Audio Computations using GPUs
Diplomarbeit
Description
The computational power of standard graphic cards with GPUs enables desktop-PCs to do real time calculations, which could be done previously only on dedicated DSP-boards.
In audio communication systems GPUs could be used to calculate concurrently for a large number of input channels:
- Audio Codecs
Audio codecs are used to compress audio signals in order to decrease the necessary bandwidth. A large number of different audio codecs are in use (i.e. G.729-, GSM-, TETRA-codecs).
- Echo Cancellers
Echo cancellers are necessary to remove electrical and acoustic echo signals.
- Conference Bridges
They are used to provide conferencing functionalities for a large number of participants.
- Other digital signal processing functions
Task
Implementation, evaluation and performance optimization of digital signal processing functions on GPUs (i.e. TETRA-codecs, echo cancellers, conferencers).C-algorithms for the DSP-functions are available, but have to be adapted and optimized for GPU processing.
A Windows kernel framework already exists which can be used to transfer audio signals between the GPU and the input/output devices and which should be extended to support GPU calculations.
Tools
Visual C++, CUDA development toolsRequirements
C++, interest in digital signal processing and CUDAAdditional Information
Contact: Michael WimmerThis diploma thesis is supervised in cooperation with the company PDTS and includes financial recompensation.
