Sys.Log

A DIY open source SpeedCoach

Tue, 01 Sep 2026 00:00:00 GMT

To see your speed and stroke-rate for rowing on the water, there is really only one device that is commonly used. For someone that knows a bit of hardware/electronics, etc., the price (~450€) seems a bit off for what this can do. In the end, it is just a waterproof box for a GPS and IMU that can record and show your speed and stroke-rate, and can also set specific intervals, etc. Basically the head unit of an ergometer but for the water.

This is an attempt to rebuild this from cheap electronics parts.

How does the NK Speed Coach work?

There are no available documents about the tech of the Speed-Coach on the Internet, so I had to guess how they solved this. The UI shows a GPS signal strength indicator, so it's obvious that they have a GPS built in. This is used to get the speed, distance, and accurate time for the device. Therefore, most of the metrics and functionalities are solved by this.

The last metric that is a bit harder to capture is the stroke-rate. For this, there are multiple options on how they do this. Some boats have a magnetic sensor mounted between the rails. This means that the seat is moving across that sensor twice per stroke. This is a very simple and robust way of measuring the stroke rate that is certainly used. But the Speed-Coach can also measure stroke-rate if such a sensor is not mounted/connected. The other option to measure this would be to use the cyclic acceleration and deceleration phases during each stroke to measure the stroke-rate. This is most likely done using an IMU. However, the boat can roll, get affected by waves, etc., which is unwanted noise when trying to isolate just the forward acceleration of the boat. Since the IMU is not always exactly mounted in the same way, and the rower does not have to do an advanced calibration movement/phase before each row, it might not always be clear what direction is along the boat movement, which is needed to define the boat acceleration. As I show for my own device later, this is possible using rolling window filters and some knowledge about the rowing motion. However, there could be another option how this is done, which would also explain the high price point. The GPS might be a GPS-RTK. Using this would make it possible to detect the movement per stroke using the GPS. That would be much more robust to the orientation of the sensor and other movement of the boat. It might be possible that both sensors are fused together (Sensor Fusion), which would be the most robust. Given that the rate of those devices is very steady and accurate as well as fast to respond, this is entirely possible.

For my Speed Coach, I am settling on a cheap GPS module and IMU. Other options I considered were using an analog circuit to detect peaks in the sound of the boat hull using a piezo contact microphone. Although this would certainly be the most fun and most challenging, it would most likely also be very unstable because of other noises, wind, and voices in the boat. Another idea would be to measure the magnetic field in such a sensitive way that the movement of the rower, oars, and the seat can be detected from any random position the device is mounted at. This would certainly be very impressive but likely not very robust, as well as probably power-hungry.

IMU-based stroke detection

To get the stroke rate using the acceleration data, we first need to identify the prominent axis of forward movement. As the forward acceleration is the prominent acceleration, we can do this by finding the vector with maximum variance over a given time frame. First, the values are smoothed using an exponential moving average. The values of the identified axis of movement are then used for a simple peak detection.

This is done by tracking the min and max values of the signal over a rolling window. These define the amplitude of the movement, which is used to set a 25% signal threshold. Every time the signal crosses this value, a new stroke is counted. Since the signal is not always perfect, a debounce window is used to ignore any second crossings for a time of 900ms.

Finally, the time between detected strokes is converted to the typical sec./500m unit, and also smoothed again to prevent quick, unreasonable changes in rate. The smoothing leads to a quite slow update time, which in the future should be reduced a bit. But since this version already uses an e-paper display, fast update times are out of the question anyway.

The following figure shows an excerpt from real rowing data and all the features detected:

Hardware

This is the hardware I am using:

NiceNano nRF52 supermini
generic cheap GPS module
MPU6050 IMU
2.7" e-paper display
Micro SD card
USB-C (for charging only)
2000mAh battery
waterproof switch
waterproof USB-C port

I decided to use the NiceNano as they are cheap, have a small form factor, come with a battery charging logic included, and also have Bluetooth. All in all, it has all the functionality I need for a very competitive size and price. Since this is just a prototype, I didn't want to spend much money on a quality GPS. After all, the goal is to make this as affordable and replicable for everyone at the cost of accuracy, which for normal training is fine for me.

