Danie Ludick MScEng Study Leader: Prof. D.B. Davidson Computational Electromagnetics Group Stellenbosch University Extended Studies of Focal Plane Arrays.

Slides:



Advertisements
Similar presentations
Numerical simulation of solute transport in heterogeneous porous media A. Beaudoin, J.-R. de Dreuzy, J. Erhel Workshop High Performance Computing at LAMSIN.
Advertisements

The Big Picture Scientific disciplines have developed a computational branch Models without closed form solutions solved numerically This has lead to.
Antennas for MeerKAT March TDP AWG, 13-14/3/08, San Francisco MeerKAT Semantics K.A.T. = Karoo Array Telescope –KAT is Afrikaans for cat –MeerKAT:
Jan Geralt Bij de Vaate DS4-T4 Wideband Integrated Antennas DS4-T4 Jan Geralt Bij de Vaate ASTRON.
SA SKA project 2010 Postgraduate Bursary Conference Prof David B Davidson SKA Research Chair Dept. Electrical and Electronic Engineering Univ. Stellenbosch,
Performance Analysis of Virtualization for High Performance Computing A Practical Evaluation of Hypervisor Overheads Matthew Cawood University of Cape.
RFI Characterisation of Karoo Array Telescope Structure P. Gideon Wiid (PhD) Howard C. Reader Riana H. Geschke Stellenbosch University December 2008.
SKA South Africa Overview Thomas Kusel MeerKAT System Engineering Manager April 2011.
Parabolic Dish Antennas. An Antenna is : An An effective interface between the radio and free space: Radio Antenna Free space a For Terrestrial Communications,
Probing the field of Radio Astronomy with the SKA and the Hartebeesthoek Radio Observatory: An Engineer’s perspective Sunelle Otto Hartebeesthoek Radio.
OpenFOAM on a GPU-based Heterogeneous Cluster
BDT Radio – 2b – CMV 2009/10/09 Basic Detection Techniques 2b (2009/10/09): Focal Plane Arrays Case study: WSRT System overview Receiver and.
1 Lecture 24: Parallel Algorithms I Topics: sort and matrix algorithms.
1 Friday, November 03, 2006 “The greatest obstacle to discovery is not ignorance, but the illusion of knowledge.” -D. Boorstin.
Distributed Processing of Future Radio Astronomical Observations Ger van Diepen ASTRON, Dwingeloo ATNF, Sydney.
Chapter 13 Finite Difference Methods: Outline Solving ordinary and partial differential equations Finite difference methods (FDM) vs Finite Element Methods.
1 Parallel Simulations of Underground Flow in Porous and Fractured Media H. Mustapha 1,2, A. Beaudoin 1, J. Erhel 1 and J.R. De Dreuzy IRISA – INRIA.
Name1 SKA(DS) System Design Aspects 4 th SKADS Workshop, Lisbon, 2-3 October 2008 SKA(DS) System Design Aspects: building a system Laurens Bakker.
1 Down Place Hammersmith London UK 530 Lytton Ave. Palo Alto CA USA.
Parallel Performance of Hierarchical Multipole Algorithms for Inductance Extraction Ananth Grama, Purdue University Vivek Sarin, Texas A&M University Hemant.
An Effective Dynamic Scheduling Runtime and Tuning System for Heterogeneous Multi and Many-Core Desktop Platforms Authous: Al’ecio P. D. Binotto, Carlos.
Ankit Jain B.Tech 5 th Sem (ECE) IIIT Allahabad. High Gain ( db) Low cross polarization Reasonable bandwidth, Fractional Bandwidth being at least.
Space-Time Digital Filtering of Radio Astronomy Signals using 3D Cone Filters Najith Liyanage 1, Len Bruton 2 and Pan Agathoklis 1 1 Department of Electrical.
Effects of RF Pulses on Circuits and Systems – Pieces UMCUUI C 1 UIUCUH Initial Problems Pieces Develop Understanding of Coupling Early Catalog of Expected.
Antennas The primary elements of a synthesis array M. Kesteven ATNF 25/September/2001.
Efficient design of a C-band aperture-coupled stacked microstrip array using Nexxim and Designer Alberto Di Maria German Aerospace Centre (DLR) – Microwaves.
Phased Array Feeds John O’Sullivan SKANZ 2012 CSIRO Astronomy and Space Science,
Abstract : The quest for more efficient and accurate computational electromagnetics (CEM) techniques has been vital in the design of modern engineering.
The Red Storm High Performance Computer March 19, 2008 Sue Kelly Sandia National Laboratories Abstract: Sandia National.
Analysis of Algorithms
