Development of a Scalable High Performance Reconfigurable Real-Time Signal Processing Platform for Dynamic Data-Driven Neural Simulations and Modeling

Montana State University, Bozeman MT

Investigators

Abstract

This proposal provides support for development of a high performance, reconfigurable signal-processing platform that will permit real-time analysis of large-scale multi-channel neurophysiologic data and subsequent use in simulation and modeling. The computational architecture will be a distributed, real-time system of modular design consisting of computational nodes connected in a three-dimensional mesh. A computational node will include a floating-point digital signal processor (DSP), a field programmable gate array (FPGA), and local memory. This system to be used is reconfigurable, so that algorithms can be directly implemented in the hardware. The FPGAs can act as communication processors, allowing significant bandwidth for communication between computational nodes. Configuring the system as a three-dimensional mesh will allow the system to scale to any number of computational nodes required to process an arbitrary number of real-time I/O data streams. The platform will be developed using the analysis of neural signal processing in a simple nervous system, that of the cricket. Specifically, the platform will be developed to allow investigation of the cooperative neural encoding schemes used to transmit information about air currents within the cricket's nervous system. The platform will enable real-time decoding of neural information, and will thus enable experimental perturbation of the encoded information while the neural signals are in transit between multiple peripheral sensors and the central processing ganglia. If the platform is successfully developed, an unprecedented degree of interactive control in the analysis of neural function will result. This could lead to major insights into the biological basis of neural computation and a new paradigm in experimental and computational neuroscience, one where experimental and theoretical neuroscientists can work together to test hypotheses of neural function in vivo. This proposal provides support for development of a high performance, reconfigurable signal-processing platform that will permit real-time analysis of large-scale multi-channel neurophysiologic data and subsequent use in simulation and modeling. The computational architecture will be a distributed, real-time system of modular design consisting of computational nodes connected in a three-dimensional mesh. A computational node will include a floating-point digital signal processor (DSP), a field programmable gate array (FPGA), and local memory. This system to be used is reconfigurable, so that algorithms can be directly implemented in the hardware. The FPGAs can act as communication processors, allowing significant bandwidth for communication between computational nodes. Configuring the system as a three-dimensional mesh will allow the system to scale to any number of computational nodes required to process an arbitrary number of real-time I/O data streams. The platform will be developed using the analysis of neural signal processing in a simple nervous system, that of the cricket. Specifically, the platform will be developed to allow investigation of the cooperative neural encoding schemes used to transmit information about air currents within the cricket's nervous system. The platform will enable real-time decoding of neural information, and will thus enable experimental perturbation of the encoded information while the neural signals are in transit between multiple peripheral sensors and the central processing ganglia. If the platform is successfully developed, an unprecedented degree of interactive control in the analysis of neural function will result. This could lead to major insights into the biological basis of neural computation and a new paradigm in experimental and computational neuroscience, one where experimental and theoretical neuroscientists can work together to test hypotheses of neural function in vivo.