Dauphin: A new statistical signal processing language

Abstract

Many software packages support scientific research by means of numerical calculations and specialised library calls, but very few support specific application domains such as signal processing at the symbolic level or at problem formulation. Translation from the natural domain-specific structure of problem description to the computer formulation is often a time consuming and error-prone exercise. As signal processing becomes more sophisticated, there is a need to codify its basic tools, thus allowing the researcher to spend more time on the challenges specific to a particular application. In this paper, we describe the design of Dauphin, a domain-specific programming language. Dauphin ultimately aims to extend the power of signal processing researchers by allowing them to focus on their research problems while simplifying the process of implementing their ideas. In Dauphin, the basic algorithms of signal processing become the standard function calls and are expressed naturally in terms of predefined signal processing primitives such as random variables and probability distributions.