Widefield Astronomy and Technologies for the SKA November 2009 at Limelette, Belgium The SKA AA-lo array; E.M. simulation and design Eloy de Lera.
ParCFD Parallel computation of pollutant dispersion in industrial sites Julien Montagnier Marc Buffat David Guibert.
Paul Alexander & Jaap BregmanProcessing challenge SKADS Wide-field workshop SKA Data Flow and Processing – a key SKA design driver Paul Alexander and Jaap.
Volume radiosity Michal Roušal University of West Bohemia, Plzeň Czech republic.
© 2011 Autodesk Freely licensed for use by educational institutions. Reuse and changes require a note indicating that content has been modified from the.
Departement Elektriese en Elektroniese Ingenieurswese  Department of Electrical and Electronic Engineering Ultra Wideband Reflector Antenna Feeds Dirk.
Electromagnetic Design of Broadband Antenna Feed Systems for the Northern Cross Radio Telescope (Bologna, Italy) Designed Broad Band Antenna Feed Systems.
On the Use of Sparse Direct Solver in a Projection Method for Generalized Eigenvalue Problems Using Numerical Integration Takamitsu Watanabe and Yusaku.
Introduction to CST MWS
Computational Time-reversal Imaging
Steerable antennas Meeting July 2005
CS591x -Cluster Computing and Parallel Programming
October 2008 Integrated Predictive Simulation System for Earthquake and Tsunami Disaster CREST/Japan Science and Technology Agency (JST)
ASKAP Update David DeBoer ASKAP Project Director 26 May 2010.
UNIVERSITY OF PATRAS ELECTRICAL & COMPUTER ENG. DEPT. LABORATORY OF ELECTROMAGNETICS A Hybrid Electromagnetic-Discrete Multipath Model for Diversity Performance.
Dr. Jacob Barhen Computer Science and Mathematics Division.
We have recently implemented a microwave imaging algorithm which incorporated scalar 3D wave propagation while reconstructing a 2D dielectric property.
DINO Communications Considerations 3 antenna ideas Timeline
Predicting the behaviour of a finite phased array from an infinite one Jacki van der Merwe MScEng University of Stellenbosch.
Fast and Efficient RCS Computation over a Wide Frequency Band Using the Universal Characteristic Basis Functions (UCBFs) June 2007 Authors: Prof. Raj Mittra*
Monte Carlo Linear Algebra Techniques and Their Parallelization Ashok Srinivasan Computer Science Florida State University
A Numerically Efficient Technique for Orthogonalizing the Basis Functions Arising in the Solution of Electromagnetic Scattering Problems Using the CBFM.
C OMPUTATIONAL R ESEARCH D IVISION 1 Defining Software Requirements for Scientific Computing Phillip Colella Applied Numerical Algorithms Group Lawrence.
1a.1 Parallel Computing and Parallel Computers ITCS 4/5145 Cluster Computing, UNC-Charlotte, B. Wilkinson, 2006.
Global Illumination (3) Path Tracing. Overview Light Transport Notation Path Tracing Photon Mapping.
Multipole-Based Preconditioners for Sparse Linear Systems. Ananth Grama Purdue University. Supported by the National Science Foundation.
Hanyang University Antennas & RF Devices Lab. ANTENNA THEORY ANALYSIS AND DESIGN Prof. Jaehoon Choi Dept. of Electronics and Computer Engineering
Netherlands Institute for Radio Astronomy 1 APERTIF beamformer and correlator requirements Laurens Bakker.
The Science Data Processor and Regional Centre Overview Paul Alexander UK Science Director the SKA Organisation Leader the Science Data Processor Consortium.
Direction dependent effects: Jan. 15, 2010, CPG F2F Chicago: S. Bhatnagar 1 Direction-dependent effects S. Bhatnagar NRAO, Socorro RMS ~15  Jy/beam RMS.
Xing Cai University of Oslo
Complexity In examining algorithm efficiency we must understand the idea of complexity Space complexity Time Complexity.
Imaging and Calibration Challenges
CHAPTER OBJECTIVES The primary objective of this chapter is to show how to compute the matrix inverse and to illustrate how it can be.
Model Order Reduction for Parametric Systems
Supported by the National Science Foundation.
Department of Intelligent Systems Engineering
EE 534 Numerical Methods in Electromagnetics
Electrical and Computer Engineering
Presentation transcript:

