A novel learning based solution for efficient data transport in heterogeneous wireless networks

There has been a spectacular growth in the use of wireless networks in recent times and consequently, adapting TCP to the wireless networks is a hot topic of current research. However, most of the existing works proposed for this problem have been designed for specific wireless networks, or they necessitate changes at either the receiver or the intermediate nodes, or at both, because of which their deployment becomes difficult. In this work, we propose a TCP variant which works over both multi-hop ad hoc wireless networks as well as single-hop (last-hop) wireless networks, like Wireless LANs, cellular, and satellite networks. We use a learning based method to dynamically change the congestion window size according to the network conditions. Our protocol does not rely on any explicit feedback from the network and requires only sender-side modifications. Through extensive simulations we show that our protocol achieves the desired goals of performance improvement in goodput, reduction in packet loss, and fairness to the competing flows. To the best of our knowledge, this is the first unified solution for both single-hop and multi-hop wireless networks.