DSP is at crossroads, again, and this time at stake are communication, voice, and vision applications in the IoT realm.

"DSP is dead; long live DSP." 

Will Strauss, president of research firm Forward Concepts, wrote this back in 2002. He pointed out that while off-the-shelf DSP (Digital Signal Processor) chips had been declining in volume, DSP cores were increasingly implemented in the custom chips for digital cameras, Bluetooth Wi-Fi, and cable modem.

Fast forward to 2016 and it's still true in the age of Internet of Things (IoT). The latest turn in the DSP evolution saga came in the mid-2010s when the IoT bandwagon started to reinvigorate a number of electronic segments: consumer, automotive, industrial, and more. Next, the CPU core kingpin ARM began integrating DSP extensions and floating-point hardware, and some industry watchers saw it akin to death knells for even the embedded DSP implementations.

However, as it turned out, such DSP-extended microcontrollers are mostly able to serve relatively simple applications like motor control and audio catering a small number of channels. In other words, system designs that need relatively low-performance floating point processing. However, in IoT designs, where power-efficient computing is a critical demand, DSP implementations still seem to hold sway for their ultra-low-power merits.

 

 
TI's TDA3x chipset for ADAS is based on the C66x DSP core

 

Take the case of a smartwatch that is running sensor fusion, sound sensing, voice activation, and music playback functions. Here, an embedded DSP consumes half the power compared to a small MCU. Next up, open and programmable DSP cores allow IoT developers to repurpose the same silicon for a slightly different application.

For example, engineers can change the software running on a DSP inside the system-on-chip (SoC) originally designed for a computer vision application and re-purpose it for a somewhat similar use-case like surveillance. That's a significant advantage for the low-volume IoT applications.

The evolution of DSP architecture is now leading the latest DSP portfolios to support the new programming frameworks like OpenCL and OpenMP. Moreover, DSP technology is heading toward 64-bit architectures while boasting greater computational capacity and features like vector processing and multi-core coherence.

In the IoT space, DSPs are being largely eyed for three key areas: communications, audio/voice, and imaging/vision. For a start, the DSP-based programmable PHY solutions can play a vital role in enabling IoT devices to support multiple connectivity standards that are operating at varying bandwidths and ranges.

 

CEVA's XM4 imaging DSP core promises a myriad of apps

 

For audio and voice sensing, DSP's ultra-low-power consumption goes a long way in “always on, always listening” voice control functions. So a growing number of wearables and sensor-enabled IoT devices are now employing DSP-enabled voice biometric analysis for applications as such speaker identification.

Then, in the imaging and computer vision domain, DSPs are playing a crucial role in facilitating compute-heavy object detection and recognition in a power efficient manner. Case in point: advanced driver assistance systems (ADAS) applications where DSPs can run deep learning algorithms like CNN and DNN at less power compared to CPUs and GPUs.

 

Tags:

 

Comments

0 Comments