Compression Programs. Compression Outline. Multimedia. Lossless vs. Lossy. Encoding/Decoding. Analysis of Algorithms
|
|
- Derek Quinn
- 6 years ago
- Views:
Transcription
1 Aalysis of Algorithms Compressio Programs File Compressio: Gzip, Bzip Archivers :Arc, Pkzip, Wirar, File Systems: NTFS Piyush Kumar (Lecture 5: Compressio) Welcome to 453 Source: Guy E. Blelloch, Emad, Tseg HDTV (Mpeg 4) Soud (Mp3) Images (Jpeg) Multimedia Compressio Outlie Itroductio: Lossy vs. Lossless Iformatio Theory: Etropy, etc. Probability Codig: Huffma + Arithmetic Codig Ecodig/Decodig Will use message i geeric sese to mea the data to be compressed Iput Message Ecoder Compressed Message Decoder Output Message CODEC The ecoder ad decoder eed to uderstad commo compressed format. Lossless vs. Lossy Lossless: Iput message = Output message Lossy: Iput message Output message Lossy does ot ecessarily mea loss of quality. I fact the output could be better tha the iput. Drop radom oise i images (dust o les) Drop backgroud i music Fix spellig errors i text. Put ito better form. Writig is the art of lossy text compressio.
2 Lossless Compressio Techiques LZW (Lempel-Ziv-Welch) compressio Build dictioary Replace patters with idex of dict. Burrows-Wheeler trasform Block sort data to improve compressio Ru legth ecodig Fid & compress repetitive sequeces Huffma code Use variable legth codes based o frequecy How much ca we compress? For lossless compressio, assumig all iput messages are valid, if eve oe strig is compressed, some other must expad. Model vs. Coder To compress we eed a bias o the probability of messages. The model determies this bias Messages Model Ecoder Probs. Coder Bits Example models: Simple: Character couts, repeated strigs Complex: Models of a huma face Quality of Compressio Rutime vs. Compressio vs. Geerality Several stadard corpuses to compare algorithms Calgary Corpus 2 books, 5 papers, bibliography, collectio of ews articles, 3 programs, termial sessio, 2 object files, geophysical data, bitmap bw image The Archive Compariso Test maitais a compariso of just about all algorithms publicly available Compariso of Algorithms Program Algorithm Time BPC Score BOA PPM Var PPMD PPM IMP BW BZIP BW GZIP LZ77 Var LZ77 LZ77? 3.94? Iformatio Theory A iterface betwee modelig ad codig Etropy A measure of iformatio cotet Etropy of the Eglish Laguage How much iformatio does each character i typical Eglish text cotai? 2
3 Etropy (Shao 948) For a set of messages S with probability p(s), s S, the self iformatio of s is: i( s) log log p( s) p ( s ) Measured i bits if the log is base 2. The lower the probability, the higher the iformatio Etropy is the weighted average of self iformatio. H ( S ) p ( s )log p ( s ) ss Etropy Example p( S) {. 25,. 25,. 25,. 25,. 25} H( S) 3. 25log log p( S) {. 5,. 25,. 25,. 25,. 25} H( S). 5log log8 2 p( S) {. 75,. 0625,. 0625,. 0625,. 0625} H( S). 75log( 4 3) log 6 3. Etropy of the Eglish Laguage How ca we measure the iformatio per character? ASCII code = 7 Etropy = 4.5 (based o character probabilities) Huffma codes (average) = 4.7 Uix Compress = 3.5 Gzip = 2.5 BOA =.9 (curret close to best text compressor) Must be less tha.9. Shao s experimet Asked humas to predict the ext character give the whole previous text. He used these as coditioal probabilities to estimate the etropy of the Eglish Laguage. The umber of guesses required for right aswer: # of guesses > 5 Probability From the experimet he predicted H(Eglish) =.6-.3 Data compressio model Iput data Reduce Data Redudacy Reductio of Etropy Etropy Ecodig Codig How do we use the probabilities to code messages? Prefix codes ad relatioship to Etropy Huffma codes Arithmetic codes Implicit probability codes Compressed Data 3
