Circuits, Systems, and Neural Networks
Fathi M. Salem | email@example.com | www.egr.msu.edu/annweb
Research in the Circuits, Systems, and Neural Networks (CSANN) Laboratory includes (deep) neural networks and adaptive predictive systems, adaptive separation and recovery of source signals in communication Systems, and integrated sensing and actuation systems for cell and protein manipulation.
Computational Deep Neural Networks, Optimal Learning Systems
This encompasses theoretical aspects of network modeling and non-convex constrained info-theoretic optimization, with subsequent applications to multitude of signals and data in communication and networked systems. Examples include detection, extraction, and classification of source information signals from streams of measurements or data in CDMA communication systems.
As an engineering application of one theoretical framework, we have developed an autonomous (or unsupervised) learning and adaptive approaches for communication systems at the receiver (e.g., a cellphone). The goal is to recover multiple original source symbols in realistic (practical) environments that may include convolution, transients, and even nonlinearity. That is, optimal recovery of (unknown) message symbols sent from a transmitter through unknown or imprecise channel or environmental characteristics.
Integrated Electronic Circuits (ICs) and Systems
The design of integrated electronic circuits (ICs) and systems implements specialized models of adaptive neural networks for information processing. We adopt integrating processing with sensing and actuation. This project demonstrates the integration of processing, precise sensing, and actuation of magnetic beads with radii in the order of a 1-4 micro-meters. Superparamagnetic beads are increasingly used in biomedical assays to manipulate, transport, and maneuver biomaterials (e.g., cells, proteins, etc.).
This effort focuses on low-cost integrated system designed in bulk CMOS—in order to leverage computational powers, precision and speed. The integrated system is to manipulate, separate, and steer biomaterial tagged with magnetic beads.
The integrated systems include insulated open cavity coil-arrays suitable for sensing and/or generating directed magnetic (fields and) forces for single biomaterial, or collaborative multi-bead manipulation, using pseudo-parallel executions.
Several lab experiments have been conducted to validate and metrically quantify this approach. The results have demonstrated the manipulation and steering effect of the generated magnetic forces on magnetized micro-beads. (see http://researchgroups.msu.edu/csann/media)
The framework has been successfully applied to Blind Multi User Detection (BMUD) in the CDMA (Code Division Multiple Access) wireless communication networks. Blind Multi User Detection (BMUD) is the process of simultaneously estimating multiple symbol sequences associated with multiple users in the downlink of a (CDMA) communication system using only the received data.