Making sure metrics are meaningful
|
|
- Horatio Montgomery
- 5 years ago
- Views:
Transcription
1 Makig sure metrics are meaigful Some thigs are quatifiable, but ot very useful CPU performace: MHz is ot the same as performace Cameras: Mega-Pixels is ot the same as quality Cosistet ad quatifiable metrics provide ope competitio Suppliers of systems always wat to use the best metrics Metrics should be defied by users, ot suppliers People will optimise for metrics (it s what they are for!) Poor metrics lead to poor desig ad optimisatio Part of the specificatio problem HPCE / dt10 / 2015 / 11.1
2 Feasibility: Learig to say o HPCE / dt10 / 2015 / 11.2
3 Feasibility studies People come up with demads I wat real-time spectral aalysis of a 0hz-1GHz sigal We must process HD video withi a latecy of 1ms This base-statio must beam-form 32 chaels Is it feasible to meet those demads? Will it be easy? Will it require optimisatio? Will it require a specialised platform? Is it fudametally impossible? You eed some estimates before you start implemetatio Executio time is the most basic check to be made HPCE / dt10 / 2015 / 11.3
4 Circuit Placemet 1 p 1 p 2 p p 4 p 5 p 6 5 p 7 p 8 p 9 Logical compoets i circuit Physical resources i device Take the graph of circuit, ad fid a valid placemet oto physical resources Make sure that all logical compoets have a uique physical locatio Make sure that all logical coectios map to a physical chael HPCE / dt10 / 2015 / 11.4
5 Circuit Placemet HPCE / dt10 / 2015 / 11.5
6 How difficult is a big problem? HPCE / dt10 / 2015 / 11.6
7 Tryig to measure complexity Wat to capture complexity of a task usig equatios Time complexity: how may steps does it require Space complexity: how much storage does it require We could derive exact equatios for each How may istructios does the task take i total? How may bytes of memory are allocated durig executio Meas we have to worry about lots of details Laguage: did you use C++, Fortra, VHDL? Compiler: what optimisatio flags were used? Architecture: are itegers 32-bit or 64-bit? Exact equatios are sometimes possible, but ofte impractical HPCE / dt10 / 2015 / 11.7
8 Complexity ad Big-O Let s assume the existece of a fuctio g() g() specifies exactly how may steps are take Note: this fuctio does t eed to be stated explicitly! is the size of the problem; e.g. a iput vector of legth Goal: fid a simple fuctio f(), such that g() ϵ O(f()) g( ) O( f ( )) iff 0, m 0: : g( ) m f ( ) c c There must exist a critical value c, ad a positive costat m, such that for all > c the relatio g() m f() holds For icreasig, evetually you ll reach a poit where f() times a costat is always bigger tha g() HPCE / dt10 / 2015 / 11.8
9 Complexity ad Big-O O(f()) is a set of fuctios with the same or lower complexity O() 2 O() O( 2 ) 2 O( 2 ) O() O( 2 ) O( 3 ) It is sort of correct to claim that everythig is O( ) But it s really ot very useful... Try to fid the smallest complexity class cotaiig a fuctio O( 2 ) O( 3 ) O(2 ) Fid the fastest growig compoet, ad choose that HPCE / dt10 / 2015 / 11.9
10 Reductio Rules fuctio F(:iteger) : iteger begi for i = 0../2 acc = acc + iit(i) for j = 64../3 tmp = tmp + fuc(j,acc) ext j ext i retur tmp Executes /2 times: O() Executes (/3-64)*/2 = 2 / 6 32 O( 2 ) O(a g()) O(g()), a is idepet of O(a + g()) O(g()) O( (a ) 2 ) O( 2 ) HPCE / dt10 / 2015 / 11.10
11 Commo complexity classes it SUM(it N) { it acc=0; for(it i=0;i<n;i++) acc=acc+d[i]; retur acc; } fuctio [R]=radMMM(N) A=rad(N,N); B=rad(N,N); R=A*B; it Ack(it N) { it A(it m, it ) { if(m==0) retur +1; if(==0) retur A(m-1,1); retur A(m-1,A(m,-1)); } } fuctio [R]=radFFT(N) A=rad(N,1); B=rad(N,1); R=fft(A,B); retur A(N,N); HPCE / dt10 / 2015 / 11.11
12 Time it SUM(it N) { it acc=0; for(it i=0;i<n;i++) acc=acc+d[i]; retur acc; } 5.E+00 4.E+00 4.E+00 SUM 3.E+00 3.E+00 2.E+00 2.E+00 1.E+00 5.E-01 0.E+00 0.E+00 1.E+07 2.E+07 3.E+07 4.E+07 5.E+07 6.E+07 7.E+07 8.E+07 HPCE / dt10 / 2015 / 11.12
13 Matrix-Matrix Multiply r 1,1 r 1,2 r 1, a 1,1 a 1,2 a 1, b 1,1 b 1,2 b 1, r 2,1 r 2,2 r 2, a 2,1 a 2,2 a 2, = * b 2,1 b 2,2 b 2, r,1 r,2 r, a,1 a,2 a, b,1 b,2 b, fuctio[r]=radmmm(n) A=rad(N,N); B=rad(N,N); R=A*B; fuctio [R]=Mul(A,B) for r=1: for c=1: R(r,c)=sum(A(r,:).*B(:,c)); HPCE / dt10 / 2015 / 11.13
14 Matrix-Matrix Multiply r 1,1 r 1,2 r 1, a 1,1 a 1,2 a 1, b 1,1 b 1,2 b 1, r 2,1 r 2,2 r 2, a 2,1 a 2,2 a 2, = * b 2,1 b 2,2 b 2, r,1 r,2 r, a,1 a,2 a, b,1 b,2 b, fuctio[r]=radmmm(n) A=rad(N,N); B=rad(N,N); R=A*B; fuctio [R]=Mul(A,B) for r=1: for c=1: R(r,c)=sum(A(r,:).*B(:,c)); HPCE / dt10 / 2015 / 11.14
15 Matrix-Matrix Multiply r 1,1 r 1,2 r 1, a 1,1 a 1,2 a 1, b 1,1 b 1,2 b 1, r 2,1 r 2,2 r 2, a 2,1 a 2,2 a 2, = * b 2,1 b 2,2 b 2, r,1 r,2 r, a,1 a,2 a, b,1 b,2 b, fuctio[r]=radmmm(n) A=rad(N,N); B=rad(N,N); R=A*B; fuctio [R]=Mul(A,B) for r=1: for c=1: R(r,c)=sum(A(r,:).*B(:,c)); HPCE / dt10 / 2015 / 11.15