4 Assumptios Commuicatio (or file) broke up ito pieces called messages. Adjacet messages might be of a differet types ad come from a differet probability distributios We will cosider two types of codig: Discrete: each message is a fixed set of bits Huffma codig, Shao-Fao codig Bleded: bits ca be shared amog messages Arithmetic codig Uiquely Decodable Codes A variable legth code assigs a bit strig (codeword) of variable legth to every message value e.g. a =, b = 0, c = 0, d = 0 What if you get the sequece of bits 0? Is it aba, ca, or, ad? A uiquely decodable code is a variable legth code i which bit strigs ca always be uiquely decomposed ito its codewords. Prefix Codes A prefix code is a variable legth code i which o codeword is a prefix of aother word e.g a = 0, b = 0, c =, d = 0 Ca be viewed as a biary tree with message values at the leaves ad 0 or s o the edges. a 0 b 0 0 c d Some Prefix Codes for Itegers Biary Uary Split May other fixed prefix codes: Golomb, phased-biary, subexpoetial,... Average Bit Legth For a code C with associated probabilities p(c) the average legth is defied as ABL ( C) p( c) l( c) cc We say that a prefix code C is optimal if for all prefix codes C, ABL(C) ABL(C ) Relatioship to Etropy Theorem (lower boud): For ay probability distributio p(s) with associated uiquely decodable code C, H( S) ABL ( C) Theorem (upper boud): For ay probability distributio p(s) with associated optimal prefix code C, ABL( C) H( S) 4
5 Kraft McMilla Iequality Theorem (Kraft-McMilla): For ay uiquely decodable code C, l( c) 2 cc Also, for ay set of legths L such that 2 l ll there is a prefix code C such that l( c ) l ( i,..., L ) i i Proof of the Upper Boud (Part ) Assig to each message a legth l( s) log p( s) We the have l( s) log / p( s) 2 2 ss ss ss ss So by the Kraft-McMilla ieq. there is a prefix code with legths l(s). 2 log / p( s) p( s) Proof of the Upper Boud (Part 2) Now we ca calculate the average legth give l(s) ABL( S) Ad we are doe. ss ss ss p( s) l( s) p( s) log/ p( s) p( s) ( log(/ p( s))) ss H ( S) p( s)log(/ p( s)) Aother property of optimal codes Theorem: If C is a optimal prefix code for the probabilities {p,, p } the p i > p j implies l(c i ) l(c j ) Proof: (by cotradictio) Assume l(c i ) > l(c j ). Cosider switchig codes c i ad c j. If l a is the average legth of the origial code, the legth of the ew code is ' la la p j ( l( ci ) l( c j )) pi ( l( c j ) l( ci )) la ( p j pi )( l( ci ) l( c j )) la This is cotradictio sice l a was supposed to be optimal Corollary The p i is smallest over the code, the l(c i ) is the largest. Huffma Codig Biary trees for compressio 5
6 Huffma Code Approach Variable legth ecodig of symbols Exploit statistical frequecy of symbols Efficiet whe symbol probabilities vary widely Priciple Use fewer bits to represet frequet symbols Use more bits to represet ifrequet symbols A A A B A A B A Huffma Codes Iveted by Huffma as a class assigmet i 950. Used i may, if ot most compressio algorithms gzip, bzip, jpeg (as optio), fax compressio, Properties: Geerates optimal prefix codes Cheap to geerate codes Cheap to ecode ad decode l a =H if probabilities are powers of 2 Huffma Code Example Symbol Dog Cat Bird Fish Frequecy /8 /4 /2 /8 Origial Ecodig Huffma Ecodig Expected size bits 2 bits 2 bits 2 bits bits 2 bits bit 3 bits Origial /82 + /42 + /22 + /82 = 2 bits / symbol Huffma /83 + /42 + /2 + /83 =.75 bits / symbol Huffma Codes Huffma Algorithm Start with a forest of trees each cosistig of a sigle vertex correspodig to a message s ad with weight p(s) Repeat: Select two trees with miimum weight roots p ad p 2 Joi ito sigle tree by addig root with weight p + p 2 Example p(a) =., p(b) =.2, p(c ) =.2, p(d) =.5 a(.) (.3) Step a(.) b(.2) c(.2) d(.5) b(.2) (.5) (.0) 0 (.3) (.5) d(.5) c(.2) 0 a(.) b(.2) (.3) c(.2) 0 Step 2 a(.) b(.2) Step 3 a=000, b=00, c=0, d= Ecodig ad Decodig Ecodig: Start at leaf of Huffma tree ad follow path to the root. Reverse order of bits ad sed. Decodig: Start at root of Huffma tree ad take brach for each bit received. Whe at leaf ca output message ad retur to root. (.0) There are eve faster methods that ca process 8 or 32 bits at a time 0 (.5) d(.5) 0 0 (.3) c(.2) a(.) b(.2) 6
