Advanced algorithms and architectures for MEMS inertial sensor platforms in orientation tracking and in fall detection applications

In this PhD thesis we carried out a research on the applications of inertial sensors to complex systems. The first part of the work deals with the development of an angular position estimation system. The goal was to obtain an integrated system within the sensor logic, in particular for a 6 DoF (Degrees of Freedom) gyro plus accelerometer sensor. Quaternions were chosen to represent the angular position for their flexibility and easy algebraic manipulation. A sensor fusion algorithm derived from the Kalman filter was developed. Several simplifications were done to reduce its computational complexity. In particular, a double stage Kalman filter was invented to improve the precision with the use of 9 DoF sensors data, with the addition of a magnetic compass. The first filter stage used the accelerometer data, while the second filter stage used the magnetic compass data to correct the angular position. The system was modeled with Matlab Simulink and the algorithms were validated with both synthetic and real acquired data. A Simulink bit true model of the system was developed, using fixed point arithmetic. The experiments shown that 20 bits are necessary to reach the target precision. It was chosen to design an ASIP (Application Specific Instruction Set Processor) to process the algorithm using a 20 bit ALU with minimal area occupation and power consumption. An FPGA prototype was realized using the SensorDynamics SD746 for the gyro plus accelerometer and the Honeywell HMC5843 for the magnetic compass. A real time 3D demo was realized. Moreover, the system was tested in the laboratory with repeatable rotations. A LabView program was designed to command the test machines. The resulting precision of the prototype was 1 deg for roll and pitch and 3 degs for the yaw. The second part of the work is about a fall detection system for elderly people. A first prototype of this system was under development in the University laboratories. It was based on an innovative algorithm that can recognize the ADL (Activities of Daily Living) from a fall to reduce the false alarms. When a fall is detected, the system launches an alarm with a GSM module. In this work, a wearable compact board with 5x8 cm2 dimensions was designed. The expected battery life is over one week.