Multi-function DSPs Keep Wearable Devices Ticking
Wearable technology is one of the few sectors able to span many markets. Manufacturing processes enable wearable tech to find a home in textiles and jewelry in the fashion sector as well as in consumer and healthcare applications.
Everyone knows smartwatches, but these are just the tip of the wearable iceberg. Although smartwatch shipments jumped from 72.2 million in 2015 to 100 million in 2016, they are expected to peak next year, as smartglasses, textiles and jewelry emerge, says Strategy Analytics in a report about wearable revenues.
Integration into clothing is particularly attractive for the health and wellness market. The healthcare market will grow 22 percent this year alone, says analyst IHS in a report on the sector. It predicts that it will be worth $2.2 billion, and to rise to $2.8 billion by 2019. As well as finding use in pedometers and other fitness monitoring bands, wearable technology can make patient monitoring more comfortable as data is collected from a discreet wristband, chest strap or even bedsheets at home, rather than cumbersome equipment in a hospital or doctor’s office.
Low Power Operation
Part of the divergence of wearable technology is attributed to the use of low power DSPs. A DSP is an efficient way to run always-on sensing and can run for several days on a single battery. This can be used in a wearable system design that is power-efficient and able to run multiple functions and applications. “There are trends in wearables, such as specific function DSPs, like image processing DSPs, which run algorithms in a segment, and the use of multiple DSPs in one product for functionality,” says Paul Garden, Product Marketing Director, for Tensilica IP, at Cadence Design Systems.
The capability of a DSP as an efficient processor on which to run always-on applications supports wearable technology development.
Multiple functions and applications can run on a single DSP, allowing multiple technologies to be integrated into an SoC without having to implement the complete system in hardware. “TeakLite-4 DSP can handle, for example, our Bluetooth low energy, always on, HD audio playback and post processing, voice communication, noise reduction and sensor fusion (context awareness),” says Richard Kingston, Vice President, Market Intelligence, CEVA.
The company launched the latest addition to its TeakLite programmable core series in April this year. It added a Power Scaling Unit (PSU) to reduce the dynamic power and leakage power, explains Kingston. Dynamic power is controlled through clock gating techniques, where the clock can be independently shut down for the DSP core, the memory subsystem, the program and data memories, and the emulation and debug modules. Leakage power is controlled through voltage domains. As a result, the system can have a separate voltage domain for the DSP core and memory subsystems, one for the program and data memories, and a separate one for the emulation and debug (which is shut down during production) and the PSU is always-on. “The result is that TeakLite-4 DSP can run always-on voice activation at less than 20µW, which is a significant power saving compared with alternative solutions,” he says.
“The DSP can run such a feature at significantly lower power than on a typical RISC CPU,” he explains. Subsequently, the battery required to run such a function can be significantly smaller than required if running on a CPU. TeakLite-4 is, says Kingston, “ideally suited to running always-on applications, such as voice activation.” Voice activation, he points out, is the most natural user interface for wearables, as the device is able to listen for a phrase or keyword to activate it.
To lower power consumption for audio playback, audio tasks can be handled by dedicated DSP engines. “Android includes native offload of all audio processing to a DSP, if there is one in the system,” says Kingston. He elaborates, saying, “This is known as audio tunnelling and it frees up the CPU for more user applications, while ensuring that DSP algorithms meet latency, real-time response and low power constraints.”
The company has developed its own Android Multimedia Framework (AMF) to allow mobile system integrators to interface the Android-based CPU systems with DSP engines in an SoC to offload audio functions to the DSP.
“For imaging,” Kingston says, “AMF ‘tunnels’ an intensive multimedia task (e.g. SuperResolution) to the DSP in the system (e.g. the CEVA-XM4 imaging and vision DSP), significantly lowering power consumption, by up to 20 times less) for running these tasks, versus running them on a CPU.”
Paul Garden was speaking at the launch of the Tensilica Fusion G3 DSP, which targets compute-intensive audio and imaging applications. It supports fixed point or floating point operation, which highlights another trend in DSP design, identified by Garden, that of customers wanting to run code ‘out of the box’ with floating point operation using the optional Vector Floating Point unit, or convert code to fixed point and run on a fixed point DSP.
In addition to support for both fixed and floating point operation, the DSP is designed to run many algorithms and maintain efficient time control. The Tensilica Fusion F1 DSP is designed for wearable or Internet of Things (IoT) applications, which require sensor fusion and which are battery powered, such as waiting for voice activation, he explains, or waiting for a wearable device to wake up. Smaller, lower power DSPs, aimed at lower power, always-on, remote, sensor, and battery powered applications, is how Garden defines the two members of the family. Both are based on the 32-bit Xtensa architecture.