How can sensors process our valuable data?
Recent advances in sensor architecture and their core technology are aligning to present a compelling new opportunity in what we term ‘edge of the edge data processing’, in turn improving the performance of wearables. Edge processing has traditionally referred to processing on a mobile phone or other reasonably powerful device, but increasingly can refer to tiny devices operating for extended periods, such as a wearable. However, we are seeing this principle taken to the next level, where the sensors inside the wearable are becoming a lot smarter.
One way that this has been facilitated is by those sensors becoming much more power efficient, but the key advancement is in the ability to conduct ‘local’ data processing, ultimately meaning that the wearable can be designed to run for longer per battery charge. Another benefit is that large amounts of sensor data can be reduced to very small payloads, comprising ‘insights’ rather than copious raw information. Just pushing these insights through the system can cut power consumption considerably, particularly if the core processing in the wearable also applies this principle, shipping only small packets of information over the airways to a mobile or cloud device. After all, transmitting data across the airways is usually the dominant drain on the battery of a connected wearable device.
There are particular applications where local processing is a huge advantage. For example, if a device relies on a phone or laptop to process data, there is a risk that application critical insights are not available in a timely fashion. In this scenario, we would want to have the processing happening locally within the wearable. Sensors tailored to provide insights from their native data streams also present the most efficient way of optimising battery use in such a scenario.
How does localisation of data help us?
The localisation of data processing at source can therefore make wearables more power efficient and hence smaller, due to the need for less battery volume, which is extremely attractive to the developers of most wearable products. Sensors detect very specific signals and process their own data in a very signal optimised way, the wearable collates and curates these processed ‘insights’ and combines these to push key product insights to the mobile and further to the cloud as required. All of these steps focus on expedient use of internal power resources: usually, minimising the use of the Bluetooth radio by reducing data rates is key.
To illustrate the efficiency of this opportunity, let’s consider a desktop computer. This is a ‘general purpose computer’ and has the ability to process all sorts of signals, including video, voice and multitasking workflows at incredible speeds. Compare this with, for example, the accelerometer in a wearable, which measures three simple values about 100 times a second. Whereas the desktop computer needs to be able to process a broad array of data types, with many things going on simultaneously, the accelerometer signals are at a fairly slow and fixed rate and are in 3 clearly defined dimensions. This means that the accelerometer can have a real time processing architecture optimised to do one job extremely efficiently and can have the mathematics of converting acceleration data to useful metrics built in. The result is that the efficiency of this system is very high, much more so than for a general purpose computer.
Each signal type used in a wearable product will have a very particular form, and processing of the associated data should be addressed in harmony with those data types. What we're seeing with these new sensors is that the manufacturers are providing onboard processing resources, highly optimised to process each specific signal type produced by the sensor. Bosch has been leading the way on this for several years, but increasingly it is becoming prevalent across the industry. ST also announced new lines of sensors recently amongst others.
So what does this mean for your wearable?
In the wearable data processing ecosystem, you can split data processing between the cloud, a mobile phone or desktop computer, and a wearable. Processing data away from a cloud or centralised repository, you gain the ability to do calculations on the raw data, reducing the volume and increasing the overall system efficiency. There's a very strong opportunity to create insights right at that primary level, and not pass on lots of the raw information up into the wider ecosystem. For example, an accelerometer has three axes of measurement, you could reduce the data by an order of 3,000, sending a single insight ‘per second’ rather than 3,000 bits of raw data. Take this to the limit and a fall detecting algorithm running on the accelerometer could only send a signal out the the phone and cloud if this event is detected, reducing the data rate to essentially zero. Throwing away most of the raw data saves a lot of power for the radio transmitter and will lead to vastly extended battery life. Of course, in some cases it will be important to send larger payloads of data, but predominantly it is not necessary to do so and adopting this mindset is essential in modern wearable technology product design.
Case study:
Wearables in extreme conditions
When we worked on a project based in the Premier League stadium of Brighton and Hove Albion Football Club, we found that the data we were collecting could not be transferred to a mobile phone because Bluetooth channels were heavily oversubscribed and hence unreliable. Having a wearable system which was able to process data locally and work independently of the mobile phone proved to be a key design constraint and had a strong influence on the product design work we conducted.
How do sensors help us in the real world?
Fall detection is an area where stand alone sensors are offering a compelling use case. A sensor processing data locally, enabled with a low power transmission system can monitor in real time and provide in the form of a short radio broadcast if a trigger is reached. Responders can provide help without the need for technical literacy. This can be an issue faced by the elderly, who tend to suffer most from falls, which can have a catastrophic impact on their health. Having intelligent mobility aids can track health long term and in some cases will be the difference between life and death. If the device can work effectively for many months without a battery replacement, this turns a burdensome device into a genuine life saver.
Concluding remarks
Wearables have moved on a lot in the last ten years. However, this has been incremental and mainly focused on improvements to each of the many building blocks making up the system. From Bluetooth evolving to the V5 standard, to advances in sensors, batteries and microprocessors now designed specifically with wearables in mind, the applications and opportunities have driven incredible progress. However, we are now entering a new phase of this journey, where silicon is being designed to process very specific sensor signals and bolted directly onto those sensors, allowing a highly efficient and optimised use of power, delivering insights and appropriate data flows right from the point of measurement. This is a game changer and coupled with advances in networks, battery technology and energy harvesting can certainly be considered at least colloquially as the dawn of wearables 3.0.
Follow us on Twitter and LinkedIn to keep up to date