Design of a wearable multi-sensor measurement platform -

Design of a wearable multi-sensor measurement platform

Presented is the complete design of a flexible multi-sensor measurement platform intended for a variety of medical, research, and recreational applications. The design considers practical constraints as a wearable device in a multitude of environments, aiming to satisfy developers’, researchers’, and users’ needs for reliability, power efficiency, ease-of-use, and a functional development environment.

Based on a high-performance 168MHz ARM-M4 digital signal controller with floating point support, the system delivers unprecedented computational throughput for a 3x5cm de- vice drawing as little as 3mW with an RTOS scheduler active.

The proposed requirements for such a device were initially focused on use in the artificial pancreas project. The focus has since shifted toward the more general- purpose research platform that it has become capable of performing complex multi- sensor fusion, computationally heavy estimation and classification, and providing a robust and flexible interface to diverse groups of both users and developers alike. This shift was performed without compromising any of the original objectives.

The design delivers several valuable contributions: Firstly, it includes an open-source hardware design for a powerful, compact, multi-sensor measurement platform with a rich user interface.

Secondly, to support data logging operations, the design incorporates a novel, extremely lightweight NAND flash file system. Thirdly, a modular, feature-rich, C++-based software framework is built on top of a real-time operating system (RTOS) to provide application developers with the utilities necessary for rapid prototyping.

Finally, this framework incorporates an automatic clock- and power-management scheme, allowing for power consumption to be kept at a minimum without burden to the application developer.

The platform is capable of measuring inertial parameters such as acceleration and angular movement rate as well as environmental parameters such as magnetic field and barometric pressure.

It also contains multiple wired and wireless interfaces to communicate with arbitrary external devices: A sub-GHz wireless interface provides for communicating with continuous glucose monitors (CGMs) as well as any of a growing multitude of Bluetooth-enabled physiological sensors such as heart-rate monitors.

Extensive experience with ST’s ARM Cortex-M4 STM32F4 microcontroller family in our NEMO designs and numerous other projects combined with its strong suitability for the project presented a compelling argument for its continued use. Cost-wise, these are among the least expensive Cortex-M4F processors considered. Since the initial design was constrained to a compact two-layer printed circuit board (PCB), the design is restricted to a 64-pin TQFP package, the smallest device that ST offers.

While NXP, a competing chip maker, offers an asymmetric dual core (Cortex-M4+ M0) at 204MHz, I feared that it would complicate the design excessively. There are also few tools available to accommodate a dual-core system where both execute a different instruction set from shared memory.

That said, it has great potential for meeting real- time constraints to have a second processor (slower, but also consuming less power and with faster interrupt servicing) handling menial tasks such as data acquisition and even scheduling for the M4 core.

While all memory usage and performance figures fit well within the constraints of the platform, there is still room for improvement and ongoing development to increase the robustness of the implementation. In addition to inode compaction, a number of potential optimizations have not yet been implemented.

Write activity can be reduced by using the on-chip flash cache to buffer multi-sector writes. This comes at a cost to the worst-case times of all operations which use the flash cache but could reduce the average cost of write operations dramatically (estimated ~20%) at high clock speeds. Also, the allocation process implementation is inefficient and should be rewritten.

Though this platform is ready for action in a number of projects in the IML, it will continue to grow. The field of body sensor networks is expanding quickly, fueled by aggressive development on many of the underlying technologies. Wearable inertial sensors are becoming commonplace on the consumer market. Though the devices are fairly primitive in their current state, it is plain to see the significance of the role that they will play in the future of health care.

To read this external content in full, download the complete document from the author archives at McGill University.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.