Project kind : Theoretical analysis, Computer simulation.

Description :

The CEO desires to gain knowledge about a sequence X(t), which cannot be observed directly but only through corrupted sequences. The CEO employs L agents who observe the independently corrupted sequences, Yi (t), for i = 1, 2, …, L and separately encode the observations for transmission to the CEO. Using this information, the CEO forms an estimate of X(t). The total data rate from the agents to the CEO is constrained by a maximum value Ri, for i=1,2, …,L. The goal is to find the minimum of the expected value of a distortion measure between X(t) and the reconstruction of X(t) for a given set of rates for agents. The challenging point in this problem is that its a combination of detection (or in case of continuous source, estimation)problem and multi user source coding problem. CEO problem is solved for special cases, but not generally.

On the other hand one might consider almost the same problem not in the case of a sequence X(t) but in case of a single shot detection problem, more precisely, a group of relays observing different noisy versions of a single object and transmitting a limited amount of information to a single destination which must fuse all the information together, the question in this scenario is what strategies relays should accept to maintain the lowest distortion at the fusion point. This problem is solved completely for the binary sources [2].

Objective :

The main aim of this project is to extend the previous results on distributed detection problem and look for any extension of this problem to the CEO problem, more precisely trying to reduce CEO problem to distributed source coding problem with the help of distributed detection problem, we will use computer simulations to find the best strategies in order to give us a clue what exactly we are looking for.

Prerequisites:

Probability theory is a must. Information theory and Detection theory is recommended. Knowledge of one programming language.

References:

[1] T. Berger, Z. Zhang, and H. Viswanathan, ”The CEO Problem,” IEEE Trans. Info. Theory, vol. 42, pp.887-902, May 1996.

[2] Shan-Yuan Ho “ Distributed Detection and Coding in Information Networks ” PhD thesis, MIT, February 2006, http://dspace.mit.edu/handle/1721.1/40082

Contact: Mohammad Karzand (LTHI) Email: mohammad.karzand@epfl.ch, Office INR-037, tel: 3 5644

Supervisors : Prof. Emre Telatar; Dr Shan-Yuan Ho

return to master thesis project