BMAR: Barometric and Motion-based Alignment and Refinement for Offline Signal Synchronization across Devices

summary

The gist

BMAR is a novel method to synchronize and align signals across devices without the need for specific user input, action…

In short

BMAR is a method to align signals from multiple wearable devices offline without needing wireless communication during recording. It uses two stages: first, it pre-aligns recordings using inexpensive barometric pressure data to reject non-simultaneous events. Second, it refines this alignment by cross-correlating accelerometer readings to account for clock skew and improve accuracy.

Key concepts

Barometric Pressure
This is an inexpensive sensor found in wearable devices that records air pressure changes over time. BMAR uses these recordings as a primary tool because the pressure readings from all devices worn by the same person are expected to be very similar, allowing the method to quickly check if two recordings happened at roughly the same time.
Pre-alignment
This is the first step where BMAR checks if two signal recordings were taken simultaneously. It uses barometric pressure observations to pre-align all data traces. This stage is crucial because it allows the system to reject any recordings that clearly did not happen together, improving the initial synchronization.
Cross-correlation
This technique is used in the second stage of BMAR to refine the alignment between sensor devices. It works by comparing two different signal traces (in this case, accelerometer data) against each other to find the best possible time shift (lag) that makes them match as closely as possible, effectively correcting for differences in device clocks.

Terminology used across episodes

This episode discusses

The paper

BMAR: Barometric and Motion-based Alignment and Refinement for Offline Signal Synchronization across Devices · Read on arXiv

MANUEL MEIER, CHRISTIAN HOLZ

ETH Zürich

A requirement of cross-modal signal processing is accurate signal alignment. Though simple on a single device, accurate signal synchronization becomes challenging as soon as multiple devices are involved, such as during activity monitoring, health tracking, or motion capture - particularly outside controlled scenarios where data collection must be standalone, low-power, and support long runtimes. In this paper, we present BMAR, a novel synchronization method that operates purely based on recorded signals and is thus suitable for offline processing. BMAR needs no wireless communication between devices during runtime and does not require any specific user input, action, or behavior. BMAR operates on the data from devices worn by the same person that record barometric pressure and acceleration - inexpensive, low-power, and thus commonly included sensors in today's wearable devices. In its first stage, BMAR verifies that two recordings were acquired simultaneously and pre-aligns all data traces. In a second stage, BMAR refines the alignment using acceleration measurements while accounting for clock skew between devices. In our evaluation, three to five body-worn devices recorded signals from the wearer for up to ten hours during a series of activities. BMAR synchronized all signal recordings with a median error of 33.4 ms and reliably rejected non-overlapping signal traces. The worst-case activity was sleeping, where BMAR's second stage could not exploit motion for refinement and, thus, aligned traces with a median error of 3.06 s.

DOI: 10.1145/3596268

Transcript

Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.

Rosa: We're looking at the paper: "BMAR: Barometric and Motion-based Alignment and Refinement for Offline Signal Synchronization across Devices".

