HiFi 3 Audio DSP

Products » Audio DSP » HiFi 3

Taking Audio Performance to the Next Level

24-bit precision has been the foundation for high quality audio. However post processing algorithms are increasingly being released in initial fixed point implementations at 32-bit precision, in part to take advantage of the higher achievable performance with the more complex functions. The HiFi 3 Audio DSP was designed to meet these escalating needs, building on the successful deployment of Tensilica's HiFi 2 Audio DSP.

Built for the Next Generation of Audio Requirements

Audio requirements are exploding as audio DSPs are being added to more devices that need to provide or simulate complex voice and audio algorithms for multi-channel sound. Smartphones need to support more complex voice and audio use cases. Voice requirements are jumping from 200 MHz today for narrow band voice codecs and basic noise suppression to 600+ MHz in a few years with the widespread deployment of wideband codecs, improved noise suppression and noise-dependent volume control.

At the same time, audio and gaming requirements are escalating. Smartphones will need to directly support multi-channel entertainment. Gaming will support up to 32 streams for immersive play. As audio post-processing complexity increases, the processing requirements for these complex algorithms continue to grow.

No longer can designers rely on the applications processor to provide the audio processing in an efficient, power-conserving way. Offloading the audio processing is one of the best ways to increase battery life and provide a better audio experience.

Optimized for Pre- and Post-Processing

The HiFi 3 Audio DSP was designed based on Tensilica's extensive experience with the popular HiFi 2 and HiFi EP DSP IP cores. By moving to a 32-bit architecture, Tensilica significantly improves the efficiency of running the complex algorithms that provide enhanced audio post processing. The 32-bit architecture also supports standard 32-bit C code, simplifying the software programming interface for companies that want to port their own audio software.

The HiFi 3 Audio DSP features improved DSP performance on post/pre-processing kernels. It provides approximately a 1.8X performance improvement on critical FFT, FIR and IIR algorithms. It also provides a 1.5X performance improvement on voice codecs.

The 3-issue architecture is very efficient, using the three slots only when necessary and avoiding the bloat of traditional VLIW architectures. Four MACs operate as four 24x24, four 32x16, two 32x32, or two 32x24 depending on that's required for the algorithm. The 64-bit load/store unit pushes data through at a fast pace. It also features a unified 16-entry, 64-bit register file.

Extensive Software Library

All of the codecs ported to Tensilica's HiFi 2 architecture will run on HiFi 3 "out of the box." Tensilica is optimizing the most popular codecs to take advantage of the architectural enhancements in HiFi 3, which will then significantly boost performance.

Tensilica's audio library includes:

The following audio, voice, and audio and voice enhancement software packages are available for the HiFi 2, HiFi EP, and HiFi 3 Audio DSPs:

AUDIO
VOICE
AUDIO AND VOICE ENHANCEMENT

Marketing Agency