmeters unfortunately and no data can be shown on it. However the same idea is taken both for GY and MM for CubETH and processed on board due the higher capacity of the microcontroller that will be mounted than the previous MSP microcontroller mounted on SwissCube. In SwissCube, the MM worked properly even if wired oscillations have been collected: these are probably due to interaction with the MTQs or with dipole generated by internal coils of wires. In this way the internal structure of CubETH has been re-designed in a no-wires configuration, avoiding unexpected (and difficult to measure) parasite dipoles and future characterization are scheduled once the MTQs and al the subsystem will be integrated togheter. Other lessons learned for Sun Sensors and the camera mounted on SwissCube are not here presented. 15 Gyros Magnetometer MSP430 Temperatures 10 5 [Celsius] X TP Y TP Z TP MM TP MSP TP 16 Jun :53: [min] Figure 3: Bias detected before flight on the SwissCube GYs integrated on board Figure 4: Temperatures detected on Gyros (x-y-z-tp) and on Magnetometer (MM- TP) and on the Microcontroller (MSP-TP) 5

6 0 GYROS 29 Jan :31:40 1 [deg/s] X Y Z X clean Y clean Z clean [min] Figure 5: Gyros angular speeds pre (dashed lines) and post processed (full lines) 4. THE COTS COMPONENTS: GYROS AND MAGNETOMETERS After a previous trade-off through the different COTS components available that could fill the requirements for the attitude determination the final choice has been done on the Gyroscopes and Magnetometers listed in the following tables. The main key features for the selection are based on the range of measurements, on the operative temperature ranges and on the interface (I 2 C). For one unit CubeSat the possible COTS magnetometers found are: HMC1043, HMC1053, HMC5883L, HMC6343, MAG3110. All these sensors utilize Anisotropic Magnetoresistive (AMR) technology. They are thus very sensitive, low-field, solid-state construction sensors compared to the coil based magnetic sensors. They are designed to measure both the direction and the magnitude of the Earth s magnetic field. The MM finally selected that can fit al the requirements are the HMC5883L * and MAG3110. For the gyros the trade-off has been made through these pre-selected COTS components: ITG (3-axis), 3 ISZ-1215 (1-axis), 3 ADXRS623 (1-axis), 3 ADXRS624 (1-axis), L3G4200D (3-axis), MPU-3300 (3-axis). A first trade-off was done to define which sensors would be tested and among all these sensors, the ITG-3200 and the L3G4200D have been chosen. They have been selected because they already have an ASIC and the communication interface is I 2 C and secondly, the other sensors are just one axis sensors, so this means that three of them would be mounted on the ADCS board and on a separate support (total of 3 gyros). This configuration was * Honeywell, HMC5883L datasheet Rev E, February [Online]. Available: [Accessed November 2013]. F. Semiconductor, MAG3110 Datasheet Rev 9.2, February [Online]. Available: [Accessed November 2013]. InvenSense, ITG-3200 datasheet1 Rev 1.7, 8 August [Online]. Available: [Accessed November 2013]. InvenSense, ITG-3200 datasheet2 Rev 1.0, 6 June [Online]. Available: [Accessed Novemeber 2013]. STMicroelectronics, L3G4200D Datasheet Rev 3, 22 December [Online]. Available: [Accessed November 2013]. 6

