Hi, I'm Joren. Welcome to my website. I'm a research software engineer in the field of Music Informatics and Digital Humanities. Here you can find a record of my research and projects I have been working on. Learn more »
This short guide will help you set up a local certificate using Caddy as the webserver to provide local TLS certificates to be able to develop websites immedately using HTTPS. Having a local HTTPS server in development can help with e.g. debugging CORS issues, accessing resources which require a HTTPS connection, or trying out analytics platforms.
1. Configure your hosts file
If you want to use a domain name, you need to first add a line to /etc/hosts which, in this case, sets localhost to correspond to example.com.
echo "127.0.0.1 example.com" | sudo tee -a /etc/hosts
2. Configure Caddy
In a directory of your choosing, create a Caddyfile with the following content, it sets Caddy to automatically generate certificates on the fly for example.com or any other domain name. Perhaps you will need to trust the main Caddy certificate on first use:
Still in the same directory as the Caddyfile and the index.html file, run the following command to start the Caddy web server: caddy run
5. Trust the locally generated certificate
In macOS this means adding the local caddy root certificate to your keychain. It can be found here /data/caddy/pki/authorities/local/root.crt In other environments a similar step is needed.
6. access the Test Site
Open your web browser and navigate to https://example.com to access the test site in the command line: open https://example.com. If you inspect the certificate it should be issued by the ‘Caddy local authority’.
Power banks have become a staple for charging smartphones, tablets, and other devices on the go. They seem ideal to power small microcontroller projects but, they often pose a problem for low-current applications. Most modern power banks include an auto-shutdown feature to conserve energy when they detect a current draw below a specific threshold, often around 50–200mA. The idea being that the power bank can shut off after charging a smartphone. However, if you rely on power banks to power DIY electronics projects or remote applications with low current draw, this auto-off feature can be a significant inconvenience.
To address this issue, consider using power banks designed with an “always-on” or “low-current” mode. These power banks are engineered to sustain power delivery even when the current draw is minimal. Look for models that explicitly mention support for low-power devices in their specifications. If replacing a power bank isn’t an option, you can add a small load resistor or a USB dummy load to artificially increase the current draw. It works, but feels wrong and dirty.
For a previous electronics project I bought a power bank randomly. After a bit of testing, I determined that the minimal power draw was around 150mA, so I added a resistor to increase current draw. Only afterwards did I check the manual of the power bank and noticed, luckily, that there was a low-current mode. I removed the resistor and improved the battery life of the project considerably. If you want to power your DIY Arduino or electronics project, first check the manual of the power bank you want to use!
Edit: after further testing it seemed that the low current mode of this specific power bank still shuts down after a couple of hours. Your mileage may vary, and the main point of this post still holds: check the manual of your power bank. Eventually I went with a solution designed for electronics projects.
There is this thing that starts playing birdsong when it detects movement. It is ideal to connect to nature while nature calls. It is a good idea, executed well but it got me thinking: this can be made less reliable, more time consuming, more expensive, and with a shorter battery life. So I started working on a DIY version.
Vid: Playing birdsong when presence is detected with an ESP32 microcontroller .
The general idea is to start playing birdsong if someone is present in a necessary room. In addition to a few of electronics components the project needs birdsong recordings. Freesound is a great resource for all kinds of environmental sounds and has a collection of birdsong which was used for this project.
For the electronics components the project needs a microcontroller and a way to detect presence. I had a laser ranging sensor lying around which measures distance but can be repurposed to detect presence in a small room: most of the time, the distance to an opposite wall is reported. If a smaller distance is measured it is probably due to a person being present. The other components:
As is often the case with builds like this, neither the software nor the hardware is challenging conceptually but, making hard and software cooperate is. Some pitfalls I encountered: the ESP32 C6 needs USB CDC set in the Arduino IDE, the non standard I2C GPIO pins. Getting the many I2S parameters right. Dealing with a nasty pop sound once audio started. A broken LiPo battery. Most of the fixes can be found in the Arduino code
I use a polling strategy to detect presence. A distance measurement is taken and then the ESP32 goes into a deep sleep until the next measurement. A sensor with the ability to wake up the microcontroller would be a better approach.
Once everything was installed it worked well enough — motion triggered a random birdsong, creating a soothing, natural vibe. It may be less practical than the off-the-shelf version but I did learn quite a lot more than I would have by simply filling in a form and providing payment details…
A discussion at work led to the question how much time it takes for a HTTP request to pass through a HTTP proxy. This blog post deals with this question by measuring a request passing through a stupid amount of HTTP proxies.
Fig: Measuring the time it takes to pass 500 proxies with Curl.
In modern development setups it is not uncommon that your HTTP request passes a few HTTP proxies before reaching a final server that actually handles the request. In our case there is a proxy which ensures an SSL certificate, which is forwarded to a proxy which automatically forwards requests to a docker container. A final HTTP proxy runs in the docker network that forwards the request to a webserver. A response follows the same way in reverse.
Fig: Configuration to pass a HTTP request through many proxies. The final response is a simple text.
To measure the time it take to pass through a HTTP proxy, I wrote a small script to start 500 separate instances of the Caddy webserver configured as a HTTP/2 proxy. Then, I measure the time it takes to pass through all 500 of the HTTP proxies or only 490, 480,… which results in the graph below.
Fig: Time it takes to pass x amount of HTTP proxies. The y-axis represents the time taken (in seconds), and the x-axis indicates the number of HTTP proxies passed.
So each proxy pass takes about 0.4 milliseconds in one of the best cases, where requests are forwarded from and to localhost. Network overhead adds to that but assuming that interconnects are fast, adding a few HTTP proxies does not affect latency in a meaningful way. Of course it is best to evaluate your situation and measure.
