Computing and Communications -- Network Coding

Similar documents
Random Access Techniques: ALOHA (cont.)

1 Minimum Cut Problem

Basic Polyhedral theory

The Matrix Exponential

The Matrix Exponential

Combating Network Pollution Attacks: A Cascaded Error-Control Coding Approach

That is, we start with a general matrix: And end with a simpler matrix:

ECE602 Exam 1 April 5, You must show ALL of your work for full credit.

Introduction to Arithmetic Geometry Fall 2013 Lecture #20 11/14/2013

Week 3: Connected Subgraphs

Searching Linked Lists. Perfect Skip List. Building a Skip List. Skip List Analysis (1) Assume the list is sorted, but is stored in a linked list.

The Equitable Dominating Graph

Lecture 6.4: Galois groups

CS 361 Meeting 12 10/3/18

Higher order derivatives

EEO 401 Digital Signal Processing Prof. Mark Fowler

cycle that does not cross any edges (including its own), then it has at least

MA 262, Spring 2018, Final exam Version 01 (Green)

Derangements and Applications

SCHUR S THEOREM REU SUMMER 2005

Problem Set 6 Solutions

Exam 1. It is important that you clearly show your work and mark the final answer clearly, closed book, closed notes, no calculator.

Math 34A. Final Review

(Upside-Down o Direct Rotation) β - Numbers

Strongly Connected Components

1973 AP Calculus AB: Section I

From Elimination to Belief Propagation

EXST Regression Techniques Page 1

On the irreducibility of some polynomials in two variables

Network Congestion Games

1 Isoparametric Concept

COMPUTER GENERATED HOLOGRAMS Optical Sciences 627 W.J. Dallas (Monday, April 04, 2005, 8:35 AM) PART I: CHAPTER TWO COMB MATH.

10. EXTENDING TRACTABILITY

A Propagating Wave Packet Group Velocity Dispersion

Brief Introduction to Statistical Mechanics

Elements of Statistical Thermodynamics

On spanning trees and cycles of multicolored point sets with few intersections

Propositional Logic. Combinatorial Problem Solving (CPS) Albert Oliveras Enric Rodríguez-Carbonell. May 17, 2018

EEO 401 Digital Signal Processing Prof. Mark Fowler

Differential Equations

Addition of angular momentum

EECE 301 Signals & Systems Prof. Mark Fowler

Addition of angular momentum

ANALYSIS IN THE FREQUENCY DOMAIN

Coupled Pendulums. Two normal modes.

Solution of Assignment #2

Ch. 24 Molecular Reaction Dynamics 1. Collision Theory

Data Assimilation 1. Alan O Neill National Centre for Earth Observation UK

perm4 A cnt 0 for for if A i 1 A i cnt cnt 1 cnt i j. j k. k l. i k. j l. i l

Roadmap. XML Indexing. DataGuide example. DataGuides. Strong DataGuides. Multiple DataGuides for same data. CPS Topics in Database Systems

A. Limits and Horizontal Asymptotes ( ) f x f x. f x. x "±# ( ).

Einstein Equations for Tetrad Fields

3 2x. 3x 2. Prepared by Vince Zaccone For Campus Learning Assistance Services at UCSB

Pipe flow friction, small vs. big pipes

The van der Waals interaction 1 D. E. Soper 2 University of Oregon 20 April 2012

CPSC 665 : An Algorithmist s Toolkit Lecture 4 : 21 Jan Linear Programming

Physical Organization

COHORT MBA. Exponential function. MATH review (part2) by Lucian Mitroiu. The LOG and EXP functions. Properties: e e. lim.

DISTRIBUTION OF DIFFERENCE BETWEEN INVERSES OF CONSECUTIVE INTEGERS MODULO P

Hardy-Littlewood Conjecture and Exceptional real Zero. JinHua Fei. ChangLing Company of Electronic Technology Baoji Shannxi P.R.

Lecture 37 (Schrödinger Equation) Physics Spring 2018 Douglas Fields

BINOMIAL COEFFICIENTS INVOLVING INFINITE POWERS OF PRIMES. 1. Statement of results

Thus, because if either [G : H] or [H : K] is infinite, then [G : K] is infinite, then [G : K] = [G : H][H : K] for all infinite cases.

