Machine Learning
Minimising the effect of WiFi interference in 802.15.4 wireless sensor networks
International Journal of Sensor Networks
Experimental Study of the Impact of WLAN Interference on IEEE 802.15.4 Body Area Networks
EWSN '09 Proceedings of the 6th European Conference on Wireless Sensor Networks
RFDump: an architecture for monitoring the wireless ether
Proceedings of the 5th international conference on Emerging networking experiments and technologies
ICC'09 Proceedings of the 2009 IEEE international conference on Communications
Sensei-uu: a relocatable sensor network testbed
Proceedings of the fifth ACM international workshop on Wireless network testbeds, experimental evaluation and characterization
Surviving wi-fi interference in low power ZigBee networks
Proceedings of the 8th ACM Conference on Embedded Networked Sensor Systems
Clearing the RF smog: making 802.11n robust to cross-technology interference
Proceedings of the ACM SIGCOMM 2011 conference
Airshark: detecting non-WiFi RF devices using commodity WiFi hardware
Proceedings of the 2011 ACM SIGCOMM conference on Internet measurement conference
Mitigating the effects of RF interference through RSSI-Based error recovery
EWSN'10 Proceedings of the 7th European conference on Wireless Sensor Networks
Hi-index | 0.00 |
With a rapidly increasing number of devices sharing access to the 2.4 GHz ISM band, interference becomes a serious problem for 802.15.4-based, low-power sensor networks. Consequently, interference mitigation strategies are becoming commonplace. In this paper, we consider the step that precedes interference mitigation: interference detection. We have performed extensive measurements to characterize how different types of interferers affect individual 802.15.4 packets. From these measurements, we define a set of features which we use to train a neural network to classify the source of interference of a corrupted packet. Our approach is sufficiently lightweight for online use in a resource-constrained sensor network. It does not require additional hardware, nor does it use active spectrum sensing or probing packets. Instead, all information about interferers is gathered from inspecting corrupted packets that are received during the sensor network's regular operation. Even without considering a history of earlier packets, our approach reaches a mean classification accuracy of 79.8%, with per interferer accuracies of 64.9% for WiFi, 82.6% for Bluetooth, 72.1% for microwave ovens, and 99.6% for packets that are corrupted due to insufficient signal strength.