Jie_VariableRate

95
Variable Rate Channel Capacity Jie Ren 2013/4/26

Transcript of Jie_VariableRate

Page 1: Jie_VariableRate

Variable  Rate  Channel  Capacity  

Jie  Ren  2013/4/26  

Page 2: Jie_VariableRate

Reference  

•  This  is  a  introduc?on  of  Sergio  Verdu  and  Shlomo  Shamai’s  paper.    

•  Sergio  Verdu  and  Shlomo  Shamai,  “Variable-­‐Rate  Channel  Capacity”,  IEEE  Transac?ons  on  Informa?on  Theory,  vol.  56,  No.  6,  June  2010.  

Page 3: Jie_VariableRate

Outline  (1st  talk)  

•  Conven?onal  Channel  Coding  •  Setup  of  Variable  Rate  Coding  

Page 4: Jie_VariableRate

Conven?onal  Channel  Capacity  

P(y|x)  X  

Page 5: Jie_VariableRate

Conven?onal  Channel  Capacity  

•  What  if  Channel  state  varies?  

P(y|x,m)  X  

! = inf!sup!!

!(!;!!)!

Page 6: Jie_VariableRate

Example:  BSC  

•  Crossover  Probability  p    

Page 7: Jie_VariableRate

Example:  BSC  

•  Crossover  Probability  p=0.5    

! = 1− ℎ 0.5 = 0!

Page 8: Jie_VariableRate

Example:  BSC  

•  Crossover  Probability  p~Uni(0,0.5)    

! = inf!1− ℎ ! = 0!

Page 9: Jie_VariableRate

Conven?onal  Channel  Capacity  

•  What  if  Channel  state  varies?  

P(y|x,m)  X  

! = inf!sup!!

!(!;!!)!

Page 10: Jie_VariableRate

Outline  

•  Conven?onal  Channel  Coding  •  Setup  of  Variable  Rate  Coding  •  Variable-­‐to-­‐Fixed  Capacity  •  Fixed-­‐to-­‐Variable  Capacity  •  Variable-­‐Blocklength  Capacity  •  Examples  

Page 11: Jie_VariableRate

Setup  of  Variable  Rate  Coding  

•  Fixed-­‐to-­‐fixed  coding  •  Variable-­‐to-­‐fixed  coding  •  Fixed-­‐to-­‐variable  coding  •  Variable-­‐blocklength  coding  

Page 12: Jie_VariableRate

Fixed-­‐to-­‐fixed  

X1…Xn   Channel   B1…Bk  B1…BK  

Page 13: Jie_VariableRate

Fixed-­‐to-­‐fixed  

X1…Xn   Channel   B1…Bk  B1…BK  

Page 14: Jie_VariableRate

Variable-­‐to-­‐Fixed  

! :!: {0,1}!! → !!!

X1…Xn   Channel   B1…Bmk  B1…Bm  

!!::!! → {0,1}!! !

Page 15: Jie_VariableRate

Variable-­‐to-­‐Fixed  

X1…Xn   Channel   B1…Bmk  B1…Bm  

Page 16: Jie_VariableRate

Fixed-­‐to-­‐variable  

!!:: {0,1}! → !!!

X1……   Channel   B1…Bk  B1…Bk  

Page 17: Jie_VariableRate

Fixed-­‐to-­‐variable  

X1……   Channel   B1…Bk  B1…Bk  

Page 18: Jie_VariableRate

Variable-­‐blocklength  

!!→: {0,1}! → !!!

!→!:!! → {0,1}! !

Page 19: Jie_VariableRate

Nota?ons  

Page 20: Jie_VariableRate

Basic  Rela?onship  

Page 21: Jie_VariableRate

Example:  BSC  

Page 22: Jie_VariableRate

Example:  BSC  

Page 23: Jie_VariableRate

Variable  Rate  Channel  Capacity  

Jie  Ren  2013/4/30  

Page 24: Jie_VariableRate

Outline  (2nd  talk)  

•  Reviews  •  Theorems  •  Examples  

Page 25: Jie_VariableRate

Review:  Defini?ons  

•  Fixed-­‐to-­‐fixed  capacity  

Page 26: Jie_VariableRate

Review:  Defini?ons  

•  Variable-­‐to-­‐fixed  capacity  

Page 27: Jie_VariableRate

Review:  Defini?ons  

•  Fixed-­‐to-­‐variable  &  upper  fixed-­‐to-­‐variable  

Page 28: Jie_VariableRate

Review:  Defini?ons  

•  ε  capacity  

lim!→!