7 Lemmas L : Let p i be the smallest over the code, the l(c i ) is the largest ad hece a leaf of the tree. ( Let its paret be u ) L2 : If p j is secod smallest over the code, the l(c j ) is the child of u i the optimal code. L3 : There is a optimal prefix code with correspodig tree T*, i which the two lowest frequecy letters are sibligs. Huffma codes are optimal Theorem: The Huffma algorithm geerates a optimal prefix code. I other words: It achieves the miimum average umber of bits per letter of ay prefix code. Proof: By iductio Base Case: Trivial (oe bit optimal) Assumptio: The method is optimal for all alphabets of size k-. Proof: Let y* ad z* be the two lowest frequecy letters merged i w*. Let T be the tree before mergig ad T after mergig. The : ABL(T ) = ABL(T) p(w*) T is optimal by iductio. Proof: Let Z be a better tree compared to T produced usig Huffma s alg. Implies ABL(Z) < ABL(T) By lemma L3, there is such a tree Z i which the leaves represetig y* ad z* are sibligs (ad has same ABL as Z). By previous page ABL(Z ) =ABL(Z) p(w*) Cotradictio! Adaptive Huffma Codes Huffma codes ca be made to be adaptive without completely recalculatig the tree o each step. Ca accout for chagig probabilities Small chages i probability, typically make small chages to the Huffma tree Used frequetly i practice Huffma Codig Disadvatages Itegral umber of bits i each code. If the etropy of a give character is 2.2 bits,the Huffma code for that character must be either 2 or 3 bits, ot
8 Towards Arithmetic codig A Example: Cosider sedig a message of legth 000 each with havig probability.999 Self iformatio of each message -log(.999)=.0044 bits Sum of self iformatio =.4 bits. Huffma codig will take at least k bits. Arithmetic codig = 3 bits! Arithmetic Codig: Itroductio Allows bledig of bits i a message sequece. Ca boud total bits required based o sum of self iformatio: l 2 s i i Used i PPM, JPEG/MPEG (as optio), DMM More expesive tha Huffma codig, but iteger implemetatio is ot too bad. Arithmetic Codig (message itervals) Assig each probability distributio to a iterval rage from 0 (iclusive) to (exclusive). e.g i f ( i) p( j) j f(a) =.0, f(b) =.2, f(c) =.7 The iterval for a particular message will be called the message iterval (e.g for b the iterval is [.2,.7)) Arithmetic Codig (sequece itervals) To code a message use the followig: l f l l s f s p s s p i i i i i i i Each message arrows the iterval by a factor of p. i Fial iterval size: s i The iterval for a message sequece will be called the sequece iterval p i Arithmetic Codig: Ecodig Example Uiquely defiig a iterval Codig the message sequece: bac The fial iterval is [.27,.3) Importat property:the sequece itervals for distict message sequeces of legth will ever overlap Therefore: specifyig ay umber i the fial iterval uiquely determies the sequece. Decodig is similar to ecodig, but o each step eed to determie what the message value is ad the reduce iterval 8