7 used for SwissCube and it was discovered that the temperature of the Y-gyro was slightly different from the others [Fig.4]. This influenced the data collected during the flight. 5. TESTS DESCRIPTION AND TEST-SET-UP This section resumes the tests that were performed, as well as the equipment that was needed and used. For further information about the tests procedure, and the interfaces a complete and detailed list can be found at [6] and [7]. As already mentioned, the tests are based on three great themes: the static tests, the dynamic tests and the thermal tests. The first concerns about the noise models of the sensors, resolutions and offset; the dynamic response of the sensors considers the output of the sensors in a dynamic environment where the measured source is not zero and the linear operative range is highlighted; the thermal tests are based on the measurements of the output when an external source of heat influences the captor as already done in [Fig.3]. All the sensors selected have an ASIC which allows choosing different parameters for the measurements, e.g. the output data rate, the cut-off frequency of a low-pass filter, the full scale and so on. To access the register of these parameters, a library for each sensor has been written. Actually, some library for the HMC5883L and the ITG-3200 already exist and are released into the public domain. They were downloaded from online open sources *. The purpose to have libraries is to have a smaller and more readable Arduino code, and faster parameter change. Furthermore, they can be reused for additional tests, or even for the coding of the ADCS software. However, they are not complete, as they do not include yet functions to change each parameter individually. Important to notice is that most of the time has taken to design a proper interface to connect and to test the sensors on the test set-up. For instance: [Fig.7], [Fig.8] in the bottom, [Fig.9] and [Fig.11] are the interfaces developed that took most of the time for the project. It s mandatory to take this backward into the project management in order to fit the time schedule. 5.1 Static tests: Test set-up For the characterization of the noise model, resolution and bias, the output of the sensor has been read without affecting its input. For the MM, due to the influence of the magnetic field of the Earth and its variation, the tests have been done both in an earth-influenced place and in a shielded room where the influence of the magnetic field is minimized. 5.2 Dynamic tests: Test set-up For the GY dynamic tests the Swiss Space Center has used a rate table shown at [Fig.8]. The connectors KPSE06F18 for the connection at the top and base of the table were not mounted with cables. Thus an interface with user-friendly connector has been developed, which allows a fast plug-in and a solid connection between the top and the bottom of the rate table. Modular Terminal Strips with push button on both sides from WAGO have been chosen and an aluminum plate to fix them on the rate table has been designed. Then, the whole system has been mounted on the top and at the table base. [Fig.8] shows the rate table with its interfaces, as well as the computer linked to the rate table (on the right) and the control unit (on the left). For the rate table, a cube to * bildr.blog, [Online]. Available: [Accessed November 2013]. google code itg-3200driver, [Online]. Available: [Accessed November 2013]. 7

8 fix the GY has been manufactured. The GY are successively fixed on three of its faces in order to test each axis Figure 6: Gyros interface designed to attaché to the top of the rate table Figure 7: Rate Table with the interfaces at the top and at the base Figure 8: Interfaces for Helmotz coils and sensors Figure 9: Helmotz coils with sensor (red PCB) attached to the interface (black, below the sensor) For the MM dynamic tests, three different pieces have been produced in order put the MM in the middle of the coils [Fig.9]. The material used is POM to avoid any possible short-circuit on the PCB. Each piece allows aligning one axis of the MM with the MF created by the coils. 5.3 Thermal tests: Test set-up Figure 10: The thermal vacuum chamber Figure 11: mechanical interface for the 8

9 at the Swiss Space Center [Fig.11] shows the Thermal Vacuum Chamber used at the Swiss Space Center to test the sensor inside a thermal environment. Since the heat transfer in the TVC is done solely by conduction through a plate, it s necessary to have the chip of the sensors leaning against it. A plate in aluminum has been developed with four interfaces that lean the sensors against heated face. The Al plate [Fig.12] is fixed in the TVC against the conductive plate. The place of the sensors (GY and MM) on the Al plate is shown in [Fig.12]. 6. TESTS PROCEDURES & RESULTS In this section, the tests procedure and the results obtained by the three campaigns of tests are presented and discussed highlighting the performances of the COTS components in comparison with the datasheet and the expected results. The lessons learned from SwissCube have been used to improve the characterization of the attitude determination sensors, and the following results show that the three tests done are extremely helpful to determine the behaviors of the sensors and their backwards. 6.1 Static Tests thermal vacuum chamber The purpose of the static tests is to determine the standard deviation of the sensors, as it is needed to know it to compute the Kalman filter for the ADCS software [1]. The standard deviation is also the noise RMS of the sensor as well as its resolution (standard deviation = 1σ). Thanks to these tests, it can be known the bias of the GY and of the MM, which is the mean value of the output over several measurements. The original histograms with the LSB scale are not shown in this section, but they can be found in [2], [6] and [7]. The graphs presented in this section have their scale in μt for the MM and in /s for the GY. The conversion from LSB to μt or /s can be done thanks to the sensitivity with the following relations: Output [μt] =!"#$"#!"#!"#$%&%'%&( [!"#/!"] Output [ /s] =!"#$"#!"#!"#$%&%'%&( [!"#/( /!)] (1) The sensitivities used for the following histograms come from dynamic tests([sec.6.2]) Mangetometers tests: The two MM (HMC5883L and MAG3110) present two different types of measurements that are internally averaged with different number of samples. Here are presented the performances of the measurements with 1 and 8 sample averaged (for the HMC5883L) and 16 and 128 samples averaged (for the MAG3110). The tests have been performed in a shielded room where the Earth magnetic field is minimized and in a normal room. The following graphs show just an example of the results obtained by the 500 samples taken during the measurement while the tables summarize all the results obtained in the test campaign. HMC5883L The most interesting results obtained come from the static measurements done in the shielded room, where the magnetic field is minimized and where the Off-set of the sensors can be better highlighted. [Fig.14] are the two tests done with 500 samples in the shielded room and averaged with 1 and 8 samples. 9