!! ≤ !!

Page 29: Jie_VariableRate

Basic  Rela?onship  

Page 30: Jie_VariableRate

State-­‐Dependent  Channels  

Page 31: Jie_VariableRate

State-­‐Dependent  Channels  

•  Finite  alphabets  

Page 32: Jie_VariableRate

Outline  (2nd  talk)  

•  Reviews  •  Theorems  •  Examples  

Page 33: Jie_VariableRate

Theorem  7  

•  Two  possible  states  

 •  Variable-­‐to-­‐fixed  capacity  in  SDC  

!!!|!! !! !! = !! !!(!!|!!)!

!!!+ !! !!(!!|!!)

!

!!!!

! = max!!"

{min ! !,!! , ! !,!!

+max{!!! !,!! ! ,!!! !,!! ! }}!

Page 34: Jie_VariableRate

Theorem  12  

•  Fixed-­‐to-­‐fixed  capacity  in  SDC  

Page 35: Jie_VariableRate

Theorem  13  

•  Fixed-­‐to-­‐variable  capacity  in  SDC  

Page 36: Jie_VariableRate

Theorem  14  

•  Suppose  

•  Then  the  fixed-­‐to-­‐variable  capacity  in  SDC  

Page 37: Jie_VariableRate

Theorem  15  

•  Upper-­‐fixed-­‐to-­‐variable  capacity  in  SDC  

Page 38: Jie_VariableRate

Outline  (2nd  talk)  

•  Reviews  •  Theorems  •  Examples  

Page 39: Jie_VariableRate

Example  1  

•  With  probability  q  

•  With  probability  1-­‐q  

C0  

Page 40: Jie_VariableRate

Example  1  

•  By  defini?on  

Page 41: Jie_VariableRate

Example  2  

•  With  probability  π0  :  BSC(δ0)  •  With  probability  π1  :  BSC(δ1),  0.5>δ1>δ0  

Page 42: Jie_VariableRate

Example  2  

! = !!!!!!!!!!!!!!!

Page 43: Jie_VariableRate

Example  2  

•  By  theorem  12  

Page 44: Jie_VariableRate

Example  2  

•  By  theorem  7  

Page 45: Jie_VariableRate

Example  2  

•  By  theorem  14  &  15  

Page 46: Jie_VariableRate

Example  3  

•  Channel  alphabet  {0…1023}  •  With  probability  0.9:  no  error  •  With  probability  0.1:  yi=(xi>0)  

Page 47: Jie_VariableRate

Example  3  

•  By  theorem  13  

Page 48: Jie_VariableRate

Example  3  

•  Subop?mal  scheme  

Page 49: Jie_VariableRate

Example  4  

•  Binary  erasure  channel    BEC(E)  – E  random  variable  – Stay  constant  during  the  dura?on  of  the  codeword  

Page 50: Jie_VariableRate

Example  4  

Page 51: Jie_VariableRate

Example  4  

•  E~Uni(0,1)  

Page 52: Jie_VariableRate

Example  5  

•  Gaussian  channel  with  nonergodic  fading  

Page 53: Jie_VariableRate

Example  5  

•  Gaussian  channel  with  nonergodic  fading  

Page 54: Jie_VariableRate

Example  5  

•  Gaussian  channel  with  nonergodic  fading  

Page 55: Jie_VariableRate

Example  5  

•  Gaussian  channel  with  nonergodic  fading  

Page 56: Jie_VariableRate

Variable  Rate  Channel  Capacity  

Jie  Ren  2013/5/7  

Page 57: Jie_VariableRate

Outline  (3rd  talk)  

•  Theorems  and  proofs  

Page 58: Jie_VariableRate

Theorems  and  Proofs  

Page 59: Jie_VariableRate

Theorem  1  

•  For  all  channels  

Page 60: Jie_VariableRate

Proof  

•  Consider  a  fixed-­‐to-­‐fixed  code  

•  Let  

Page 61: Jie_VariableRate

Theorem  2  

•  If  the  channel  input  alphabet  is  finite  

Page 62: Jie_VariableRate

Proof  

•  Prove  by  contradic?on  

Page 63: Jie_VariableRate

Theorem  3  

•  State-­‐dependent  channel,  suppose  each  component  sa?sfies  strong  converse  