The e-paper display was chosen because it provides great readability under sunlight outdoors as well as very low power consumption. In hindsight, the refresh rate is a bit too slow, and using only partial updates over a long period of time leads to washed-out and corrupted numbers on the display. The real Speed-Coach uses a custom segment LCD for this. They provide good readability and excellent power efficiency as well as fast refresh times. The only downside, and the reason why I could not use this, is that they are custom and fixed to the exact layout of the display. So they are reserved for companies that will actually sell thousands of items.

Here are some more pictures of the pretty crammed and not tidy hardware as well as the main screen:

Comparison

The following video shows a side-by-side display of the NK Speed Coach as well as my DIY one. It is clear to say that the update time for mine is way higher, and due to the heavy use of rolling avg filters along the processing pipeline, there is a second-long lag in the update. Also, the accuracy of the pace is not spot-on, which is due to the cheap GPS not finding the real position of the boat.

All in all, I think this shows that the concept of using cheaper parts is definitely possible, and some better tuning could make this a very stable platform for further development.

The Code for this project is available here

How much wind shelter does a snow wall give your tent?

Tue, 18 Aug 2026 00:00:00 GMT

High-altitude and polar winds can intensify quickly. Building a snow wall protects your tent by blocking the wind. It reduces fabric flapping, increases wind tolerance, and helps retain heat.

There are several excellent sources that discuss sleeping in snow and all its peculiarities:

These sources state that the snow wall should be between two-thirds and the full height of the tent, and placed at least two meters upwind of it. In this article we put every combination of a snow wall and a tent into a simulated wind tunnel, to predict the best snow-wall height and distance for each tent type and size.

TL;DR: Build a snow wall roughly 60% of your tent's height and pitch the tent about 1.5 or 2.8 m away from it. The best setup differs a lot between tent types and sizes though, and is quite sensitive.

Interactive 3D wind tunnel

The first test I ran is a 3D simulation for the Staika from Hilleberg. The fluid simulation is run for different wall heights between 0 m (no snow wall) and 1.2 m as well as different distances from the tent (0.5 m to 5 m). It lets you interactively drag the snow wall and the tent to get a feeling for the wind dynamics at play. You can clearly see that wind velocity and air pressure drop behind the wall as it provides shelter. The plots also show that a tall enough wall, combined with a tent close enough to it, can hide the tent fully in this low-pressure zone. That is unfavorable though, since it would lead to snow accumulating behind the wall and burying the tent. If the tent is placed too far behind the wall, the wind hits it at full speed and the wall has no effect. Furthermore, there are special combinations of wall height and tent distance where a lot of turbulence builds up between wall and tent, while other combinations lead to much smoother airflow. For example, set the wall to about 0.5 m and slide the tent towards it. You will see that at a distance of 4 m and higher you will observe smooth airflow in the 3D scene, while it is very turbulent at a distance of 3 m. This shows that the optimal snow wall placement is not trivial and slight changes in setup can lead to big differences, due to the turbulence and vortexes building behind the wall.

Now is probably also a good time to mention that this is a simulation, and I am no expert in fluid dynamics. Take all results with a grain of salt, as they provide a "feeling" or an estimate rather than the hard truth. In reality, a lot of parameters influence the results, and a real-world setup will never be perfect or match a wind tunnel. For example, snow flakes carried by the wind, temperature, viscosity, or snow build-up over time all have an influence.

