channel upgradation for non binary input alphabets and
play

Channel Upgradation for Non-Binary Input Alphabets and MACs Uzi - PowerPoint PPT Presentation

Channel Upgradation for Non-Binary Input Alphabets and MACs Uzi Pereg and Ido Tal Technion, EE () June 25, 2014 1 / 16 Problem W : X Y |Y| S upgrade ( W , S ) W : X Y |Y | S Goal: make I ( W


  1. Channel Upgradation for Non-Binary Input Alphabets and MACs Uzi Pereg and Ido Tal Technion, EE () June 25, 2014 1 / 16

  2. Problem W : X → Y |Y| ≫ S � � � upgrade ( W , S ) � � W ′ : X → Y ′ |Y ′ | ≤ S Goal: make I ( W ′ ) − I ( W ) as small as possible, where I ( · ) is symmetric capacity. X need not be binary W need not be symmetric W can be a t -user MAC W : X t → Y Concurrent work A. Ghayoori and T. A. Gulliver, Upgraded Approximation of Non-Binary Alphabets for Polar Code Construction, on arXiv () June 25, 2014 2 / 16

  3. Upgradation Vs. Degradation original another channel channel W P � �� � degraded channel Q upgraded another channel channel Q ′ P � �� � original channel W W ( y | u ) is the probability of receiving y ∈ Y , given that u ∈ X was sent. The input is assumed to have symmetric distribution over X . () June 25, 2014 3 / 16

  4. Motivation: polar codes Assure Eve’s channels are very bad (wiretap setting) Upper bound on rate (error correcting code) Theorem Let W : X → Y be a given channel, with |X| = q . Let µ ≥ max { 5 , q ( q − 1 ) } be a given fidelity parameter. We can construct W ′ : X → Y ′ such that W ′ is upgraded with respect to W I ( W ′ ) − I ( W ) ≤ q − 1 µ ( 2 + q · ln q ) |Y ′ | ≤ S ( q , µ ) � ( 2 µ ) q + q - impractical for big q (i.e. large input alphabet.) () June 25, 2014 4 / 16

  5. Key ideas Use bins: “close” output symbols of W go into the same bin “merge” all symbols in bin to one symbol Use “boost” symbols (sparingly): New output alphabet will have q = |X| special symbols, { κ u } u ∈X Symbol κ u received on W ′ only if u sent Small probability of receiving κ u () June 25, 2014 5 / 16

  6. W y 1 y 2 . . . u=0 W ( y 1 | 0 ) W ( y 2 | 0 ) u=1 W ( y 1 | 1 ) W ( y 2 | 1 ) u=2 W ( y 1 | 2 ) W ( y 2 | 2 ) Q y 1 y 2 . . . κ 0 κ 1 κ 2 u=0 α 0 ( y 1 ) · W ( y 1 | 0 ) α 0 ( y 2 ) · W ( y 2 | 0 ) ǫ 0 0 0 u=1 α 1 ( y 1 ) · W ( y 1 | 1 ) α 1 ( y 2 ) · W ( y 2 | 1 ) 0 ǫ 1 0 u=2 α 2 ( y 1 ) · W ( y 1 | 2 ) α 2 ( y 2 ) · W ( y 2 | 2 ) 0 0 ǫ 2 � ǫ u = ( 1 − α u ( y )) W ( y | u ) , y ∈Y 0 ≤ α u ( y ) ≤ 1 , α u ( y ) ≈ 1 () June 25, 2014 6 / 16

  7. Example W Consider a bin { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 u = 0 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 u = 1 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 u = 2 0 . 2375 0 . 004 0 . 0525 Q We choose α u ( y ) such that the columns are multiples of one another: y 1 y 2 y 3 y 4 κ 0 κ 1 κ 2 1 . 901 · 10 − 12 u = 0 0 . 2459 0 . 0038 0 . 0483 0 . 0185 0 0 5 . 8 · 10 − 12 u = 1 0 . 75 0 . 0118 0 . 1475 0 0 0 1 . 806 · 10 − 12 u = 2 0 . 2336 0 . 0036 0 . 0459 0 0 0.0107 Q ( y 1 | u ) = 63 . 55 · Q ( y 2 | u ) = 5 . 084 · Q ( y 3 | u ) = 0 . 1293 · 10 12 · Q ( y 4 | u ) () June 25, 2014 7 / 16

  8. Functions η ( x ) and R ( x ) for µ = 17 . 2 η ( x ) 6 /µ 1 + 5 /µ = e p o l s 4 /µ 3 /µ 2 /µ s l 1 /µ o p e = − 1 x 0 1 1 / e 2 1 / e R ( x ) 1 2 3 4 5 20 6 7 8 9 10 11 12 13 14 15 16 17 18 19 η ( x ) = − x · ln x () June 25, 2014 8 / 16

  9. For y ∈ Y and u ∈ X , define the APP W ( y | u ) ϕ W ( u | y ) = v ∈X W ( y | v ) . � Two symbols y 1 , y 2 ∈ Y are in the same bin, if i ( u ) = R ( ϕ W ( u | y 1 )) = R ( ϕ W ( u | y 2 )) , for all u ∈ X . In our example earlier y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 i ( 0 ) = 6 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 i ( 1 ) = 13 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 i ( 2 ) = 6 () June 25, 2014 9 / 16

