Product Update - March 2018
This month's update is pretty dense, and includes a video update from our Founder. Enjoy!
Jacob Flood, March 19, 2018
Coming off of Chinese New year, we’ve had a very productive month on product development back in Montreal. Having both the hardware and software teams working under the same roof on test protocols sped up some of the decision making that’s difficult to coordinate when we’re twelve time zones apart!
We focused this month on two very important mechanical developments, alongside some crucial signal quality testing on electrical - these make up the crux of this update. We will also address a few comments we’ve been getting our backers regarding comfort, SDK, beta units, and timelines.
This update has a ton of great content. To start, we have a short video summary from our Founder Jacob, about the current state of the company:
In our previous update, we talked at length about designing different aspects of the headphones – in particular, how we were optimizing for comfort and fit on the head. This month we have evaluated the results of this work, testing the headphones on the users’ head. The test had two components: a virtual one and a real one.
For the virtual test, we used MRI data to generate 2D slices from 3D renders of more than one hundred heads, to test the mechanics of the headphones as they open, close, adjust the arm length, and pivot the ear cup angle. In our Computer Aided Design (CAD) software, we were able to evaluate the quality of the fit on each head, determining how well the headphones would sit, and the effect this would have on the sensor contact.
The second test involved creating physical head cut-outs from rigid foam, over which we tested the real headphones. This physical comparison was used to validate the virtual model without having to rely on each of our subjective opinions – we use behavior in the real world to confirm the accuracy of the simulation tests, which we then rely on for design optimization. This saves us a lot of time, allowing us to quickly tweak various subtle parameters in the headphones, without needing to reiterate on the physical prototypes.
2D Virtual Fit test
One of the most difficult (read: exciting) part of developing a product as a start-up is finding cheap, fast, yet effective solutions to the various challenges. In order to compete with larger corporations on speed and quality, despite a fraction of the budget, we need to be creative in the way we resolve problems. Often, these constraints ultimately lead to better processes - one such example is how we decided to pre-validate the fit of the headphones on the different user’s head.
We used an open MRI data bank of brain scans to model both in 2D and 3D the shapes of many heads. This allowed to do two things:
- Obtain statistically significant range expectations for the height, width, and curvature of a normal head
- Use these model heads to validate the design tweaks we make on the headphones virtually, without having to physically build prototypes every time we tweak a parameter
Several datasets are available online that discuss normal ranges for head size in terms of two parameters: height and width. Most headphones are optimized for these parameters – if it fits on your ear, and touches your head, the headphone is considered “good enough.” The problem is that this simplification doesn’t capture the variance in the curvature of the top of your head and leads to designs where the headphone contacts your scalp at only one point. Not only is the resulting weight distribution uncomfortable, but this wouldn’t allow our sensors to read a proper signal.
For this reason, we created a model that takes head curvature into account. We used the publicly available datasets to validate our MRI data, and constructed our own database from the results. The output from our head shape analysis allowed us to formalize the constraints that ensure that the headphones fit on everyone’s head comfortably, and in a distributed way.
To measure using MRI-generated drawings, we modeled a simplified 2D version of our latest headphone. The model is parametric, implying that it can be opened and closed just like the real headphones. This allows us to test how the headphones would fit on the various heads we modeled. You can see below what it looks like to fit the headphones virtually to a head.
Fitting the headphones allowed us to measure the gap between the top band electrodes, and the surface of the head where they should contact. The goal is to minimize this gap for all heads to avoid needing complex adjustment mechanisms within headphones, while ensuring the electrodes all contact properly to get a good signal.
Results from 2D Virtual Fit Test
Using the models described above, we were able to validate the overall fit of the headphones – we made a few mechanical tweaks from the previous design, and we’ll be validating those in our upcoming iteration. The questions that remained related to the fit of the electrodes on the head.
We already knew from previous experiments that fitting 3 electrodes to each person’s head was a difficult challenge – in our previous updates, we’ve described solutions we came up with to address these issues. At the core of the problem lies a geometric law: only one circle exists that will fit 3 points perfectly.
Given this understanding, over the past few months we designed what we refer to colloquially as the “electrode mechanism”, which flexes to adapt the electrode position to individual’s head shapes. Theoretically, this solved the problem – the mechanism in our latest prototype works, comfortably contacting people’s heads when worn.
In practice, this turns out to be less-than-ideal. The fit tests described above determined that the required movement from the mechanism in order to adapt to all head shapes was very large – almost exceeding the overall thickness of the top band of the headphone. The result was that the bottom half of the band became very thick, as seen in the picture above.
The other issue is that the mechanism turns out to require many moving parts, making it very complex to assemble. Reducing assembly difficulties is a huge priority in our current design stage – these issues can creep up and cause catastrophic problems later in production. The mechanism, in its current form, is a design risk that we aren’t comfortable taking.
Due to this complexity, we brought back an old idea of using two electrodes instead of three. In all honesty, we were nervous about the idea - the amount of effort already put in the design, and the possible reduction in signal captured from one-fewer electrodes, seemed like potentially huge issues. Having the hardware and software teams together in Montreal this month, we decided to test these premises, rather than assume conclusions.
