Design and implementation of real-time signal processing applications on heterogeneous multiprocessor arrays

Processing structures based on arrays of computational elements form an important class of architectures, which includes field programmable gate arrays (FPGAs), systolic arrays, and various forms of multicore processors. A wide variety of design methods and tools have been targeted to regular processing arrays involving homogeneous processing elements. In this paper, we introduce the concept of field programmable X arrays (FPXAs) as an abstract model for design and implementation of heterogeneous multiprocessor arrays for signal processing systems. FPXAs are abstract structures that can be targeted for implementation on application-specific integrated circuits, FPGAs, or other kinds of reconfigurable processors. FPXAs can also be mapped onto multicore processors for flexible emulation. We discuss the use of dataflow models as an integrated application representation and intermediate representation for efficient specification and mapping of signal processing systems on FPXAs. We demonstrate our proposed models and techniques with a case study involving the embedding of an application-specific FPXA system on an off-the-shelf FPGA device.