Machine Learning Takes UWB Localization to the Next Level

Imec this week said it has developed next generation ultra-wideband (UWB) technology that uses digital RF and machine learning to achieve a ranging accuracy of less than 10cm in challenging environments while consuming 10 times less power than today’s implementations.

The research and innovation hub announced two new innovations from its secure proximity research program for secure and very high accuracy ranging technology. One is hardware-based, with a digital-style RF circuit design such as its all-digital phase locked loop (PLL), to achieve a low power consumption of less than 4mW/20mW (Tx/Rx), which it claims is up to 10 times better than today’s implementations. The second is software-based enhancements which utilize machine learning based error correction algorithms to allow less than 10cm ranging accuracy in challenging environments.

Explaining the context imec said ultra-wideband technology is currently well suited to support a variety of high accuracy and secure wireless ranging use-cases, such as the ‘smart lock’ solutions commonly being applied in automotive; it automatically unlocks a car’s doors as its owner approaches, while locking the car when the owner moves away.

UWB benefits and challenges (Image: imec)

However, despite its benefits such as being inherently more difficult to compromise than some alternatives, its potential has largely remained untapped because of its higher power consumption and larger footprint. Hence imec said the hardware and software innovations it has introduced mark an important step to unlocking the technology’s full potential, and opens up the opportunity for micro-localization services beyond the secure keyless access that it’s been widely promoted for so far, to AR/VR gaming, asset tracking and robotics.

Christian Bachmann, the program manager at imec, said, “UWB’s power consumption, chip size and associated cost have been prohibitive factors to the technology’s adoption, especially when it comes to the deployment of wireless ranging applications. “Imec’s brand-new UWB chip developments result in a significant reduction of the technology’s footprint based on digital-style RF-concepts: we have been able to integrate an entire transceiver – including three receivers for angle-of-arrival measurements – on an area of less than 1mm².”

He added this is when implemented on advanced semiconductor process nodes applicable to IoT sensor node devices. The new chip is also compliant with the new IEEE 802.15.4z standard supported by high-impact industry consortia such as the Car Connectivity Consortium (CCC) and Fine Ranging (FiRa).

Complementing the hardware developments, researchers from IDLab (an imec research group at Ghent University) have come up with software-based enhancements that significantly improve UWB’s wireless ranging performance in challenging environments. This is particularly in factories or warehouses where people and machines constantly move around, and with metallic obstacles causing massive reflection – all of which impact the quality of UWB’s localization and distance measurements.

Using machine learning, it has created smart anchor selection algorithms that detect the (non) line-of-sight between UWB anchors and the mobile devices that are being tracked. Building on that knowledge, the ranging quality is estimated, and ranging errors are corrected.  The approach also comes with machine learning features that enable adaptive tuning of the network’s physical layer parameters, which allows appropriate steps to then be initiated to mitigate those ranging errors – for instance by tuning the anchors’ radios.

Obstacles and non line of sight effects can impact the quality of UWB’s localization and distance meaurements. On chip ML (machine learning) can correct errors as shown in these two examples. (Image: imec)

Professor Eli De Poorter from IDLab said, “We have already demonstrated an UWB ranging accuracy of better than 10cm in such very challenging industrial environments, which is a factor of two improvement compared to existing approaches. Additionally, while UWB localization use-cases are typically custom-built and often depend on manual configuration, our smart anchor selection software works in any scenario – as it runs in the application layer.”

Through these adaptive configurations, the next-generation low power and high-accuracy UWB chips can be utilized in a wide range of other applications such as improved contact tracing during epidemics using small and privacy-aware devices.

In fact, imec has already licensed the technology to its spin-off Lopos, which has released a wearable that enables enforcement of Covid-19 social distancing by warning employees through an audible or haptic alarm when they are violating safe distance guidelines while approaching each other.

Choosing UWB instead of Bluetooth, Lopos’ SafeDistance wearable operates as a standalone solution which weighs 75g and has a battery life of 2-5 days. The UWB-technology based device enables safe, highly accurate (< 15cm error margin) distance measurement. When two wearables approach each other, the exact distance between the devices (which is adjustable) is measured and an alarm is activated when a minimum safety distance is not respected.

Since it is standalone, no personal data is logged and there is no gateway, server or other infrastructure required. Lopos has already ramped up production to meet market demand, with multiple large-scale orders received over the last few weeks from companies active in a wide range of different sectors.

The post Machine Learning Takes UWB Localization to the Next Level appeared first on EE Times Asia.

from EE Times Asia https://ift.tt/2M4Qryc