Data synchronization and time-keeping Thorbjrn Jrger Lehrstuhl fr - PowerPoint PPT Presentation

Data synchronization and time-keeping Thorbjörn Jörger Lehrstuhl für Systemtheorie Institut für Mikrosystemtechnik - IMTEK Albert-Ludwigs-Universität Freiburg

Groundstation - 2 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Flight Controller - 3 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Clock synchronization  Reference pulse which occurs simultaneous in all systems  1PPS signal of GPS is the natural choice for outdoor experiments no information exchange between systems necessary Absolute and jump-free GPS time reference is also available Small jitter (15-50 ns) with cheap equipment, < 1 ns with better equipment Indoor Environment usually has to low signal strength t N t N t N GS FC ARM t N is a synchronous timing pulse - 4 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Time keeping  Each measurement gets a timestamp in a unified time frame  Need for synchronous clocks Flight Controller:  Self designed timing framework  1PPS pulse for synchronization  Intervals between pulses measured with precise quartz oscillator  Absolute deviation below 10 µs - 5 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Time keeping  Each measurement gets a timestamp in a unified time frame  Need for synchronous clocks Groundstation:  Synchronized with UTC via GPS  Less than 1 µs deviation from true time  Essentially a Stratum 1 time server  Automatic synchronization on startup < 2 min  GS references ticks and PPS pulses to UTC  Keeps track of all quartz oscillators in the system - 6 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Time keeping GPS devices – Furuno GT-87 / GF-8701:  Lock on GPS, GLONASS, Galileo, QZSS, SBAS  Precise PPS generation < 15 ns  For PPS generation, only one satellite necessary  Sensitive enough for indoor operation (-161 dBm)  After startup and lock-on, separate devices generate PPS pulses with less than 100 ns difference  Deviation gets smaller over time, as clock disciplination progresses  When lock is lost, drifts with 1µs/100s (hold-over) - 7 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Result: Nicely timed measurements 22 1,00E+009 20 5,00E+008 18 Number of ticks 0,00E+000 PPS counter 16 -5,00E+008 14 12 -1,00E+009 10 -1,50E+009 8 -2,00E+009 380000 381000 382000 383000 Package number - 8 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Synchronous control system  Measurements taken at a fixed frequency  All clocks in the system need to be synchronous Synchronization method:  Measurements need to be triggered externally  Not all sensors support this  Difficult to implement and maintain  Manual calibration Change one thing Repeat previous step - 9 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Thor likes his hammer, so let’s look for an easy and “swift”, but effective solution - 10 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Methods of “Forced Synchronization” Resampling method: Just ignore the problem  Simply take the latest available value  Easiness:  Brutality:  Effectiveness: ? (Should work well, if the signal is slowly changing)  Introduces discretization noise and potential systematic errors Resampling method: Ignore the problem a little less  Same as above, but increase the sampling rate  Easiness:  Brutality:  Effectiveness: ? (Linearly reduces the time since last measurement)  That is what we do/did but lets try a more subtle approach - 11 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Methods of “Forced Synchronization” Resampling method: Linear approximation  Take the last two values and approximate the current value  Easiness:  Effectiveness: ? (Should work well, if the signal is slowly changing) - 12 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Methods of “Forced Synchronization” Resampling method: Moving average aka. FIR filter  The N last samples are averaged with a windowing function  Higher sampling rates are beneficial  Necessary for fulfilling Nyquist criterion and to prevent aliasing  Reduces Gaussian noise  Introduces delay  Discontinuity of rectangular windowing function introduces side- lobes to the signal spectrum  Rectangular window not optimal - 13 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Methods of “Forced Synchronization” Resampling method: FIR filter with Hamming window  Much better suppression of side-lobes  Higher computational cost  Possible overkill Discussion - 14 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Delay overview t SPI , t UART significant, jitter Airspeed AoA TAS through interrupts I²C SPI I²C RS485 RS485 SPI GS FC ARM IMUs IMU (Cable) (Cable) UART UART SPI IMU GPS GPS t RS485 biggest contribution t AS , t AoA , t LAS , t IMU small, constant - 15 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Sensor fusion  It is desirable to have all sensors fused at one point  Time keeping and synchronisation much easier  Easier to maintain  Fast and low jitter connection to controller  More time for NMPC calculations in each time step Past: IMU Airspeed AoA I²C SPI I²C SPI SSH Ethernet SPI GS BB FC IMU (Cable) UART UART SPI LAS GPS - 16 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger

Sensor fusion  It is desirable to have all sensors fused at one point  Time keeping and synchronisation much easier  Easier to maintain  Fast and low jitter connection to controller  More time for NMPC calculations in each time step Present: Airspeed AoA I²C I²C SSH RS485 SPI GS FC IMUs IMU (Tether) UART SPI LAS GPS - 17 - 7. September 2016 Group Retreat 2016 – Botanischer Garten – Thorbjörn Jörger