CCA Functional Structure & Architecture
Contents
1. FDK Coorporation June 2005 http www fdk com hatsnew e release050606 e htmi FIE10 Fiete R Modeling the Imaging Chain of Digital Cameras Spie Press 2010 LIO8 Li F Yu J Chai J A hybrid camera for motion deblurring and depth map super resolution IEEE Conference on Computer Vision and Pattern Recognition 2008 LUPO7 LUPA 300 CMOS Image Sensor Cypress 2007 LUPO9 LUPA 300 Demo Kit User Manual Cypress 2009 MAO6 Ma Y Soatto S Kosecka J Sastry S An Invitation to 3 D Vision Springer 2006 ISBN 10 0 387 00893 4 MARO2 Martinez K Cupitt J Sounders D and Pilly R Ten years of Art Imaging Research Proceedings of the IEEE 90 1 2002 pp 28 41 MOHO9 Mohammadi G Dufaux F Minh T H Ebrahimi T Multi view video segmentation and tracking for video surveillance SPIE Proc Mobile Multimedia Image Processing Security and Applications 2009 NGUO9 Nguyen D T Canny J More than Face to Face Empathy Effects of Video Framing CHI2009 Telepresence and Online Media Boston April 6 2009 NIDO9 PPD6 6mm Stepping Motor Nidec Copal November 2009 http www nidec copal usa com PDFs LPD6 20data pdf P109 Micropositioning Fundamentals page 128 in Micropositioning Precision Linear Rotary Multi Axis Positioning Systems PI October 2009 14 RONOYa R nningen L A The Camera Cluster Array El2. 300 data sheet for details sensor to evaluate the specification and quality LUPO9 The Printed Circuit Board PCB The PCB is a four layer surface mount board The highest frequencies are 80 MHz Some analog parts are sensitive to noise and are separated from digital signals Separate earth planes are used for the analog and digital parts To minimize interference through power supply leads all analog power supplies are implemented by separate regulators The image sensor is placed on one side of the PCB while all regulators connectors R and C are placed on the other side Two shielded FFCs connect the PCB and a trans connector on the Common Stand The power supply is 3 3 V regulated DC one lead for digital supplies and one for analog supplies An image of the circuit board is shown in Figure 3 Figure 3 The camera PCB with the image sensor mounted on bottom side Size 15 5 x 15 5 mm The Camera Cluster optics The Lens System is provided by Lensagon the model used is BSM16016S12 BSMO9 The main technical features and specifications are given in Table 2 below Figure 4a shows a perspective image of the lens system In Figure 4b the mechanical drawing with measures is shown Figure 4a Lensagon BSM16016S12 lens Table 2 Technical Specifications Format 1 2 Mount M12 0 5 Focal length 16mm Aperture 1 1 6 Focus Extent 0 2m infinite Back Focal 6 58mm Len
3. Note that CCAB 01 boards can be connected in series via a backplane switch The same can be done with ml605 boards and combinations of various boards Note Camera modules of a CC or several CCA can be synchronized This requires a common 80 MHz clock generator After power on VDDD on minimum 500 nano seconds are needed to upload LUPA 300 internal registers before the input signal RESET N determines the timing and sequence of operation Short review of multiple view object oriented camera systems In MAO06 Ma et al invite to 3D vision and cover the classical theory of two view epipolar geometry as well as extensions to rank based multiple view geometry The textbook also reviews facts from linear algebra image formation and photometrics Sonka et al present in their textbook SON10 a thorough basis for video segmentation 3D vision geometry and applications are also treated Video segmentation of scenes into background and foreground objects has been a major research area for 2 3 decades When objects are identified in the static case the next natural step is to track moving objects Some papers addressing segmentation are reviewed below G Mohommadi et al MOHO9 classifies multi camera tracking into three categories gt Geometry based gt Color based gt Hybrids Another dimension is whether the cameras are calibrated or not In MOHO9 a hybrid version is proposed using two views and th
4. parameters for non square and skewed pixels changed image origin and lens distortions Mathematically this can be described by projective geometry Given a known calibration surface chessboard in 3D space and the shot 2D image of it the calibration parameters can be calculated See MA06 SON10 for detailed mathematical descriptions and BRAO8 MAT11 for software calibration tools Camera pose The camera pose can be second version be realized using microposition techniques as described in an earlier section Shaper The Shaper in addition performs several functions as shown in Figure 11 Possible parallel and sequential processing is suggested White balance RR gt Bayer gt x Object Sub Cn Stitching detection amp object y segmentation division Depth map gt Ni Object tracking JPEG2000 encoding AppTraNet packing Figure 11 Functional blocks of the Shaper White balance The white balance can be corrected by changing the contribution of R G and B components of the image pixels by shooting a known white surface Most cameras have white balance correction CAM11 automatic or manual Bayer interpolation The Bayer interpolation is normally performed immediately after the whole image is grabbed from the sensor and stored In our case Bayer interpolation is performed just before images are shown stored for future use or used in a scene compositi
