Research visit to Ericsson, Programming Languages for DSP
- Reference number
- SM08-0026
- Start and end dates
- 090101-101231
- Amount granted
- 1 548 933 SEK
- Administrative organization
- Chalmers University of Technology
- Research area
- Information, Communication and Systems Technology
Summary
The main high level objective is improved efficiency of baseband SW development through increasing the level of abstraction for the developers. Two applications will be studied more in detail: DSP domain specific languages and functional languages as a means of extracting the parallelism in the baseband design. In both sub-projects, the work will be case-study driven. 1) DSL for DSP (with the product group) We have chosen the AMR speech and H.263 video codecs as suitable examples for the DSL work. I will work as part of a larger project (see www.cs.chalmers.se/~ms/DSPDSL.pdf), along with a postdoc from Chalmers who is funded by Ericsson. I will contribute to the following tasks: Study of the needs of DSP algorithm designers, with emphasis on finding a suitable style of algorithm description Prototype implementation of description language Development of method of adding annotations to control date and time of computation and data representation and precision choices. The overall aim is to develop a standard language for DSP algorithm design, giving the desired separation from hardware implementation. 2) Investigation of functional languages for controlling parallelism, aiming at future multicore systems for media processing(with Baseband Research) This is a longer term project. A simple prototype implementation of a DSL for parallel programming, embedded in Haskell, is the natural first step. Our recent work on GPU programming is relevant.
Popular science description
Ericsson has been extremely successful in designing, programming, building and selling the base stations that allow our mobile phones to communicate not only voice data, but also much greater data volumes for 3G internet communcation. Modern base stations contain many computer processors of varying kinds, including some that are specialised to doing digital signal processing. Without all these processors, it would be impossible to process all the data that needs to be processed. But it is very hard to program these multiprocessor systems and get everything right, and in particular it is hard to make changes to the overall program if one wants to move parts of it around onto different processors in order to reach the required performance. We need new programming methods for these systems, so that Ericsson can continue to be a leader in this field. And things are only going to get worse, as we move from single to dual- to quad- to multicore processors. How to deal with this development, and still get programs written that are both correct and make good use of the hardware resources is a big research question for the whole of academic computer science, and for the entire information technology industry. Prof. Sheeran is interested in special modern forms of programming language called functional languages, and has worked with Intel on using such languages for hardware design. Now, she will investigate the use of such languages in the development of base stations -- working with collaborators from product and research groups at Ericsson, Lindholmen in Göteborg. Prof. Sheeran will escape from her daily grind to work on interesting research problems with highly skilled researchers and developers at Ericsson. The problems are real and challenging, and successful results will have real practical impact. This one year project will also lay the ground-work for a research collaboration that will continue after her return to Chalmers, with the emphasis being on multicore programming problems. During the visit to Ericsson, Prof. Sheeran will also learn how Swedish industry works, and how best to enable fruitful collaboration between Chalmers and Ericsson, and more generally between academia and industry. People who know about and understand the ways of both academia and industry are few and far between, but are vital to continued transfer of ideas, problems and solutions and even people across the divide. Here, we aim to produce one more such person.