Last week three vendors introduced licensable cores targeting consumer audio applications. In the first of these announcements, MIPS announced the MIPS24KEc, the first core to implement the "DSP ASE" signal-processing instruction-set extensions. (For a discussion of these extensions, see the October 2004 edition of Inside DSP.) Although the MIPS24KEc is not specifically tuned for audio applications, MIPS claims that the addition of the DSP ASE provides a 15%-40% boost in performance across a range of audio algorithms. BDTI has not verified this claim, but it has begun an analysis of the processor's signal-processing performance. The preliminary results of this analysis suggest that the MIPS24KEc is one of the faster embedded general-purpose processor cores on the market—certainly more than fast enough for most consumer audio applications.
The MIPS architecture is widely used in applications that include sophisticated audio features, such as DVD recorders and set-top boxes. The MIPS24KEc will likely help solidify MIPS' position in these markets. For example, the signal-processing capabilities of MIPS24KEc will eliminate the need for a separate audio processor in some of these applications.
In the second announcement, ARM introduced its AudioDE coprocessor. The AudioDE is a highly specialized coprocessor targeting audio encoding and decoding in low-power devices such as portable audio players. The coprocessor can be programmed using C, allowing AudioDE to support multiple audio codecs. However, the coprocessor is not intended to perform functions other than audio encoding and decoding. For example, the coprocessor is not intended to perform equalization or other post-processing functions.
AudioDE is based on ARM's configurable OptimoDE architecture. This architecture allows system designers to build a processor using pre-defined execution units. System designers can also customize execution units for their applications. By tuning the architecture specifically for audio decoding, ARM claims it has achieved a very low-power solution. For example, ARM claims that an AudioDE implemented in a 0.13-micron process requires just 0.8 mW to decode a 320 Kbit/sec stereo MP3 stream.
In the final announcement, Tensilica previewed its forthcoming audio processor. Like its first-generation audio processor, the new processor will be based on an Xtensa customizable core. However, the new processor will use the newer Xtensa LX architecture instead of the Xtensa V architecture. According to Tensilica, moving to the Xtensa LX architecture will enable the new processor to achieve at least a 40% improvement in audio performance at a given clock speed while maintaining the same gate count as the first-generation processor.
The Tensilica processor resembles the MIPS24KEc in some respects, and it resembles the AudioDE in other respects. The Tensilica processor is based on a RISC architecture like that of the MIPS24KEc. Yet the Tensilica processor will be highly specialized: Tensilica claims it will include over 300 audio-specific instructions. In this respect, the Tensilica processor will be a highly-specialized solution like AudioDE.
It may seem surprising that so many new cores are targeting audio applications, considering many of today's mainstream cores are fast enough for these applications. The motivation for the new cores is not increased speed, but rather improved cost effectiveness and energy efficiency. While many cores may be able to run a given audio workload, the silicon area and energy required to do so varies widely. As digital audio becomes common in a wide range of consumer products—many of which are battery-powered—SoC designers will seek out the most cost-effective and energy-efficient solutions.
Add new comment