Functional model-based design of embedded systems with UniTi

Abstract

Advancing the field of embedded systems requires a rigorous approach to their design. This is because embedded systems are complex, diverse and challenging. Although many tools exist, none support the following four essential features: (i) the modelling of multiple domains, (ii) accurate inclusion of time, (iii) mathematical definitions, and (iv) model transformations. In addition, such a tool must underlie a sound design flow that adequately supports the complexity of designing embedded systems. In this thesis we propose a design flow and a modelling and simulation framework called UniTi that manages complexity in a top-down fashion; a problem is split up into sub-problems that are solved individually and then combined. This design flow and framework is based on model-based design, i.e. a single reference model is iteratively and incrementally developed and refined during the design process. Our approach is a functional approach, not only because it is practical and useful, but also because it has a mathematical basis supported by a functional language, i.e. computations are considered as evaluations of mathematical functions. In this work we specialise the design for the application domain of beamforming applications, for which we propose a generic platform. Two adaptive algorithms for tracking are developed in the context of this platform. A tiled reconfigurable architecture is used, as the tiles provide scalability and reconfigurability provides flexibility. The environment and analogue hardware are represented in the continuous time (CT) domain, while digital hardware is represented in the discrete time (DT) domain and software in the dataflow (DF) domain. We formally define the CT, DT, and DF domains for UniTi. It also supports exact time delays in the CT domain by representing signals as functions of time. Model components, represented as signal transformations, are composed using function composition instead of value-passing, with unified sequential, parallel and feedback composition by re-defining the dataflow model to match with CT and DT components and signals. As a consequence, mixed-domain models are executable for simulation. Finally, UniTi provides support for model transformations. The result of this work is a functional model-based design approach for designing, modelling, and simulation of embedded systems.