9 Arithmetic Codig: Decodig Example Decodig the umber.49, kowig the message is of legth 3: RealArith Ecodig ad Decodig RealArithEcode: Determie l ad s usig origial recurreces Code usig l + s/2 trucated to +-log s bits RealArithDecode: Read bits as eeded so code iterval falls withi a message iterval, ad the arrow sequece iterval. Repeat util messages have bee decoded. The message is bbc. Boud o Legth Theorem: For messages with self iformatio {s,,s } RealArithEcode will geerate at most 2 s i i bits. log s log pi i log pi i si i 2 s i i Applicatios of Probability Codig How do we geerate the probabilities? Usig character frequecies directly does ot work very well (e.g. 4.5 bits/char for text). Techique : trasformig the data Ru legth codig (ITU Fax stadard) Move-to-frot codig (Used i Burrows-Wheeler) Residual codig (JPEG LS) Techique 2: usig coditioal probabilities Fixed cotext (JBIG almost) Partial matchig (PPM) Ru Legth Codig Code by specifyig message value followed by umber of repeated values: e.g. abbbaacccca => (a,),(b,3),(a,2),(c,4),(a,) The characters ad couts ca be coded based o frequecy. This allows for small umber of bits overhead for low couts such as. Facsimile ITU T4 (Group 3) Stadard used by all home Fax Machies ITU = Iteratioal Telecommuicatios Stadard Ru legth ecodes sequeces of black+white pixels Fixed Huffma Code for all documets. e.g. Ru legth White Black Sice alterate black ad white, o eed for values. 9
10 Move to Frot Codig Trasforms message sequece ito sequece of itegers, that ca the be probability coded Start with values i a total order: e.g.: [a,b,c,d,e,.] For each message output positio i the order ad the move to the frot of the order. e.g.: c => output: 3, ew order: [c,a,b,d,e, ] a => output: 2, ew order: [a,c,b,d,e, ] Codes well if there are cocetratios of message values i the message sequece. Residual Codig Used for message values with meaigfull order e.g. itegers or floats. Basic Idea: guess ext value based o curret cotext. Output differece betwee guess ad actual value. Use probability code o the output. JPEG-LS JPEG Lossless (ot to be cofused with lossless JPEG) Just completed stadardizatio process. Codes i Raster Order. Uses 4 pixels as cotext: NW N NE W * Tries to guess value of * based o W, NW, N ad NE. Works i two stages JPEG LS: Stage Uses the followig equatio: mi( N, W ) if NW max( N, W ) P max( N, W ) if NW mi( N, W ) N W NW otherwise Averages eighbors ad captures edges. e.g * * * 40 JPEG LS: Stage 2 Uses 3 gradiets: W-NW, NW-N, N-NE Classifies each ito oe of 9 categories. This gives 9 3 =729 cotexts, of which oly 365 are eeded because of symmetry. Each cotext has a bias term that is used to adjust the previous predictio After correctio, the residual betwee guessed ad actual value is foud ad coded usig a Golomblike code. Usig Coditioal Probabilities: PPM Use previous k characters as the cotext. Base probabilities o couts: e.g. if see th 2 times followed by e 7 times, the the coditioal probability p(e th)=7/2. Need to keep k small so that dictioary does ot get too large. 0
11 Ideas i Lossless compressio That we did ot talk about specifically Lempel-Ziv (gzip) Tries to guess ext widow from previous data Burrows-Wheeler (bzip) Cotext sesitive sortig Block sortig trasform LZ77: Slidig Widow Lempel-Ziv a a c a a c a b c a b a b a c Dictioary (previously coded) Cursor Lookahead Buffer Dictioary ad buffer widows are fixed legth ad slide with the cursor O each step: Output (p,l,c) p = relative positio of the logest match i the dictioary l = legth of logest match c = ext char i buffer beyod logest match Advace widow by l + Lossy compressio Scalar Quatizatio Give a camera image with 2bit color, make it 4-bit grey scale. Uiform Vs No-Uiform Quatizatio The eye is more sesitive to low values of red compared to high values. Vector Quatizatio How do we compress a color image (r,g,b)? Fid k represetative poits for all colors For every pixel, output the earest represetative If the poits are clustered aroud the represetatives, the residuals are small ad hece probability codig will work well. Trasform codig Trasform iput ito aother space. Oe form of trasform is to choose a set of basis fuctios. JPEG/MPEG both use this idea.