16 Matrix-Matrix Multiply r 1,1 r 1,2 r 1, a 1,1 a 1,2 a 1, b 1,1 b 1,2 b 1, r 2,1 r 2,2 r 2, a 2,1 a 2,2 a 2, = * b 2,1 b 2,2 b 2, r,1 r,2 r, a,1 a,2 a, b,1 b,2 b, fuctio[r]=radmmm(n) A=rad(N,N); B=rad(N,N); R=A*B; fuctio [R]=Mul(A,B) for r=1: for c=1: R(r,c)=sum(A(r,:).*B(:,c)); HPCE / dt10 / 2015 / 11.16
17 Matrix-Matrix Multiply r 1,1 r 1,2 r 1, a 1,1 a 1,2 a 1, b 1,1 b 1,2 b 1, r 2,1 r 2,2 r 2, a 2,1 a 2,2 a 2, = * b 2,1 b 2,2 b 2, r,1 r,2 r, a,1 a,2 a, b,1 b,2 b, fuctio[r]=radmmm(n) A=rad(N,N); B=rad(N,N); R=A*B; fuctio [R]=Mul(A,B) for r=1: for c=1: R(r,c)=sum(A(r,:).*B(:,c)); HPCE / dt10 / 2015 / 11.17
18 Quick-Sort void Qsort(it N, it *data) { partitio(n, data); // O() if(n>1){ Qsort(N/2, data); Qsort(N/2, data+n/2); } } /2 /2 /4 /4 /4 / log 2 HPCE / dt10 / 2015 / 11.18
19 How fast do fuctios grow? O(1) O( 4 ) O() O(1.01 ) O(!) O(2 ) O( log ) HPCE / dt10 / 2015 / 11.19
20 Polyomial Algorithms Polyomial-time algorithms are cosidered easy That s mathematically easy, ot ecesarily practical O(1) Costat time The best complexity class! Not much iterestig i it though... Read a item from RAM O() Liear time Vector additio Search through a u-ordered list O( 2 ) Quadratic time Matrix-vector multiply O( 3 ) Cubic time Dese matrix-matrix multiply Gaussia elimiatio I theory it s lower, but i practise it ofte is t see Strasse s algorithm HPCE / dt10 / 2015 / 11.20
21 Log ad Log-Liear Algorithms Algorithms which recursively sub-divide some space These are actually easy, ot just mathematically O(log ) Logarithmic time Fid a elemet i a sorted list Root fidig through bi-sectio See if a elemet belogs to a set / add a elemet to a set O( log ) Log-Liear time (also called liearithmic) Sortig a vector of items Fast-Fourier-Trasform (FFT) Both are huge improvemets over closest polyomial O(log ) preferred to O() O( log ) much better tha O( 2 ) HPCE / dt10 / 2015 / 11.21
22 Expoetial Algorithms Algorithms which explore some multi-dimesio space Class of algorithms with complexity O(a ) for a>1 Brute-force search of all legth- biary patters : O(2 ) Brute-force search of all legth- decimal strigs : O(10 ) Expoetial time algorithms are geerally very bad Scale extremely poorly with Occasioally useful as log as you are careful O(2 ) algorithm ca be useful for <32 O(a ) algorithm with a close to 1 is sometimes feasible HPCE / dt10 / 2015 / 11.22
23 Combiatorial Algorithms Lookig at permutatios ad combiatios of thigs Lots of specific sub-classes, but geerally O(!) Optimal mappig: bid abstract resources to physical oes Fid the best sub-set from a larger set of resources May iterestig egieerig problems are combiatorial HPCE / dt10 / 2015 / 11.23
24 Avoid aythig more tha polyomial Copyright Fox HPCE / dt10 / 2015 / 11.24
25 Thought Experimet We have four applicatios All curretly ru i oe secod All curretly hadle problems of size 16 Each applicatio has differet complexity Log-liear: O( log ) Cubic: O( 3 ) Expoetial: O(2 ) Combiatorial: O(!) The customer wats to hadle problems of twice the size How much do we eed to accelerate the existig applicatios? HPCE / dt10 / 2015 / 11.25
26 Time Secods log Speedup by about 2.5 times probably use multi-core HPCE / dt10 / 2015 / 11.26
27 Time Secods log Speedup by about 8 times maybe use a GPU? HPCE / dt10 / 2015 / 11.27
28 Time Secods 5 x log Umm... HPCE / dt10 / 2015 / 11.28
29 Time Secods log Speedup by times err, cluster of FPGAs? Cloud? HPCE / dt10 / 2015 / 11.29
30 Time Secods log 3 2! Speedup by times tur every atom i 1kg of iro ito a Petium? HPCE / dt10 / 2015 / 11.30
31 The limits of computatio New systems are ot magic Multi-core CPUs: ~16x speedup GPUs: ~500x speedup FPGAs: ~1000x speedup Cloud: ~10,000x speedup O(!), from =16 to =32 ~10 22 required Lots of problems are completely itractable Travellig Salesma: fid shortest path to visit cities Bi packig: pack objects ito the miimum umber of bis Boolea satisfiability: fid values to make equatio true HPCE / dt10 / 2015 / 11.31
32 How to deal with itractable problems? Measureable metrics! Circuit place-ad-route has ridiculously high complexity But we regularly create desigs with millios of logic gates... Must make decisio about quality versus rutime Wait 1 hour : desig rus at 250MHz Wait 10 hours : desig rus at 310MHz Wait? hours : desig rus at 317 MHZ Some algorithms are progressive ad approximate Quality of solutio improves as more compute time applied Mote-Carlo, Geetic Algorithms, Simulated Aealig,... No guaratee of optimality but at least you get a aswer HPCE / dt10 / 2015 / 11.32
33 Why parallelism fails: Amdahl s Law Split a give compute task X ito two portios A : The parts that caot be easily optimised or accelerated B : The parts that ca be sped up sigificatly Assume we achieve a speed-up of S B times to part B What is the speed-up S X for the etire task? T T X X' T T A A T T B B / S B Origial executio time New executio time S X T X / T TA TB T TB A SB TA TB as T A X' Achieved speedup SB HPCE / dt10 / 2015 / 11.33
34 Maximum Speed-up Maximum speed-up is limited by the serial fractio : T A /(T A +T B ) Need a tiy serial fractio to achieve big speed-ups Are 1000x speed-ups realistic the? T A /(T A +T B ) HPCE / dt10 / 2015 / 11.34