Dev: Here's the gist of the paper. BMAR is a novel method to synchronize and align signals across devices without the need for specific user input, action, or explicit synchronization through wired or wireless communication (e.g., WiFi or BLE) BMAR is capable of synchronizing (a) independently recorded signals after the fact by (b) first pre-aligning recordings using air pressure as an inexpensive sensing modality that simultaneously allows us to reject non-simultaneous recordings (c) In a second step, BMAR produces a refined signal alignment across sensor devices by cross-correlating accelerometer observations BMAR synchronizes signal traces individually recorded by wearable devices in two stages: a pre-alignment based on air pressure observations and a refinement step that processes accelerometer observations In Stage 1, BMAR derives a pre-alignment of signal traces based on air-pressure observations, simultaneously leveraging these measurements for the rejection of non-concurrent recordings Both are possible because air pressure recordings are unique over time but near-identical across all devices worn on the same body In Stage 2, BMAR refines alignment based on accelerometer measurements by cross-correlating accelerometer observations This refinement accounts for clock skew between devices and computes these computations for patches of data Signal Alignment Techniques The paper discusses various techniques for aligning discrete finite signals, including cross-correlation, cross-covariance, and signal delta Cross-correlation is used to determine the best alignment of two functions by finding the lag m0 of g that maximizes Rf g The quality of this approximation decreases at the edges with shorter window lengths l when m is close to Ng or-Nf Signal delta is used to align sensor signals with a long characteristic timescale in scenarios where the sensors have recorded the same underlying signal at the same time bar A loss function Ld is formulated which can be minimized to find the lag m0 that aligns two sensor traces the best Evaluation Results The evaluation showed that BMAR synchronizes signals with a median error of less than 40 ms during sessions with activity This is well below the period of human activity and physiological signals The worst-case activity was sleeping, where BMAR’s second stage could not exploit motion for refinement and, thus, aligned traces with a median error of 3.06 s For active sessions, the synchronization achieved a median alignment error of 33.4 ms for recording durations of 15 min and longer Contributions and Limitations The contributions include BMAR, a novel method for signal synchronization that processes recordings offline through a pre-alignment and refinement stage without relying on assumptions about signal content Particular support is given for wearable low-power platforms requiring no wireless connectivity or communication during operation A limitation noted is that the improvement in refining synchronization based on accelerometer signals depends on the signal characteristics, and if recordings contain no events or asynchronous patterns, our refinement may worsen the result of the pre-alignment BMAR is currently designed to synchronize signals captured on devices worn by a single person BMAR leverages barometric pressure sensors and accelerometers, which are inexpensive, low-power, and commonly found in wearable devices BMAR’s two-stage process comprises pre-alignment, which we use to avoid latching on to local but not a global optimum in alignment Our method achieved accurate synchronization with a median error of 33.

Paper summary: Rosa: So we're looking at this paper called "BMAR: Barometric and Motion-based Alignment and Refinement for Offline Signal Synchronization across Devices". The authors are Manuel Meier and Christian Holz from ETH Zürich, Switzerland. Basically, they’re tackling the problem of getting signals from different wearable devices to line up accurately when you look at them later.

Dev: Yeah, it sounds like a synchronization method that doesn't need any wireless connection while the devices are running. What's the main idea here for what this paper is proposing?

Rosa: The core idea is that BMAR works purely on recorded signals and doesn't need any wireless communication between devices during runtime, which makes it good for offline processing. It uses barometric pressure and acceleration, which are pretty cheap and low-power sensors you find in wearables.

Dev: So it’s using those two things—air pressure and motion—to figure out when the recordings actually happened together, right? And what's the big claim they make about how it works?

Rosa: In its first stage, BMAR verifies that two recordings were acquired at the same time and pre-aligns all those data traces using air pressure. It uses that pressure because it changes over time but is pretty similar across all devices worn by one person.

Dev: That sounds like a clever way to filter out recordings that just didn't happen simultaneously, which is important for filtering out bad data right away. What happens after that initial alignment step?

Rosa: Then in the second stage, BMAR refines this alignment further using the acceleration measurements while also accounting for any clock skew between those devices. They cross-correlate those accelerometer observations to get a better fit.

Dev: So it’s not just one step; they have a two-stage process for getting that final signal alignment, focusing on using different sensor strengths sequentially. That sounds like it handles some real messiness in the data collection environment.

Taro: I'm curious about how robust this is when things go wrong, especially since they mention offline processing. If you can't communicate during runtime, what happens if the world misbehaves and you need to recover gracefully?

Rosa: Well, their evaluation involved three to five body-worn devices recording data for up to ten hours during various activities. They found that BMAR synchronized all the signal recordings with a median error of thirty-three point four milliseconds during sessions with activity.

Dev: Thirty-three milliseconds sounds pretty tight, especially when you think about how fast human movements are—that's less than one hundredth of a second. What were the caveats they found, like where it might fail?

Rosa: The worst-case scenario was sleeping. In that situation, BMAR’s second stage couldn't use motion to refine the alignment at all because there was no movement to exploit. So in that case, they aligned traces with a median error of three point zero six seconds.