10 Figure 12: Histogram of the MM HMC5883L output in shielded room (at ambient temperature, 8 samples averaged) The sensor seems to have a better resolution than what claimed by the datasheet. Furthermore, the noise level is all the time lower with an internal average of 8 samples than with only one measure: there is an improvement by a factor of 2. The sensor thus has been used with the internal average for the next tests. The mean value of the sensor s output when there is no external excitation is shown in [Tab.3]. However, these values are not really representative since, for the measurements in a normal room, the component of the Earth s MF is present (which adds an important offset to the output), and in the shielded room, there is always small remnant of MF from objects or tools. Table 2: Resolution of the HMC5883L obtained by 500 samples at ambient temperature in normal and shielded room X axis Y axis Z axis Datasheet Number of samples averaged Resolution in normal room [nt] Resolution in shielded room [nt] Table 3: Offset of the HMC5883L without external excitation obtained by 500 samples at ambient temperature in normal and shielded room X axis Y axis Z axis In Datasheet Number of samples averaged Offset in normal room [µt] Offset in shielded room [µt] MAG3110 The resolution of the x and y axes with an internal sample average set to 16 seems to be as good as or even better than the expected one ([Tab.4]). However, when there are 128 samples averaged, the noise level is sometimes higher than with only 16 samples, and it is worse than what is expected from the datasheet. The z axis has a worse resolution as it can be seen in (the noise level is almost twice higher than these of the two other axes). Thus, it is not necessary to use this MM with 128 samples averaged as it does not bring any improvement, but it consumes more power: 34.4μA typically for 16 samples averaged against 275μA typically for 128 samples averaged (from (Semiconductor, 2013)). 10

11 Figure 13: Histogram of the MAG3110 output in shielded room (at ambient temperature, 16 samples averaged) The mean value of the sensor s output when there is no external excitation is shown in [Tab.9]. However, these values are not really representative since, for the measurement in a normal room, the component of the Earth s MF is present (which adds an important offset to the output), and in the shielded room, there is always small remnant of MF from objects or tools. Table 4: Resolution of the MAG3110 obtained by 500 samples at ambient temperature in normal and shielded room X axis Y axis Z axis In Datasheet Number of samples averaged Resolution in normal room [nt] Resolution in shielded room [nt] Table 5: Offset of the MAG3110 without external excitation obtained by 500 samples at ambient temperature in normal and shielded room X axis Y axis Z axis In Datasheet Number of samples averaged Offset in normal room [µt] Offset in shielded room [µt] Gyros tests: The two GY selected (ITG-3200 and L3G4200D) have been tested fixed without any movement and here are presented the results and the issues encountered. During the test of the ITG-3200 it has been found that the output of the sensor is mechanically sensible to the deformation of the interface where is fixed. For instance: it was observed that the bias of the sensor was changing when the PCB was screwed on the cube support [Fig.7]. It has been fixed with 3 screws on the boarder of the PCB, which creates a flexion of the PCB and some constraints on the chip. This has the effect to change the bias. As it can be seen in the following results, the difference for the x axis is about 5 /s and is lower for the y and z axes. The L3G400D gives the possibility to change the Full Scale range (FS): the 250dps (minimum) and the 2000dps (maximum) have been chosen for the tests. ITG-3200 For the GY, it is interesting to compare the noise density measured with the one from datasheet. For that, the noise (the output s standard deviation in /s in [Tab.10]) is divided by the square root of the low-pass filter. For the tests on this sensor, the bandwidth was set to 20 Hz. 11

