Low-complexity Data Aggregation Methods for Energy-constrained IoT Networks using Compressed Sensing Framework

Amarlingam, M and Rajalakshmi, P (2019) Low-complexity Data Aggregation Methods for Energy-constrained IoT Networks using Compressed Sensing Framework. PhD thesis, Indian institute of technology Hyderabad.

Full text not available from this repository. (Request a copy)

Abstract

Internet of things (IoT) is an ecosystem consisting of networks of sensors, actuators, and smart objects whose purpose is to interconnect all things, including everyday and industrial objects, in such a way to make them intelligent, programmable, and more capable of interacting with humans. In an IoT ecosystem, the wireless sensor network (WSN) takes an active part in the integration of physical world information. WSN comprises of a huge number of tiny sensor devices which are deployed in the sensing area of interest to measure and collect the data. IoT enables access to a wide variety of WSNs for providing a plethora of digital services. In most of the IoT applications such as remote sensing and environmental monitoring, seamless data aggregation from the sensor nodes is the fundamental requirement for data processing (data analytics) to facilitate the user with a useful interface and ubiquitous access to the network data. Sensor nodes employed in typical monitoring applications of IoT are constrained by available onboard resources such as energy, memory, computational capability. The existing data aggregation algorithms have proven that compressed sensing (CS) is promising for energy efficient data aggregation. However, these methods compromise either on on-node computational complexity or recovery fidelity. In this thesis, we propose data aggregation algorithms to offer better performance with respect to energy consumption, recovery fidelity, and on-node computational complexity jointly. In particular we propose two low-complex data aggregation algorithms which make use of clustering and CS for aggregating the data. The first method is named as light-weight compressed data aggregation (LWCDA) algorithm which exploits random non-overlapping clustering and compressed sensing for low-complex and energy efficient data aggregation. The second algorithm is called as low-complexity compressed data aggregation (LCCDA) algorithm which utilizes constrained overlapped clustering and CS for data aggregation to provide a better tradeoff between on-node computational complexity, energy consumption and recovery fidelity. We also propose data aggregation approaches for two real time scenarios: 1) The scenario wherein sensor nodes measure the data from multiple and different sensors (i.e., different type of sensors). 2) The scenario wherein data sparsity of sensors vary with the time. The random non-overlapping clustering offers two important advantages: 1) energy efficiency, as each node has to send its measurement only to its cluster head, 2) highly sparse measurement matrix, which leads to a practically implementable framework with low complexity. We analyze the properties of our measurement matrix using restricted isometry property, the associated coherence and phase transition. Through extensive simulations on practical data, we show that the measurevi ment matrix can reconstruct data with high fidelity. Further, we demonstrate that the LWCDA algorithm reduces energy consumption significantly when compared with state of art approaches, thereby implying the enhancement of the network lifetime. While in case of LCCDA, we show that the measurement matrix constructed from constrained overlapped clustering possesses the restricted isometry property (RIP) that guarantees the recovery of the aggregated data. We make use of weight adjacency matrix based graph Laplacian eigenbasis to find sparse representation for the measured data from a randomly deployed network that enables the high recovery fidelity for aggregated data at the sink node. Through numerical experiments, we demonstrate that the proposed LCCDA method is capable of delivering the data to the sink with high recovery fidelity while achieving significant energy savings. The existing approaches consider single sensor per node for data aggregation in CS-aided IoT network, specifically for the IoT network that uses CS for data aggregation. Most of the monitoring applications employ sensor nodes that require to measure the data from multiple and different sensors. The data aggregation methods proposed in the literature which consider one sensor per node are not optimum with respect to energy consumption for usage in scenario wherein multiple sensors are present. To address this issue, we propose an algorithm for aggregating the data from a network where each sensor node measures data from multiple and different sensors using dictionary learning and CS. We construct a dictionary from multiplexed data for providing sparse representation of the data from different sensors that avoid usage of different bases for different sensors. This provides significant energy savings for data aggregation in CS-aided IoT networks as learned dictionary can provide the sparse representation of the data from different sensors. In CS-aided IoT networks, the size of the measurement matrix used for data aggregation depends on the predefined sparsity value which is usually computed using testing data sets. In the literature most of the CS-aided data aggregation algorithms optimize the network for energy efficiency and high fidelity recovery by making use of the structural properties of the measurement matrix. Every change in sparsity leads to change in the size of the measurement matrix and reconfiguration of the network. In the existing approaches, for every change in sparsity the sink node broadcasts the sparsity variation information to the network and the network reconfigures accordingly. Sparsity information sharing and network reconfiguration requires significant number of transmissions over the network which makes data aggregation energy inefficient. To address this issue, we propose an optimal method for CS-aided IoT network where we use dictionary as a sparsifier that is learnt with the data that has sparsity variations with respect to time. The proposed method achieves minimum reconstruction error by accounting for changes in the data sparsity while delivering the data to the vii sink with low energy consumption thereby improving network lifetime. In addition, the placement of relay nodes also play a vital role in the network lifetime of the relay based IoT networks. Optimal deployment of relay nodes can improve the network lifetime significantly. The existing approaches are too theoretical to use in a real world setting. They assume ideal scenarios to estimate the location of relay nodes which are impractical. Accurate placement of the nodes at these estimated locations is difficult, often due to poor location information. To address this issue, we propose a mobile phone based deployment adviser tool which is robust as well as practically implementable. The tool advises a layman (deployer) to create an optimized network by placing the nodes according to application requirements. Also we propose an algorithm which helps in distributing the power consumption among the nodes in the network, thus, increasing the network lifetime.

[error in script]

IITH Creators: