Self-organization and data compression in wsn of extreme scalesapplication to environmental monitoring, climatology and bioengineering

Resumen

Wireless Sensor Networks (WSNs) aim for accurate data gathering and representation of one or multiple physical variables from the environment, by means of sensor reading and wireless data packets transmission to a Data Fusion Center (DFC). There is no comprehensive common set of requirements for all WSN, as they are application dependent. Moreover, due to specific node capabilities or energy consumption constraints several tradeoffs have to be considered during the design, and particularly, the price of the sensor nodes is a determining factor. The distinction between small and large scale WSNs does not only refers to the quantity of sensor nodes, but also establishes the main design challenges in each case. For example, the node organization is a key issue in large scale WSNs, where many inexpensive nodes have to properly work in a coordinated manner. Regarding the amount of data and the required accuracy, there is also a significant difference between small and large scale networks, and different in-network data processing techniques are essential, either to protect the data quality or to compress it in order to transmit less bytes. Finally, irreplaceable power supplies that are common in large scale WSNs, lead to energy consumption issues and the energy efficient operation becomes a relevant design requirement. The main goal of this thesis is the development of distributed processing algorithms for WSNs of extreme scales, for node organization and accurate data gathering. The term extreme scales is used in this work to denote both small and large scale, from network size point of view. To achieve this goal the research takes into account the requirements and design challenges in each case. The considered areas of application are the environmental monitoring, climatology, and bioengineering, fields of current relevance and with high economic and social impact. Due to the diverse nature of the WSNs and the selected applications, the following specific objectives are established. The first specific objective comprises the development of a self-organized clustering algorithm, named Second-Order Data-Coupled Clustering (SODCC). This algorithm uses the data measured by the nodes for clustering the network such that the autocorrelation matrix of each cluster is invertible. With this configuration, the in-network data processing algorithm implemented is optimally conditioned to compress the data. The experimental evaluation of SODCC uses actual air temperature data gathered by a large scale WSN and confirms the expected power-law behavior of the cluster size statistical distribution. Moreover, the combination of SODCC with the Compressive-Projections Principal Component Analysis (CPPCA) in-network processing algorithm achieves a perfect balance between the quality of the reconstructed data at the DFC and the energy consumption of the data gathering process. This objective is related to the environmental monitoring application. The second specific objective refers to the climatology application. Historical temperature data gathered by weather stations spread all over Europe and Western Asia are analyzed with the SODCC algorithm. The motivation is to identify the existing types of spatio-temporal data correlations based on their geographical extent. This enables the possibility to detect changes in the data correlation over the decades, and to link them with known climate issues. With this analysis it is possible to identify a change in the trend of air temperature for the stations in the Iberian Peninsula and Southern France. This change can be associated with a increased risk of extinction for plant species, and points to an evidence of a climate-change pattern. The third specific objective focuses on the analysis of the human gait within the bioengineering field, the first application of this thesis for small scale WSNs. The main contribution towards this goal is the analysis of the variability of the human gait, and the development of an ambulatory gait measurement system. The potential of this system is shown, by means of two proofs of concept that include actual experiments of diverse temporal duration. The signal variability analysis performed over the acceleration data shows that this system may be used as a diagnostic support system, as different issues as the gait symmetry can be analyzed. The fourth and last specific objective of this thesis is related to the small scale WSNs for Electrocardiogram (ECG) measurement. This bioengineering application requires a high signal quality, as the consequences of a possible incorrect diagnostic issued by a physician would be severe. Therefore, the Compressed Sensing (CS) and Wavelet Transform (WT) techniques with a suitable parameter configuration are used to increase the diagnostic capabilities of the measured ECGs. The results show that quality metrics based on central tendency statistics are not appropriate to evaluate this kind of systems due to their high variance, even among healthy subjects.