Last Updated: 20050321, meiSunday, May 29, 11:00 — 13:00
W5 - Differences in Implementation of Audio Algorithms when Moving Back and Forth between Native (PC, Mac) and Dedicated DSP Platforms
Christoph M. Musialik, Algorithmix - Waldshut-Tiengen, Germany
Ulrich Hatje, Algorithmix - Waldshut-Tiengen, Germany
Manfred Lutzky, Fraunhofer Institut IIS - Erlangen, Germany
Søren H. Nielsen, TC Electronic - Risskov, Denmark
Shai Shasag, Waves - Tel Aviv, Israel
Today audio engineers are continually confronted with different platforms. On the one hand, there are native systems, PC or Mac, and on the other hand, dedicated DSP chips. Due to extremely fast CPUs working with a high degree of accuracy (double floating-point) and a lot of memory, PCs or Macs are able to run complex, high-quality audio algorithms in real-time. Even using schoolbook algorithms written in C language good results can be expected. In spite of this rapid development in this native area, there are still enough reasons to use modern DSP processors: lower latency (live sound systems), portability, and reliability (unfortunately today's native operating systems have still tendency to crash). Advanced signal processors can cope well with professional audio standards, but dependent on architecture and price, they have a lot of limitations when compared to native processors: less accuracy, less memory, less speed, not always floating point ALU. The situation is even more complicated if high-quality audio algorithms have to be ported to the new generation of low-cost processors usually being something between microprocessor and RISC machine. They are very fast, have a lot interfaces, even DSP typical instructions, but mostly only 16-bit fix point ALU. For young engineers "spoiled" by practically no limits when working on a PC or Mac such processors mean absolute back-to-the-roots exercise.
Experienced development engineers will share their experiences concerning problems, rules, limitations, and the development effort when implementing algorithms on different platforms or porting them among PC, Mac, dedicated DSPs, and RISC processors. In many cases a straightforward implementation without algorithm changing (e.g., filter structure) is impossible if high audio quality has to be provided.
©2005 Audio Engineering Society, Inc.