Show simple item record

dc.contributor.authorQuan, Yili
dc.date.accessioned2016-03-21T20:43:52Z
dc.date.available2016-03-21T20:43:52Z
dc.date.issued2008
dc.identifier.isbn9780549765592
dc.identifier.other304405456
dc.identifier.urihttp://hdl.handle.net/10477/43561
dc.description.abstractIn this dissertation a decision making model called the Lens Model, is implemented with hardware-software hybrid-mode design. The system combines a software implementation of an Adaptive Linear Neuron (ADALINE) network to perform a part of the Lens Model computation, and hardware in the form of a specialized analog integrated circuit to complete the calculations. The majority of the research efforts have concentrated on the development of correlator chip, which calculates the correlation coefficient (consistence) between the predicted judgment and actual judgment. We propose three novel circuits for this: a Difference Squaring circuit, and Square Root circuit, and a Ratio circuit. The novel Difference Squaring circuit improves a previous design wherein the DC offset is cancelled, or removed. The Square Root circuit and the Ratio circuit are implemented with transistors working in the subthreshold mode with current exponential dependence on gate voltage. This learning chip has been fabricated using the CMOS 1.5μm AMI ABN process available through the MOSIS integrated circuit fabrication service. The ADALINE (Adaptive Linear Neuron) was simulated using Matlab with a given set of profiles to produce a best-fitting network with least error between the estimated judgment and actual judgment. Different learning rates and training epochs were applied to monitor the effect of these factors. Total error between the estimated judgment and actual judgment is compared for these cases. The estimated judgment and actual judgment are stored and sent to the correlator chip via Labview to monitor the consistency between these two parameters after epochs of training. We present the experimental results obtained for the correlator block and the whole system. Test results show very good agreement between theory and measurements and thus indicate a successful design. The system can be used to determine the factors in individual performance in complex judgment tasks. This is the first time that the Lens Model was implemented in a hybrid software and hardware system.
dc.languageEnglish
dc.subjectApplied sciences
dc.subjectLens Model
dc.subjectAdaptive linear neuron network
dc.subjectCorrelator chips
dc.subjectSoftware
dc.subjectHardware
dc.titleHardware-software implementation of Lens Model
dc.typeDissertation/Thesis


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record