In wireless networks, data packets are lost due to channel interference and fading rather than network congestion. Thus, TCP designed for wired networks, cannot achieve a satisfying performance in wireless networks. Utilizing a cross-layer mechanism to identify the course of packet loss, we have proved that a mobile-host-centric transport protocol (MCP) can achieve higher throughput than TCP Reno and New Reno in static wireless environment. In this paper, we extend the cross-layer feedback mechanism and congestion control algorithm based on previous work. Furthermore, we implement the extended MCP+ in NS-2 simulator and evaluate its performance in the mobile multi-hop WLAN scenarios. The simulation results show that the proposed protocol improves the end-to-end throughput more effectively than TCP Reno and New Reno.
The cumulative sum (CUSUM) algorithm is proposed to detect the selfish behavior of a node in a wireless ad hoc network. By tracing the statistics characteristic of the backoff time between successful transmissions, a wireless node can distinguish if there is a selfish behavior in the wireless network. The detection efficiency is validated using a Qualnet simulator. An IEEE 802.11 wireless ad hoc network with 20 senders and 20 receivers spreading out randomly in a given area is evaluated. The well-behaved senders use minimum contention window size of 32 and maximum contention window size of 1 024, and the selfish nodes are assumed not to use the binary exponential strategy for which the contention window sizes are both fixed as 16. The transmission radius of all nodes is 250 m. Two scenarios are investigated: a single-hop network with nodes spreading out in 100 m×100 m, and all the nodes are in the range of each other; and a multi-hop network with nodes spreading out in 1 000 m×1 000 m . The node can monitor the backoff time from all the other nodes and run the detection algorithms over those samples. It is noted that the threshold can significantly affect the detection time and the detection accuracy. For a given threshold of 0.3 s, the false alarm rates and the missed alarm rates are less than 5%. The detection delay is less than 1.0 s. The simulation results show that the algorithm has short detection time and high detection accuracy.