12 Other Trasform codes Wavelets Fractal base compressio Based o the idea of fixed poits of fuctios. 2
Intermediate 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 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 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 informationH(X,Y) = H(X) + H(Y X)
Today s Topics Iformatio Theory Mohamed Hamada oftware gieerig ab The Uiversity of Aizu mail: hamada@u-aizu.ac.jp UR: http://www.u-aizu.ac.jp/~hamada tropy review tropy ad Data Compressio Uiquely decodable
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 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 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 informationWavelet Transform. CSEP 590 Data Compression Autumn Wavelet Transformed Barbara (Enhanced) Wavelet Transformed Barbara (Actual)
Wavelet Trasform CSEP 59 Data Compressio Autum 7 Wavelet Trasform Codig PACW Wavelet Trasform A family of atios that filters the data ito low resolutio data plus detail data high pass filter low pass filter
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 informationx 1 + x x n n = x 1 x 2 + x x n n = x 2 x 3 + x x n n = x 3 x 5 + x x n = x n
Sectio 6 7A Samplig Distributio of the Sample Meas To Create a Samplig Distributio of the Sample Meas take every possible sample of size from the distributio of x values ad the fid the mea of each sample
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 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 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 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 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 informationAlignment 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 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 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 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 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 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 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 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 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 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 informationPROJECT #2 GENERIC ROBOT SIMULATOR
Uiversity of Missouri-Columbia Departmet of Electrical ad Computer Egieerig ECE 7330 Itroductio to Mechatroics ad Robotic Visio Fall, 2010 PROJECT #2 GENERIC ROBOT SIMULATOR Luis Alberto Rivera Estrada
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 informationIV054 IV054 IV054 IV054 LITERATURE INTRODUCTION HISTORY OF CRYPTOGRAPHY
IV5 CODING, CRYPTOGRAPHY ad CRYPTOGRAPHIC PROTOCOLS CONTENTS Prof. Josef Gruska DrSc.. Liear codes 3. Cyclic codes. Classical (secret-key) cryptosystems 5. Public-key cryptography 6. RSA cryptosystem.