5. the screen Adjust the projector so that 5 x 5 sub pixels are shown behind each lens of the array Watch at a distance of more than tbd meters the multi 13 view effect Study the possibility of using a convex lens array on the back side of the screen in order to smooth out the discrete nature of the sub pixels References BENOO Ben Ezra M Segmentation with Invisible Keying Signal Proc IEEE Computer Vision and Pattern Recognition CVPR Hilton Head Island South Carolina 1 32 37 June 2000 BRAO8 Bradski G Kaehler A OpenCV Computer Vision with OpenCV Library O Reilly 2008 BSM09 Lens BSM16016S12 Lensagon 2011 CAM11 Cambridge in Colors A Learning Community For Photographers August 2011 http www cambridgeincolour com tu torials white balance htm CONO6 Conaire C O Connor N E Cooke E Smeaton A F Multispectral Object Segmentation and Retrieval in Surveillance Video ICIP 2006 13 International Conference on Image Processing Atlanta GA 8 11 October 2006 DAV98 Davis J Bobick A A Robust Human Silhouette Extraction Technique for Interactive Virtual Environments MIT Media Lab Springer 1998 FAN11 Yu Cheng Fan Wei Lun Chien Jan Hung Shen Depth Map Measurement and Generation for Multi view Video System Instrumentation and Measurement Technology Conference IEEE 2011 FDKO5 Ultra compact Stepper Motor SM4 3 series
6. the x and y axis are obtained A second building block with slightly different mounting plates as shown in Figure 4 in RONO7b is introduced to obtain rotation around the z axis and translation in the z direction Four lead screws take care of the z translation while the same stepper motor with gear as above can be used for the rotation When two first and one second building blocks are joined together rotation around x y and z axis are obtained by electronic control and in addition z axis translation can be done manually Mount plate Mount plate Mount olate Motor Gear box Mount plate Figure 6 The building block Table 1 SM4 3 stepper motor specification Product name Stepper Motor SM4 3 series Dimensions 4 3mm dia x 5 15mm long Shaft diameter 0 7mm Weight 0 4g No of steps step 20 1Step angle 18 angle Coil resistance 23 ohm Phase Voltage between 3V terminals Pull in torque 0 07mN m 2 Phase 500pps hour Stepper motors provide motion in discrete angular steps Positioning sensors or servo loops are not needed A position is always predictable and it takes a time proportional to the number of steps to go to the actual position The controllers control electronics needed are then quite simple Drawbacks with stepper motors compared to the alternatives are high power consumption when holding a position lower dynamic performance and vibrati
7. Technical report CCA Functional Structure amp Architecture Item NTNU Version 1 06 Project Collaboration Surfaces Author Leif Arne R nningen Date 29Aug11 File CCAFuStruct amp Arch1 The Camera Cluster Array A Camera Cluster Array consists of an array of Clusters A Cluster consists of a number of densely placed cameras A number of Camera Cluster Arrays are used in Collaboration Surfaces as described below Collaboration Surfaces DMP the Hems Lab Collaboration Surfaces are essential parts of the DMP three layer systems architecture RON11a DMP provides near natural virtual networked autostereoscopic multiview video and multi channel sound collaboration between users and users and servers Users use Collaboration Surfaces to interact with the DMP system The end to end time delay is guaranteed by allowing the quality of audio visual content and the scene composition vary with time and traffic The adaptation scheme is called Quality Shaping The scheme uses traffic classes controlled dropping of packets traffic measurements feedback control and forecasting of traffic Adaptive parameters are the end to end delay the number of 3D scene sub objects and their temporal and spatial resolution adaptive and scalable compression of sub objects and the number of spatial views The scheme also includes scene object behaviour analysis The architecture supports pertinent security and graceful degradation of qu