Danie Ludick MScEng Study Leader: Prof. D.B. Davidson Computational Electromagnetics Group Stellenbosch University Extended Studies of Focal Plane Arrays for the SKA and the MeerKAT

Focal Plane Array Feed-structures for the SKA Parabolic Dish Reflector Antennas for the SKA Typical Radio Astronomy Receivers Typical Radio Astronomy Receivers MeerKAT – already built a prototype MeerKAT – already built a prototype - the XDM (15 m dish) - the XDM (15 m dish) Requires a feed-structure at the focal point Requires a feed-structure at the focal point to collect the focused energy Focal Plane Arrays Focal Plane Arrays Some Applications of SKA requires whole sky Some Applications of SKA requires whole sky imaging – very time consuming with conventional single pixel feeds, such as horn antennas Solution: Reduce survey time by using focal plane Solution: Reduce survey time by using focal plane arrays → multiple beams can be formed that can be scanned independently of each other Focal Point XDM

Focal Plane Array Design Difficulty … Simulating the FPA Structure Large FPA designs, such as an array Large FPA designs, such as an array of Vivaldi Antennas → memory intensive, simulations can take several hours even days ! Need to investigate effective simulation techniques based on the Method of Moments Using FEKO in a High Performance Parallel Computing Environment Developing efficient solution techniques using the Characteristic Basis Function Method (CBFM)

The Method of Moments The Method of Moments (MoM) … Computational EM method for determining unknown surface current distribution on an electromagnetic scatterer, when illuminated by various excitations Computational EM method for determining unknown surface current distribution on an electromagnetic scatterer, when illuminated by various excitations Computational Cost of ~ O(N 2 ) memory storage and ~ O(N 3 ) for solving linear equations [A]{x} = {y} (size N x N) N = Unknowns, i.e. current associated with non-boundary edges of triangles Discritise the problem Solve for {x}

Using FEKO in a High Performance Computing (HPC) environment The Benchmarking Study … Simulating various sized Vivaldi Arrays on a Cluster at the CHPC in Cape Town Simulating various sized Vivaldi Arrays on a Cluster at the CHPC in Cape Town CHPC Infrastructure … IBM e1350 Linux Cluster … the iQudu 160 Compute Nodes → 2 dual-core AMD Opteron 2.6 GHz Processors Compute Nodes → 2 dual-core AMD Opteron 2.6 GHz Processors + 16 GByte of RAM per Node In Total 640 Processors In Total 640 Processors ~ 2.5 Teraflops Processing Power ~ 2.5 Teraflops Processing Power 2 Available Interconnects … 2 Available Interconnects … 1 GByte Ethernet and 10 GBit Infiniband Number of elements in array Number of Unknowns (N) 5504,3688,73617,47233,62438,34063,400 the iQudu

Example of HPC FPA Simulation 64 Element Vivaldi Array 32 Nodes 32 Nodes 33,624 Unknowns 33,624 Unknowns 25 Frequency Points Analyzed 25 Frequency Points Analyzed (400 MHz – 2 GHz) Results On 1 node, estimated runtime is ~ 40 hours (1.6 days) ! SimulationPrediction Total Runtime ~ 4 hours ~ 4.5 hours Memory Usage GByte 8.4 GByte |S 11 | [dB] Frequency [GHz]

What we have in terms of simulation power Fast … but not fast enough

What we want … To achieve this … improve algorithms

The Characteristic Basis Function Method (CBFM) Conventional Solvers … Direct Methods Direct Methods (Fast, N is limited) Iterative Solvers Iterative Solvers (Big problems, very long runtimes) The CBFM … Reduces the size of the matrix equations Reduces the size of the matrix equations Solve using direct methods Solve using direct methods [A]{x} = {y} (size N x N) MoM Solve

The Characteristic Basis Function Method (CBFM) … introductory example Z X Y 3λ3λ 3λ3λ N i unknowns ([A]{x} = {y}) NxN N ~ 1000’s 2 HE Z (1)Sub-domains (1 … 4) (2)Primary CBFs, J i (i) i = (3)Secondary CBFs, J k (i) i = 1..4; k = 1.. i-1, i+1,.. 4 (4)Entire solution to the problem … (5) Reduced [A]{x} = {y} (6) Solve this equation for the complex α coefficients with direct methods α coefficients with direct methods                 )( 1 )( 1 )2( )2( 1 )1( )1( 1 : : 0 0 : 0 M k M k M k M k k k M k k k N J J J x 

The Characteristic Basis Function Method (CBFM) – Numerical Results Results show Current Distribution on 4 x 1.5λ plates (λ/12 discritization) Number of Unknowns (N) = 3,744 Number of CBFM Sub-domains (M) = 4 Number of CBFs (M 2 ) = 16 Runtime reduces by factor of ~ 13 Iterative Methods CBFM Iterative Methods CBFM MethodCGSGMRESCBFM Solution Time (s) Iterative Solutions include Preconditioning

The Way Forward HPC resources (e.g. the iQudu) and methods such as CBFM provides basis for simulating very large Focal Plane Arrays CBFM is highly parallelizable … implement on cluster such as iQudu … dramatically reduce simulation time Apply efficient numerical techniques to various FPA’s (Vivaldi, Checkerboard, New Ideas …) Goal: Provide SKA Engineers with a means of analyzing large FPA feed-structures efficiently

Questions ?