12 The noise density is given in [Tab.11] and it is decreased by a factor of 3 than what is expected from the datasheet and almost seen for all axes. The resolution of the sensor is also very good since it is below 0.05 /s. What is very important to note is that the noise does not change when the PCB is screwed, but just the biases are different from the two tests with the PCB fixed and not. The biases are within the tolerance given in the datasheet, even if the x axis presents an higher value. Figure 14: Histogram of the GY ITG-3200 output (at ambient temperature, PCB screwed) The double test with the screwed PCB was not expected, however this gave the possibility to find an interesting behavior of the sensor that are mechanically dependent and in this way the thermal variance that can cause displacement of the interfaces should generates biases. The final conclusion is to repeat the tests for the characterization of the biases with the final configuration on the ADCS board and focusing on the thermal characterization due to the possible thermal deformations of the boards on which the sensors will be fixed. Table 6: Resolution of the ITG-3200 X axis Y axis Z axis In Datasheet Resolution [ /s] (PCB free) Resolution [ /s] (PCB screwed) Table 7: Noise Density of the ITG-3200 X axis Y axis Z axis In Datasheet Noise Density [ /s/ Hz] (PCB free) Noise Density [ /s/ Hz] (PCB screwed) Table 8: Bias of the ITG-3200 X axis Y axis Z axis In Datasheet Bias [ /s] (PCB free) ±40 Bias [ /s] (PCB screwed) ±40 L3G4200D 12

13 Figure 15: Histogram of the GY L3G4200D output (at ambient temperature, FS 250 dps) The noise (the output s standard deviation in /s in [Tab.9]) is divided by the square root of the low-pass filter. For the tests on this sensor, the bandwidth was set to 25 Hz. The noise density is given in [Tab.10] and it is close to the expected values from the datasheet. There is just one value which is bit higher for the z axis with a FS of 2000 dps. The resolution presents values not exceeding /s. It is a little bit better with a FS of 250 dps. The biases ([Tab.15]) are within the tolerance given in the datasheet. Table 9: Resolution of L3G4200D X axis Y axis Z axis In Datasheet Resolution [ /s] (FS 250 dps) Resolution [ /s] (FS 2000 dps) Table 10: Noise density of L3G4200D X axis Y axis Z axis In Datasheet Noise density [ /s/ Hz] (FS 250 dps) Noise density [ /s/ Hz] (FS 2000 dps) Table 11: Bias of L3G4200D X axis Y axis Z axis In Datasheet Bias [ /s] (FS 250 dps) ±10 Bias [ /s] (FS 2000 dps) ± Dynamic tests The purpose of the dynamic tests is to identify the sensitivity of the sensor and their nonlinearity when the measured source is varied along the acquisition range. The dynamic tests have been done within Helmholtz coils for the MM and on the rate table for the GY. The range of the magnetic field created is between [-66.08μT; 66.08μT] with a step of 1.12μT. The range of angular speed is between [-1000 /s; 1000 /s] with step of 50 /s when the FS range is 2000dps, and between [-250 /s; 250 /s] with step of 10 /s when the FS range is 250dps. 100 data are collected from the MM for each MF created and 200 data are collected from the GY for each angular speed. The mean is then calculated for each excitation. It must be noticed that due to the lack of time some modes (for example the 16-sample averaged mode for the MAG31110 and the 1-sample averaged mode for the HMC5883L) has not been performed and they are previewed in the next months Mangetometers tests: HMC5883L 13

