Microcontroller Embedding of Support Vector Machines and its Application to Signal Processing for Smart Sensor Systems
Conference: edaWorkshop 08 - Workshop 2008 Electronic Design Automation (EDA)
05/06/2008 - 05/07/2008 at Hannover, Germany
Proceedings: edaWorkshop 08
Pages: 7Language: englishTyp: PDF
Personal VDE Members are entitled to a 10% discount on this title
Authors:
Mielenz, Holger; Graf, Alexander; Dölling, Rolando (Robert Bosch Group, Reutlingen, Germany)
Gerlach, Gerald (Technical University of Dresden, Department of Theoretical Physics, Dresden, Germany)
Rosenstiel, Wolfgang (University of Tübingen, Department of Computer Engineering, Tübingen, Germany)
Abstract:
Functional behavioral modeling is an important step for the simulation-based verification of complex systems. A data-based approach for functional model generation by support vector machines (SVM) has been shown to speed-up time-critical components of analog-mixed signal systems significantly. In this paper, we present an algorithm and computational results for discretized support vector machines on regression problems and motivate an embedding of the discretized models into an ATMEGA microcontroller. We briefly analyze the quantization effects on the performance of the SVM in regression problems in order to show its robustness in the feedforward phase. Since SVMs are not limited to any special physical disciplines the presented methodology introduces the embedding of SVMs into external hardware as a flexible method for system design. The advantage of our data-based modeling procedure is given by an efficient way to integrate a signal processing unit of a smart sensor product in order to become a totally integrated device. This is shown by the modeling of the signal processing part of an automotive pressure system and the embedding of the quantized model on an ATMEGA microcontroller.