FPGA-based implementation of audio effects for ultralow-latency Networked Music Performance applications

A Networked Music Performance (NMP) is a real-time musical interaction that aims at allowing musicians to play together from remote locations. One of the greatest challenges of networked performances is keeping latency, i.e. the mouth-to-ear delay between users, as low as possible. In recent years, several solutions, mainly software ones, have been developed to ensure a satisfactory communication. A very promising approach to further reduce latency is the implementation of a dedicated audio processor. Indeed, this solution permits improvements in terms of speed, while keeping some degree of flexibility through an implementation based on a Field-Programmable Gate Array (FPGA). Among the many architectures available, one of the most suitable for an audio processor with these characteristics is the Transport Triggered Architecture (TTA). This processor design is a one-instruction set architecture that is based on different function units connected by buses. The only available operation is the move one, which allows the execution of more complex instructions by moving data from one unit to another. This thesis proposes a hardware implementation of new function units to perform two of the most widely used functionalities in the music environment: audio mixing and reverb effect. Specifically, the former is the process of combining different audio sources to generate a single output sound. This feature is crucial in NMPs because it allows multiple input channels and different users to be heard together in real-time. The audio mixing implementation is described starting from its algorithm, going through a first firmware implementation and achieving a dedicated hardware design by exploiting a toolset called TTA-based Co-design Environment (TCE). The second functionality added to the system is the reverb effect, an audio unit to simulate the acoustic phenomenon of reverberation. Since the goal of NMP applications is also to provide natural listening conditions to musicians, adding reverberation artificially can give the sense of being part of a shared acoustic space. The algorithm chosen for the reverb implementation is the one theorized by Manfred Robert Schroeder. Regarding its development, a different approach is considered: MATLAB HDL Coder is employed to generate a hardware implementation starting from a Simulink model. Finally, the entire audio processor is implemented on FPGA exploiting Xilinx Vivado, a synthesis tool for hardware design.