The last stage of my PhD work is to investigate retransmission scheme with the PET framework. The paper is available here . (It has been accepted for publication in the IEEE Transaction on Image Processing with minor revisions) The experiments are based on the IID packet loss in an underlying BSC (Binary Symmetric Channel) model.
I have continued working on this project, aiming to run some experiments in bursty channels, such as Gilbert-Elliott (GE) channel model. This work will be submitted to the International Conference on Image Processing 2004. The official website is here.
When I started working on this, I asked several professional electrical engineers for some advice to “kick-start” some research about GE model. (Professional means non-student) To my surprise, no one has encountered this model. This model was proposed back in the 60’s (Yep – by Gilbert and later modified by Elliott) and has been used widely in the research community. (Yep, there are heaps of papers out there – I wasn’t sure where to start. That’s why I needed some preliminary advice.)
Well….after several weeks “wondering in the dark”, reading several papers, I found two very useful papers which are written by Michele Zorzi. One of them derives the channel parameters (i.e. transition probabilities) at the packet level given some channel parameters at the bit/symbol level. The other derives probability of error correction channel coding at the symbol level.
My contribution is to combine the two ideas together and formulate the probability of erasure detection channel coding. This is to be used with joint source-channel coding to find an optimal channel code allocation.
I have written a summary to achieve this. Its main purpose is a personal reminder but it may be useful for those “wondering in the darkness” of GE model.
Download file