|
Activity Number:
|
608
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 6, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #305359 |
|
Title:
|
Decentralized Sequential Hypothesis Testing
|
|
Author(s):
|
Georgios Fellouris*+ and George V. Moustakides
|
|
Companies:
|
Columbia University and University of Patras
|
|
Address:
|
124 La Salle St., #1C, New York, NY, 10027,
|
|
Keywords:
|
Sequential ; Testing ; SPRT ; Decentralized ; Asymptotic ; Optimality
|
|
Abstract:
|
We discuss the problem of decentralized sequential hypothesis testing in the case that the sensors have full local memory. We adopt a scheme- known as Decentralized Sequential Probability Ratio Test (D-SPRT)- which entails asynchronous communication of the sensors with the fusion center at some random times. We prove that the D-SPRT is asymptotically optimal and we show that in a certain sense this asymptotic optimality can be of order-2; thus, for small type-I and type-II error probabilities, the expected time for a decision of the D-SPRT differs from that of the optimal centralized test by a constant. These results have important implications on the design of the suggested scheme. Finally, simulation experiments reveal that the D-SPRT is very efficient and outperforms the order-1 asymptotically optimal scheme suggested by J. Mei (2008).
|