35 Practical example Fiite differece applicatios (fluid-mechaics, physics) Discretise cotiuous space ito cells Discretise cotiuous time ito distict time-steps Goal of acceleratio is to support fier resolutio solutios Usually icrease resolutio of space ad time axis together Let s take as the resolutio alog each axis Tasks withi fiite-differece Iitialisatio: iitialise the items i the first colum Processig: advace each colum through steps i time Collectio: retrieve aswers from fial colum HPCE / dt10 / 2015 / 11.35
36 Space g[1,1]=f1(0) for s=2.. g[s,1]=f1(g[s-1,1]) for t=1..-1 g[1,t+1]=g[1,t] for s=2..-1 g[s,t+1]=f2(g[s-1,t],g[s,t],g[s+1,t]) g[,t+1]=g[,t] Time retur F3(g[1..,t]) HPCE / dt10 / 2015 / 11.36
37 Total work: (+1)(+2) + C O( 2 ) HPCE / dt10 / 2015 / 11.37
38 Total work: (+1)(+2) + C O( 2 ) Critical path Critical path: logest path through depecy graph Assume ifiite processors, ad zero commuicatio overhead HPCE / dt10 / 2015 / 11.38
39 Double the resolutio of the grid Icrease resolutio i both time ad space Model thigs of smaller size Model thigs that happe faster Solutio is better, but more compute itesive Need High-Performace Computig! HPCE / dt10 / 2015 / 11.39
40 Double the resolutio of the grid Icrease resolutio i both time ad space Model thigs of smaller size Model thigs that happe faster Solutio is better, but more compute itesive Need High-Performace Computig! More... HPCE / dt10 / 2015 / 11.40
41 Serial executio: (+2) + (+2) + C Parallel executio: (+2) + + C Speedup: [ C] / [ C] Speedup is O() - icreases liearly with problem size HPCE / dt10 / 2015 / 11.41
42 Why parallelism works: Gustafso s Law Split a task X ito two portios A ad B A caot be accelerated, while B ca be parallelised But ow the executio time of A ad B deps o problem size T A () : time to perform part A for problem of size T T X X' ( ) ( ) T T A A ( ) T ( ) T B B () ( )/ S B Origial executio time New executio time S X ( ) T TA ( ) TB ( ) T T B( ) A( ) S S X B ( )/ T as X' ( ) B if O(T Achieved speedup A ()) O(T B ()) How precise is this? HPCE / dt10 / 2015 / 11.42
Alignment in linear space
Sequece Aligmet: Liear Space Aligmet i liear space Chapter 7 of Joes ad Pevzer Q. Ca we avoid usig quadratic space? Easy. Optimal value i O(m + ) space ad O(m) time. Compute OPT(i, ) from OPT(i-1, ). No
More informationAPPLICATION NOTE UNDERSTANDING EFFECTIVE BITS
APPLICATION NOTE AN95091 INTRODUCTION UNDERSTANDING EFFECTIVE BITS Toy Girard, Sigatec, Desig ad Applicatios Egieer Oe criteria ofte used to evaluate a Aalog to Digital Coverter (ADC) or data acquisitio
More informationCHAPTER 5 A NEAR-LOSSLESS RUN-LENGTH CODER
95 CHAPTER 5 A NEAR-LOSSLESS RUN-LENGTH CODER 5.1 GENERAL Ru-legth codig is a lossless image compressio techique, which produces modest compressio ratios. Oe way of icreasig the compressio ratio of a ru-legth
More informationRoberto s Notes on Infinite Series Chapter 1: Series Section 2. Infinite series
Roberto s Notes o Ifiite Series Chapter : Series Sectio Ifiite series What you eed to ow already: What sequeces are. Basic termiology ad otatio for sequeces. What you ca lear here: What a ifiite series
More informationLecture 4: Frequency Reuse Concepts
EE 499: Wireless & Mobile Commuicatios (8) Lecture 4: Frequecy euse Cocepts Distace betwee Co-Chael Cell Ceters Kowig the relatio betwee,, ad, we ca easily fid distace betwee the ceter poits of two co
More information}, how many different strings of length n 1 exist? }, how many different strings of length n 2 exist that contain at least one a 1
1. [5] Give sets A ad B, each of cardiality 1, how may fuctios map A i a oe-tooe fashio oto B? 2. [5] a. Give the set of r symbols { a 1, a 2,..., a r }, how may differet strigs of legth 1 exist? [5]b.
More informationLogarithms APPENDIX IV. 265 Appendix
APPENDIX IV Logarithms Sometimes, a umerical expressio may ivolve multiplicatio, divisio or ratioal powers of large umbers. For such calculatios, logarithms are very useful. They help us i makig difficult
More informationSingle Bit DACs in a Nutshell. Part I DAC Basics
Sigle Bit DACs i a Nutshell Part I DAC Basics By Dave Va Ess, Pricipal Applicatio Egieer, Cypress Semicoductor May embedded applicatios require geeratig aalog outputs uder digital cotrol. It may be a DC
More informationEECE 301 Signals & Systems Prof. Mark Fowler
EECE 3 Sigals & Systems Prof. Mark Fowler Note Set #6 D-T Systems: DTFT Aalysis of DT Systems Readig Assigmet: Sectios 5.5 & 5.6 of Kame ad Heck / Course Flow Diagram The arrows here show coceptual flow
More informationOn Parity based Divide and Conquer Recursive Functions
O Parity based Divide ad Coquer Recursive Fuctios Sug-Hyu Cha Abstract The parity based divide ad coquer recursio trees are itroduced where the sizes of the tree do ot grow mootoically as grows. These
More informationResolution. Learning Objectives. What are the four types of resolution?
