PCR A PC based wave front reconstructor for MMT-AO

Transcription

1 PCR A PC based wave front reconstructor for MMT-AO Vidhya Vaitheeswaran 1, Phil Hinz 1, Guido Brusa 2, Doug Miller 2, Thomas Stalcup 3 1. Steward Observatory/Univ. of Arizona, 933 N Cherry Ave., Tucson AZ, 85721, USA 2. Large Binocular Telescope, 933 N Cherry Ave, Tucson, AZ 85721, USA 3. Multiple Mirror Telescope, 933 N cherry Ave, Tucson, AZ 85721, USA ABSTRACT PC Reconstructor (PCR) is the control software for the natural guide star and the laser guide star systems at the 6.5m Multiple Mirror Telescope (MMT) operating with the adaptive secondary mirror on Mt. Hopkins south of Tucson, AZ. The PCR computes and corrects atmospheric turbulence featuring a common interface between the wave front sensor camera control link and the deformable mirror, diagnostic data management, vibration control, closed-loop data distribution & saving routines, and housekeeping modules. We report here on the development, use and the on-sky performance of the PCR. Keywords: Adaptive Optics, Deformable Mirror, and Software 1. SOFTWARE GOALS PC Reconstructor (PCR) was developed as the new PC based software control for the adaptive options program at MMT, which was operating on a VME architecture that required more flexibility in supporting stable science observation, and to enable fast prototyping of new features for on sky experiments. Observing with adaptive optics involves a bright source (natural or artificial guide star), a fast, low-noise wave front sensor to sense atmospheric turbulence, a deformable surface to adapt the wave front to correct the error arising from the turbulence, and a low latency control system to compute the wave front shape to be applied. We describe PCR software as an operational bridge in the adaptive optics based observation at the MMT. 2. GUDE STAR MODES MMT adaptive optics system supports both the Natural Guide Star (NGS) observing mode and the Laser Guide Star (LGS) observing mode. The NGS mode uses the NGS top box mounted at the Cassegrain focus. A dichroic directs the visible light to the NGS wave front sensor and the channels the infrared light directly to the science camera. The LGS AO system consists of a beam projector, a laser wave front sensor, a tip tilt sensor and a host of optics at the Cassegrain package. The beam projector is mounted behind the adaptive secondary and it projects five Rayleigh LGSs as a pentagon of beacons on the sky. The Laser wave front sensor camera is similar to the NGS wave front sensor camera in hardware interface and device driver modules. All the software components that are common to both the observing modes are grouped together to form the PCR software. (a) Figure 1. The Laser guide star pentagon (a) and the natural guide star image (b) as seen by AO real time displays. (b) Advanced Software and Control for Astronomy, edited by Alan Bridger, Nicole M. Radziwill Proc. of SPE Vol. 7019, 70190E, (2008) X/08/$18 doi: / SPE Digital Library -- Subscriber Archive Copy Proc. of SPE Vol E-1

2 3. WAVE FRONT SENSOR Atmospheric turbulence for the adaptive corrections in the NGS mode is measured using a Shack Hartmann based CCD sensor with 4-channel output with 80x80 pixel dimensions. The camera is configured to perform on-chip 3x3 binning to increase the CCD readout rate, and a PC interface card is used in the control computer to read frames at 527Hz. The PC driver employs multiple buffering schemes to increase the speed of the array read out. The binned image forms 2x2 pixel quad cell sub apertures, and the slope error signal is computed by the summing and differencing light in each quad cell. This signal is then linear-ised using a slope look up table resulting in144 pairs of slopes from the 12x12 sub apertures. The LGS mode utilizes a 128x128 pixel camera with 16-channel output and is read at 400Hz 460Hz. A prism array splits pupil into 60 sub apertures. n the ground layer adaptive optics (GLAO) mode, the spots are averaged from the five beacons to correct just the ground-layer turbulence. n the Laser tomography mode (LTAO) all spots from each beacon are used to cancel the turbulence. The interface card and the driver modules are common with the NGS setup. 4. MMT ADAPTVE SECONDARY MRROR MMT adaptive secondary mirror, commissioned in 2003 is a 2-mm thin aspheric shell mounted on a 40mm thick reference plate. There are 336 electromagnetic voice coil actuators controlled by 168 DSP boards housed in 3 separate crates. The crates are placed just above the reference plate and the communications to these crates are routed through a pair of fibers. The position control loop runs at 40kHz and the outer optical loop runs at 527Hz. We have successfully completed 27 observing runs with the MMT adaptive optics system supporting science observations from 4 nfrared cameras. A high-speed communication with the adaptive secondary is crucial to reduce phase lag between the time the error is sensed, and the time the error is corrected. Since any new technology implementation would require an expensive and time consuming upgrade to the mirror controller DSP boards with new commands, we adopted the Large Binocular Telescope (LBT) communication equipment, the Basic Communication Unit (BCU), and developed system libraries to enable gigabit Ethernet network to the BCU. The BCU acts as bridge between the control computer and the adaptive secondary mirror, and converts Ethernet packets from the control computer to mirror commands that were already being supported under the VME architecture. Figure 2. MMT adaptive secondary mirror is shown. The spider arms seen here is part of the support structure. Proc. of SPE Vol E-2

