Abstract

Wireless communication is inherently vulnerable in nature and packets can be corrupted due to various reasons. The network performance can potentially be improved if one is able to identify the reasons for packet corruption and react to them accordingly. However, none of the current wireless protocols do so. In this paper, we design a novel scheme to diagnose packet corruption in OFDM wireless networks by statistically analyzing certain available physical layer information, based on the observation that different causes of corruption result in different per-symbol-SINR patterns within a packet. Our approach introduces no additional traffic overhead. By running experiments on a GNU radio test bed in an electromagnetic anechoic chamber, we study and demonstrate that one can separate three reasons for packet corruption: weak signal, interference from transmitters using the same band, and interference from transmitters using an adjacent band. We study three statistical classification methods and compare their accuracy.

Citations