A CMOS Mixed Mode Non-Linear Processing Unit for Adaptive Sensor Conditioning in Portable Smart Systems

详细信息    查看全文

• 作者：A. Martí ; nez-Nieto^a ; M.T. Sanz-Pascual^a ; A. Má ; rquez^b ; J. Pé ; rez-Bailó ; n^b ; B. Calvo^b ; N. Medrano^b ; ^{nmedrano@unizar.es}
• 关键词：artificial neural networks ; CMOS ; sensor calibration
• 刊名：Procedia Engineering
• 出版年：2016
• 出版时间：2016
• 年：2016
• 卷：168
• 期：Complete
• 页码：1689-1692
• 全文大小：710 K
• 卷排序：168

文摘

This paper presents the architecture of a novel non-linear digitally programmable analog unit for sensor output conditioning in battery-operated smart systems. Designed in an 180 nm 1.8 V standard CMOS technology, by properly setting the 6-bit registers in the arithmetic unit, the voltage inputs are weighted before being processed by a non-linear circuit. Thus, a processing system consisting of a set of these devices suitably tuned and interconnected can be applied to condition a non-linear sensor, improving its behavior both in linearity and operating range, while reducing the effects of cross sensitivity. The robustness of the digital weight tuning is tested simulating a chip-on-the-loop training using a Levenberg-Marquardt-based algorithm. Electric simulations of the proposed unit and the results of its application in a complete neural network-based processing system to improve the linear operating range of a thermistor are presented.