Ladsat ETM+ image Resolutio Learig Objectives Be able to ame ad defie the four types of data resolutio. Be able to calculate the umber of pixels i a give area. Uderstad the trade-offs betwee differet types
More informationData Mining the Online Encyclopedia of Integer Sequences for New Identities Hieu Nguyen
Slide 1 of 18 Data Miig the Olie Ecyclopedia of Iteger Sequeces for New Idetities Hieu Nguye Rowa Uiversity MAA-NJ Sectio Sprig Meetig March 31, 2012 2 MAA-NJ Sprig Meetig Data Miig OEIS.b ü Ackowledgemets
More informationCh 9 Sequences, Series, and Probability
Ch 9 Sequeces, Series, ad Probability Have you ever bee to a casio ad played blackjack? It is the oly game i the casio that you ca wi based o the Law of large umbers. I the early 1990s a group of math
More informationH2 Mathematics Pure Mathematics Section A Comprehensive Checklist of Concepts and Skills by Mr Wee Wen Shih. Visit: wenshih.wordpress.
H2 Mathematics Pure Mathematics Sectio A Comprehesive Checklist of Cocepts ad Skills by Mr Wee We Shih Visit: weshih.wordpress.com Updated: Ja 2010 Syllabus topic 1: Fuctios ad graphs 1.1 Checklist o Fuctios
More informationGeometrical Optics Fiber optics
Phys 322 Lecture 15 Chapter 5 Geometrical Optics Fiber optics First optical commuicatio Alexader Graham Bell 1847-1922 1880: photophoe 4 years after ivetig a telephoe! Fiberoptics: first lightguide 1870:
More informationCOS 126 Atomic Theory of Matter
COS 126 Atomic Theory of Matter 1 Goal of the Assigmet Video Calculate Avogadro s umber Usig Eistei s equatios Usig fluorescet imagig Iput data Output Frames Blobs/Beads Estimate of Avogadro s umber 7.1833
More informationELEC 204 Digital Systems Design
Fall 2013, Koç Uiversity ELEC 204 Digital Systems Desig Egi Erzi College of Egieerig Koç Uiversity,Istabul,Turkey eerzi@ku.edu.tr KU College of Egieerig Elec 204: Digital Systems Desig 1 Today: Datapaths
More informationGeneral Model :Algorithms in the Real World. Applications. Block Codes
Geeral Model 5-853:Algorithms i the Real World Error Correctig Codes I Overview Hammig Codes Liear Codes 5-853 Page message (m) coder codeword (c) oisy chael decoder codeword (c ) message or error Errors
More information1. How many possible ways are there to form five-letter words using only the letters A H? How many such words consist of five distinct letters?
COMBINATORICS EXERCISES Stepha Wager 1. How may possible ways are there to form five-letter words usig oly the letters A H? How may such words cosist of five distict letters? 2. How may differet umber
More informationCP 405/EC 422 MODEL TEST PAPER - 1 PULSE & DIGITAL CIRCUITS. Time: Three Hours Maximum Marks: 100
PULSE & DIGITAL CIRCUITS Time: Three Hours Maximum Marks: 0 Aswer five questios, takig ANY TWO from Group A, ay two from Group B ad all from Group C. All parts of a questio (a, b, etc. ) should be aswered
More informationCombinatorics. Chapter Permutations. Reading questions. Counting Problems. Counting Technique: The Product Rule
Chapter 3 Combiatorics 3.1 Permutatios Readig questios 1. Defie what a permutatio is i your ow words. 2. What is a fixed poit i a permutatio? 3. What do we assume about mutual disjoitedess whe creatig
More informationDepartment of Electrical and Computer Engineering, Cornell University. ECE 3150: Microelectronics. Spring Due on April 26, 2018 at 7:00 PM
Departmet of Electrical ad omputer Egieerig, orell Uiersity EE 350: Microelectroics Sprig 08 Homework 0 Due o April 6, 08 at 7:00 PM Suggested Readigs: a) Lecture otes Importat Notes: ) MAKE SURE THAT
More informationObjectives. Some Basic Terms. Analog and Digital Signals. Analog-to-digital conversion. Parameters of ADC process: Related terms
Objectives. A brief review of some basic, related terms 2. Aalog to digital coversio 3. Amplitude resolutio 4. Temporal resolutio 5. Measuremet error Some Basic Terms Error differece betwee a computed
More information8. Combinatorial Structures
Virtual Laboratories > 0. Foudatios > 1 2 3 4 5 6 7 8 9 8. Combiatorial Structures The purpose of this sectio is to study several combiatorial structures that are of basic importace i probability. Permutatios
More informationCHAPTER 6 IMPLEMENTATION OF DIGITAL FIR FILTER
CHAPTER 6 IMPLEMENTATION OF DIGITAL FIR FILTER 6.1 INTRODUCTION The digital FIR filters are commo compoets i may digital sigal processig (DSP) systems. There are various applicatios like high speed/low
More informationCS3203 #5. 6/9/04 Janak J Parekh
CS3203 #5 6/9/04 Jaak J Parekh Admiistrivia Exam o Moday All slides should be up We ll try ad have solutios for HWs #1 ad #2 out by Friday I kow the HW is due o the same day; ot much I ca do, uless you
More informationx y z HD(x, y) + HD(y, z) HD(x, z)
Massachusetts Istitute of Techology Departmet of Electrical Egieerig ad Computer Sciece 6.02 Solutios to Chapter 5 Updated: February 16, 2012 Please sed iformatio about errors or omissios to hari; questios
More information7. Counting Measure. Definitions and Basic Properties
Virtual Laboratories > 0. Foudatios > 1 2 3 4 5 6 7 8 9 7. Coutig Measure Defiitios ad Basic Properties Suppose that S is a fiite set. If A S the the cardiality of A is the umber of elemets i A, ad is
More informationPERMUTATION AND COMBINATION
MPC 1 PERMUTATION AND COMBINATION Syllabus : Fudametal priciples of coutig; Permutatio as a arragemet ad combiatio as selectio, Meaig of P(, r) ad C(, r). Simple applicatios. Permutatios are arragemets
More informationX-Bar and S-Squared Charts
STATGRAPHICS Rev. 7/4/009 X-Bar ad S-Squared Charts Summary The X-Bar ad S-Squared Charts procedure creates cotrol charts for a sigle umeric variable where the data have bee collected i subgroups. It creates
More information信號與系統 Signals and Systems
