On dropping noisy packets in Kalman filtering over a wireless fading channel

We consider the wireless estimation problem in fading environments which is inherent to many mobile sensor networks. Previous works on wireless estimation in sensor networks have extended the same design principles of data networks by carrying out their analysis under the assumption that the erroneous packets are simply dropped. In this paper, we show that proper use of the information contained in all the packets, including the erroneous ones, can improve the performance considerably. We derive an analytical expression to evaluate the performance improvement, for one class of communication noise characteristics, when the information of all the packets is used in the receiver. We also prove that keeping all the packets can prevent the potential instability that is introduced if erroneous packets are dropped. This suggests that applying the same design principles of data networks to sensor networks can lead to performance degradation as sensor network applications are typically delay sensitive and therefore may require different design strategies. The work also highlights the importance of sharing information between the physical and application layers since extracting the information contained in all the packets would necessitate a cross layer design approach.