Splitting frequency-domain processing between multiple dsp cores

Abstract

An audio processing system may split frequency-domain processing between multiple DSP cores. Processing multi-channel audio data—e.g., from devices with multiple speakers—may require more computing power than available on a single DSP core. Such processing typically occurs in the frequency domain; DSP cores, however, typically communicate via ports configured for transferring data in the time-domain. Converting frequency-domain data into the time domain for transfer requires additional resources and introduces lag. Furthermore, transferring frequency-domain data may result in scheduling issues due to a mismatch between buffer size, bit rate, and the size of the frequency-domain data chunks transferred. However, the buffer size and bit rate may be artificially configured to transfer a chunk of frequency-domain data corresponding to a delay in the communication mechanism used by the DSP cores. In this manner, frequency-domain data can be transferred with a proper periodicity.