3 5. DGTAL CONTROL A block diagram of the AO system s digital control is presented in Figure 3. The WFS module reads the camera frame, the SSC module computes the slopes, and the RTR module is the real time reconstructor that computes the delta command while the DM is the communication interface module that talks to the adaptive secondary mirror. Each of the modules is connected to the data archiving procedure over UDP network. The MMT adaptive optics system provides extensive continuous data saving at full frame rates for the both NGS and LGS observing modes. spots slopes delta cmd composcur Telemetry Figure 3. Block diagram of the real time control flow is shown. 6. ADAPTVE OPTCS OPERATON AT THE TELESCOPE UML MODEL UML modeling was employed to understand the roles of the astronomer and the AO operators. Use case models and the state transition model were developed using Enterprise Architect, a commercially available package for UML modeling. The division of labor led to the grouping of modules and module properties, which led to the thread functions. All the state variables and the state context blocks were transitioned to the thread context blocks. The header files that are included in our C programs are derivatives of this state context blocks. The development specification document of the PCR details each and every use case in detail including the pre and post condition, steps for each use case, success condition, and exception handling routines. A use case model depicting the observing sequence using adaptive optics is presented in Figure 4. The first step in AO observation is to create an optically flat wave front by setting the adaptive secondary mirror shell at 40-micron gap from the reference plate. This step involves configuration of the DSP boards. The astronomer selects the science target and the guide star, the telescope operator points the telescope, and the adaptive optics operator aligns the f/15 optics to image the wave front sensor camera to the guide star. Once the star is aligned, the optical loop is closed. Several AO related parameters such as the loop gain are tuned to get diffraction-limited images once the loop is closed and a stable AO loop is achieved. The astronomer then observes the science target with the science instrument. Low order modal errors such as tip, tilt and focus are offloaded to the telescope mount and the hexapod in close loop operation. Figure 5 shows the state transition graph for the AO operation on sky. Proc. of SPE Vol E-3

4 Operator Astronomer Figure 4. Use case diagram of AO observing procedure is shown UML model done in Enterprise Architect software. 7. SOFTWARE ARCHTECTURE 7.1 The architecture model shown in Figure 6 depicts the order of computation and the operation sequence for AO control. The sequence begins with reading the camera frames using the EDT PC interface controller card connected to the CCD camera. The EDT driver facilitates the digital readout from the camera. Then the slopes are computed and static offsets are applied. The static offsets can be of the form: Static offsets: to dial in a static shape e.g. astigmatism on the deformable mirror Modulation offsets: to put in a modulation signal e.g. to compute reconstructor on sky Science camera offsets: to provide tip tilt corrections on faint guide star that wave front sensor is not able to sense. NGS tip/tilt offsets: To include tip-tilt signal from NGS while observing in LGS mode. Vibration offsets: To compensate for structural vibrations of the telescope. Once the static offsets are applied, the actual reconstruction of the slopes is calculated using a reconstructor matrix. Matrix multiplication is performed using CBLAS routines to maximize the performance and increase the speed of the computation. The output of the matrix multiplication is a 336-element command vector that is ready to be applied to the adaptive secondary mirror through the DSP boards. Any command offsets are also applied at this time. The DSP boards that control the adaptive secondary mirror operate in fixed-point arithmetic while the control computer operates in floating point arithmetic to preserve accuracy. Any high frequency errors arising from floating point truncation errors are filtered using a modal matrix. The output of the modal matrix multiplication is then subtracted from the command vector. Proc. of SPE Vol E-4