14 The dynamic tests for this MM were only done for the case when the output is the average of 8 samples. This average is done internally to the sensor. Due to a lack of time and the long period needed to do these tests, the 1-sample averaged dynamic characteristics were not performed during this project. A typical characterization on one axis can be seen in [Fig.16]. The same tests done for [Fig.16] have been done for Y and Z axes and for each graph, a linear approximation of the samples has been drawn (in red) and the determination coefficient (R 2 ) has been calculated, giving an idea of how well the data fit the line. From these data, it is also possible to calculate the sensitivity of each axis which is the slope of the curve, and the non-linearity in percentage of the full-scale which has been given by the maximum output deviation from the best fit line divided by the full-scale output. [Tab.12] gives the values of these parameters Figure 16: Dynamic characterization for MM HMC5883L X-axis (at ambient temperature).the sensitivity of the z axis is closer to the datasheet s data, whereas the x and y axes are a bit higher. The z axis has also better performance in term on linearity. It is important to note that the non-linearity is here calculated as the maximum deviation from the ideal line of the output in the range of test, i.e. ±68.08 μt, divided by the full scale of the sensor, i.e. 800 μt Table 12: Dynamic characteristics for MM HMC5883L Axis of the sensor X Y Z Datasheet Sensitivity [LSB/μT] R Non-linearity (in % of FS) The lack of linearity can come from the magnetic field, which may has not been at the expected value during the testing phases due to a lack of precision on the ammeter. Another possibility is that the sensor is not as good as what the supplier says in the datasheet. Or this specific MM has some small defaults. To know, it would be great to perform tests on other MM HMC5883L. MAG3110 The dynamic tests for this MAG3110 have been done only done for the case where the output is the averaged every 128 samples. This average is done internally to the sensor. Due to a lack of time and the long period needed to do these tests, the 16-sample averaged dynamic characteristics were not performed during this project. As shown in [Fig.16] the same graphs for the three axes have been done and the following table resume the performances and the results obtained from that analysis. 14

15 Table 13: Dynamic characteristics for MAG3110 Axis of the sensor X Y Z Given in the datasheet Sensitivity [LSB/μT] R Non-linearity (in % of FS) The sensitivity of the z axis is a little worse than the other axes, but the characteristics of each axis remains in general very good. Again, the non-linearity is here calculated as the maximum deviation from the ideal line of the output in the range of test, i.e. ±68.08 μt, divided by the full scale of the sensor, i.e μt Gyros tests: ITG-3200 The dynamic tests for this GY were only done for the first range, i.e. from /s to 1000 /s since the FS range is unique for this sensor, i.e. a FS range of 2000dps. The three graphs obtained (for the three axes) with the same approach as for the MM are here summarized in the performances shown in [Tab.14], where the performances are estimated. Table 14: Dynamic characteristics for ITG-3200 Axis of the sensor Given in the datasheet X Y Z Sensitivity [LSB/( /s)] R Non-linearity (in % of FS) The performances of this GY are very good, and close to the characteristics given in the datasheet by the supplier. The linearity is even much better than what is written in the datasheet. L3G4200D The dynamic tests for this GY were done for both ranges, i.e. from /s to 1000 /s and from -250 /s to 250 /s with a FS range of 2000dps, respectively 250dps. It is possible to calculate the sensitivity of each axis and the non-linearity in percentage of the full-scale in the same way done with the ITG3200. [Tab.15] gives the values of these parameters. Table 15: Dynamic characteristics for GY L3G4200D (250dps FS) Axis of the sensor Given in the datasheet X Y Z Sensitivity [LSB/( /s)] R Non-linearity (in % of FS) The performances of this GY in FS = 250 dps close to the characteristics given in the datasheet by the supplier. There is just the non-linearity which is slightly different than what the supplier suggests. Table 16: Dynamic characteristics for GY L3G4200D (2000dps FS) Axis of the sensor Given in the datasheet X Y Z Sensitivity [LSB/( /s)] R Non-linearity (in % of FS)

16 With the FS set to 2000 dps, the characteristics remain close to the datasheet Dynamics tests summary: Gyros and Magnetometers After the dynamic tests were done, it was decided to reuse the data collected to see if the noise of the sensor is constant, or if it is varying with the input (the Magnetic Field or the angular speed). Each point on the graphs below corresponds to the standard deviation for 100 data points for the MM and for 200 data points for the GY. The noise of the MM HMC5883L is constant for any value of MF, except for few points. These higher values of the noise might come from different errors that occur during the tests: maybe someone entered in the shielded room during one measurement, the sensor was thus sensing the Earth s MF which implies a variation of the output and a higher standard deviation. Another possibility is that an object was moved to close to the sensor, modifying the MF lines and thus the output of the sensor. Figure 17: noise variation of HMC5883L over test range Figure 18: Noise variation of MAG3110 over test range In the case of the MAG3110, it is clear that the z axis is worse than the two other axes, as it was already observed in the static tests. However, this sensor also shows a constant noise level over the range of test. From [Fig.19], it is clear that the noise level of an axis is increasing with the angular speed around this axis (blue star for x axis, green plus for y axis and red x-mark for z 16