Paper summary: Dev: three point zero six seconds is a big difference from thirty-three milliseconds, so the refinement really depends on having some kind of signal change happening. How much does this mean for someone who just wants to track their activity without worrying about perfect synchronization?

Taro: It means if you're sleeping, the system defaults to a much coarser alignment because it can't rely on motion patterns for that refinement step. You have to accept that trade-off in accuracy based on what the sensors are actually picking up during that specific time.

Rosa: Exactly. The authors note a limitation: the improvement they get from refining with accelerometer signals depends heavily on the signal characteristics. If you don't have those events or asynchronous patterns, their refinement step might not help at all, and in some cases it could even make the result worse than just using the initial air pressure alignment.

Dev: So BMAR is designed to be a powerful offline tool for single-person tracking that relies on the physical properties of air pressure and motion, but it has this dependency on having meaningful patterns in the data for its second stage to shine. What's the bigger picture here?

Rosa: It’s about making sure that even when you are collecting data without constant wireless links, you can still get a reasonably accurate picture of what happened across multiple devices worn by one person. This is relevant for things like health tracking or motion capture where you need to combine different sensors' perspectives.

Dev: So to wrap up this paper on "BMAR: Barometric and Motion-based Alignment and Refinement for Offline Signal Synchronization across Devices", it’s a method that uses air pressure for a first pass, then uses acceleration to fine-tune the alignment while managing clock skew. It achieves a median error of thirty-three point four milliseconds during active sessions but can degrade significantly when the data is stationary like sleeping.

Taro: The implication for us is that if we're building systems that need this level of offline synchronization, we have to design our systems to handle the possibility that one of those refinement steps might not yield significant improvement depending on what the user is actually doing.

Rosa: Right. It’s a method built on simple, inexpensive sensors like barometers and accelerometers to achieve a level of accuracy that beats just assuming everything happened at once without any sensor input. That's what BMAR does, and it’s all about getting those recordings lined up for later analysis.

Dev: So we see the authors proposing this two-stage approach using pressure first, then motion refinement, which gives us a baseline for how to tackle offline synchronization problems across multiple sensors worn by one person.

Conclusion: Rosa: So they’re looking at how to get those recordings to match each other after the fact using just barometric pressure and movement. That’s the core idea here, right?

Dev: Yeah, it’s about taking those two simple sensors—pressure and acceleration—and using them in a two-step process to align all the data traces offline. It avoids needing any wireless communication during runtime, which is big for low-power stuff.

Taro: The methodology they use involves first pre-aligning everything based on the air pressure readings, which helps them reject recordings that just didn't happen at the same time. Then they refine that alignment using the acceleration measurements to fix any clock skew between the devices.

Rosa: And when they tested it, during active sessions, BMAR achieved a median error of thirty-three point four milliseconds for recordings longer than fifteen minutes. That’s pretty tight, especially considering how fast things move during activity.

Dev: Thirty-three milliseconds is low enough that it’s below the typical period of human activity and physiological signals. But they also found a limitation, which is important for the safety case, right?

Taro: Yeah, the worst-case scenario was sleeping. In that situation, BMAR’s second stage couldn't use motion to refine anything at all because there was no movement to exploit. So in that worst case, they got a median error of three point zero six seconds.

Rosa: So the paper shows that while it works well for active tracking, you have to be aware that when things are stationary, like sleeping, the refinement step doesn't help much at all.

Dev: Exactly. The authors say the improvement from using acceleration depends on whether there are actual events or patterns in the data; if there aren't any motion events, their refinement might actually make things worse than just using that initial pressure alignment.

Taro: What this means for autonomy is that you can’t assume you’ll always have meaningful motion data to fix the timing later. You have to design your system to handle that lack of refinement gracefully.

Rosa: So in conclusion, BMAR is a practical approach using cheap sensors like barometers and accelerometers for offline signal synchronization, but it has these clear boundaries based on what the user is actually doing during the recording.

More episodes

← Home