Abstract Interpretation: concrete and abstract semantics

Finite element discretization of Laplace and Poisson equations

NEW APPLICATIONS OF THE ABEL-LIOUVILLE FORMULA

The pn junction: 2 Current vs Voltage (IV) characteristics

On the Hamiltonian of a Multi-Electron Atom

Quasi-Classical States of the Simple Harmonic Oscillator

1997 AP Calculus AB: Section I, Part A

Function Spaces. a x 3. (Letting x = 1 =)) a(0) + b + c (1) = 0. Row reducing the matrix. b 1. e 4 3. e 9. >: (x = 1 =)) a(0) + b + c (1) = 0

y = 2xe x + x 2 e x at (0, 3). solution: Since y is implicitly related to x we have to use implicit differentiation: 3 6y = 0 y = 1 2 x ln(b) ln(b)

BINOMIAL COEFFICIENTS INVOLVING INFINITE POWERS OF PRIMES

Random Process Part 1

Self-Adjointness and Its Relationship to Quantum Mechanics. Ronald I. Frank 2016

1997 AP Calculus AB: Section I, Part A

Middle East Technical University Department of Mechanical Engineering ME 413 Introduction to Finite Element Analysis

First derivative analysis

SER/BER in a Fading Channel

What are those βs anyway? Understanding Design Matrix & Odds ratios

Homework #3. 1 x. dx. It therefore follows that a sum of the

5. B To determine all the holes and asymptotes of the equation: y = bdc dced f gbd

Division of Mechanics Lund University MULTIBODY DYNAMICS. Examination Name (write in block letters):.

Gradebook & Midterm & Office Hours

Economics 201b Spring 2010 Solutions to Problem Set 3 John Zhu

3 Finite Element Parametric Geometry

u x v x dx u x v x v x u x dx d u x v x u x v x dx u x v x dx Integration by Parts Formula

2008 AP Calculus BC Multiple Choice Exam

4037 ADDITIONAL MATHEMATICS

Lorentz force rotor formulation.

ME 321 Kinematics and Dynamics of Machines S. Lambert Winter 2002

u 3 = u 3 (x 1, x 2, x 3 )

Where k is either given or determined from the data and c is an arbitrary constant.

Mutually Independent Hamiltonian Cycles of Pancake Networks

Superposition. Thinning

Hydrogen Atom and One Electron Ions

International Journal of Scientific & Engineering Research, Volume 6, Issue 7, July ISSN

Partial Derivatives: Suppose that z = f(x, y) is a function of two variables.

Statistical Thermodynamics: Sublimation of Solid Iodine

Sundials and Linear Algebra

Transcription:

89 90 98 00 Computing and Communications -- Ntwork Coding Dr. Zhiyong Chn Institut of Wirlss Communications Tchnology Shanghai Jiao Tong Univrsity China Lctur 5- Nov. 05 0

Classical Information Thory Sourc has ntropy rat H bits/sampl. Channl has capacity C bits/sampl. Rliabl communication is possibl iff H < C. Information is lik fluid passing through a pip. How about for ntworks? A Mathmatical Thory of Communication 98

Gnral Problm Each sourc is obsrvd by som nods and nds to b snt to othr nods Qustion: Undr what conditions can th sourcs b rliably snt to thir intndd nods?

Motivation Givn th wirlss ntwork as blow whr two nods A and C ar too far away to communicat dirctly. If transmitting on packt costs tim unit how many tim units do w nd to transmit on packt from A to C and on packt from C to A? A B C Traditionally intrmdiat nods in ntworks just forward data. Ntwork coding dviats from this paradigm in th sns that intrmdiat nods ar allowd to procss data bfor forwarding!

Ntwork Coding Savs Transmissions [Christina Fragouli EPFL] 5

Th Classic Exampl Givn two sourcs ach with a GB fil and two rcivrs. Each dirctd (wir-lin!) link can forward MB/s. How long dos it tak until both rcivrs hav rcivd both fils?

Without Ntwork Coding? Wll th naïv solution would first dlivr th first fil to both rcivrs thn th scond. Th total tim ndd is 000s. Can w do bttr (without ntwork coding)? First it sms that thr is a bttr forwarding-only solution. Th pictur shows that w can dlivr a total of MB/s. Howvr this is not tru. Indd crossing traffic must go through th bottlnck link A-B; to dlivr th GB information through this MB/s link w nd 000s What about with ntwork coding?

Max-Flow Min-Cut Thorm (From Wiki) Th max-flow min-cut thorm is a statmnt in optimization thory about maximal flows in flow ntworks Th maximal amount of flow is qual to th capacity of a minimal cut. In layman trms th maximum flow in a ntwork is dictatd by its bottlnck. Graph G(VE): consists of a st V of vrticsand a st E of dgs. V consists of sourcs sinks A and othr B nods A mmbr (uv) of E has a to snd information S capacity c(uv) T from u to v D C 8

Max-Flow Min-Cut Thorm Cuts: Partition of vrtics into two sts Siz of a Cut = Total Capacity Crossing th Cut Min-Cut: Minimum siz of Cuts = 5 Max-Flows from S to T Min-Cut = Max-Flow S A A B S D D C B T C T 9

Multicast Problm Buttrfly Ntworks: Each dg s capacity is. Max-Flow from A to D = Max-Flow from A to E = Multicast Max-Flow from A to D and E =.5 Max-Flow for ach individual connction is not achivd. B A F G C D E 0

With Ntwork Coding With ntwork coding w can indd dlivr all th data in 000s. Simply lt th bottlnck link transmit th XOR of th two packts (or bits) and rconstruct vrything at th rcivrs. Ntwork coding savs a factor! In this xampl this is optimal. In gnral? BTW: Sam xampl with on sourc only is known bttr: [Yunnan Wu]

Ntwork coding Nw rsarch ara (Ahlswd t. al. 000) Bnfits many aras Ntworking Communication Distributd computing Uss tools from Information Thory Algbra Combinatorics Has a potntial for improving: throughput robustnss rliability and scurity of ntworking and distributd systms.

Ntwork coding Introduc nods that do mor than forwarding + duplication. Each outgoing packts is a function of incoming packts. m F (m m m ) m F (m m m ) m ncoding

Ntwork coding Linar Coding ovr finit filds Exampls m m m +m +m m m m +m +m m m

Applications: Ntwork Bottlncks Nod B in th ntwork blow is a bottlnck bcaus it will nd to forward traffic for two flows (A to C and D to E). Howvr thanks to ovrharing it is nough if B transmits th XOR. In this xampl all nods hav th sam amount of traffic.

Applications: Scurity [Christina Fragouli EPFL] Without ntwork coding an avsdroppr may gt half of th information. With ntwork coding gtting usful information is hardr.

Digital Fountain A digital fountain strams data continuously and consumrs gt th full contnt aftr a fixd numbr of rcivd packts. Fil Transmission Clint Clint +

Digital Fountain Discussion With th right cods arbitrary n + o(n) out of n packts ar sufficint to rconstruct th complt fil. Th digital fountain ida is slightly oldr than th othr ntwork coding applications and may b sn as th original work on ntwork coding. Howvr in both th digital fountain and th scurity xampls intrmdiat nods simply forward th data without modification. As such it may b outsid th nw scop of ntwork coding. Digital fountains may also b usd to mak data mor availabl. Indd in pr-to-pr ntworks thanks to coding data may b availabl long aftr th sourc did.

Physical Layr Ntwork Coding Rmind -station xampl: Instad nod B may just rpat th rcivd physical signal saving on mor slot: [Christina Fragouli EPFL]

Linar Ntwork Coding Random Procsss in a Linar Ntwork Sourc Input: Info. Along Edgs: Sink Output: X Rlationship procsss among gnratd at thm v Y ( ) l X ( v l) ( v l) x0( v l) x ( v l)... Wightd Combination of Wightd Combination of Y( ) y ( ) ( )... Th 0 y indx procsss is a from adjacnt tim indx dgs of Z( v l) z ( ) ( )... 0 v l z v l Wightd Combination from all incoming ( v) dgs coms out of v Y ( l??? had (? tail( ) Z( v j)? jy(?? had(? v 0

Transfr Matrix Lt X X x ( X ( v) X ( v) X ( v)) z ( v ) ( v ) X ( v ) ( Z( v) Z ( v) vz ( v z x M )) 5 v 5 M A 5 v 0 5 5 5 AB Y ( ) X ( v) ( ) ( X v X v Y ( ) X ( v) ( ) ( X v X v Y ( ) X ( v) ( ) ( X v X v Z( v ) Y ( ) ( ) ( Y Y v Z ( v ) Y ( 5 ) Y ( ) Y ( 5 5 Z( v ) B Y ( ) Y ( ) Y ( Y ( ) Y ( ) Y ( Z( v) Y ( 5 ) Y ( ) Y ( 5 Z( v) Y ( 5 ) Y ( ) Y ( 5 Z( v) Y ( 5 ) Y ( ) Y ( 5 ) ) ) ) ) ) ) ) ) )

Ntwork Coding Solution W want Choos to b an idntity matrix. Choos B to b th invrs of 5 5 5 B A B A M 5 5 0 M x z z x A 5 5 0 NETWORK CODING SOLUTION EXISTS IF DETERMINANT OF M IS NON-ZERO

Connction btwn an Algbraic Quantity and A Graph Thortic Tool Kottr and Mdard (00): Lt a linar ntwork b givn with sourc nod sink nod and a dsird connction c of rat. Th following thr statmnts ar quivalnt.. Th connction c is possibl.. Th Min-Cut Max-Flow bound is satisfid. Th dtrminant of th ( c) R( c) transfr matrix is non-zro ovr th Ring F... l...?...? j R(c) R M...

Random Ntwork Coding Ho Kottr Mdard Kargr Effros (00/0) Random cofficints for linar ntwork coding Can dcod w.p. providd that th bas fild is sufficintly larg Enabls ntwork coding in unknown ntwork topologis Subspac coding: Kottr and Kschischang (00/08)

zhiyongchn@sjtu.du.cn http://iwct.sjtu.du.cn/prsonal/zychn/ 5