8. ality The Hems Lab is a downscaled DMP realization RON11a While the quality of the scenes intends to approach near natural quality in the long term the present stereoscopic implementation version 1 0 is of good quality The environment combines virtual from digital library and remote collaborators and live musical theatre drama elements and sequences including actors singers playing singing roles and scenography selected from musical theatres operas Other applications of Hems Lab are TV productions games education and virtual meetings The Camera Cluster The Camera Cluster version v1 0 consists of nine LUPA 300 Image Sensors LUP07 By means a sophisticated mechanical arrangement the Cluster enables each camera to be oriented to cover a certain part of the scene in front The cameras can synchronously shoot nine non overlapping images partly overlapping images or completely overlapping images The shots are synchronized at a maximum rate of 250 images per second A Camera Cluster provides a stitched image of maximum spatial resolution 9 x 640 x 480 pixels 2 764 800 pixels and a temporal resolution of 250 images per second In the other extreme the Cluster can shoot images of 640 x 480 pixels interlaced at a rate of 9 x 250 2250 images per second The next version of the Camera Cluster v2 0 should use image sensors with the same shutter and features as the LUPA 300 but hopefully with 150 fps temporal
9. and 2k x 1k spatial resolution The sensor should use the same package and pinning 48 pin LCC Several configurations are possible 4 sensors 4k x 2k 6 sensors 6k x 2k or 9 sensor 6k x 3k pixels at 150 fps One to three sensors should be capable of filtering out defined spectral bands from the infrared range This can be used to support video segmentation The Camera Cluster mechanics The Camera Cluster consists of up to nine Camera Modules The Camera Modules are mounted on a Common Stand as shown in Figure 1 Several Camera Clusters can be grouped to constitute a Camera Cluster Array Figure 1 Three Camera Modules placed on the Common Stand The upper part is the Lens System detailed in the next section The lens system is screwed into the Sensor House from top to adjust focus The Sensor House is fastened to a Pose Mechanism The Sensor is soldered to the Sensor PCB which has 48 leads through vias to the other component side of the FCB Flexible Flat Cables interconnects the PCB to an FPGA card see other sections for more details The Pose Mechanism can pose the Camera Module to point in any direction into the scene in front within limits 3 axis adjustment is needed and can be accomplished by a manual ball joint arrangement a micro robot or other In documents RON09a and RONO9b the overall mechanical construction and the micropositioning system are shortly described The LUPA 300 CMOS Image Sensor Th
10. e camera is started by a signal from the Host Computer operator with a reset of the Imager When the camera generates images it sends 10 bits in parallel to the Calibration and Shaper blocks synchronized by frame valid line valid and clock signals The Calibration block starts the FPN and PRNU imager correction procedures see below sends the correction and configuration data to the Configure block which in turn configures the Imager Then the Calibration block runs the camera Calibration procedures estimating five distortion parameters and four intrinsic parameters using chess board images in known pose in front of the camera The calibration data is used by the Shaper block when real time generation of images takes place FPN and PRNU FPN noise stems from individual variations of the transistors from pixel to pixel and PRNU noise stems from variations of column amplifier of the image chip Corrections can be performed by using known optical inputs inspecting the resulting image and writing correction data into chip registers as described in LUPO9 Software for this correction is available A linear regression model built from measured intensities for given calibration scenes can also be used see FIE10 Camera calibration The camera calibration is normally based ona frontal pinhole camera model which describes idealized projection of points in 3D space onto an image surface and is extended with corrections
11. e following steps gt Recover the homography relation between camera views gt Find corresponding regions of objects between different views by use of regional descriptors of spatial and temporal features gt Track objects in space simultaneously across multiple viewpoints Video segmentation with invisible illumination can be regarded as a hybrid category Ifrared IR light is then added to the RGB light in the scene to support the segmentation This can also be regarded as an extension of chroma keying known from movie production for decades Support of IR light has been tested by several researchers e g DAV98 WU08 CONO6 In WU08 Wu et al perform bilayer video segmentation by combining IR video with color video Their contrast preserving relaxation labeling algorithm is shown to perform better than graph cut algorithms A problem with IR supported segmentation arises when scene objects are made of glass To avoid this Ben Ezra has in a paper BENO2 proposed the Catadiopric camera image sensor design using beamsplitters and prisms to polarize light on backgrounds while the foreground objects are illuminated by non polarized light There are today several cameras in the market but expensive that perform depth 12 measurement and depth map generation in addition to normal RGB image shooting LIO8 Advanced Imaging As pointed out in the introductory section the camera module can be built us