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 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 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 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 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 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 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 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 informationDesign of FPGA Based SPWM Single Phase Inverter
Proceedigs of MUCEET2009 Malaysia Techical Uiversities Coferece o Egieerig ad Techology Jue 20-22, 2009, MS Garde,Kuata, Pahag, Malaysia MUCEET2009 Desig of FPGA Based SPWM Sigle Phase Iverter Afarulrazi
More informationCounting on r-fibonacci Numbers
Claremot Colleges Scholarship @ Claremot All HMC Faculty Publicatios ad Research HMC Faculty Scholarship 5-1-2015 Coutig o r-fiboacci Numbers Arthur Bejami Harvey Mudd College Curtis Heberle Harvey Mudd
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 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 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 informationEstimation of non Distortion Audio Signal Compression
Estimatio of o Distortio Audio Sigal Compressio M. BAK, S. PODOXI, V. TSIGOUZ Departmet of Commuicatio Egieerig Ceter for Techological Educatio Holo 5 Golomb street, POB 35 Holo 58, Tel: +97-3-5669 ISRAEL
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 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 informationThe Institute of Chartered Accountants of Sri Lanka
The Istitute of Chartered Accoutats of Sri Laka Postgraduate Diploma i Busiess ad Fiace Quatitative Techiques for Busiess Hadout 02:Presetatio ad Aalysis of data Presetatio of Data The Stem ad Leaf Display
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 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 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 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 informationIntroduction to Markov Models
Itroductio to Markov Models But first: A few prelimiaries o text preprocessig Estimatig the probability of phrases of words, seteces, etc. What couts as a word? A tricky questio. How to fid Seteces?? CIS
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 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 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 informationMath 140 Introductory Statistics
6. Probability Distributio from Data Math Itroductory Statistics Professor Silvia Ferádez Chapter 6 Based o the book Statistics i Actio by A. Watkis, R. Scheaffer, ad G. Cobb. We have three ways of specifyig
More informationName Class. Date Section. Test Form A Chapter Chapter 9 Infinite Series. 1 n 1 2 n 3n 1, n 1, 2, 3, Find the fourth term of the sequence
8 Chapter 9 Ifiite Series Test Form A Chapter 9 Name Class Date Sectio. Fid the fourth term of the sequece,,,,.... 6 (a) (b) 6 (c) 8 6. Determie if the followig sequece coverges or diverges: If the sequece
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 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 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 informationSubband Coding of Speech Signals Using Decimation and Interpolation
3 th Iteratioal Coferece o AEROSPACE SCIENCES & AVIATION TECHNOLOGY, ASAT- 3, May 6 8, 9, E-Mail: asat@mtc.edu.eg Military Techical College, Kobry Elkobbah, Cairo, Egypt Tel : +() 459 43638, Fax: +() 698
More informationCHAPTER 8 JOINT PAPR REDUCTION AND ICI CANCELLATION IN OFDM SYSTEMS
CHAPTER 8 JOIT PAPR REDUCTIO AD ICI CACELLATIO I OFDM SYSTEMS Itercarrier Iterferece (ICI) is aother major issue i implemetig a OFDM system. As discussed i chapter 3, the OFDM subcarriers are arrowbad
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 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 informationRecovering low frequencies for impedance inversion by frequency domain deconvolution
