May 24, 2005STOC 2005, Baltimore1 Limits to List Decoding Reed-Solomon Codes Venkatesan Guruswami Atri Rudra (University of Washington)

Slides:



Advertisements
Similar presentations
PRG for Low Degree Polynomials from AG-Codes Gil Cohen Joint work with Amnon Ta-Shma.
Advertisements

The Method of Multiplicities Madhu Sudan Microsoft New England/MIT TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A.
Russell Impagliazzo ( IAS & UCSD ) Ragesh Jaiswal ( Columbia U. ) Valentine Kabanets ( IAS & SFU ) Avi Wigderson ( IAS ) ( based on [IJKW08, IKW09] )
Multiplicity Codes Swastik Kopparty (Rutgers) (based on [K-Saraf-Yekhanin ’11], [K ‘12], [K ‘14])
Approximate List- Decoding and Hardness Amplification Valentine Kabanets (SFU) joint work with Russell Impagliazzo and Ragesh Jaiswal (UCSD)
15-251: Great Theoretical Ideas in Computer Science Error Correction Lecture 17 October 23, 2014.
May 5, 2010 MSRI 1 The Method of Multiplicities Madhu Sudan Microsoft New England/MIT TexPoint fonts used in EMF. Read the TexPoint manual.
Group 5: Daryl, Etkin, Supartha, Rajendra and Aarthi 1.
296.3Page :Algorithms in the Real World Error Correcting Codes II – Cyclic Codes – Reed-Solomon Codes.
L. J. Wang 1 Introduction to Reed-Solomon Coding ( Part I )
15-853:Algorithms in the Real World
Information and Coding Theory
Probabilistic verification Mario Szegedy, Rutgers www/cs.rutgers.edu/~szegedy/07540 Lecture 4.
(speaker) Fedor Groshev Vladimir Potapov Victor Zyablov IITP RAS, Moscow.
List decoding Reed-Muller codes up to minimal distance: Structure and pseudo- randomness in coding theory Abhishek Bhowmick (UT Austin) Shachar Lovett.
CHANNEL CODING REED SOLOMON CODES.
The RSA Cryptosystem and Factoring Integers (II) Rong-Jaye Chen.
On Uniform Amplification of Hardness in NP Luca Trevisan STOC 05 Paper Review Present by Hai Xu.
Deciding Primality is in P M. Agrawal, N. Kayal, N. Saxena Slides by Adi Akavia.
Correcting Errors Beyond the Guruswami-Sudan Radius Farzad Parvaresh & Alexander Vardy Presented by Efrat Bank.
The Goldreich-Levin Theorem: List-decoding the Hadamard code
6/20/2015List Decoding Of RS Codes 1 Barak Pinhas ECC Seminar Tel-Aviv University.
CS151 Complexity Theory Lecture 10 April 29, 2004.
Variable-Length Codes: Huffman Codes
Foundations of Privacy Lecture 11 Lecturer: Moni Naor.
15-853Page :Algorithms in the Real World Error Correcting Codes I – Overview – Hamming Codes – Linear Codes.
CS151 Complexity Theory Lecture 9 April 27, 2004.
Ger man Aerospace Center Gothenburg, April, 2007 Coding Schemes for Crisscross Error Patterns Simon Plass, Gerd Richter, and A.J. Han Vinck.
exercise in the previous class (1)
Hamming Codes 11/17/04. History In the late 1940’s Richard Hamming recognized that the further evolution of computers required greater reliability, in.
DIGITAL COMMUNICATION Error - Correction A.J. Han Vinck.
Cyclic codes 1 CHAPTER 3: Cyclic and convolution codes Cyclic codes are of interest and importance because They posses rich algebraic structure that can.
CS151 Complexity Theory Lecture 9 April 27, 2015.
Of 29 August 4, 2015SIAM AAG: Algebraic Codes and Invariance1 Algebraic Codes and Invariance Madhu Sudan Microsoft Research.
The Polynomial Time Algorithm for Testing Primality George T. Gilbert.
Great Theoretical Ideas in Computer Science.
Combinatorial Algorithms Reference Text: Kreher and Stinson.
BCH_Code 2004/5/5Yuh-Ming Huang, CSIE, NCNU1 BCH Code a larger class of powerful random error-correcting cyclic codes a remarkable generalization of the.
15-853Page :Algorithms in the Real World Error Correcting Codes II – Cyclic Codes – Reed-Solomon Codes.
Reed Solomon Code Doug Young Suh Last updated : Aug 1, 2009.
Great Theoretical Ideas in Computer Science.
CS717 Algorithm-Based Fault Tolerance Matrix Multiplication Greg Bronevetsky.
§6 Linear Codes § 6.1 Classification of error control system § 6.2 Channel coding conception § 6.3 The generator and parity-check matrices § 6.5 Hamming.
Probabilistic verification Mario Szegedy, Rutgers www/cs.rutgers.edu/~szegedy/07540 Lecture 5.
One algorithm to rule them all One join query at a time Atri Rudra University at Buffalo.
Some Computation Problems in Coding Theory
1 Asymptotically good binary code with efficient encoding & Justesen code Tomer Levinboim Error Correcting Codes Seminar (2008)
Secret Sharing Non-Shannon Information Inequalities Presented in: Theory of Cryptography Conference (TCC) 2009 Published in: IEEE Transactions on Information.
List Decoding Product and Interleaved Codes Prasad Raghavendra Joint work with Parikshit Gopalan & Venkatesan Guruswami.
Secure Computation Lecture Arpita Patra. Recap >Three orthogonal problems- (n,t)-sharing, reconstruction, multiplication protocol > Verifiable Secret.
Secret Sharing in Distributed Storage Systems Illinois Institute of Technology Nexus of Information and Computation Theories Paris, Feb 2016 Salim El Rouayheb.
Reed-Solomon Codes Rong-Jaye Chen.
1 List decoding of binary codes: New concatenation-based results Venkatesan Guruswami U. Washington Joint work with Atri Rudra.
Class Report 林格名 : Reed Solomon Encoder. Reed-Solomom Error Correction When a codeword is decoded, there are three possible outcomes –If 2s + r < 2t (s.
RS – Reed Solomon Error correcting code. Error-correcting codes are clever ways of representing data so that one can recover the original information.
1 Tolerant Locally Testable Codes Atri Rudra Qualifying Evaluation Project Presentation Advisor: Venkatesan Guruswami.
Tali Kaufman (Bar-Ilan)
15-853:Algorithms in the Real World
Sublinear-Time Error-Correction and Error-Detection
Error-Correcting Codes:
Algebraic Codes and Invariance
General Strong Polarization
Maximally Recoverable Local Reconstruction Codes
Local Error-Detection and Error-correction
RS – Reed Solomon List Decoding.
Block codes. encodes each message individually into a codeword n is fixed, Input/out belong to alphabet Q of cardinality q. The set of Q-ary n-tuples.
How many deleted bits can one recover?
General Strong Polarization
Soft decoding, dual BCH codes, & better -biased list decodable codes
Zeev Dvir (Princeton) Shachar Lovett (IAS)
Presentation transcript:

