Multi-channel RSS measurements on TWIST

Internet of Things (IoT) systems, including Wireless Sensor Networks (WSNs), are getting integrated into virtually all aspects of life faster than before. These systems are used from agricultural monitoring and actuation to manufacturing plants. Such wireless networks play a critical role in home and health applications as well as security and asset tracking. The knowledge of physical position of the nodes is important for many applications of WSNs, but this information is often not readily available. In the past, a plethora of different solutions has been proposed that focus on recovering the positions of the sensor nodes, often at the cost of high complexity, preconfiguration, training or limited accuracy. The precision and computational complexity of such "positioning" algorithms is still a big issue. However, there are cases where the objects are placed in one of a few possible predetermined positions, especially indoors. In those cases, the set of potential locations of the objects is limited and computing the relative positions of those objects in relation to each other might be sufficient to determine their real positions and a precise location in the x,y,z coordinates is not necessary. This thesis focuses on determining the relative positions of nodes in a WSN by utilizing only readily available Received Signal Strength (RSS) information that is provided by the radio chips they are equipped with. Contrary to common belief, the "closeness" information can be extracted with high confidence among one sender and multiple receiver nodes. An RSS sampling technique for extracting closeness information is introduced as the initial step of position discovery. This technique utilizes the frequency diversity features of the radio modules. Combining the frequency diversity with statistical reasoning allowed us to demonstrate how RSS information can be used for detecting closer nodes to a transmitting node with high confidence, where this information cannot be extracted by connectivity information. The closeness information can then be utilized to discover node positions by having only one or few nodes at known places on a grid-like setting. For relative node position discovery, two types of grid settings were considered: one-dimensional and two-dimensional. The only prerequisite for introduced position discovery algorithms is the knowledge of one reference node at one head of a one-dimensional grid and the knowledge of two reference nodes at two corners of a two-dimensional grid. In this study, a successful position discovery is defined as mapping all nodes to the cells of the grid perfectly, otherwise a result is considered unsuccessful even when only two nodes are mapped to swapped cells. The proposed techniques result in up to 100 % successful position discovery in the repeated real-world experiments and in the simulations. For self-detecting whether a run of position discovery was successful, reliability analyses are developed. Using these analyses each result is mapped into one of the high, medium or low reliability categories. Over 99 % of the results that are assigned to high reliability category were perfectly correct and the unsuccessful computations could be assigned to the remaining categories. It has been discussed that the suggested frame of procedures has significant advantages over other systems that are commonly used for indoor position discovery, such as accuracy, time-complexity and independence from an infrastructure. The results are compared to Multi-Dimensional Scaling (MDS) and fingerprinting-based systems