Firstly, we looked at the mechanical. Using two electrodes on the top band – removing the center electrode – solves the geometric problem described above. It’s possible to fit two points to an infinite number of circles, making it easier to adapt the upper band to different head shapes. This implies that by using two electrodes, we wouldn’t need the mechanism at all – a pivot-like adjustment on each electrode would suffice to make a quality contact for each individual.
In parallel, distributing the weight of the headphones over two electrodes instead of three would result in a more even contact on each electrode as well. Since electrode contact correlates directly with signal quality, we could get two better-quality signals instead of three lower-quality ones. This premise gave us renewed hope in the idea.
Following this, we turned to the software team. Over the last year, we have collected a significant amount of data using three electrodes and developed our concentration algorithms around that. To validate the idea, we decided to re-run our algorithms, removing the data from the center electrode, and verify how this affected the ability for our software to predict a user’s concentration. The result, it turns out, was that removing the center electrode had a very insignificant effect on the quality of the predictions.
In retrospect, the conclusion makes sense. EEG has notoriously poor spatial resolution – two electrodes close to each other capture practically identical signals. When we ran a comparison analysis, it turned out that the center and side electrodes were very closely correlated – they contained almost identical information about the user’s state. Given the modern artificial intelligence algorithms we use to identify neural states, this redundant information provided very little overall value.
The final conclusion is that we have decided to remove the center electrode from the design. This change dramatically reduces the overall manufacturing cost and complexity, while keeping the functionality intact. The resulting fit of the headphones on user’s heads is dramatically improved, and the product is less likely to encounter issues in late-stage production that could slow us down.
Making the three electrodes fit through the complex mechanism shown above has been our greatest hardware risk for the past few months. Having found a simple, effective solution, we are now extremely confident in every facet of the headphone design.
Once we settled on using two electrodes, our team sat together to rethink the optimal placement of the remaining two electrodes. The goal is to facilitate the mechanical fit, while recording over the largest portion of the brain to capture as differentiated data as possible. The result of our analysis has shown that the point of diminishing return for proper mechanical fit and signal differentiation across the two top electrodes ends up being midway between C3/C1 and C4/C2 for the two electrodes. Closer than that, the signal between the two electrodes become too similar; farther than that, the gap between the side electrodes and the head becomes too large.
With two electrodes, only the angle – rather than the position – of the electrodes needs to change to adapt to heads of different curvatures. You can see below an isolated portion of the top band below showing the difference in how the electrodes are adjusted between a large and small head. You can see how the smaller head “compresses” the foam of the top band, while still touching the electrodes at an adjusted angle.
This solution provides a simple, yet effective way to maintain a very high signal quality at each electrode, while distributing the weight of the band on the head more effectively than before. We’re very happy about this change, and we can’t wait to get the latest prototype to test out the effects in-person.
While the mechanism was changed to incorporate the revised electrodes configuration, additional changes for proper functioning of the ANC were implemented. These include the addition of a proper seal, some tolerancing adjustments to avoid air leakage, and a slight increase in internal ear cup cavity volume. Each of these minor tweaks should yield major sound improvements, while causing no delay in production.
Tweaks for Comfort
We’re received many questions about comfort of the headphones concerning the way the headphones touch the ears, the interface material and so on. As we’ve mentioned, comfort is a major priority in designing the headphones. The following list describes a few of the design decisions we’ve taken to ensure that Mindset is incredibly comfortable:
- The design of the ear cups is circumaural (over-ear). This means that your ear entirely fits inside the earcups, without being compressed
- The plate covering the speaker is curved to be parallel to the ear, to avoid the external portion of your ear contacting any hard surfaces inside the earcup
- The material contacting your head was chosen to make sure it breathes well, dissipating heat in a more effective manner
- The weight of the headphones is evenly distributed on the top of the head, as described above.
- The electrodes were designed to apply the minimum force required to get a proper contact. In other words, the force between the electrodes and your head will not exceed a tested “comfort threshold”
- The clamping force on the ears is designed to be close to the minimum required for proper ANC function, in order to reduce the pressure on the head
- The ear cups are by default tilted to match the slight natural backwards inclination of your ear
- The ear cups can pivot along two axes to match the angle on the side of your head, where it contacts the scalp. This means that the ear cup will distribute pressure on your head very evenly.
- The adjustment of the arm bands is continuous, rather than discrete, to allow you to get it just right for you
- The buttons were chosen to avoid a loud clicking noise resonating in your ear when they are pressed
- The top band was made larger, and the radius was tuned to distribute its weight over a larger region of your head
With the recent unforeseen - albeit necessary - changes to the mechanical design, the tooling process has not started begun. We are finishing our final version of the engineering prototypes this month, which will become the standard against which we produce the tooling. As discussed in the video, based on the current development, it has become apparent that the scheduled shipping date is not realistic anymore. Our current timeline aims to ship the products in September of this year.
Our focus remains on shipping a high-quality product that exceeds the high expectations set. We are putting extra effort and taking steps to ensure that the product will survive in the market and provide as much value as possible. The hardware changes above are consistent with this belief – every design change is deliberate, and intended to improve the overall experience of the product.
Moving through the manufacturing process, it’s difficult to cut corners to ship quicker. Moving too quickly through validation steps often results in catastrophic problems downstream, that require entire redesigns. We’re trying to make today the difficult choices today that will ultimately bring more value tomorrow.
In addition to the mechanical complexity, the mitigation of small issues that creep up has also been a major difficulty. The complexity of a product (number of parts, assembly steps, manufacturing processes, tolerancing…) correlates directly with the potential for problems when mass producing. As a result, reducing this complexity early on has been a primary goal, to avoid bigger problems down the line. Similarly, since EEG requires very high-end electronics, we must be extra careful in sourcing and tracing the components, ensuring proper quality control at each step. This has involved significant due diligence, since small purchasing errors can yield an entire batch of counterfeit chipsets. Finally, because the product was sold in more than 80 countries, and directly interfaces with your body, we need to be extremely careful in ensure that all safety requirements and certifications are met. This involves additional design robustness, which take time to perfect. While none of these issues have been a significant impasse, dealing with each step, as a small team, has taken longer than we anticipated.
While it’s always difficult to make major changes this late in the process, we are not willing to compromise the quality and functionality of the product. The whole team is hard at work tackling every aspect of the product to mitigate all risks. As the time-cost of making changes increases exponentially the closer we get to production we’ve made the conscious decision to fix these issues immediately, by adding an additional prototype iteration to our timeline pre-tooling.
As our main supporters, none of this would be possible without you. This is exactly why we are committed to handing a device to you that will exceed your expectation. While we aren’t happy with the delay, we firmly believe that this is the right decision to deliver on our promises of providing you a took that will foundationally improve the way you work. The result will be worth the wait.
In parallel with the mechanical changes, we’ve been continuing the testing process for our electrode, electronics, and software pipeline. Moving closer to the production, we wanted to design a more robust and meaningful ways of validating the quality of the signal stack - both to quantify the effects of design changes, but also to set a quality baseline for production, to ensure that all units meet the requirements for proper functionality.
Together, the hardware, software, and neuroscience teams have designed, and are currently running a test that evaluates every prototypes’ ability to record and use the brain signals to predict concentration. This is a full-stack test that uses each component of the final product together, including the software pipeline.
To test the units in a real world setting and validate its efficacy, we test it with real users: we ask several users to participate into a data acquisition experiment designed to elicit a certain response in the brain. We then evaluate how well the prototype can recognize this specific the brain response. This becomes the metric against which we compare Mindset with a gold standard (usually, a medical grade EEG system). We will talk more about the results of these tests in our next update, once they are completed.
In addition to internal development tests, we have also been running an academic study with McGill University, through the Montreal Neurological Institute, on measuring attention while driving, and during a computer task. This study provides us with a high-quality dataset, against which we can compare our real-world data. We will be using this to evaluate the quality of our prediction algorithms, in order to make decisions about future product directions.
Finally, we are focusing a significant amount of attention to user experience testing, as we get closer to shipping the final units. Warren has spent the last few months putting together the final architecture for the app, through which we’ll be running our experiments with users both in-house, and in the wild. This will be a primary focus for the software team over the next few months – we’ll have more news on the results soon!
SDK AND EARLY BETAS
Lately, we’ve gotten several questions about the sort of data that will be available through the SDK. The goal of the SDK is to allow developers to collect and visualize real-time data coming out of the headphones, in order to build applications running on top on Mindset. This data will include some filtering done in hardware to remove power-line noise, as well as other un-wanted frequencies - the final details of these hardware filters will be provided soon.
Data can be streamed via both USB-C and Bluetooth, depending on your application. Details on the headers, examples of raw data files, transfer speeds, and sampling rates will be finalized once we start beta tests. In addition, we’ll be providing a Mindset Marketplace, through which you’ll be able to download apps build to be compatible with Mindset – anything from brain training for creativity, to solutions for faster learning, to pomodoro timers, and anything else you can think of. We’ve already had several developers reach out to share with us the apps they’re building to be compatible with Mindset, and they’re really exciting. If you’re a developer, designer, or just have ideas for an app, we’d love to hear from you – please send us a shout at firstname.lastname@example.org, and we’ll follow up to see how we can bring your ideas to life!
In parallel, while the final units are scheduled to arrive later, we will still be manufacturing sample units for beta testing. We really want to engage with our most hardcore backers – those of you who are interested in testing the headphones, helping us debug the beta versions and supporting the development of the SDK. We will be making a private Facebook group for beta users to engage with us over the next month. If you’re interested in contributing in any way to this process, please contact us at email@example.com, and we’ll reach out to follow up!
We’ll be doing a Q&A over the next few days in the comments below – submit your questions, and we’ll answer them as quickly as we can! Give us your feedback, ask us any questions, or even share any stories you have about how you’ll use Mindset!
As always, lots of love,
- The Mindset Team