EasiPLED: Discriminating the causes of packet losses and errors in indoor WSNs

Abstract

It is well known that there are two kinds of causes, namely channel-errors and collisions, which lead to high probability of packet losses and errors in wireless networks. The ability of discriminating the above two causes provides many opportunities for implementing high efficient networking protocols in wireless sensor networks (WSNs). This paper presents EasiPLED, a discriminator that can accurately and timely predict these two causes. EasiPLED has three salient features. First, it investigates F-BER patterns and statistic characteristics of RSSI in different indoor environments through extensive experimental studies. F-BER is the Frame-level Bit Error Rate measured at the receiver side by a coarse-grained method without incurring any overhead. An adaptive RSSI estimator based on error-based filter is proposed to mitigate effects of noise on RSSI readings for successfully received packets. Second, EasiPLED designs an off-line dominant-factor classifier using machine learning method. The classifier takes a combination of F-BER and RSSI features as input and outputs the probability of dominant causes of failed transmissions. Finally, it presents a lightweight on-line discriminator which diagnoses the root cause of a packet loss or error when it occurs at the receiver side. Experimental results show that EasiPLED achieves an accuracy by up to 95.4%. We evaluate the effectiveness of EasiPLED by applying it to link-layer retransmission scheme, which yields a reduction of single-hop transmission delay by up to 47%, and provides high packet delivery ratios as compared to the existing retransmission methods.