Título: E-AFTER: Estimating performance in dense IEEE 802.11 networks
Autor: Juan Lucas do Rosário Vieira
Resumo: Performance estimation can be used to improve IEEE 802.11 networks. Not only can it be used when designing the network to find a suitable number of APs to cover an area, but it can also be applied to several performance-maintaining tasks, such as load-balancing and interference control. MAPE is a framework that can provide reasonable throughput estimations in multi-hop IEEE 802.11 networks. However, dense, interference-prone scenarios have an inherently higher complexity due to the number of interactions between the transmitting nodes. Since the original proposal of MAPE does not consider the interference between concurrent transmissions, its accuracy tends to decrease in such scenarios. This work focuses on enhancing MAPE by proposing several changes that model extra network interactions to improve its accuracy in dense IEEE 802.11 networks while maintaining short execution times. The evaluation of this enhanced version, called E-AFTER, shows a significant increase in correlation between the estimates and the actual network performance and the reduction of estimation error compared to the original MAPE.
Título: Identifying People Using Channel State Information on Low-Cost Wi-Fi Networks
Autor: Julio César Huarachi Soto
Resumo: With the development of smart devices, human detection and location have become important tasks for various applications, including security, health care monitoring, entertainment, etc. Thus, the detection of the identity of a person for the mentioned applications emerges as a main purpose to be developed. In this work we propose the identification of people using low-cost devices in a Wi-Fi network. We start from the detection of the presence of a person in a room, to then proceed to the identification of the identity of the person involved in the experiment. The proposal includes the application of features of the Dynamic Time Warping (DTW) algorithm to compare the differences between empty rooms and full rooms, and then detect human presence in a room. We then use the amplitudes of the various CSI signals for the identification of a person. We train classification models and use various machine learning algorithms. The proposed architecture and approach achieve acceptable accuracy in identifying people in a room.
