A Flexible, Low-Power, Programmable Unsupervised Neural Network Based on Microcontrollers for Medical Applications

We present an implementation and laboratory tests of a winner takes all (WTA) artificial neural network (NN) on two microcontrollers (μC) with the ARM Cortex M3 and the AVR cores. The prospective application of this device is in wireless body sensor network (WBSN) in an on-line analysis of electrocardiograph (ECG) and electromyograph (EMG) biomedical signals. The proposed device will be used as a base station in the WBSN, acquiring and analysing the signals from the sensors placed on the human body. The proposed system is equiped with an analog-todigital converter (ADC), and allows for multi-channel acquisition of analog signals, preprocessing (filtering) and further analysis.