Page 64: Jie_VariableRate

Proof  

•  The  bound  would  be  ?ght  if  the  encoder  knows  the  channel  state  

•  Hence  it’s  an  upper  bound  in  general  

Page 65: Jie_VariableRate

Theorem  4  

•  Capacity  region  or  2-­‐user  DMBC  with  degraded  message  sets  

Page 66: Jie_VariableRate

Proof  

•  Thomas  M.  Cover,  “Elements  of  Informa?on  Theory”,  page  568  

Page 67: Jie_VariableRate

Theorem  5  

•  In  state-­‐dependent  channels  

Page 68: Jie_VariableRate

Proof  

•  Assume  iden?ty  permuta?on  •  Show  

Page 69: Jie_VariableRate

Theorem  6  

•  Theorem  5  holds  with  equality  if  the  K-­‐user  broadcast  channels  sa?sfy  the  strong  converse  

Page 70: Jie_VariableRate

Proof  

•  Number  the  channel  states  in  order  of  increasing  expected  number  of  recovered  bits  

Page 71: Jie_VariableRate

Proof  

•  For  an  arbitrarily  small  δ  

Page 72: Jie_VariableRate

Proof  

•  Hence,  

Page 73: Jie_VariableRate

Theorem  7  

•  Suppose  the  channel  has  finite  input/output  alphabets  and  is  memoryless.  

Page 74: Jie_VariableRate

Proof  

•  When  the  cons?tuent  channels  are  discrete  memoryless,  the  broadcast  channel  with  degraded  message  sets  sa?sfies  the  strong  converse.  

•  Theorem  7  follows  from  4  and  6  

Page 75: Jie_VariableRate

Theorem  8  

•  For  all  channels  

Page 76: Jie_VariableRate

Proof  

•  Consider  a  fixed-­‐to-­‐fixed  code  

•  Define  fixed-­‐to-­‐variable  code  

Page 77: Jie_VariableRate

Theorem  9  

•  Suppose  the  channel  has  finite  memory    

Page 78: Jie_VariableRate

Proof  

•  Similar  proof  as  theorem  8  

Page 79: Jie_VariableRate

Theorem  10  

•  Assume  either  the  input  or  output  alphabets  are  finite  

Page 80: Jie_VariableRate

Proof  

•  Jensen’s  inequality  •  Counterpart  of  Theorem  2  

Page 81: Jie_VariableRate

Theorem  11  

•  For  all  channels  

Page 82: Jie_VariableRate

Proof  

•  Assume    

Page 83: Jie_VariableRate

Proof  

•  Non-­‐an?cipatory  sesng  •  Compa?ble  variable-­‐to-­‐fixed  encoder  

Page 84: Jie_VariableRate

Proof  

•  Construct  a  sequence  of  rateless  encoders  – Let  the  ith  component  be  equal  to  the  ith  component  of  variable-­‐to-­‐fixed  code    

– For  n  sufficiently  large  

Page 85: Jie_VariableRate

Proof  

Page 86: Jie_VariableRate

Proof  

!!! !≤ !!!

!!! !≤ !

!!!!!

!

Page 87: Jie_VariableRate

Theorem  12  

•  In  state-­‐dependent  channel  

Page 88: Jie_VariableRate

Proof  

•  Shannon  capacity  •  Has  to  sa?sfy  the  worst  channel  

Page 89: Jie_VariableRate

Theorem  13  

•  In  state-­‐dependent  channel  

Page 90: Jie_VariableRate

Proof  

•  Achievability  proof  •  Scheme:  – Fixed-­‐to-­‐variable  – Repi?on  auer  the  first  transmission  

Page 91: Jie_VariableRate

Theorem  14  

•  In  state-­‐dependent  channel,  suppose  

•  Then  theorem  13  holds  with  equality  

! = min!!

!!!(!!,!!)

!

!!!

!!

!

Page 92: Jie_VariableRate

Proof  

•  Converse  proof  •  Data-­‐processing  inequality  

Page 93: Jie_VariableRate

Theorem  15  

•  In  state-­‐dependent  channel  

Page 94: Jie_VariableRate

Proof  

•  Achievability  – Assume  the  decoder  has  knowledge  of  the  ergodic  mode  

– Apply  theorem  8    

! ≥ ! = max!!

!!!(!!,!!)!

!!!!

Page 95: Jie_VariableRate

Proof  

•  Converse