May 24, 2005STOC 2005, Baltimore1 Limits to List Decoding Reed-Solomon Codes Venkatesan Guruswami Atri Rudra (University of Washington)

May 24, 2005 STOC 2005, Baltimore 2 Error-Correcting Codes Linear Code C : GF(q) k ! GF(q) n Hamming Distance or  (u,v) for u,v 2 GF(q) n  Number of positions u and v differ Distance of code C, d=min x,y 2 GF(q k )  (C(x),C(y)) C is an [n,k,d] GF(q) code  Relative distance  =d/n This talk is about Reed-Solomon (RS) Codes

May 24, 2005 STOC 2005, Baltimore 3 List Decoding Given r 2 GF(q) n and 0 · e · 1  Output all codewords c 2 C such that  (c,r) · en Combinatorial Issues  How big can the list of codewords be ?  LDR(C) largest e such that list size is poly(n) Algorithmic issues  Can one find list of codewords in poly(n) time ?  Cannot have poly time algo beyond LDR(C) errors

May 24, 2005 STOC 2005, Baltimore 4 List Recovery Related to List Decoding Problem Given C: GF(q) k ! GF(q) n and L i µ GF(q), 1 · i · n Find all codewords c= h c 1, ,c n i s.t. c i 2 L i 8 i |L i | · s LRB(C) largest s for which # of codewords is poly(n) L1L1 L2L2 L3L3 LnLn 12n3

May 24, 2005 STOC 2005, Baltimore 5 Reed-Solomon Codes RS [n,k+1] GF(q)  Message P a poly. of degree · k over GF(q)  S µ GF(q)  RS(P) = h P(a) i a 2 S  n= |S|  d = n – k  For this talk S = GF(q) 9 poly time algo for list decoding of RS codes till error bound J(  )=1-(1-  ) 1/2= 1- (k/N) 1/2

May 24, 2005 STOC 2005, Baltimore 6 The Big Picture for RS Polynomial Reconstruction RS List RecoveryRS List Decoding 9 poly time algo for error bound · J(  ) Negative Result ) above algo optimal

May 24, 2005 STOC 2005, Baltimore 7 Talk outline Our main result is about combinatorial limitation of List Recovery of Reed Solomon Codes Motivation of the problem Main Result and Implications Proof of the main result

May 24, 2005 STOC 2005, Baltimore 8 Combinatorial Limitations- I For any C  LDR(C ) ¸  /2 Unique decodability Relative Distance (  ) Error Bound Half Distance

May 24, 2005 STOC 2005, Baltimore 9 Combinatorial Limitations- II LDR(C) ¼  is the best one can hope for  e ¸  can’t detect errors Lots of “good” codes with LDR(C) ¼   Random Linear Codes  2x improvement over unique decoding  Difficulty: getting explicit codes Relative Distance (  ) Error Bound Half Distance Full Distance

