PERFORMANCE TESTING FOR ON BOARD PROCESSOR

Transcription

1 PERFORMANCE TESTING FOR ON BOARD PROCESSOR Arvind Vishnubhatla* *Gokaraju Rangaraju institute of engineering and technology,hyderabad Abstract We use Terahertz imaging arrays to get synthetic aperture imaging data. As the data rates are extremely high we intend to make a custom processor for an unmanned vehicle taking due care of weight and DC power restrictions. Figure 1. Basic Block Diagram 1. Introduction We use the time-of-flight measurement technique in which we first beam a laser pulse and calculate range as a function of the propagation delay of a laser pulse. The measured round trip time and the corresponding signal processing helps evaluate a targets s terrain characteristics. To adhere to the weight restriction constraint we use a low power eventhough this calls for a bigger receiver aperture to get minimum S/N after the detector. The receiver collects the reflected signal and focuses it on a photodetector. Our Custom processor is in fact a Central Processing Unit which is suitably designed with parts like memory caches for instructions and data, instruction decoders, and arithmetic logic units. To get enhanced performance we use a multicore CPU where the workload is suitably decomposed into task graphs which can work in parallel, an architecture which has been proved to enhance system performance. 532

2 However, when a large portion of the cores are idle Figure 2. the system performance falls. Imaging In this paper a method of testing the performance of the processor in the laboratory is illustrated. The real time tasks are run on Texas instruments C6678 board whose architecture is as shown below. Figure 4 C6678 Evaluation Module An EM406 at 4800 baud rate and NMEA protocol is used for communication with the GPS. Three LISY300AL gyros provide the roll,yaw and pitch signals. Figure 3. C6678 Architecture Figure 5. A board showing gyro signals 533

3 Initially we theoretically assign values to the four tasks 1.Beamforming(bf) 2.Flight Control(fc) 3.Image Analysis(ia) 4.Communications(cm) And play with the task times and periodicities and study schedulability on one processor.then we get into the physics behind the implementation. We observe that the burst time of the beamforming task depends on the the number of adaptive array elements,the number of nulling weights, nulling bandwidth, adaptive antenna diameter and the number of interfering sources and jammers. Likewise we observe that burst time of the flight control task depends on the strictnesss of the flight guidance and control, mission control strategies and stochastic processing of the aerodynamics and the complexity of the probabilistic models. We also observe that the burst time of the image analysis task depends on the number of point clouds in the 3D point cloud map, density of the map,the feature geometries to be classified,the fineness of the topographic details, resolution of the earth surface and the amount of decision made on board the aircraft vis a vis the ground station. We see a set of execution graphs with task 1 beamforming(blue) task 2Communication (Green) task 3 flight control (Red) and image analysis(yellow). As this is only a scaled representative model it will be very difficult to fix bounds on the execution times of each of these tasks. It is very clear that if the beamforming gets to be tight and the there is lot of interference and jamming,the other tasks will starve. Like wise if a lot of image analysis is to be performed on the processor,then only image analysis will run. Like wise if the surveillance activity requires lots of tactical maneuvering under tough weather conditions then the flight control task will overshadow the other tasks. To get a better performance we use a three processor model with Beamforming on one processor, communication and flight control on the other, and image analysis on the third. Figure 6.7. The load testing has been planned on a scaled model and the performance results reiterate the claims that have been made in the design. We run the tasks on one processor and get the execution graphs. Figures 6.3,6.4,6.5 and

4 Figure 6.1 Scheduling periodic tasks 1. bf (20,50) fc(10,50) ia(15,50)cm(5,50) 2 bf(2550) fc(10,50) ia(15,50) cm(5,50) 3 bf (20,50) fc(10,45)ia(15,50) cm(5,50) 4 bf (20,40) fc(10,45) ia(15,50) cm(5,50) 5 bf (20,30) fc(10,45) ia(15,50) cm(5,50) 6 bf (20,20) fc(10,45) ia(15,50) cm(5,50) 535

5 Figure 6.2 Scheduling Periodic tasks 7. bf (20,10) fc(10,45) ia(15,50) cm(5,50) 8 bf(20,50) fc(10,35) ia(15,50) cm(5,50) 9. bf (20,50) fc(10,25)ia(15,50) cm(5,50) 10. bf (20,50) fc(10,15) ia(15,50) cm(5,50) 11. bf (20,30) fc(10,45) ia(15,50) cm(5,50) 12. bf (20,50) fc(10,45) ia(15,40) cm(5,50) 536

6 Figure 6.3 Execution Graphs on One Processor Figure 6.4 Execution Beamforming task 537

7 Figure 6.5 Execution Communications task Figure 6.6 Execution flight control task 538

8 Figure 6.7 Execution Graph Beamforming(Green),Flight Control (Magenta) Image analysis (Blue) 539

9 References [1] [Ahuja93] Network Flows, Ravindra K. Ahuja, Thomas L. Magnanti, and James B. Orlin, Prentice Hall, February 18, 1993, 864, X. [Bemporad04] Mixed Integer [2] Quadratic Program (MIQP) solver for Matlab, Alberto Bemporad and Domenico Mignone, Automatic Control Laboratory, ETH Zentrum, Zurich, Switzerland,2004. About the Author Arvind Vishnubhatla is currently a professor in Gokaraju Rangaraju Institute of Engineering and Technology. He has M.S in electrical engineering from Northern Illinois University. He is pursuing his Ph.d with JNTU (Hyderabad). [3] [Berkelaar05] lp solve (Open source (Mixed- Integer) Linear Programming system) [4] Michel Berkelaar, Kjell Eikland, and Peter Notebaert, [5][Bru76] Sequencing unit-time jobs with treelike precedence on m processors to minimize maximumlateness, P.J. Brucker, Proc. IX International Symposium on Mathematical Programming, Budapest, [6] [Brucker99] A branch and bound algorithm for a single-machine scheduling problem with positive and negative time-lags, P. Brucker, T. Hilbig, and J. Hurink, Discrete Applied Mathematics,1999. [7] [Butazo97] Hard Real-Time Computing Systems, G. C. Butazo, Kluwer Academic Publishers, 1997, [8] [B la zewicz01] Scheduling Computer and Manufacturing Process, J. B la zewicz, K. H. Ecker, E. Pesch, [9] G. Schmidt, and J. W eglarz, Springer, 2001, sdk.asp, 2005, (Accessed April 2005). [10]. [11]. J. A. Kamalakar. "Lunar ranging instrument for Chandrayaan-1", Journal of Earth System Science, 12/