The 2k2: a modular computational toolkit for embedded signal processing

Computational hardware has become prevalent in recent years for signal processing applications for reasons such as flexibility, implementation efficiency and cost-effectiveness. The large variety of architectures available from many vendors has enabled a profusion of competing architectures, assembly languages and feature sets to match many individual applications. Although a single computational device may well suit a particular application, it seldom suits very many other applications closely. From an industrial perspective, this is inefficient since either system developers who are familiar with one processor from a previous project choose to use it for the next project, where it may be less suitable or they are faced with the long and costly learning curve implied in the adoption of an new architecture. Even when a new architecture is demonstrably better than older architectures for a given project, engineers will tend to work around the deficiencies in the older architecture rather than accept the learning curve for the new architecture. A preferable approach would be to allow computational hardware to be adapted at a microarchitecture level to match the application within a particular project, whilst maintaining a common instruction set, development tool chain and system framework. The paper introduces the 2k2, a flexible and modular computational system that allows developers to standardise on a signal processing solution for a wide variety of applications without requiring either a costly learning exercise, or necessitating a compromise between actual processor capability and requirement. The proposed system also allows late changes to processor capability without either costly hardware changes or lengthy software redesign