Measuring the force on the tent

Next we plot the measured forces on the tent for each wall height and distance, always using the best possible value of the other factor and averaged over all tent shapes. So for the two left plots we find the best wall distance for each wall height and show the median over all tent shapes of the resulting force. We differentiate between dome and tunnel tents. The black dotted lines show the force measured in the tent without any wall. This shows that building a wall is almost always helpful if the wind is strong. In our simulations we observe up to a 68% reduction in force on the tent fabric. The blue line shows the median force on the tent for different setups, and the pink dashed lines show the measured forces for one specific common tent size as an example. Since this data was obtained from a much faster 2D simulation, the force is given in Newton per meter of tent. It can be interpreted as the force acting on the fabric, therefore the lower the value the better. From this we can see that the lowest forces on the tent are observed for a snow wall that is about 0.6 times the tent height. It is clear that a wall that is too low leads to higher forces, but a wall that is too high also increases the forces due to turbulence. For the wall distance the results are less clear. For dome tents the optimal distance averaged over all tent shapes is about 4 m, however our single tent example shows that placing the tent closer to the wall is generally better. However, snow deposition is not considered here, so a minimum distance of 1.8 m should not be crossed in practice. For tunnel tents the same trends hold, although for our example shape we can see some oscillation in the tent force. There are local minima at 1.5 m, 2.8 m and 4 m. This indicates that vortices from the wall are hitting the tent, and placing the tent strategically can decide between a good and a bad setup. Based on these results I would recommend pitching your tent about 1.5 or 2.8 meters behind the wall (but not something in between!), depending on how much snow is in the air. Placing it further away reduces the shelter effect, while placing it closer risks snow burial.

Customized Predictions

Last but not least we can use this data to build a simple prediction for each tent size and type. The little applet below lets you input a tent type and size, and based on the simulation data and a fitted interpolated curve it provides the best snow-wall setup.

Note that for tunnel tents longer than 3.5 meters this oscillating effect is very pronounced.

Building a DIY ultra light versatile tent

Tue, 28 Jul 2026 00:00:00 GMT

When doing any type of self-supported outdoor trips I am generally looking for the most compact and lightest solution. Unfortunately packing size and weight are among the things that let the price for sleeping-systems skyrocket. Since I already have a number of tents for various conditions I didn't want to get another one. That is why, two years ago, I decided to build my own one.

Design

My requirements for the tent:

As small as possible in packed state.
As light as possible.
Fully rain proof and somewhat weather robust.
Good ventilation (I hate bivi-bags).
Glexible and adaptive so I can pitch it in forests, on hard rocks, and et it up for various levels of weather protection.

These are the features I ended up with:

Flysheet should be just a tarp. That way I'm super flexible in how to set it up and where to mount it to.
Needs a pole, so I'm not reliant on trees etc. This can also be replaced with a tracking pole in case it's already brought along.
Fully closed inner liner with insect mesh.
Some form of vestibule to store stuff in a dry place.

So I came up with this design which is inspired by a typical tarp setup:

Specs

Here are the dimensions I used. It's just about enough space to be able to sit somewhat upright (useful for changing shirts/jackets etc.):

The loops for guylines or pegs are always halving the side length they are on. I added 3 loops per side but only on one side per opposite pair. This way I'm saving loops, sewing and structural weakness points in the fabric.

All lops are doubled sewed and around them I also used double layers of fabric to distribute the load better. Since I couldn't buy the silnylon in these dimensions there is a seem going across the tarp. I waterproofed that one with special silnylon glue. The seam turned out to be a bit ugly but very watertight and functional.

Including the packaging this is how much everything weighs:

Tent: 290g
Tarp: 300g
Pole: 79g

In total the sleeping system weights 669g. For a tent that is not even using dyneema and (partly because of that) costs a fraction of the typical price for tents in that weight class, I think this is a very good weight and design.

The whole setup in it's packed state looks like this:

The material can be ordered from shops like Ripstop-by-the-roll or Extremtextil. For materials I decided on:

Silicone-Coated Ripstop Nylon for the flysheet, and some parts of the inner liner
- 20den, 36g/qm
Heavy-Duty PU-Coated Ripstop Nylon for the floor
- 10.000mm hydrostatic head, 90g/qm
Ultra-Light Polyamide Mesh for most part of the inner liner
- 155 meshes/qcm, 25g/qm
EASTON Aluminium 7075
- 40.6cm and 23g per segment, outer diameter 9mm
- As the poles are so short, using ligher material is not worth it
and these clamps to tighten guy-lines