Recoverig low frequecies for impedace iversio by frequecy domai decovolutio Sia. Esmaeili*, CREWES, Uiversity of Calgary, sesmaeil@ucalgary.ca Gary. Frak. Margrave, CREWES, Uiversity of Calgary, margrave@ucalgary.ca
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 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 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 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 informationINCREASE OF STRAIN GAGE OUTPUT VOLTAGE SIGNALS ACCURACY USING VIRTUAL INSTRUMENT WITH HARMONIC EXCITATION
XIX IMEKO World Cogress Fudametal ad Applied Metrology September 6, 9, Lisbo, Portugal INCREASE OF STRAIN GAGE OUTPUT VOLTAGE SIGNALS ACCURACY USING VIRTUAL INSTRUMENT WITH HARMONIC EXCITATION Dalibor
More informationBOUNDS FOR OUT DEGREE EQUITABLE DOMINATION NUMBERS IN GRAPHS
BULLETIN OF THE INTERNATIONAL MATHEMATICAL VIRTUAL INSTITUTE ISSN 2303-4874 (p), ISSN (o) 2303-4955 www.imvibl.org/bulletin Vol. 3(2013), 149-154 Former BULLETIN OF THE SOCIETY OF MATHEMATICIANS BANJA
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 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 informationEMPIRICAL MODE DECOMPOSITION IN AUDIO WATERMARKING BY USING WAVELET METHOD
EMPIRICAL MODE DECOMPOSITION IN AUDIO WATERMARKING BY USING WAVELET METHOD G.PAVAN KUMAR (PG Scholor) S.RAHUL (M.Tech) Departmet of ECE, S.V.College of Egieerig ad Techology, JNTUH Assistat Professor,
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 informationDirect Link Networks - Encoding
Direct Lik Networks - Ecodig Readig: Peterso ad Davie, Chapter 2 Sprig 2018 CS 438 Staff, Uiversity of Illiois 1 Where are we? Applicatio Layers BitTorret (P2P) HTTP (Web) Skype (VOIP) IPTV (streamig media)
More informationCS 201: Adversary arguments. This handout presents two lower bounds for selection problems using adversary arguments ëknu73,
CS 01 Schlag Jauary 6, 1999 Witer `99 CS 01: Adversary argumets This hadout presets two lower bouds for selectio problems usig adversary argumets ëku73, HS78, FG76ë. I these proofs a imagiary adversary
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 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 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 informationPerformance Limits and Practical Decoding of Interleaved Reed-Solomon Polar Concatenated Codes
Performace Limits ad Practical Decodig of Iterleaved Reed-Solomo Polar Cocateated Codes Hessam Mahdavifar, Mostafa El-Khamy, Jugwo Lee, Iyup Kag 1 arxiv:1308.1144v1 [cs.it] 6 Aug 2013 Abstract A scheme
More informationShuffling Cards. D.J.W. Telkamp. Utrecht University Mathematics Bachelor s Thesis. Supervised by Dr. K. Dajani
Shufflig Cards Utrecht Uiversity Mathematics Bachelor s Thesis D.J.W. Telkamp Supervised by Dr. K. Dajai Jue 3, 207 Cotets Itroductio 2 2 Prerequisites 2 2. Problems with the variatio distace................
More informationRadar emitter recognition method based on AdaBoost and decision tree Tang Xiaojing1, a, Chen Weigao1 and Zhu Weigang1 1
Advaces i Egieerig Research, volume 8 d Iteratioal Coferece o Automatio, Mechaical Cotrol ad Computatioal Egieerig (AMCCE 7) Radar emitter recogitio method based o AdaBoost ad decisio tree Tag Xiaojig,
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 informationRun-Time Error Detection in Polynomial Basis Multiplication Using Linear Codes
Ru-Time Error Detectio i Polyomial Basis Multiplicatio Usig Liear Codes Siavash Bayat-Saramdi ad M.A. Hasa Departmet of Electrical ad Computer Egieerig, Uiversity of Waterloo Waterloo, Otario, Caada N2L
More informationAdaptive Fuzzy Color Interpolation
Joural of Electroic Imagig Vol. (3), July 00. Adaptive Fuzzy Color Iterpolatio Pig-Sig Tsai, Tiku Acharya,, Ajay K. Ray 3 Itel Corporatio Desktop Architecture Lab 5000 West Chadler Boulevard Mail Stop:
More informationAnalysis of SDR GNSS Using MATLAB
Iteratioal Joural of Computer Techology ad Electroics Egieerig (IJCTEE) Volume 5, Issue 3, Jue 2015 Aalysis of SDR GNSS Usig MATLAB Abstract This paper explais a software defied radio global avigatio satellite
More informationDecode-forward and Compute-forward Coding Schemes for the Two-Way Relay Channel
Decode-forward ad Compute-forward Codig Schemes for the Two-Way Relay Chael Peg Zhog ad Mai Vu Departmet of Electrical ad Computer Egieerig McGill Uiversity Motreal, QC, Caada H3A A7 Emails: peg.zhog@mail.mcgill.ca,
More informationMaking sure metrics are meaningful
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
More informationThe Detection of Abrupt Changes in Fatigue Data by Using Cumulative Sum (CUSUM) Method
Proceedigs of the th WSEAS Iteratioal Coferece o APPLIED ad THEORETICAL MECHANICS (MECHANICS '8) The Detectio of Abrupt Chages i Fatigue Data by Usig Cumulative Sum (CUSUM) Method Z. M. NOPIAH, M.N.BAHARIN,
More informationTO DETERMINE THE NUMERICAL APERTURE OF A GIVEN OPTICAL FIBER. 2. Sunil Kumar 3. Varun Sharma 4. Jaswinder Singh
TO DETERMINE THE NUMERICAL APERTURE OF A GIVEN OPTICAL FIBER Submitted to: Mr. Rohit Verma Submitted By:. Rajesh Kumar. Suil Kumar 3. Varu Sharma 4. Jaswider Sigh INDRODUCTION TO AN OPTICAL FIBER Optical
More informationUsing Color Histograms to Recognize People in Real Time Visual Surveillance
Usig Color Histograms to Recogize People i Real Time Visual Surveillace DANIEL WOJTASZEK, ROBERT LAGANIERE S.I.T.E. Uiversity of Ottawa, Ottawa, Otario CANADA daielw@site.uottawa.ca, lagaier@site.uottawa.ca
More informationFault Diagnosis in Rolling Element Bearing Using Filtered Vibration and Acoustic Signal
Volume 8 o. 8 208, 95-02 ISS: 3-8080 (prited versio); ISS: 34-3395 (o-lie versio) url: http://www.ijpam.eu ijpam.eu Fault Diagosis i Rollig Elemet Usig Filtered Vibratio ad Acoustic Sigal Sudarsa Sahoo,
More informationThe Fundamental Capacity-Delay Tradeoff in Large Mobile Ad Hoc Networks
The Fudametal Capacity-Delay Tradeoff i Large Mobile Ad Hoc Networks Xiaoju Li ad Ness B. Shroff School of Electrical ad Computer Egieerig, Purdue Uiversity West Lafayette, IN 47907, U.S.A. {lix, shroff}@ec.purdue.edu
More informationON THE FUNDAMENTAL RELATIONSHIP BETWEEN THE ACHIEVABLE CAPACITY AND DELAY IN MOBILE WIRELESS NETWORKS
Chapter ON THE FUNDAMENTAL RELATIONSHIP BETWEEN THE ACHIEVABLE CAPACITY AND DELAY IN MOBILE WIRELESS NETWORKS Xiaoju Li ad Ness B. Shroff School of Electrical ad Computer Egieerig, Purdue Uiversity West
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 information(2) The MOSFET. Review of. Learning Outcome. (Metal-Oxide-Semiconductor Field Effect Transistor) 2.0) Field Effect Transistor (FET)
EEEB73 Electroics Aalysis & esig II () Review of The MOSFET (Metal-Oxide-Semicoductor Field Effect Trasistor) Referece: Neame, Chapter 3 ad Chapter 4 Learig Outcome Able to describe ad use the followig:
More informationThe Eye. Objectives: Introduction. PHY 192 The Eye 1
PHY 92 The Eye The Eye Objectives: Describe the basic process of image formatio by the huma eye ad how it ca be simulated i the laboratory. Kow what measuremets are ecessary to quatitatively diagose ear-sightedess
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 informationOutline. Motivation. Analog Functional Testing in Mixed-Signal Systems. Motivation and Background. Built-In Self-Test Architecture
Aalog Fuctioal Testig i Mixed-Sigal s Jie Qi Dept. of Electrical & Computer Egieerig Aubur Uiversity Co-Advisors: Charles Stroud ad Foster Dai Outlie Motivatio ad Backgroud Built-I Self-Test Architecture
More informationApplication of Image Fusion to Wireless Image Transmission
Applicatio of Image Fusio to Wireless Image Trasmissio Liae C. Ramac ad Pramod K. Varshey EECS Departmet, Syracuse Uiversity 121 Lik Hall, Syracuse, NY 13244 Email: lcramac@mailbox.syr.edu, varshey@cat.syr.edu
More information