May 24, 2005 STOC 2005, Baltimore 10 Combinatorial Limitations- III Johnson Bound  For any code C  LDR(C) ¸ J(  )=(1-(1-  ) 1/2 ) Exists codes for which Johnson Bound is tight  Non-linear codes [GRS00]  Linear codes [G02] Relative Distance (  ) Error Bound Half Distance Full Distance Johnson Bound

May 24, 2005 STOC 2005, Baltimore 11 Going beyond the Johnson Bound Go beyond Johnson Bound  Choice of code matters  Random Linear codes get there  What about well studied codes like RS codes ?  Motivation of our work Relative Distance (  ) Error Bound Half Distance Full Distance Johnson Bound

May 24, 2005 STOC 2005, Baltimore 12 Algorithmic Status of RS Unique decoding  [Peterson60] List Decoding  Johnson Bound  [Sud97, GS99] Unknown beyond JB  Some belief that LDR(RS)=(1-(1-  ) 1/2 ) ? ? ? ? ? ? ? ? Half Distance Full Distance Johnson Bound

May 24, 2005 STOC 2005, Baltimore 13 General setup for GS algorithm Polynomial Reconstruction  Pairs of numbers {(a i,b i )}, i=1..N  Finds all degree k poly P at most N-(Nk) 1/2 indices i, P(a i )  b i  a i  distinct ) List Decoding of RS  a i not necessarily distinct ) List Recovery of RS a1a1 a2a2 a3a3 aiai anan b2b2 b3b3 bibi bnbn LiLi b1b1

May 24, 2005 STOC 2005, Baltimore 14 Main Result of this talk Version of Johnson Bound implies LRB(RS) ¸ d n/k e -1 (GS algo works in poly time in this regime) We show LRB(RS) = d n/k e -1 ) For Polynomial reconstruction GS algo is optimal

May 24, 2005 STOC 2005, Baltimore 15 Implication for List Decoding RS Polynomial Reconstruction  In List Recovering setting N=n ¢d n/k e  Number of disagreements = N-n w (Nk) 1/2 With (little more than) N-(Nk) 1/2 disagreement have super poly RS codewords  GS algo works for disagreement · N-(Nk) 1/2  Improvement “must” use near distinctness of a i s a1a1 a2a2 a3a3 aiai anan d n/k e

May 24, 2005 STOC 2005, Baltimore 16 Main Result n=q=p m  D=(p m -1)/(p-1) =p m-1 +p m-2 +  +p+1 Consider RS [n,k=D+1] GF(q m ) * For each i=1, ,n the list L i = GF(p)  d n/k e = p Number of deg D polys over GF(p m ) which take values in GF(p) is p 2 m a1a1 a2a2 a3a3 aiai anan d n/k e

May 24, 2005 STOC 2005, Baltimore 17 Explicit Construction of Polys P b (z) =  i=0 b i ( za i + 1) D where b i 2 GF(p)  a is a generator of GF(p m )  D=(p m -1)/(p-1)=p m-1 +  +p+1 Poly over GF(p m ) Takes values in GF(p)  Norm function: for all x 2 GF(p m ), x D 2 GF(p) Will now prove for distinct b, P b (z) are distinct polys over GF(p m ) 2 m -1

May 24, 2005 STOC 2005, Baltimore 18 Proof Idea By Linearity,need to show P b (z) =  i=0 b i ( za i + 1) D  0 ) b  = b  =  = b 2 m -1 =0 Coefficients of all z j must be 0 ( )  i=0 b i (a i ) j =0 for j=0..D D+1 eqns and 2 m vars (some of them trivial) D j 2 m -1 D=p m-1 +  +p+1

May 24, 2005 STOC 2005, Baltimore 19 Lucas’ Lemma p prime and integers a and b  a=a 0 +a 1 p+  +a r p r  b=b 0 +b 1 p+  +b r p r ( )  ( ) ( )  ( ) mod p D=1+p+  p m-1, j=j 0 +j 1 p+  +j m p m-1 ( )  0 iff for all i, j i 2 {0,1} ) 2 m equations and 2 m var a bb0b0 a0a0 a1a1 b1b1 arar brbr D j

May 24, 2005 STOC 2005, Baltimore 20 Wrapping up the proof 2 m equations in 2 m variables  T= { j 0 +j 1 p+  j m-1 p m-1 | j i 2 {0,1} }   i=0 b i (a i ) j =0 for j 2 T Coefficient matrix is Vandermonde 2 m a j 0 (a j0 ) 2 … (a j0 ) 2 m -1 a j 1 (a j1 ) 2 … (a j1 ) 2 m -1 b0b0 b1b1 = 0

May 24, 2005 STOC 2005, Baltimore 21 Other Results in the Paper Use connection with BCH codes to get an exact estimate Show existence of explicit received word with super poly “close by” RS codewords for certain parameters  Uses ideas from [CW04]

May 24, 2005 STOC 2005, Baltimore 22 Open Questions Is Johnson Bound the true list decoding radius of Reed Solomon codes ? Show RS of rate 1/L cannot be list recovered using lists of size L which are not prime powers.  What RS codes on prime fields ?