Abstract: Based on the atmospheric pressure and temperature measured by a piezoresistive sensor and a temperature sensor, and their relations with altitude, a novel algorithm for sensor design is proposed.The pressure and temperature sensors are treated as two inputs of a multi-layered perceptron in an artificial neural network, and the altitude as the output.To get weights and offsets for each layer, the ANN is trained with standard data using the MOBP algorithm, which can be implemented on DSP.It can produce accurate altitude at any given pressure and temperature.Experiments indicate that the new sensor provides a higher accuracy than using the curve fitting technique, and have advantages in suppression of sensors' temperature shift and time shift.

Key words: altitude sensor, error compensation, MOBP algorithom, multilayered perceptron