Design of Iterative Message-Passing Receivers with Sparse Channel Estimators
Translated title
Design af Iterative Modtagere via Approksimative Bayesiansk Inferens
Author
Jørgensen, Peter Bjørn
Term
4. term
Publication year
2013
Submitted on
2013-06-06
Pages
100
Abstract
I denne afhandling udvikler vi en ny algoritme til modtageren i et OFDM-baseret trådløst system. Den udnytter, at radiokanalen ofte er sparsom, dvs. at dens impulssvar kun har få dominerende komponenter. Derfor formuleres kanalestimeringen som et problem i at finde et sparsomt signal. Vi anvender sparse Bayesian learning med et hierarkisk prior og beskriver hele modtagersystemet som en faktorgraf, så kanalestimering kan kobles tæt til de øvrige opgaver. På denne graf udleder vi en iterativ metode, der samtidigt estimerer kanalen og dekoder data, ved at lade meddelelser cirkulere mellem noder efter principperne belief propagation og mean-field. Vores numeriske resultater viser lavere bitfejlrate end en tilsvarende modtager, der ikke udnytter kanalens sparsitet. Da kanalestimeringen er beregningstung, foreslår vi to måder at reducere kompleksiteten: 1) en heuristisk ændring af opdateringsplanen, så kun en del af den “bløde” (sandsynlighedsbaserede) information bruges i kanalestimeringen, og 2) gruppering af kanalvariabler i vektorer af en given størrelse.
This thesis presents a new receiver algorithm for OFDM-based wireless systems. It takes advantage of the fact that the wireless channel is often sparse, meaning its impulse response has only a few dominant paths. We therefore pose channel estimation as a sparse signal estimation problem. Using sparse Bayesian learning with a hierarchical prior, we represent the entire receiver as a factor graph so that channel estimation is tightly integrated with the other receiver tasks. On this graph, we derive an iterative method that jointly estimates the channel and decodes the data by passing messages between nodes using a combination of belief propagation and mean-field inference. Our numerical results show a lower bit-error rate than a comparable receiver that does not exploit channel sparsity. Because channel estimation is computationally demanding, we propose two ways to reduce complexity: 1) a heuristic change to the message-update schedule so that only part of the soft (probabilistic) information is used for channel estimation, and 2) grouping channel variables into vectors of a chosen size.
[This abstract was generated with the help of AI]
Keywords