12. e main features of the LUPA 300 are shown in Table 1 Figure 2 shows the imaging surface the bonding the pin layout and the size of the LUPA 300 integrated circuit LUPO7 Table 1 Main parameters of LUPA 300 Image Sensor source LUPA 300 data sheet Parameter Typical View Optical Format inch Active Pixels 640 H x 480 V Pixel Size 9 9 um x 9 9 um Shutter Type Electronic Snapshot Shutter Maximum Data Rate Master Clock 80 MPS 80 MHz Frame Rate 250 fps 640 x 480 ADC Resolution 10 bit on chip Responsivity 3200 V m2 W s 17 V lux s Dynamic Range 61 dB Supply Voltage Analog 2 5V 3 3V Digital 2 5V I O 2 5V Power Consumption 190 mWatt Operating Temperature 40C to 70C Color Filter Array RGB Bayer Pattern Packaging 48 pins LCC The snapshot shutter is a must to avoid distortion of fast moving objects of a scene Rolling shutters do not meet our requirements Adapted from the LUPA 300 data sheet The VGA resolution CMOS active pixel sensor features synchronous shutter and a maximal frame rate of 250 fps in full resolution The readout speed can be boosted by means of sub sampling and windowed Region of Interest ROI readout High dynamic range scenes can be captured using the double and multiple slope functionality User programmable row and column start stop positions allow windowing and sub sampling reduces resolution while
13. ectromechanical Design Technical Report Item NTNU October 2009 RONO9b R nningen L A Micropositioning System for the Camera Cluster Array Electromechanical Design Technical Report Item NTNU October 2009 RON11a R nningen L A The Distributed Multimedia Plays Architecture Technical Report Item 2011 http www item ntnu no people perso nalpages fac leifarne the dmp archite cture RON11b Ronningen L A DMP Processing Architectures based on FPGA and PCIe NTNU August 2011 http www item ntnu no media peo le personalpages fac leifarne pciearc h2 docx SONO8 Sonka M Hlavac V Boyle R Image Processing Analyses and Machine Vision Thomson 2008 ISBN 10 0 495 24438 4 WIK09 Wikipedia Stepper Motor November 2009 http www wikipedia http en wikipedia org wiki Stepper motor WIKO9 Wikipedia Epicyclic gearing October 2009 http en wikipedia org wiki Epicyclic earing WIK11 Image stitching 2011 http en wikipedia org wiki lmage stit ching W0O08 Wu Q Boulanger P Bischof W F Automatic Bi Layer Video Segmentation Based on Sensor Fusion Proceeding of the 16th ACM international conference on Multimedia Vancouver British Columbia Canada 2008 15
14. es with known reference points and images This can first of all be used to calibrate each camera the Camera Clusters CC and CCA The reference information can also be used to avoid too large position deflections of the moving parts of the positioning system Furthermore the sine cosine microstepping can be made adaptive If wanted the control loop can be used for moving the cameras when tracking objects This also gives the possibility to increase temporarily the resolution in time and or space for important objects which are tracked Figure 8 The Camera Module assembled using a mechanical positioning system The size of house with lens is approx 17 x 17 x 38 mm The Camera Cluster Array Functional Structure and Behavior The CCA Functional Structure is shown in Figure 9 The Camera Module CM Processing 10G PCle x4 Ethernet Switch Other fiber TE Collaboration CM CCA Space gt Processing Processing j Processing CM Camera Module CCA Camera Cluster Array Figure 9 The Camera Cluster Array Functional Structure Figure 10 shows the functional structure of the CM processing block Calibration lt Camera Shaper Optics e PE imager lt Camera Configure Pose 8 Camera Pose Ctrl Figure 10 Camera Module Processing In the first version the Camera Pose sub system will be static and mechanical and manually adjusted Initially calibration of th
15. fect colors MAR02 In case B the infrared bands support video segmentation in 3D space When three infrared sub bands are available the method described in the VASARI project MARO2 can be extended to help finding the contour of objects seen from three different directions Case C is an extension of the VASARI project It is of course possible to obtain higher spatial resolutions e g 4k x 2k pixels as in Case D Three spectral sub bands RGB are realistic One camera could be used for IR Other interesting problems we intend to study are related to integrating a camera array witha display to avoid the Mona Lisa effect NGUO09 Small lenses of 1 3 mm diameter can be applied and used in a full duplex two way manner shooting scene images inwards and showing remote scene image pixels outwards In this way the cameras will be nearly invisible but will be placed close to the areas of the display that the viewer focuses on Due to diffraction effects small lenses with a diameter of less than 1 mm limit the spatial resolution Using a number of such lenses to shoot one object can improve the spatial resolution To start with an up scaled test bed can be built for experimentation Install a projector and a back projection screen Use the pixels of the projector as sub pixels for a multiple view five or more experiment Place a 3 x 3 lens array with lenses of diameter 92 mm cut down to 65 mm squares in front of
