DocumentCode
3282852
Title
Coding vs. Packet Retransmission over Noisy Channels
Author
Baron, Dror ; Sarvotham, Shriram ; Baraniuk, Richard G.
Author_Institution
Dept. of Electr. & Comput. Eng., Rice Univ., Houston, TX
fYear
2006
fDate
22-24 March 2006
Firstpage
537
Lastpage
541
Abstract
In many packet-based communication systems such as TCP/IP-based systems, packets are communicated over a noisy physical layer (a channel), and if a packet cannot be decoded correctly, then the transport layer retransmits it. Of course, retransmissions consume significant resources and their use should be limited. However, decreasing the likelihood of retransmission requires to encode the packets with strong channel codes in the physical layer, which also requires additional channel resources. In this paper, we study the cross-layer tradeoff between coding and packet retransmissions, and optimize over the total channel resource consumption. We show that as the packet length k increases, the redundancy r beyond the k/C channel uses implied by Shannon´s channel capacity C is Theta(radickln(k)) extra channel uses. Moreover, as k increases we must use stronger channel codes. We then apply these results to universal coding over a piecewise memoryless channel with transitions between unknown i.i.d. statistics. Our constructive universal algorithm has redundancy r=O(k2/3radicln(k)) using packets of polynomially increasing lengths while accounting for possible packet drops caused by transitions in the statistics.
Keywords
channel capacity; channel coding; redundancy; Shannon´s channel capacity; coding; noisy channel; packet retransmission; redundancy; Channel capacity; Channel coding; Communication systems; Decoding; Delay; Information theory; Memoryless systems; Physical layer; Redundancy; Statistics; Channel coding; cross-layer design; non-asymptotic information theory; packet networks; piece-wise memoryless channels; universal channel coding;
fLanguage
English
Publisher
ieee
Conference_Titel
Information Sciences and Systems, 2006 40th Annual Conference on
Conference_Location
Princeton, NJ
Print_ISBN
1-4244-0349-9
Electronic_ISBN
1-4244-0350-2
Type
conf
DOI
10.1109/CISS.2006.286528
Filename
4067869
Link To Document