Conclusion

After 2 years of bringing it to various trips I can say that I am super happy. For the price and effort I put into this, the tent holds up really well and I can reliably trust it also in bad conditions. The only thing I would change is the type of fabric for the part at the feet of the inner liner. On very humid nights this is too rain-proof and it traps condensation water at the end of the sleeping bag. A different more breathable fabric would have been good here. The tent-pole construction is a bit fiddly, but I got used to it and now setting it up is super fast. Since there is just one vertically standing pole it is not the most stable construction but so far it held up very well (also tested in quite strong winds, snow, rain, ...). The stability is very dependent of the pole having a god contact with the floor. If the floor is very hard (angled rocks) or slippery, the downwards pressure on the pole can lead to it sliding away (never happened to me so far though).

All of these are compromises I'm willing to take for the ultra light and ultra compact setup that this tent is.

Example Setups

Using a more open V-shape:

A fully closed down rain proof setup:

Only fixed in place by rocks:

Selecting my Analog Camera Lineup

Sat, 04 Jul 2026 00:00:00 GMT

Why Analog

Shooting analog film enforces discipline. Having a restricted number of photos makes each shot valuable, and I tend to think before taking the photo. I noticed that for me, this leads to higher quality compositions and more meaningful pictures most of the time. Not seeing the pictures until months have passed and the film is developed really makes me rediscover the moments and trips I've been on, which is a nice extra. It makes photography a skill again, which is a welcome challenge on vacations and trips for me.

Initial Selection

For me, I knew that my camera had to be somewhat compact and sturdy, as I want to take it on outdoor trips as well as just drop it in some bags without having to worry. This is why I settled for rangefinder cameras instead of SLRs. They offer the compactness I need while still providing enough settings and adjustments to differentiate them from simple automatic point-and-shoots. That way, I keep control of the picture and can also take more complex shots (long exposure, special light conditions, etc.).

Here is a list of initial candidates I considered after a bunch of research on YouTube and the web:

Good sites to get an overview of the different models are:

Half-Frame

I considered half-frame cameras like the Olympus Pen / Pen FT and Minolta Repo for quite a while. They yield 72 shots per roll, lowering the cost per photo. I would argue that I will most likely never develop my film in such a high resolution that 35mm is really needed, and also I do prefer and embrace the grainy film look and lower resolution. However, half-frame models lack the classic aesthetic of standard rangefinders, and there are just not enough models to choose from for me. I didn't find the half-frame camera with all the features I really wanted, which is why I decided against them. There are people modifying a 35mm camera like a Lomo LC-A for half-frame, which could be interesting for some of the other classic 35mm models.

My Criteria

To narrow the list, the camera had to meet specific operational targets:

Compact rangefinder with a silver finish (it just looks so cool).
Fully manual controls for shutter speed, aperture, and focus.
A light meter and fully automatic mode available for quick action shots.
A viewfinder indicating accurate focus and composition.
A light meter that functions in both auto and manual modes (for some models, the light meter only works in auto mode. But I want to see if a scene has the right exposure and adjust my settings based on that).
Fully operational without battery dependence.

Closer Selection

Before arriving at my final three, here are the cameras I rolled out and why:

Rollei 35: Too small, almost too hard to use, and the controls are weird. The build quality is very good, but it's a tradeoff between compactness and image quality, and I want higher quality. There is also no auto light meter, and focusing relies on zone focusing (guessing the distance). I want to be exact and precise with where I focus to.
Canonet QL17 GIII: I had this camera for a while and liked it a lot, but it is missing some manual controls. I want full control over all settings (shutter speed, aperture).
Konica Auto S: It has a shutter priority mode. This means the user sets the shutter speed, and the camera sets the aperture. I don't like this, as I care more about the aperture to control depth of field. For shutter speed, I usually just care that it's fast enough so the picture isn't blurry, meaning any value above that threshold is fine. That's why I have more constraints on the aperture than the shutter speed.
Fujica GA: Does not have the full manual mode.
Fujica 35 ML: This camera does not have an automatic mode.
Fujica Compact Deluxe: I don't like the design that much, especially the film advance lever and the viewfinder. It also doesn't seem as robust, and I just didn't vibe with it. Other models offer the same capabilities at a similar price point, and I liked them more.
Yashica Electro 35 GS: This camera cannot be used in full manual mode.

With those ruled out, three cameras met almost all of my criteria:

1. Konica C35 FD / Auto S3
Features a bulb mode and can be hacked for manual control. However, it relies entirely on a battery, even when shooting manually.
Reference: Konica Auto S3

2. Olympus 35 RC
Includes a bulb mode, and the light meter functions in manual. The downsides: shutter speed must always be set manually, and the aperture ring is positioned too close to the body, making it difficult to operate (especially in colder conditions when wearing gloves).
Reference: Olympus 35 RC

3. Minolta Hi-Matic 7sII
Features bulb mode, fully automatic operation, and fully manual capability. The only compromise is that the light meter turns off in full manual mode.
Reference: Minolta Hi-Matic 7sII

The Final Verdict: A Two-Camera Setup

After reviewing the options, the Minolta Hi-Matic 7sII is the best choice for me. It offers solid build quality, robust mechanics, and excellent optics. It provides the flexibility of fully automatic shooting or full manual control when there is time to compose a shot carefully.

However, after shooting with it for a while, I noticed that it is a bit heavy and still too large for my needs. Especially for ultralight outdoor trips, the Olympus XA was really appealing. It has a built-in cover, still provides all manual settings, a light meter, and good optics. I tested it for a while, but noticed that the picture quality is noticeably lower than with the Minolta. I still like it a lot, but most of my best pictures have been shot with the Minolta.

So, in the end, I landed on a dual setup. Most of the time, I pick the Minolta, but for some special occasions, the sturdiness and size of the Olympus are a big advantage. Having both options means I never have to compromise between image quality and true pocketability. Whether I am packing light for a rigorous hike or taking my time on a dedicated photo walk, this combination ensures I always have the right tool for the job.

FIFA WC 2026 Hydration Break Study

Thu, 02 Jul 2026 00:00:00 GMT

Following the FIFA World Cup 2026, I noticed that most people don't like the hydration breaks. My feeling is that they slow the gameplay down and break the momentum of the winning team.

So here is a data-driven analysis of the hydration break.

The hydration break is called by the referee roughly after the 22nd and 67th minute. It still counts as normal gameplay time and takes about 3 min.

The figure above shows the match momentum of the play for the full length. We can clearly see dips in momentum for each hydration break as well as the half- and full-time breaks. The momentum is shown in absolute value here, so a dip in momentum means that the team with the upper hand loses its advantage and the game is more balanced. We can also see that the extra time is mostly very chaotic with a lot of changes in momentum.

For the half- and full-time breaks, we can also see that the momentum just before the break increases. This makes sense as the teams give all they have just before the break to decide the game here (and they know they can rest soon). The same can not be observed for the hydration breaks because these are not pre-defined times but rather unexpectedly announced by the referee.

So judging from this, we can see that the hydration break has a similar impact on the momentum as normal half- and full-time breaks.

Instead of looking at the momentum, we can also look at the number of events like goals, assists, and cards:

From this figure, we can see that there is a difference between the hydration and normal breaks. As previously noted, we can see an increase of events towards the end of the 90 minutes. We can also see that teams are more likely to score goals just after the half-time break. Presumably because the coach gave them valuable tips, or they are just recovered and know what to expect in the opponent's style of football.

The hydration breaks, however, see a significant drop in events just after the break. This shows that compared to the normal breaks (they don't see this drop), the hydration breaks really disrupt the game flow and lead to fewer events happening.

Also interesting to see is that the first hydration break does not lead to fewer goals shot just after it; however, the second hydration break has a big drop in goals right after. This difference in effect of the first and second hydration break could indicate that players get more tired over the time of the match and therefore are more influenced by game interruptions.

I couldn't find any free and available data on the exact moment the hydration breaks were called. In my experience, they are not called exactly at minute 22. So there might be some noise in the data of games that had the break at minute 28 instead of 22, for example.