Sprig 2 信號與系統 Sigals ad Systems Chapter SS- Sigals ad Systems Feg-Li Lia NTU-EE Feb Ju Figures ad images used i these lecture otes are adopted from Sigals & Systems by Ala V. Oppeheim ad Ala S. Willsky,
More informationSHORT-TERM TRAVEL TIME PREDICTION USING A NEURAL NETWORK
SHORT-TERM TRAVEL TIME PREDICTION USING A NEURAL NETWORK Giovai Huiske ad Eric va Berkum Dept. of Civil Egieerig - Uiversity of Twete - 7500 AE Eschede - The Netherlads E-mail: g.huiske@ctw.utwete.l ad
More informationDensity Slicing Reference Manual
Desity Slicig Referece Maual Improvisio, Viscout Cetre II, Uiversity of Warwick Sciece Park, Millbur Hill Road, Covetry. CV4 7HS Tel: 0044 (0) 24 7669 2229 Fax: 0044 (0) 24 7669 0091 e-mail: admi@improvisio.com
More informationPRACTICAL FILTER DESIGN & IMPLEMENTATION LAB
1 of 7 PRACTICAL FILTER DESIGN & IMPLEMENTATION LAB BEFORE YOU BEGIN PREREQUISITE LABS Itroductio to Oscilloscope Itroductio to Arbitrary/Fuctio Geerator EXPECTED KNOWLEDGE Uderstadig of LTI systems. Laplace
More informationFPGA Implementation of the Ternary Pulse Compression Sequences
FPGA Implemetatio of the Terary Pulse Compressio Sequeces N.Balaji 1, M. Sriivasa rao, K.Subba Rao 3, S.P.Sigh 4 ad N. Madhusudhaa Reddy 4 Abstract Terary codes have bee widely used i radar ad commuicatio
More informationPermutation Enumeration
RMT 2012 Power Roud Rubric February 18, 2012 Permutatio Eumeratio 1 (a List all permutatios of {1, 2, 3} (b Give a expressio for the umber of permutatios of {1, 2, 3,, } i terms of Compute the umber for
More information信號與系統 Signals and Systems
Sprig 24 信號與系統 Sigals ad Systems Chapter SS- Sigals ad Systems Feg-Li Lia NTU-EE Feb4 Ju4 Figures ad images used i these lecture otes are adopted from Sigals & Systems by Ala V. Oppeheim ad Ala S. Willsky,
More informationMeasurement of Equivalent Input Distortion AN 20
Measuremet of Equivalet Iput Distortio AN 2 Applicatio Note to the R&D SYSTEM Traditioal measuremets of harmoic distortio performed o loudspeakers reveal ot oly the symptoms of the oliearities but also
More informationDiscrete Mathematics and Probability Theory Spring 2014 Anant Sahai Note 12
EECS 70 Discrete Mathematics ad Probability Theory Sprig 204 Aat Sahai Note 2 Probability Examples Based o Coutig We will ow look at examples of radom experimets ad their correspodig sample spaces, alog
More informationA study on the efficient compression algorithm of the voice/data integrated multiplexer
A study o the efficiet compressio algorithm of the voice/data itegrated multiplexer Gyou-Yo CHO' ad Dog-Ho CHO' * Dept. of Computer Egieerig. KyiigHee Uiv. Kiheugup Yogiku Kyuggido, KOREA 449-71 PHONE
More informationGrade 6 Math Review Unit 3(Chapter 1) Answer Key
Grade 6 Math Review Uit (Chapter 1) Aswer Key 1. A) A pottery makig class charges a registratio fee of $25.00. For each item of pottery you make you pay a additioal $5.00. Write a expressio to represet
More informationASample of an XML stream is:
1 Efficiet Multichael i XML Wireless Broadcast Stream Arezoo Khatibi* 1 ad Omid Khatibi 2 1 Faculty of Computer Sciece, Uiversity of Kasha, Kasha, Ira 2 Faculty of Mathematics, Uiversity of Viea,Viea,
More informationFast Sensor Deployment for Fusion-based Target Detection
Fast Sesor Deploymet for Fusio-based Target Detectio Zhaohui Yua*, Rui Ta*, Guoliag Xig*, Cheyag Lu, Yixi Che *Departmet of Computer Sciece, City Uiversity of Hog Kog Departmet of Computer Sciece ad Egieerig,
More informationA SELECTIVE POINTER FORWARDING STRATEGY FOR LOCATION TRACKING IN PERSONAL COMMUNICATION SYSTEMS
A SELETIVE POINTE FOWADING STATEGY FO LOATION TAKING IN PESONAL OUNIATION SYSTES Seo G. hag ad hae Y. Lee Departmet of Idustrial Egieerig, KAIST 373-, Kusug-Dog, Taejo, Korea, 305-70 cylee@heuristic.kaist.ac.kr
More informationChapter 3 Digital Logic Structures
Copyright The McGraw-HillCompaies, Ic. Permissio required for reproductio or display. Computig Layers Chapter 3 Digital Logic Structures Problems Algorithms Laguage Istructio Set Architecture Microarchitecture
More informationSpread Spectrum Signal for Digital Communications
Wireless Iformatio Trasmissio System Lab. Spread Spectrum Sigal for Digital Commuicatios Istitute of Commuicatios Egieerig Natioal Su Yat-se Uiversity Spread Spectrum Commuicatios Defiitio: The trasmitted
More informationA New Space-Repetition Code Based on One Bit Feedback Compared to Alamouti Space-Time Code
Proceedigs of the 4th WSEAS It. Coferece o Electromagetics, Wireless ad Optical Commuicatios, Veice, Italy, November 0-, 006 107 A New Space-Repetitio Code Based o Oe Bit Feedback Compared to Alamouti
More information3. Error Correcting Codes
3. Error Correctig Codes Refereces V. Bhargava, Forward Error Correctio Schemes for Digital Commuicatios, IEEE Commuicatios Magazie, Vol 21 No1 11 19, Jauary 1983 Mischa Schwartz, Iformatio Trasmissio
More informationApplication of Improved Genetic Algorithm to Two-side Assembly Line Balancing
206 3 rd Iteratioal Coferece o Mechaical, Idustrial, ad Maufacturig Egieerig (MIME 206) ISBN: 978--60595-33-7 Applicatio of Improved Geetic Algorithm to Two-side Assembly Lie Balacig Ximi Zhag, Qia Wag,
More informationA New Design of Log-Periodic Dipole Array (LPDA) Antenna
Joural of Commuicatio Egieerig, Vol., No., Ja.-Jue 0 67 A New Desig of Log-Periodic Dipole Array (LPDA) Atea Javad Ghalibafa, Seyed Mohammad Hashemi, ad Seyed Hassa Sedighy Departmet of Electrical Egieerig,
More informationProblem of calculating time delay between pulse arrivals
America Joural of Egieerig Research (AJER) 5 America Joural of Egieerig Research (AJER) e-issn: 3-847 p-issn : 3-936 Volume-4, Issue-4, pp-3-4 www.ajer.org Research Paper Problem of calculatig time delay
More informationMassachusetts Institute of Technology Dept. of Electrical Engineering and Computer Science Fall Semester, Introduction to EECS 2.
Massachusetts Istitute of Techology Dept. of Electrical Egieerig ad Computer Sciece Fall Semester, 006 6.08 Itroductio to EECS Prelab Exercises Pre-Lab#3 Modulatio, demodulatio, ad filterig are itegral
More informationVoice Command Recognition System Based on MFCC and VQ Algorithms
Voice Commad Recogitio System Based o MFCC ad VQ Algorithms Mahdi Shaeh, ad Azizollah Taheri Abstract The goal of this project is to desig a system to recogitio voice commads. Most of voice recogitio systems
More informationHigh Speed Area Efficient Modulo 2 1
High Speed Area Efficiet Modulo 2 1 1-Soali Sigh (PG Scholar VLSI, RKDF Ist Bhopal M.P) 2- Mr. Maish Trivedi (HOD EC Departmet, RKDF Ist Bhopal M.P) Adder Abstract Modular adder is oe of the key compoets
More informationEMCdownload. Acknowledgements. Fair use
EMC_Sulight.idd 1 28/03/2013 09:06 Ackowledgemets Writte by Aa Sarchet, with Kate Oliver Edited by Kate Oliver Frot cover: Rebecca Scambler, 2013 Published by The Eglish ad Media Cetre, 2013 for EMCdowload.co.uk
More informationComputational Algorithm for Higher Order Legendre Polynomial and Gaussian Quadrature Method
Computatioal Algorithm for Higher Order Legre olyomial ad Gaussia Quadrature Method Asif M. Mughal, Xiu Ye ad Kamra Iqbal Dept. of Applied Sciece, Dept. of Mathematics ad Statistics, Dept. of Systems Egieerig
More informationOPTIMIZATION OF RNS FIR FILTERS FOR 6-INPUTS LUT BASED FPGAS
OPTIMIZATION OF RNS FIR FILTERS FOR 6-INPUTS LUT BASED FPGAS G.C. Cardarilli, M. Re, A. Salsao Uiversity of Rome Tor Vergata Departmet of Electroic Egieerig Via del Politecico 1 / 00133 / Rome / ITAL {marco.re,
More information202 Chapter 9 n Go Bot. Hint
Chapter 9 Go Bot Now it s time to put everythig you have leared so far i this book to good use. I this chapter you will lear how to create your first robotic project, the Go Bot, a four-wheeled robot.
More informationSampling. Introduction to Digital Data Acquisition: Physical world is analog CSE/EE Digital systems need to
Itroductio to Digital Data Acuisitio: Samplig Physical world is aalog Digital systems eed to Measure aalog uatities Switch iputs, speech waveforms, etc Cotrol aalog systems Computer moitors, automotive
More informationFingerprint Classification Based on Directional Image Constructed Using Wavelet Transform Domains
7 Figerprit Classificatio Based o Directioal Image Costructed Usig Wavelet Trasform Domais Musa Mohd Mokji, Syed Abd. Rahma Syed Abu Bakar, Zuwairie Ibrahim 3 Departmet of Microelectroic ad Computer Egieerig
More informationHOW BAD RECEIVER COORDINATES CAN AFFECT GPS TIMING
HOW BAD RECEIVER COORDINATES CAN AFFECT GPS TIMING H. Chadsey U.S. Naval Observatory Washigto, D.C. 2392 Abstract May sources of error are possible whe GPS is used for time comparisos. Some of these mo
More informationNovel pseudo random number generation using variant logic framework
Edith Cowa Uiversity Research Olie Iteratioal Cyber Resiliece coferece Cofereces, Symposia ad Campus Evets 011 Novel pseudo radom umber geeratio usig variat logic framework Jeffrey Zheg Yua Uiversity,
More informationCCD Image Processing: Issues & Solutions
CCD Image Processig: Issues & Solutios Correctio of Raw Image with Bias, Dark, Flat Images Raw File r x, y [ ] Dark Frame d[ x, y] Flat Field Image f [ xy, ] r[ x, y] d[ x, y] Raw Dark f [ xy, ] bxy [,
More informationEfficient Feedback-Based Scheduling Policies for Chunked Network Codes over Networks with Loss and Delay
Efficiet Feedback-Based Schedulig Policies for Chuked Network Codes over Networks with Loss ad Delay Aoosheh Heidarzadeh ad Amir H. Baihashemi Departmet of Systems ad Computer Egieerig, Carleto Uiversity,
More informationMcStas. An introdution to. Kim Lefmann, Peter Willendrup. Kim Lefmann, Peter Willendrup, DTU-RISØ August 28th, 2007
A itrodutio to Kim Lefma, Peter Willedrup Kim Lefma, Peter Willedrup, DTU-RISØ 1 itroductio Flexible, geeral simulatio utility for eutro scatterig experimets. Origial desig for Mote carlo Simulatio of
More informationOn the Number of Permutations on n Objects with. greatest cycle length
Ž. ADVANCES IN APPLIED MATHEMATICS 0, 9807 998 ARTICLE NO. AM970567 O the Number of Permutatios o Obects with Greatest Cycle Legth k Solomo W. Golomb ad Peter Gaal Commuicatio Scieces Istitute, Uiersity
More informationPERMUTATIONS AND COMBINATIONS
www.sakshieducatio.com PERMUTATIONS AND COMBINATIONS OBJECTIVE PROBLEMS. There are parcels ad 5 post-offices. I how may differet ways the registratio of parcel ca be made 5 (a) 0 (b) 5 (c) 5 (d) 5. I how
More informationSapana P. Dubey. (Department of applied mathematics,piet, Nagpur,India) I. INTRODUCTION
IOSR Joural of Mathematics (IOSR-JM) www.iosrjourals.org COMPETITION IN COMMUNICATION NETWORK: A GAME WITH PENALTY Sapaa P. Dubey (Departmet of applied mathematics,piet, Nagpur,Idia) ABSTRACT : We are
More informationlecture notes September 2, Sequential Choice
18.310 lecture otes September 2, 2013 Sequetial Choice Lecturer: Michel Goemas 1 A game Cosider the followig game. I have 100 blak cards. I write dow 100 differet umbers o the cards; I ca choose ay umbers
More informationECONOMIC LOT SCHEDULING
ECONOMIC LOT SCHEDULING JS, FFS ad ELS Job Shop (JS) - Each ob ca be differet from others - Make to order, low volume - Each ob has its ow sequece Fleible Flow Shop (FFS) - Limited umber of product types
More informationDesign of FPGA- Based SPWM Single Phase Full-Bridge Inverter
Desig of FPGA- Based SPWM Sigle Phase Full-Bridge Iverter Afarulrazi Abu Bakar 1, *,Md Zarafi Ahmad 1 ad Farrah Salwai Abdullah 1 1 Faculty of Electrical ad Electroic Egieerig, UTHM *Email:afarul@uthm.edu.my
More informationSummary of Random Variable Concepts April 19, 2000
Summary of Radom Variable Cocepts April 9, 2000 his is a list of importat cocepts we have covered, rather tha a review that derives or explais them. he first ad primary viewpoit: A radom process is a idexed
More informationLossless image compression Using Hashing (using collision resolution) Amritpal Singh 1 and Rachna rajpoot 2
Lossless image compressio Usig Hashig (usig collisio resolutio) Amritpal Sigh 1 ad Racha rajpoot 2 1 M.Tech.* CSE Departmet, 2 Departmet of iformatio techology Guru Kashi UiversityTalwadi Sabo, Bathida
More informationLaboratory Exercise 3: Dynamic System Response Laboratory Handout AME 250: Fundamentals of Measurements and Data Analysis
Laboratory Exercise 3: Dyamic System Respose Laboratory Hadout AME 50: Fudametals of Measuremets ad Data Aalysis Prepared by: Matthew Beigto Date exercises to be performed: Deliverables: Part I 1) Usig
More informationCAEN Tools for Discovery
Applicatio Note AN2506 Digital Gamma Neutro discrimiatio with Liquid Scitillators Viareggio 19 November 2012 Itroductio I recet years CAEN has developed a complete family of digitizers that cosists of
More informationCross-Entropy-Based Sign-Selection Algorithms for Peak-to-Average Power Ratio Reduction of OFDM Systems
4990 IEEE TRASACTIOS O SIGAL PROCESSIG, VOL. 56, O. 10, OCTOBER 2008 Cross-Etropy-Based Sig-Selectio Algorithms for Peak-to-Average Power Ratio Reductio of OFDM Systems Luqig Wag ad Chitha Tellambura Abstract
More informationDesign of a Mixed Prime Factor FFT for Portable Digital Radio Mondiale Receiver
590 Desig of a Mixed Prime Factor FFT for Portable Digital Radio Modiale Receiver Dog-Su Kim, Member, IEEE, Sag-Seol Lee, Jae-Yeo Sog, Kyu-Yeul Wag, ad Duck-Ji Chug Abstract To achieve better soud quality
More informationIntroduction to Wireless Communication Systems ECE 476/ECE 501C/CS 513 Winter 2003
troductio to Wireless Commuicatio ystems ECE 476/ECE 501C/C 513 Witer 2003 eview for Exam #1 March 4, 2003 Exam Details Must follow seatig chart - Posted 30 miutes before exam. Cheatig will be treated
More informationProcedia - Social and Behavioral Sciences 128 ( 2014 ) EPC-TKS 2013
Available olie at www.sciecedirect.com ScieceDirect Procedia - Social ad Behavioral Scieces 18 ( 014 ) 399 405 EPC-TKS 013 Iductive derivatio of formulae by a computer Sava Grozdev a *, Veseli Nekov b
More informationIntermediate Information Structures
Modified from Maria s lectures CPSC 335 Itermediate Iformatio Structures LECTURE 11 Compressio ad Huffma Codig Jo Roke Computer Sciece Uiversity of Calgary Caada Lecture Overview Codes ad Optimal Codes
More informationCOMPRESSION OF TRANSMULTIPLEXED ACOUSTIC SIGNALS
COMPRESSION OF TRANSMULTIPLEXED ACOUSTIC SIGNALS Mariusz Ziółko, Przemysław Sypka ad Bartosz Ziółko Departmet of Electroics, AGH Uiversity of Sciece ad Techology, al. Mickiewicza 3, 3-59 Kraków, Polad,
More informationE X P E R I M E N T 13
E X P E R I M E N T 13 Stadig Waves o a Strig Produced by the Physics Staff at Colli College Copyright Colli College Physics Departmet. All Rights Reserved. Uiversity Physics, Exp 13: Stadig Waves o a
More informationADSP ADSP ADSP ADSP. Advanced Digital Signal Processing (18-792) Spring Fall Semester, Department of Electrical and Computer Engineering
ADSP ADSP ADSP ADSP Advaced Digital Sigal Processig (8-79) Sprig Fall Semester, 7 Departmet of Electrical ad Computer Egieerig OTES O RADOM PROCESSES I. Itroductio Radom processes are at the heart of most
More informationEncode Decode Sample Quantize [ ] [ ]
Referece Audio Sigal Processig I Shyh-Kag Jeg Departmet of Electrical Egieerig/ Graduate Istitute of Commuicatio Egieerig M. Bosi ad R. E. Goldberg, Itroductio to Digital Audio Codig ad Stadards, Kluwer