17 axis). This behavior seems to be linear. With a FS of 250 dps, the standard deviation of the GY L3G4200D is insensitive to the variation of the angular speed. The noise of the y axis is lower than for the other axes as it was already observed during the static tests. However, when the FS is set to 2000 dps, the noise varies with the angular speed (blue star for x axis, green plus for y axis and red x-mark for z axis in [Fig.20]). The linear dependence of the noise with the input is more marked for this sensor than for the GY ITG It can also be observed that noise level increases slightly even if the angular speed is not along the axis (as for the red plus). This is due to the misalignment of the GY with the rate table. Thus, the two other axes also have a small angular velocity, and thus their noise becomes bigger. Figure 19: Noise variation of ITG-3200 over test range Figure 20: Noise variation of GY L3G4200D over test range (FS of 2000 dps) 6.3 Thermal Tests The thermal tests were unfortunately done only for the GY ITG-3200, due to a lack of time. Indeed, the temperature of the Al plate on which the sensors are fixed has its temperature changing very slowly (about 10 C per hour), although the TVC s bath temperature is changing much faster. This means a measurement period of about hours per sensor. Thus, it will be more 17

18 appropriate to perform these tests in a thermal chamber where there is convection (and not only conduction as in the TVC). Despite the fact that one wanted to perform the test on the range from -30 C to 70 C, the tests were finally done on the range from -20 C to 70 C because it was not possible to obtained a temperature below -20 C inside the TVC. The graphs presented in this section have their scale in /s for the GY. The conversion from LSB to /s is explained in [sec.6.1]. The sensitivities used for the following histograms come from dynamic tests [sec.6.2]. Figure 21: Thermal characterization for GY ITG-3200 X-axis The Z-axis results are not presented because they follow the linear approximation. Only the Z- axis follows a line, except for the point at -15 C which may be due to an error during the measurement. This behavior was expected for the three axis of the GY, in order to know the bias sensitivity in function of the temperature. However, the X and Y axes present some kinds of discontinuities: from -20 C to about 10 C, we almost have fixed bias, and this bias is increasing more or less linearly from 35 C to 70 C. One possible explanation to this phenomenon is that the fixation of the PCB with screws creates some constraints in the PCB during the thermal cycle. These constraints affect more drastically the X and Y axis. When the GY is screwed on the Al plate, it bends slightly the PCB around the X-axis. This bending probably does not vary over the temperature change, as it should without the PCB screwed. This may affect more the measurements of the X-axis than the ones of the Y-axis. Furthermore, the GY chip is leading against the Al plate, which provides more constraints on the plane XY for temperature change. This might be the reason why the ZRO of the X-axis is highly nonlinear and the ZRO of the Y-axis is a bit curved. The cavities of the masses for the Z-axis on the opposite are situated on the left of the left image above the poly capacitors, one above the other (not shown here). This axis may thus be insensitive to the bending variation of the PCB over the temperature. Figure 22: Thermal characterization for GY ITG-3200 Y-axis 18

