The purpose of this work is to show the advantages of implementing digital signal processing for high quality audio applications in custom floating-point. We consider the trade-offs dynamic range vs. precision (i.e., quantization) by comparing standard floating-point (namely, binary32) to custom floating- point. Moreover, by resorting to Tunable Floating-Point (TFP) hardware units, we can adjust the dynamic range and the precision in different parts of the algorithms to explore several alternatives. Results show that 16-bit floating-point formats can give a good compromise between quality and energy efficiency.
Cardarilli, G.c., Di Nunzio, L., Fazzolari, R., Cesa, R.l., Nannarelli, A., Re, M. (2023). Tunable Floating Point for high quality audio systems: the sound of numbers. In Conference Record of the fifty-seventh Asilomar Conference on Signals, Systems & Computers (pp.1547-1551). New York : IEEE [10.1109/IEEECONF59524.2023.10476741].
Tunable Floating Point for high quality audio systems: the sound of numbers
Cardarilli G. C.;Di Nunzio L.;Fazzolari R.;Re M.
2023-01-01
Abstract
The purpose of this work is to show the advantages of implementing digital signal processing for high quality audio applications in custom floating-point. We consider the trade-offs dynamic range vs. precision (i.e., quantization) by comparing standard floating-point (namely, binary32) to custom floating- point. Moreover, by resorting to Tunable Floating-Point (TFP) hardware units, we can adjust the dynamic range and the precision in different parts of the algorithms to explore several alternatives. Results show that 16-bit floating-point formats can give a good compromise between quality and energy efficiency.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
