Some Numbers...

• 42 months research project funded by the French national research agency
• Starting date: March 2021
• 3 labs (all based in Lyon) are involved: GRAME-CNCM (Faust and embedded systems for audio), CITI Lab @ INSA Lyon (FPGAs and embedded systems), and LMFA @ Centrale Lyon (active control of acoustics)
• 8 permanent researchers, 1 PhD student, 1 postdoc, 1 research engineer, a couple of interns, etc.

Goals of FAST

• Facilitate the programming of FPGAs (Field Programmable Gate Array) for real-time audio signal processing applications through the use of the Faust programming language
• Design high-performance FPGA-based multichannel (we're aiming for 2x2 and 32x32 modules) audio processors with ultra-low latency (<10us)
• Exploit our system in the context of active control of room acoustics and musical instruments

What We've Done so Far

• faust2fpga command line tool to program a Digilent Zybo Z7 (Xilinx Zynq-7000 FPGA)¹
• This tool relies on High Level Synthesis (HLS): Faust -> Optimized C
  -> Intellectual Property (IP)
• Generated IPs could be significantly optimized if Faust could generated fixed-point C code
• Dealing with memory is a complex task when aiming for ultra-low latency

1. T. Risset, R. Michon, Y. Orlarey, S. Letz, G. Müller, and A. Gbadamosi, "Faust2FPGA for Ultra-Low Audio Latency: Preliminary Work in The Syfala Project," in Proceedings of the International Faust Conference (IFC-20), Paris, France, 2020

faust2fpga Overview

Challenges to Come

• Implementing internal fixed-point support in Faust
• Implement a faust2vhdl tool and compare its performances with IPs generated through HLS
• Optimize Faust DSP algorithms to be used on FPGAs
• Design custom FPGA-based audio processing units
• Use these tools in the context of active control

Overview of FAST

Thanks! :)

https://fast.grame.fr

Contact: michon@grame.fr

---

Slides of this presentation:

https://ccrma.stanford.edu/~rmichon/talks/ccrma-openhouse-21