19 6.4 Single Events Among all these tests, a lot of data were collected for each sensor. But, sometimes a single event error occurred, which means that the output was totally beside the expected value. As it can be seen in [Tab.17], the percent of error is below 0.2% for each sensor. This is a very important parameter as it proves that the sensors are reliable. Table 17: Single Event Error of the sensors Sensor N data collected N single event error Percent of error MM HMC5883L % MM MAG % GY ITG % GY L3G4200D % 7. DISCUSSION As explained above, offsets of the MM calculated during this work are not important. The parameters that would be interesting to know are the hard-iron interference and the zero-flux offset for compensation in the ADCS software. However, it was not useful to calculate them here since they depend on the device soldering on the PCB. Thus, they will be totally different once the MM will be mounted on the final ADCS board. These parameters need to be calculated when the integration of the board will be completed. The thermal tests have to be performed once the integration of the board is finished, and preferably in a thermal chamber (not a TVC) where the temperature of the board will reach faster the equilibrium. Indeed, the ADCS board will not these constraints due to the screwing as the small PCB of the sensors had. Thus the sensors chip will also have fewer constraints and they will perhaps present some linearity of their output over temperature (as the z axis for the GY ITG-3200). Several other possible tests can be considered for future work: - The bias stability of the GY over time: it would be great to know if the offset may vary over a long period, and in such a case, if there is a relation between the bias and the time - The power consumption of the sensor: since the satellite will have a limit of power for each subsystem, it is important to know the consumption of the device. To perform this test, the only thing to do is to measure the current that the device draws in operating mode. The power is then the multiplication of the current with the voltage. 8. CONCLUSIONS This paper presents results of the test campaign to characterize the sensors for the attitude determination on a cubesat. We have based our tests on SwissCube heritage and more than 4 years of in-flight data. In particular for the Attitude and Determination Control System, the paper underlines that vigorous testing procedures for the sensors is essential to implement systems to support a science experiment. After the analyses of the on-orbit data collected from SwissCube [3], interesting behaviors underestimated during the ADCS tests for SwissCube have been found. These gave the feedback necessary to improve the test campaign for the sensors characterizations that here have been presented. Static, Dynamic and Temperature tests are the focus for the CubETH s ADCS in order to define the necessary characteristics for the Control Algorithms and, moreover, to preview and process those behaviors detected with the flight data of SwissCube. Important issues have been underlined in the paper, such as the unpredicted amount of time used to prepare the test set-up and either the time to perform the tests. These must be two fundamentals key points in the CubeSat s ADCS, because, in order to have an optimum and reliable 19

20 test campaign, the time constraints to the preparation and the execution must be taken into account. The objectives of this study has been the test and the characterization of magnetometers and gyroscopes from different suppliers, and see which one could fit the best the requirements for CubETH. Thanks to the results obtained during this work, it is already possible to make a comparison between the sensors under tests, and a first choice: - The magnetometer HMC5883L satisfied requirements for CubETH as it has less noise: about nt (with a sample average of 8) against about 250 nt for the magnetometer MAG3110. Furthermore, it has a higher sensitivity (14 LSB/µT). The only disadvantage found is the linearity (non-linearity between % of FS), which is worse than what is said in the datasheet (non-linearity of 0.1%). However, its linearity is comparable to the MAG3110 s linearity ( % of FS). - For the gyroscopes: the ITG-3200 shows an average resolution of 0.04deg/s, whereas L3G4200D has the possibility to change the full-scale range and has a lower bias (less than 1.5deg/s for the L3G4200D and more than 15deg/s). Both of them present linearity close to the expected values of the data-sheet (non-linearity below the 0.2%) REFERENCES [1] Bonnet Frank, Rossi Stefano, Anton Ivanov Application of Kalman Filtering to Control Systems of Small Spacecraft CubETH Project, Swiss Space Center Library [2] Burri Gaetan, Rossi Stefano, Anton Ivanov CubETH Attitude Determination Sensors Qualification, Swiss Space Center Library [3] Stefano Rossi, Anton Ivanov, Muriel Richards, Volker Gass The Swisscube s Technologies Results After Four Years Of Flight, IAC Beijing IAC-13-B4.6B.5 [4] Daniel Selva, David Krejci A survey and assessment of the capabilities of Cubesats for Earth observation, ActAstronautica Volume 74, May June 2012, Pages [5] J. Bouwmeester, J. Guo Survey of worldwide pico- and nanosatellite missions, distributions and subsystem technology, Acta Astronautica Volume 67, Issues 7 8, October November 2010, Pages [6] Burri Gaetan, Rossi Stefano, Anton Ivanov Gyroscopes Test Specification Procedure and Report, Swiss Space Center Library [7] Burri Gaetan, Rossi Stefano, Anton Ivanov Mangetometers Test Specification Procedure and Report, Swiss Space Center Library 20