Abstract:
Most of the existing received signal strength indicator (RSSI) based wireless video capsule endoscope (VCE) localization algorithms employ trilateration or triangulation positioning techniques which require accurate estimation of channel parameters and distance or angle and their performance decreases significantly due to the estimation errors. Firstly, the estimation errors in distance or angle arise due to the highly random path loss caused by severe multipath propagation and heavy shadow fading effects of RF propagation through nonhomogeneous medium of human body. On the other hand, due to the complex environment of human body, it is quite difficult to estimate the channel parameters accurately. Hence, a non-parametric path loss based localization approach is preferable for VCE. Weighted centroid-based localization (WCL) is such an approach which computes the position of the target by applying weighted average of the positions of sensor receivers. In this dissertation, we propose path loss based localization approach where a three dimensional array of sensor receivers is used to localize the capsule transmitter while it travels through the small intestine. We model the path loss for medical implant communication services (MICS) and ultra wideband (UWB) RF propagation channel using the statistics of deep-tissue implant to body surface scenario. To address the randomness issue of path loss, we propose both parametric and non-parametric approaches of path loss estimation. As parametric approach, we propose linear least square based local channel parameter dependent path loss estimation. As non-parametric approach of path loss estimation, we propose moving averaging, Savitzky-Golay filtering, local weighted regression, local weighted averaging, maximum likelihood (ML) estimation and MIMO diversity schemes to simplify the localization methods by avoiding any prior knowledge of channel parameters.
Next, we propose WCL based localization algorithms using weighted average of the reference positions of the sensor receivers to find the position of the capsule considering human body channel characteristics. We apply different variants of path loss namely estimated distance, randomly scattered path loss, estimated path loss and degree based path loss to determine the weight of the sensors position. Additionally, a calibration stage is developed to determine the location more accurately. We propose heuristic and sub-optimal methods of finding the degree and calibration coefficient. We also compute the optimal values of degree and calibration coefficient analytically to set the benchmark of accuracy. In each of the above cases, we develop the localization algorithms and analyze the impact of the number of neighboring points, number of packets, number of sensor receivers, sensor network topology, network dimension and the channel parameter errors on the accuracy of localization.
Finally, we develop a 3D visualization simulation platform to visualize the results of VCE localization using MATLAB considering real characteristics of human body channel. The simulation platform includes RF propagation channel statistics of deep-tissue implant to body surface scenario, an intestine model with 3D position maps, a number of sensor receivers placed at reference positions and a capsule transmitter. We also verify the accuracy of localization using different performance metrics as localization error (LE), average localization error (ALE), root mean square error (RMSE), standard deviation (STD), normalized RMSE (NRMSE) and P-value. To validate the performance, the proposed algorithms are compared to the benchmark and also to other works in the existing literature. It is observed that our proposed localization algorithms approach the proposed benchmark and can localize the capsule with significantly improved accuracy as compared to the other.