Abstract
The main objective for which body centered systems are being developed is to obtain and process biological signals in order to monitor and, in some cases even treat, a physical condition, either a disease or the athletic performance in the case of sports. Since the core of body-centered systems is sensing and processing, signal processing algorithms play a central role in the system’s functioning. This thesis is focussed on those real-time algorithms that are needed to obtain the relevant information from the sensed signals. In the initial part, the types of sensors and algorithms are reviewed, after that, the thesis deals with two different applications and the related real-time signal processing algorithms are designed and implemented. 
The first study case (Chapter 3) is glucose monitoring in diabetes patients. The objective was to detect therapeutically wrong measurements of Medtronic’s Minimed CGMS using learning algorithms for pattern classification. The methodology used was the following: Patients were monitored using CGMS and simultaneously blood samples were taken in a clinical study. Gaussian SVM classifiers were tuned to detect wrong glucose estimations making use of monitor’s electrical signal, the CGMS glucose estimation and the real blood-glucose measure obtained from the blood samples. The results showed that the classifiers were indeed able to learn the data structure and an overall good detection of wrong measures was obtained in spite of the somewhat low sensitivity of the detector. The classifiers were able to detect the time intervals where the monitor’s glucose profile shouldn’t be trusted because of wrong measurements. This was illustrated with the detection of hypoglycaemic episodes missed by the CGMS. From this analysis it was concluded that detection of therapeutically wrong measurements given by the continuous glucose monitor Minimed CGMS is feasible through the use of SVM classifiers. For all patients, missed hypoglycaemic states were detected, as well as other therapeutically wrong measurements. The presence of False Positives did not alter the conclusions drawn out from the analysis of time profiles. This tool could thus support the clinician in the interpretation of continuous glucose monitor readings. 
The second application of body centered systems, included in Chapter 4 is neural signal monitoring. Recent medical advances have demonstrated the benefits that such monitoring can bring to medicine and even to other areas as entertainment. That is why, nowadays, there are many research groups dedicated to develop wireless implantable brain monitors. In this work neural spike detection classification and compression algorithms have been implemented and evaluated together with wireless transmission techniques. Such combination will enable the implementation of the wireless brain monitors. A new method for adaptive threshold spike detection was applied that successfully adapts to different input SNRs eliminating the need for manual threshold setting. For the classification algorithm, PCA pattern recognition techniques were used and a performance of 92% of correctly classified spikes was accomplished. Detection and classification were used together with a compression and resource management algorithm for efficient wireless transmission of neural signals. The frame-based algorithm was capable of adapting the compression of the 60 input channels according to: the neural activity present, the priority set to each channel and the bandwidth available at each processing frame. As a result, signals were compressed and multiplexed in a single transmission frame that fits in the available transmission bandwidth. The reconstruction algorithm at the receiving side was able to demultiplex and decode the received frame to reconstruct the neural spiking patterns. The conjunction of detection, sorting and compression algorithms produced a scheme for neural monitoring system that self-adapts to the signal conditions (adaptive threshold detector) and to the transmission bandwidth.
Finally, although the main topic of the thesis is signal processing, a chapter (Chapter 5) has been dedicated to wireless transmission technologies and more precisely to on-body UWB transmission. UWB was selected in this thesis as the most promising transmission technology for body-centered systems because of the combination of low-power, short-range and high data rates that characterize such technology. There are some additional considerations to be regarded. UWB allows only short distance communications with these high transmission rates, which is perfectly assumable for body area networks but therefore it raises the need of a bridge between the close body field and the remote monitoring stations. In this work, the objective was to evaluate UWB for the particular application of real-time neural signal monitoring. As for the methodology, a channel measuring campaign was designed and performed in order to characterize the head-to-body channel. From such measures, model parameters were extracted. Additionally, and for performance evaluation, neural signals were transmitted through a UWB evaluation kit and the spiking characteristics of the received signals were compared to those of the transmitted signals for different experimental set-ups. The study concluded that for real-time neural signal monitoring, UWB seems to offer best transmission conditions in a near-body environment up to 2m. It allows high-fidelity signal transmission at extremely high data rates with low power consumption.