16. gth F O V DxHXV 24 x 19 7 x 16 2 Lens 4 Components 4 Group Construction Weight 6g Note Megapixel IR corrected A V 7 4 al farter N S I Figure 4b BSM16016S12 lens mechanical drawing source Lensagon Balancing optics and sensor resolution As described by Fiete in FIE10 the resolution of the optics and the sensor should be balanced In theory the ratio of the detector sampling frequency to the optics diffraction cutoff should be one giving the so called Q value equal to two In practice 0 5 lt Q lt 1 5 is often chosen to obtain a sharper and brighter image of large scenes In this design the imaging chip has been chosen to meet the frame rate of 250 Hz The reason is that the perceived quality of high frame rates 50 250 Hz is to be tested The lens is then selected to match the imaging chip In the final design it is desirable to have a lens diameter of 1 mm and then the optical system will be diffraction limited When and if imaging chips with say 150 Hz frame rate and 2k x 1 k resolution will be available is uncertain The Camera Cluster The Common Stand supports normally nine Camera Modules per Cluster The Camera Modules are placed as dense as practical for adjustments see Figure 5 The distance between the centers of Clusters are normally 6 5 cm the normal distance between human eyes The Common Stand can be mounted on a standard camcorder tripod two
17. ing various imaging sensors if the 48 pin LCC package is used If the pinning is different from LUPA PCBs and the control HW SW also must be redesigned Great possibilities open up if the imaging sensors say 150Hz 2k x 1k resolution basically are monochrome but have different light filters in front In addition to using IR for video segmentation we here propose to use UV light as well and maybe combine IR UV and polarization If a certain amount of cerium is added to phosphate glass efficient blocking of UV light is obtained This is certainly for the future but also silicate glass normal glass today with cerium shows improved blocking effect of UV light How various transparent plastic materials behave have to be studied Each of the cameras in a CCA 36 cameras can in principle apply different optical filters If we to start with limit the different filter to within a cluster we can have A nine spectral sub bands in the visible frequency band 150Hz 2k x 1k or B six spectral sub bands in the visible band and three infrared or and ultraviolet sub bands 150Hz 2k x 1k or C seven spectral sub bands i e seven Gaussian filters with 50 nm interval between 400 and 700 and a bandwidth of 70 nm 6 plus two infrared or and ultraviolet sub bands 150Hz 2k x 1k or D several 150Hz 4k x 2k configurations or E other In case A we can with resolution of 12 16 bits per sub band pixel shoot images with nearly per
18. maintaining the constant field of view and an increased frame rate The programmable gain and offset amplifier maps the signal swing to the ADC input range A 10 bit ADC converts the analog data to a 10 bit digital word stream The sensor uses a 3 wire Serial Parallel SPI interface It operates with a 3 3V and 2 5V power supply and requires only one master clock for operation up to 80 MHz pixel rate It is housed in a 48 pin ceramic LCC package 11 17620 13 P 1 016X11 gt 1 016 0 05 1 5520 15 20 076 2 5620 076 i pr YA r pe trorerorcd ro toy EA S mm j I a Gd s m a a a amp R N Figure 2 The LUPA 300 CMOS Image Sensor source LUPA 300 data sheet The sensor used here is the Bayer RGB patterned color filter array version Note that the image rate for a single sensor can be increased to 1076 images per second but with 256 x 256 pixels per image The LUPA evaluation kit The LUPA Demo Kit consists of two main parts a box with an image sensor lens and hardware and image software running on a PC under Windows XP The box contains a RAM buffer which can receive and store short videos at maximum shooting rate When the buffer is full the video can be transferred to the PC at a much lower data rate for further processing and inspection The video can be stored in different file formats The PC can configure the sensor on the fly see LUPA
19. on That means that all interpolation is done at the receiver From the camera the R G and B components are low pass filtered From the original G component edges transitions features and objects are segmented At the receiver the Bayer 10 interpolation is performed and un sharp edges in the objects are sharped using the edges from the original G component FIE10 Stitching Stitching is to align slightly overlapping images together covering a scene shot from various positions and viewing angles Typically images from one cluster will be stitched together See WIK11 Object detection and segmentation A major feature of DMP is to detect and recognize objects as important and then segment them from other objects and the background The start of the process is to find features like corners and edges that can be boundaries of the objects The theory behind can be found in SON10 MA06 and software tools in MAT11 BRAO8 Object tracking When an object has been segmented it can be of interest to track it for some time over the visual field to let cameras focus and follow the object The intention can be to increase the spatial resolution of the images of the object or send data of the object only leaving the background out Theory techniques and tools can be found in MAT11 BRAO8 MAO6 SON10 Depth map generation It is intractable to shoot and send all possible views of a scene and a good compromi
20. on if not handled adequately Microstepping for example sine cosine microstepping can be used to obtain higher position resolution WIK09a Assume that the step of a motor can be reduced from 18 to 18 256 0 070 If a gear of rate 10 is added the resolution could theoretically be 0 007 Another advantage with microstepping is that smoother motion is obtained In Figure 7 the principle behind a planetary or epicyclic gear is illustrated WIKO9a Figure 7 Planetary gear principle It is used here to decrease output speed The planet gear carrier green is fixed to the motor house The sun gear yellow is the motor shaft while the ring gear red is rotating output adapted from WIK09a Pose Control System The pose of the Camera Module attached to the positioning combination as explained above can be controlled by four screws manually for the z translation and the z y and z rotation by three parallel stepper motor controllers It is desirable to use the pose control system to calibrate the cameras and maybe to track and follow moving objects in space and the controllers should therefore be integrated with the image processing system FPGA implementation As pointed out the controllers should be of the sine cosine microstepping type In our application the micropositioning systems will be part of the closed control loop of the Camera Cluster Array CCA The cameras will shoot images of the scen
21. se is to use depth maps for reconstruction of lacking views Depth map generation are described in FAN11 MAT11 BRAO8 MA06 SON10 L108 Sub object division The DMP sub objects are described in RON11a JPEG2000 encoding See RON11a AppTraNet packing See RON11a The Camera Cluster Array Processing Architecture In Figure 12 a possible CCA processing architecture is shown As described earlier it consists of Camera Modules processing boards CCAB 01 ML605 boards and PCs When the Pose System is mechanical manual and static the processing requirements to a number of functions are small All imager and camera calibration procedures can then be run in software on the Host Computer LUP10 Although the functions in Figure 11 should be performed by one ml605 on a complete stitched image from a cluster of cameras it should be investigated if some functionality could be shared by CCAB 01s and ml605 In the following it is assumed that Bayer interpolation is not performed by the sender Cluster of nine cameras r CCABO1 Virtex 6 FPGA Boards To host PC PCle Root Complex One Stop Systems MaxExpress PCle Backplane PClex4 1 0 Fiber SM 1550nm 40km Plugged into Backplane PClex8 1 0 lobal Hitech 10 3 Gbit s SFP Tansceiver Xilinx MI605 Figure 12 CCA Processing Architecture 11 See RON11b for description of hardware architecture for CCA processing
22. standard holes or other One Cluster includes three CCAB 01 boards with a Virtex 6 FPGA onboard That is one CCAB 01 can serve three Camera Modules Flexible flat cables each 22 lead connect the Camera PCBs with the CCAB 01 The CCAB 01 provides power 3 3 V DC for the Camera Modules The PCle standard is used for external communication One Xilinx ML605 board with a three input PCle Mezzanine Card can serve several Camera Clusters A high end PC can include three ML605 boards Backplanes with PCle switches can take 10 20 PCle x4 and some x8 boards Figure 5 A Cluster on the Common Stand Micropositioning Electro mechanics This section describes a micropositioning stage with rotation motion around the x y and z axis and translation in the z direction It is based on the principle of hinges as shown in Figure 5 in RONO7a but has got ultra compact stepper motors with gears to provide accurately controlled motion The basic building block is the combination of a small cylindrical stepper motor house with a mounting plate rigidly attached and a cylindrical gear house with another mounting plate rigidly attached The two cylinders are mounted end to end and rotate relative to each other by means of the stepper motor and the gear and with a roller bearing in between This is shown in Figure 6 When two building blocks are joined together by fastening the mounting plates 90 to each other rotation around
Download Pdf Manuals
Related Search