More informationCompound Controller for DC Motor Servo System Based on Inner-Loop Extended State Observer
BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 6, No 5 Special Issue o Applicatio of Advaced Computig ad Simulatio i Iformatio Systems Sofia 06 Prit ISSN: 3-970; Olie ISSN:
More informationECE 333: Introduction to Communication Networks Fall Lecture 4: Physical layer II
ECE 333: Itroductio to Commuicatio Networks Fall 22 Lecture : Physical layer II Impairmets - distortio, oise Fudametal limits Examples Notes: his lecture cotiues the discussio of the physical layer. Recall,
More informationTehrani N Journal of Scientific and Engineering Research, 2018, 5(7):1-7
Available olie www.jsaer.com, 2018, 5(7):1-7 Research Article ISSN: 2394-2630 CODEN(USA): JSERBR 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38
More informationA Novel Small Signal Power Line Quality Measurement System
IMTC 3 - Istrumetatio ad Measuremet Techology Coferece Vail, CO, USA, - May 3 A ovel Small Sigal Power Lie Quality Measuremet System Paul B. Crilly, Erik Leadro Boaldi, Levy Ely de Lacarda de Oliveira,
More information11.11 Two-Channel Filter Banks 1/27
. Two-Chael Filter Baks /7 Two-Chael Filter Baks M We wat to look at methods that are ot based o the DFT I geeral we wat to look at Fig..6 rom Porat ad igure out how to choose i & i to get Perect Reco
More informationCOMBINATORICS 2. Recall, in the previous lesson, we looked at Taxicabs machines, which always took the shortest path home
COMBINATORICS BEGINNER CIRCLE 1/0/013 1. ADVANCE TAXICABS Recall, i the previous lesso, we looked at Taxicabs machies, which always took the shortest path home taxipath We couted the umber of ways that
More informationarxiv: v2 [math.co] 15 Oct 2018
THE 21 CARD TRICK AND IT GENERALIZATION DIBYAJYOTI DEB arxiv:1809.04072v2 [math.co] 15 Oct 2018 Abstract. The 21 card trick is well kow. It was recetly show i a episode of the popular YouTube chael Numberphile.
More informationAdaptive Resource Allocation in Multiuser OFDM Systems
Adaptive Resource Allocatio i Multiuser OFDM Systems Fial Report Multidimesioal Digital Sigal Processig Malik Meherali Saleh The Uiversity of Texas at Austi malikmsaleh@mail.utexas.edu Sprig 005 Abstract
More informationUnit 5: Estimating with Confidence
Uit 5: Estimatig with Cofidece Sectio 8.2 The Practice of Statistics, 4 th editio For AP* STARNES, YATES, MOORE Uit 5 Estimatig with Cofidece 8.1 8.2 8.3 Cofidece Itervals: The Basics Estimatig a Populatio
More informationLecture 28: MOSFET as an Amplifier. Small-Signal Equivalent Circuit Models.
hites, EE 320 ecture 28 Page 1 of 7 ecture 28: MOSFET as a Amplifier. Small-Sigal Equivalet Circuit Models. As with the BJT, we ca use MOSFETs as AC small-sigal amplifiers. A example is the so-called coceptual
More informationAkinwaJe, A.T., IbharaJu, F.T. and Arogundade, 0.1'. Department of Computer Sciences University of Agriculture, Abeokuta, Nigeria
COMPARATIVE ANALYSIS OF ARTIFICIAL NEURAL NETWORK'S BACK PROPAGATION ALGORITHM TO STATISTICAL LEAST SQURE METHOD IN SECURITY PREDICTION USING NIGERIAN STOCK EXCHANGE MARKET AkiwaJe, A.T., IbharaJu, F.T.
More informationTHE LUCAS TRIANGLE RECOUNTED. Arthur T. Benjamin Dept. of Mathematics, Harvey Mudd College, Claremont, CA Introduction
THE LUCAS TRIANLE RECOUNTED Arthur T Bejami Dept of Mathematics, Harvey Mudd College, Claremot, CA 91711 bejami@hmcedu 1 Itroductio I 2], Neville Robbis explores may properties of the Lucas triagle, a
More informationIndicator No mark Single preset Dual preset DIN W144 H72mm DIN W48 H96mm No mark DIN W72 H72mm (4 digit) (6 digit) Counter/Timer
FX/FX/FX Series DIN W7 7, W8 96, W 7mm er/timer Features 6 iput modes ad output modes ig speed: cps/cps/kcps/kcps Selectable voltage iput (PNP) or No voltage iput (NPN) dditio of Up/Dow iput mode Wide
More informationExtra Practice 1. Name Date. Lesson 1.1: Patterns in Division
Master 1.22 Extra Practice 1 Lesso 1.1: Patters i Divisio 1. Which umbers are divisible by 4? By 5? How do you kow? a) 90 b) 134 c) 395 d) 1724 e) 30 f) 560 g) 3015 h) 74 i) 748 2. Write a 5-digit umber
More informationDesign and Construction of a Three-phase Digital Energy Meter
Desig ad Costructio of a Three-phase Digital Eergy Meter D.P.Chadima, V.G.R.G. Jayawardae, E.A.E.H. Hemachadra, I.N.Jayasekera, H.V.L.Hasaraga, D.C. Hapuarachchi (chadima@elect.mrt.ac.lk, geethagaj@gmail.com,era.hem@gmail.com,ishaivaka@gmail.com,lahiru_hasaraga@yahoo.com,diya_elect.uom@gmail.com)
More information4. INTERSYMBOL INTERFERENCE
DATA COMMUNICATIONS 59 4. INTERSYMBOL INTERFERENCE 4.1 OBJECT The effects of restricted badwidth i basebad data trasmissio will be studied. Measuremets relative to itersymbol iterferece, usig the eye patter
More information