Markovian Models for Medical Signals on Wireless Sensor Networks

Medical monitoring is a very important wireless sensor network (WSN) application, because collecting vital information (such as heart rate, temperature, blood oxygen saturation level, electrocardiogram (ECG) and electroencephalogram (EEG) signals) using wireless sensors and making it available on time can improve the safety of the patient's care and save lives. Medical support networking has very stringent performance requirements and a complete WSN design must rely on accurate traffic models: being able to efficiently model data generated by each WSN node can be fundamental for correctly simulating network traffic, network congestion, interference between nodes, energy expended by each node and selecting the most appropriate routing protocols. This paper proposes two different Markovian models, the Markov modulated Poisson process (MMPP) and the Markov modulated deterministic process (MMDP), to model the signal generated by a single WSN node monitoring electroencephalogram (EEG) data. Since some medical WSN nodes will employ source coding to reduce the amount of data that must be transmitted, thus conserving energy, the first model is found to be appropriate for modeling the EEG signal while the second one is more suitable for fitting the power that is used to transmit the encoded EEG signal. The accuracy of the models was tested using empirical data from public domain medical signal databases and the results obtained show that the proposed models are able to accurately fit the probability density function and the autocovariance function of EEG signals and the power that is used to transmit them.