  10. New APP Denote all the output letters in bin z as B ( z ) Define the leading input and output of the bin as ( u ∗ , y ∗ ) = arg max ϕ W ( u | y ) u ∈X , y ∈B ( z ) Define a new APP measure by for all u � = u ∗ ψ ( u | z ) = min y ∈B ( z ) ϕ W ( u | y ) and ψ ( u ∗ | z ) = 1 − � ψ ( u | z ) u � = u ∗ ϕ W ( u | y ) · ϕ W ( u ∗ | y ) ψ ( u | z ) α u ( y ) � ψ ( u ∗ | z ) () June 25, 2014 10 / 16

  11. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 () June 25, 2014 11 / 16

  12. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 () June 25, 2014 11 / 16

  13. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 The leading input is u ∗ = arg max u ∈{ 0 , 1 , 2 } � � max y ∈B ( z ) ϕ W ( u | y ) = () June 25, 2014 11 / 16

  14. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 The leading input is u ∗ = arg max u ∈{ 0 , 1 , 2 } � � max y ∈B ( z ) ϕ W ( u | y ) = 1 . () June 25, 2014 11 / 16

  15. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ψ ( 0 | z ) = 0 . 20 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 ψ ( 2 | z ) = 0 . 19 The leading input is u ∗ = arg max u ∈{ 0 , 1 , 2 } � � max y ∈B ( z ) ϕ W ( u | y ) = 1 . () June 25, 2014 11 / 16

  16. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ψ ( 0 | z ) = 0 . 20 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ψ ( 1 | z ) = 1 − ( 0 . 20 + 0 . 19 ) = 0 . 61 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 ψ ( 2 | z ) = 0 . 19 The leading input is u ∗ = arg max u ∈{ 0 , 1 , 2 } � � max y ∈B ( z ) ϕ W ( u | y ) = 1 . () June 25, 2014 11 / 16

  17. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ψ ( 0 | z ) = 0 . 20 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ψ ( 1 | z ) = 1 − ( 0 . 20 + 0 . 19 ) = 0 . 61 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 ψ ( 2 | z ) = 0 . 19 The leading input is u ∗ = arg max u ∈{ 0 , 1 , 2 } � � max y ∈B ( z ) ϕ W ( u | y ) = 1 . E.g. α 0 ( y 1 ) = 0 . 20 0 . 21 · 0 . 60 0 . 61 ≈ 0 . 936. Thus, we subtract ( ≈ 0 . 064 · W ( y 1 | 0 )) and pass over to our boost symbol κ 0 . () June 25, 2014 11 / 16

  18. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 004 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ψ ( 0 | z ) = 0 . 20 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ψ ( 1 | z ) = 1 − ( 0 . 20 + 0 . 19 ) = 0 . 61 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 ψ ( 2 | z ) = 0 . 19 y 1 y 2 y 3 y 4 0 . 20 0 . 21 · 0 . 60 0 . 21 · 0 . 59 0 . 20 0 . 20 0 . 20 · 0 . 59 0 . 20 0 . 21 · 0 . 58 α 0 ( y ) 0 . 61 0 . 61 0 . 61 0 . 61 α 1 ( y ) 1 1 1 1 0 . 19 0 . 19 · 0 . 60 0 . 19 0 . 20 · 0 . 59 0 . 19 0 . 21 · 0 . 59 0 . 19 0 . 21 · 0 . 58 α 2 ( y ) 0 . 61 0 . 61 0 . 61 0 . 61 () June 25, 2014 12 / 16

  19. Example (Cont.) Consider a bin B ( z ) = { y 1 , y 2 , y 3 , y 4 } , where y 1 y 2 y 3 y 4 2 . 1 · 10 − 12 W ( y | 0 ) 0 . 2625 0 . 0042 0 . 0500 5 . 8 · 10 − 12 W ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 2 . 1 · 10 − 12 W ( y | 2 ) 0 . 2375 0 . 0040 0 . 0525 y 1 y 2 y 3 y 4 ϕ W ( 0 | y ) 0 . 21 0 . 21 0 . 20 0 . 21 ψ ( 0 | z ) = 0 . 20 ϕ W ( 1 | y ) 0 . 60 0 . 59 0 . 59 0 . 58 ψ ( 1 | z ) = 1 − ( 0 . 20 + 0 . 19 ) = 0 . 61 ϕ W ( 2 | y ) 0 . 19 0 . 20 0 . 21 0 . 21 ψ ( 2 | z ) = 0 . 19 y 1 y 2 y 3 y 4 1 . 901 · 10 − 12 Q ( y | 0 ) 0 . 2459 0 . 0038 0 . 0483 5 . 8 · 10 − 12 Q ( y | 1 ) 0 . 75 0 . 0118 0 . 1475 1 . 806 · 10 − 12 Q ( y | 2 ) 0 . 2336 0 . 0036 0 . 0459 () June 25, 2014 13 / 16

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend