Wireless Sensor Network Localization with Neural Networks Student name: Amit Geron Supervisors: Dr. Zvi Lotker Eng. Yaniv Weizman
Outline Introduction Project Goals Implementation Experiments
Outline Introduction Project Goals Implementation Experiments
Introduction What is a WSN? What is Localization and why is it an important issue? What are Neural Networks?
Introduction - WSN Unique characteristics of a WSN include: Limited power they can harvest or store Ability to withstand harsh environmental conditions Ability to cope with node failures Mobility of nodes Dynamic network topology Communication failures Heterogeneity of nodes Large scale of deployment Unattended operation
Introduction - WSN Topology
Introduction - Localization Localization- Location Estimation. Many methods, mainly distinguished according to either one-hop / multi-hop and centralized / distributed estimation algorithm. Main disadvantages- high complexity, long headers, many messages between units.
Introduction – Neural Networks Neural Network- A graph with weighted edges and linear and non-linear functions at nodes. 3 main usages: Function Approximation. Classification. Data Processing. Training set has a major effect over performance.
Outline Introduction Project Goals Implementation Experiments
Project Goals Neural networks computation & learning algorithms concept. High and Low level design for algorithmic implementation under sensors networks characteristics. Algorithm Implementation within real sensor network framework.
Outline Introduction Project Goals Implementation Experiments
Implementation The tools and methods that have been used to implement the project: Motes and Sensors – Tmote Sky. Communication method – Local. Measurement method – RSSI. IDE – Sentilla. Neural Network: encog-java package by Jeff Heaton. Function Approximation. FeedForward. BackPropagation. Multiple hidden layers.
Implementation – Messages structure Tmote Tmote Tmote Tmote Tmote Sampled Tmote Gateway Tmote Tmote Neural Network
Implementation – Planned Vs. Implemented S R G
Outline Introduction Project Goals Implementation Experiments
Experiments Performance and error rate will be examined under the variation of: Training set Characteristics: Number of samples. Samples’ Locations. WSN Characteristics: Number of motes. Area size. NN Characteristics: Number and size of hidden layers. Classification Vs. Approximation.
Experiments - Example
Experiments - Example
Experiments - Example