E-textiles are entering the validation era
New research compares a smart garment-integrated e-textile heart rate monitor with 3-lead ECG during running, showing strong agreement across speeds.
E-textiles are starting to be judged against the technology they replace
E-textiles have spent a long time proving that sensors can be put into clothing.
The more interesting question now is whether they can perform against the conventional technology they are supposed to replace.
A new study published in Sensors does exactly that, comparing a garment-integrated e-textile heart rate system against a 3-lead ECG (electrocardiogram) during incremental running.
Twenty-six participants were tested continuously from rest through running speeds of up to 12 km/h and back into recovery. Rather than validating the garment at a single steady state, the researchers compared both systems while heart rate, movement and therefore the conditions around the textile electrodes were changing.
The legacy benchmark
ECG is still the established reference for measuring the electrical activity used to derive heart rate. New wearable systems are therefore normally validated either against ECG itself or against another device that has already been validated against it.
For an e-textile, this is a much harder test than demonstrating that a conductive fabric can pick up a signal.
The sensing surface is part of something that stretches, moves and changes contact pressure as the wearer moves. The textile electrodes used in the device are dry electrodes, relying on direct skin contact and compression from the garment rather than the adhesive gel electrodes used by the reference ECG.
Running makes all of that more difficult.
Motion artefact, electrode displacement and changes in skin contact have been persistent problems in garment-based physiological sensing. Previous smart garment studies have shown good performance at rest only for accuracy to deteriorate as exercise intensity increases.
So testing an e-textile while somebody is actually moving is rather more useful than proving it works while they stand still.
How well did it do?
Across the complete test, agreement between the raw e-textile signal and the reference ECG was 0.94. When heart rate produced by the device’s own processing algorithm was compared with the ECG, that increased to 0.96.
More importantly, that relationship held as speed increased.
Across rest, running, cooldown and recovery, agreement remained between 0.82 and 0.93, including at 10 and 12 km/h.
The Bland–Altman analysis reported an average heart-rate bias of −0.6 bpm, with 95% of differences falling between approximately −8.47 and +7.27 bpm. There are limitations, as there should be with any validation study. This was a controlled treadmill test in healthy, active adults. An e-textile sensing system is being evaluated using the same sort of comparison we expect from its conventional counterpart.
The device was built as one system
Our connection to this work is the device rather than the research. Our founder was responsible for the system architecture and led the team that developed the device used in the study. One of the important decisions in that development was not treating the different technical disciplines as separate work streams.
Hardware, firmware, garment integration, digital signal processing, data science and QA all worked as one cohesive team around one system. This may sound obvious, but in wearable development it is often not the case.
A biosignal moves through a chain. The electrode has to acquire it. The analogue electronics have to preserve it. Firmware has to sample and move it correctly. DSP has to separate useful information from noise and motion artefact. The algorithms sitting above that signal then have to produce something meaningful from what is left.
Optimising one of those layers without understanding the others can simply move a problem further down the chain.
A ‘poor’ signal cannot always be fixed in software. Equally, adding more analogue circuitry is not necessarily the right answer if the same problem can be handled more effectively in DSP. A change in electrode geometry might alter both the front-end electronics and the characteristics of the data being presented to an algorithm. The system level architecture must account for all of this.
Why multidisciplinary development matters particularly in e-textiles
This becomes even more pronounced once the sensor itself becomes part of a garment.
Traditional electronics development already has dependencies between hardware, firmware and software. E-textiles add another dimension on top of that: the textile.
Fit affects the electrode pressure. Electrode pressure affects signal quality. Signal quality affects DSP. DSP affects what the algorithm can reliably infer. Changes made to improve one layer can have consequences several disciplines away.
It is the same broader problem we found in our research into why e-textiles struggle to scale: projects start to fail at the joins between disciplines rather than necessarily within the individual disciplines themselves.
The solution isn’t having more talented team members, it is making them work as one. For this device, firmware, electronics, signal processing and data science could iterate around the same underlying problem rather than each discipline receiving the output of the previous one as a finished specification. That matters when what you are ultimately trying to build something with many unknowns.
E-textiles are moving past the demonstration stage
For years a lot of e-textile work has understandably focused on feasibility, or limited washability. Those are necessary questions. For a commercial product, the challenge is around developing something that performs as well as the technology it is replacing. If e-textiles are going to move beyond academic demos, or limited prototypes, increasingly they need to be validated against those incumbents rather than simply compared with other e-textiles. That is why this research is so interesting to us.
Read the full paper
A Speed-Dependent Assessment of E-Textile-Based Sensor Technology
Burgess, L.C., Armstrong, M., Beresford, L., Callaway, A.J.
Sensors (2026), 26(11), 3378
DOI: 10.3390/s26113378