Design
Tensilica’s HiFi 3 DSP IP Core Provides Over 1.5x Better Performance for Audio Post Processing and Voice in Smartphones and Home Entertainment
Tensilica,Inc. today announced the HiFi 3 audio/voice DSP (digital signal processor) IP (intellectual property) core for SOC (system-on-chip) design. This fourth-generation audio DSP offers higher performance with lower power for high-performance audio post-processing and voice processing algorithms used in smartphone and home entertainment systems and expands Tensilica’s market-leading HiFi architecture to quad 24/32-bits. Tensilica has already licensed HiFi 3 to a tier one smartphone OEM and a tier one semiconductor manufacturer.
“TThe HiFi 3 audio/voice DSP is upwards software compatible with the proven HiFi 2 and HiFi EP audio/voice DSPs, so the entire existing library of over 90 HiFi-optimized audio and voice codecs, and audio enhancement software packages, will run on HiFi 3 cores with improved performance. In addition, Tensilica has further optimized the most demanding multi-channel audio software packages to run even more efficiently on the HiFi 3. As an example, HiFi 3 can support DTS-HD Master Audio for Blu-ray Disc in only 233 MHz. HiFi 3 carries over the assets of HiFi 2 to enable easy programmability for companies wanting to port their own proprietary audio software packages and builds on this to add improved code porting productivity from a further simplified programming model.
The Growing Requirements in Audio Post Processing and Voice Most of the innovation occurring in audio today is in post processing with sophisticated algorithms such as volume leveling, dialog clarity, volume boost, spatial expansion, equalization, and gaming with requirements to support 32 or more simultaneous audio streams and VoIP (voice over Internet protocol) for interactive game play. These complex algorithms are being designed into smartphones, home entertainment systems, and gaming consoles.
Voice requirements in smartphones are significantly outpacing Moore’s Law and battery technology. While today’s narrow band voice codecs and basic noise suppression can be satisfied with about 200 MHz of audio DSP horsepower, these requirements will exceed 600 MHz in a few years with the widespread deployment of the AMR-WB (adaptive multi-rate wide band) voice codec, the Skype SILK super wideband codec for VoIP, improved noise suppression, and other pre-processing functions such as noise-dependent volume control.
Meeting these Requirements
The HiFi 3 audio/voice DSP addresses these new requirements with over an 80 percent increase in performance for the FFT, FIR and IIR algorithms commonly used in post processing resulting for example in a 1.8x performance improvement on DTS-HD Master Audio post processing compared to HiFi EP. For speech it offers direct support for common ITU operations enabling greater than 50 percent performance improvement on many voice codecs, with significantly reduced porting effort.
The HiFi 3 audio/voice DSP is a 3-way VLIW style DSP based on Tensilica’s Xtensa Dataplane Processor (DPU) architecture with four MACs (multiply accumulate units), a 64-bit load/store unit, and a unified 16-entry 64-bit audio register file.
Availability
HiFi 3 has already been delivered to lead customers with general availability in March.