A wireless sensor network monitors the environment at a macroscopic level. It comprises interconnected sensor nodes that sense a physical parameter of interest. The sensed data is first digitized and then transmitted over a multi-hop network to a base station or sink. The main challenge of data collection in a wireless sensor network is to keep the transmissions volume within the limited bandwidth of the sensor nodes. Several efficient in-network processing techniques are developed to reduce the network traffic. These techniques are based on the fact that the raw data sensed by the sensor nodes are often compressible.

This thesis focuses on efficient sampling techniques based on the theory of compressed sensing. Compressed sensing or compressive sampling is a lossy signal compression technique for robust and efficient data collection by applying a simple network coding mechanism. The main advantage of compressed sensing over the existing methods is that it guarantees balanced load on the sensor nodes in a normal operation of the wireless sensor network. This effectively avoids exhaustion of overloaded sensor nodes.

In this thesis, we extend compressed sensing techniques for wireless sensor networks in the following ways:

Reordering for better compressibility: The effectiveness of compressed sensing depends very much on the compressibility of the sensed data. The more compressible the raw data is, the less transmissions are needed to collect the data. We show that compressibility is affected by the order or permutation of the samples. The samples recorded by a wireless sensor network are conventionally indexed by the sensor node id's. This indexing does not necessarily lead to the most compressible order of the samples. We propose an algorithm that maps the physical indexing of the samples to a logical indexing that is more compressible. This mapping does not require reprogramming or relocating the sensor nodes.

Spatiotemporal compressive sampling: Several studies show that the data sensed by a wireless sensor networks are both spatially and temporally compressible. We propose an extension to compressive sampling that takes advantage of spatiotemporal compressibility of the sensed data.

Handling link and node failures: In practice, the sensor nodes often experience occasional failures or link disconnections. We propose a novel variation of compressed sensing that is more tolerant to network perturbations. Our method detects the sensor nodes that are facing node or link failures and isolate their sensor reading to preserve the accuracy of genuine data.

Data dissemination via network coding: We introduce a novel network coding method that disseminates a particular linear combination of the sensed data to all sensor nodes of a wireless sensor networks. We show that the originally sensed data are recoverable from any arbitrarily chosen sensor node that receives an error-free linear combination. Our dissemination allows accessing the globally sensed data from any sensor node by performing local data exchanges.