5 p. p - - fl[d d_ ttig - r& r & ig. - it : - rb&l d :&o i r ft _ft! + fr >1 + b kg dk: i C i d pj! E - + t,y d 4 (r -p_: r&.bp. thb: + d + d ppy pfft ) tt + ffly + d ::=z=t:t: ar. 'U.' y_d Adt OJl.Sp!M. >f 4 dft h t - - thoo,_flo: + Y + LtUL_fo!ffl ft t!d tht! : php,: i - pig. lt.. i*: - p 'b: m ppy p, h pffi dpy DSF. - f _ppid: i -,jpjs: i THCt h dfft itthd i thgt c2;), 2 2 Figure 5. Stare transition diagram of open loop and close loop sequence is shown. The command vector is then formatted to form UDP packets that will be sent through the high-speed peer-peer communication channel to the mirror. The internal DSP control, programmed in the mirror electronics computes and applies the feed forward forces needed to travel to the new desired position specified in the command vector. Once the shell settles down, the position of the 336 actuators are sampled using the capacitive sensors. Boundary conditions for the position range and forces range check are performed on this data to ensure safety limits and ensure ranges are not exceeded. The data memory of the mirror DSP is then populated with the telemetry consisting of the new position, current and the absolute command vectors. The downlink of the high-speed communication is used to read the telemetry at full frame rate back to the control computer. This data is shared with the AO environment and other users through a distributed DDS computer. 7.2 The interface to the external modules from the control computer is through the socket service, and the closed loop error messages are sent to the subsystem through an interrupt/exception port. The camera frames are sent to a real time display both in open and close loop mode through a stream port. The slope offsets mentioned in section 7.1 are received through the service ports. A dedicated command port handles the entire operational request from the AO operator s interface as all commands and events are logged. Proc. of SPE Vol E-5

6 Matrix Multiplication Camera Driver Modal Fiters WFS camera readout high frequency modes limit checks ii safety checks Date archives & event logs jjlimit checks Apply correction Safety checks to avoid bending the mirror capacitive sensor & feed forward force command DM to move to a new position Teemetry Figure 6. PCR software architecture diagram is shown 7.3 There are ten threads that are spawned to support multi-threading multi tasking of the AO close loop operations. The thread switching is done based on the conditional variables. The threads are dedicated to the following tasks: 1. Camera thread: To continuously read frames from the CCD camera 2. WFSC thread: To stream wave front sensor data to the AO operator s real time display 3. DM read-thread: To read telemetry data from adaptive secondary DSP memory locations 4. DM write-thread: To write the command vector to the adaptive secondary mirror 5. Command Thread: To listen for AO commands from the AO operator s interface 6. Error Thread: To broadcast error messages 7. Tip-Tilt Thread: To listen to tip-tilt port 8. Science camera thread: To listen to tip-tilt information from the science camera 9. Vibration thread: To receive vibration tip-tilt signals from the accelerometer sub system 10. Data thread: To forward close loop telemetry to the DDS computer The DM read-thread and DM write-thread access the same high-speed link to the adaptive secondary mirror and therefore are synchronized and protected using pthread mutex variables. The DDS spawns three threads to support data distribution to the data clients, to save continuous data and to process requests from data clients. The spots, slopes, delta commands, mirror positions, mirror forces and the mirror absolute commands are stored every 20 seconds in close loop mode. The files are stored in FTS format and are available to the observers through a dedicated file system The development environment for the PCR development is as follows: The hardware architecture to support compute-intensive operations is based on dual quad core Xeon X5355 with 2GB memory. The matrix multiplication is performed using CBLAS, a basic linear algebra package for C. All the vector operations for the DM communication are performed using GSL (Gnu Scientific Library) routines. The PC driver for EDT is referenced as a static library from the camera acquisition thread. The design for the DM communication device was adopted from the Large Binocular Telescope architecture. A hardware unit called BCU (Basic Communication Unit) was retrofitted to interface with the MMT adaptive optics control electronics. Microgate, taly developed the firmware and software used in this box. The communication to the DM is carried over UDP stack. The top-level application performs packet acknowledgment operations, as UDP is a connection-less protocol. Proc. of SPE Vol E-6

7 8. LATENCY The latency of the PCR systems plays a critical role in the atmospheric turbulence corrections of the AO system and the quality of science images. The time lag between the last pixel of the frame arrival and the last command sent to the DM was measured using logic analyzer and the JTAG debug interface of the communication hardware interface. The time lag for each module is shown in Figure 7. The camera integrates on the star for a period of 1800 microseconds. The slopes calculations take 60 microseconds. The matrix-vector multiplication takes 165 microseconds. The UDP uplink to command the DM to travel to a new position takes 100 microseconds. The DM write thread hands over the control to the DM read thread. While the DM thread waits to read back the telemetry, the camera thread continues with the next frame to minimize the time lag. The DM read and the write thread are protected by mutex and are governed by the safety check flag set by the DM read thread. f the read thread finds the DM to be close to or at the safety boundary conditions, the safety flag is latched on and the loop is opened. The DM write thread is locked from commanding the DM. n>l camera integration & readout 1.8 ms reconstructor Matrix C Slope Calculations Multiplication Write DM commands 6Ops l6sps loops Read DM telemetry (UDP jumbo frame traffic ( uplink for data request + downlink for data) C 100 ps (uplink request) ps Figure 7. NGS computation latency in closed loop state is shown. 9. ON SKY PERFORMANCE The first closed loop tests on sky with PCR was conducted in March of The VME system was completely decommissioned in April of There have been total of 12 observing runs scheduled with PCR to support NGS and LGS observations between April of 07 and June of 08. The PCR system has performed well in all of those observing runs. The adaptive optics system at MMT supports three science cameras in NGS mode and one science camera in LGS mode. They are listed below: ARES: 1-2.5µm imager and spectrograph BLNC/MRAC: 10 µm imager with 2µm nuller CLO: 3-5µm imager PSECS: 2-5µm wide field imager Proc. of SPE Vol E-7

8 The structural vibrations on the MMT telescope are limiting the performance of the AO system drastically. The tip tilt residuals shown in Figure 8 demonstrate the 20Hz telescope vibrations. The vibrations are outside the close loop bandwidth of 20Hz and as a result not corrected by the AO loop. The average amplitude of this vibration is found to be around 20 mille arcs seconds depending on the wind direction and the elevation of the telescope. This contributes the reduction in the strehl ratio significantly. Even with this vibration PCR is shown to achieve up to 40% strehl in K band and 25% strehl in H band. To mitigate the issue of vibration, four accelerometers are mounted on the back of the cold plate of the DM Work is in progress to integrate the vibration feed forward signal into the close loop. The will allow for dynamic correction of structural vibration of any frequency in the future. Op.n Loop Tp flit Dath May27200&05101L Co..d Loop np flit Data May27200&051019_ tip - tip tilt 0.5 0_DO l l 0 2 (a) (b) (c) Figure 8. Tip-Tilt corrections for 20 s time scales in open loop (a) and close loop (b) are shown. The 2 Hz vibrations seen in open loop is removed in close loop mode, but the 20 Hz vibrations can still be seen. (c). Close loop image of FK taken with CLO during the same time is shown. 10. FUTURE WORK The PCR system is fully operational supporting both the NGS and the LGS mode of AO. Observing with the adaptive secondary mirror using the PCR has become a standard mode of operation. There are few developments planned for the future to make the system more robust and efficient. Close loop data analysis has shown oscillations in the loop during certain user gain. This oscillation is undetectable due to lack of tools to analyze the close loop data. A new PD control Proc. of SPE Vol E-8

9 algorithm is being currently being developed to include the proportional and derivative path to the existing integral control. This new control should remove the overshoots and oscillations in the AO system. ncluding the accelerometer feed forward is also under development. Work is also underway to support Laser Tomography Adaptive Optics mode. We have demonstrated that control of adaptive optics can be done with simple hardware and software setup backed by well-designed and well-organized software programming techniques. 11. ACKNOWLEDGEMENTS Observations reported here were made at MMT, a joint facility of the University of Arizona and the Smithsonian nstitution. We thank Manny Montoya, Matthew Kenworthy, Richard Sosa and the MMT operators for their valuable contributions. 12. REFERENCES 1. Guido Brusa, Armando Riccardi, Francois P.Wildi, Michael Lloyd-Hart, Hubert M. Martin, Richard Allen, Donald L. Fisher, Douglas L. Miller, Roberto Biasi, Daniele Galliei, Fabio Zocchi, MMT adaptive secondary: first AO closed-loop results, Proc. SPE 5169, pp.26, Christoph Baranac, Michael Llyod-Hart, N. Mark Milton, Thomas Stalcup, Miguel Snyder, Vidhya Vaitheeswaran, Don McCarthy, and Roger Anger, Astronomical imaging using ground-layer adaptive optics, Proc.SPE Proc. of SPE Vol E-9