In my house, I have a few smart home features: to control ventilation, to open and close solar screens, and to switch a few smart sockets. Up until a couple of days ago, the ventilation and screen controllers operated using custom software running on an ESP32. However, configuring, maintaining, upgrading, and integrating with this custom software gradually became a headache.
Recently, I switched from custom software to Tasmota, an open-source smart home platform targeting ESP32 devices. Tasmota includes a web UI, flexible configuration options, OTA upgrades, and scripting features. The scripting functionality allows devices to be extended with additional commands, which is especially practical for controlling my solar screens. These screens use pulses to toggle between up-stop-down-stop states. By default, Tasmota only supports enabling or disabling a relay, not enabling it for a very brief period (e.g., 150 milliseconds). With a short ‘Berry’ script, such functionality is quickly added.
I appreciate the effort of the Tasmota team to lower the entry barrier for users. They provide ample documentation and a web installer, making setup straightforward. Simply connect your ESP32 via USB, flash it with Tasmota, and configure it—all from your browser. It’s a surprisingly simple process compared to installing a dedicated toolchain. While this might not be what Tim Berners-Lee envisioned 35 years ago, it certainly simplifies the user experience. Lowering the entry barrier even further, some manufacturers even offer smart home devices with Tasmota preinstalled, such as the Nous A1 smart sockets. Eternal september is here.
If you’re managing custom ESP32 smart home devices, consider switching to Tasmota. Its robust features, ease of setup, and active community support make it an excellent choice for both beginners and advanced users.
The research day of the faculty of Arts and Philosophy of Ghent University took place last November. The theme of the day was ‘From Source to Understanding’ and the program gave an overview of the breadth of research at our faculty with topics as logic, history, archeology, chemistry, geography, language studies, … There were several contributions by our group: the Ghent Center for Digital Humanities. The contribution by me and my close colleagues was a poster about a reusable text annotation building block.
Fig: Poster on a text annotation component.
At GhentCDH we support several text annotation projects and have extracted a text annotation component for reuse. The abstract reads:
“Text annotation is essential for analyzing ancient texts, identifying entities in texts, or documenting evolving grammar. There is a need for reusable annotation methods which copes with challenges such as overlapping annotations, filtering annotation types, and enabling large-scale collaboration and computational analysis on text annotation work.
We present a reusable text annotation component built with TypeScript and Vue 3. It provides an intuitive interface for creating, visualizing, and editing annotations, it allows component users to enrich annotations with complex metadata, and facilitates flexible annotation filtering. This solution meets many needs of researchers in digital humanities and ancient language studies and will be used in several GhentCDH projects.”
Imagine you want to stream a movie at home but also want to keep things quiet to avoid disturbing others. Evidently, this what headsets were invented for. Connecting one wireless Bluetooth headset is typically straightforward - aside from the occasional Bluetooth pairing issues. But what if you want to watch that movie with someone else, and you both want to use headsets? Connecting two Bluetooth headsets, or even combining wired and wireless headsets to share the same audio, isn’t as simple as it sounds. This blog post shows how to achieve this on modern Linux distributions.
Fig: Connecting an audio source - Spotify - to multiple output devices by using audio routing with PipeWire and `qpwgraph`.
During the last years, several Linux distributions have started to support the PipeWire audio server. It is even the default audio server in Debian 12 and Ubuntu 22.10. With PipeWire, managing audio devices has become much easier. PipeWire enables flexible audio setups and supports audio routing: sending out audio from a single source to several output devices. This is exactly what we need to stream audio to multiple headsets.
If you use PipeWire on your system, qpwgraph provides an intuitive graphical interface that lets you visualize and control audio routing. To connect multiple headsets:
First install qpwgraph e.g via apt install qpwgraph
Startup qpwgraph which should show your current audio routing graph.
Pair your Bluetooth headsets to your machine. They will appear in the audio routing graph once paired successfully.
Connect the audio source to your headsets by connecting ‘wires’ from your media player to the headsets.
I was surprised how robust audio has become on Linux and how easy and user friendly it is to set up even more complex audio / MIDI configurations. Give it a try!
The publication of this paper seemed an almost sisyphean task, but it is now finally in print after about four years since first submission.
All’s well that ends well and it is well indeed: the paper contributes a fundamental insight around the resultant peak tibial acceleration (PTA) in forefoot running: contrary to what is often presumed, the resultant PTA is higher in forefoot running! The paper combines two separate experiments into a single analysis framework which ensures robustness in the finding. The conclusions of the article can be found below:
ConclusionsMany coaches and practitioners presume that forefoot striking decreases impact severity and prevents overuse injuries; however, our data show that instructed and habitual forefoot strikes have greater resultant but not axial PTA than habitual rearfoot strikes in level running at a submaximal speed. The forefoot strikes had a sharp decrease in the antero-posterior velocity of the shank following touchdown and, therefore, a greater antero-posterior acceleration, which resulted in the greater resultant peak tibial acceleration compared to the rearfoot strikes. Conclusively, the foot strike pattern differently affected PTAs and should be taken into account when evaluating 3D impact severity in distance runners.
A couple of days ago, OnTracx launched their first product: a system to measure impact during running with the aim to become ‘The Future of injury-free running’. Next to the launch event itself, OnTracx was featured in the nationalmedia as well.
OnTracx is a Ghent University spin-off and their product is based on a couple of scientific studies. I had the chance to collaborate on some of these studies:
Van den Berghe, P., Lorenzoni, V., Derie, R., Six, J., Gerlo, J., Leman, M., & De Clercq, D. (2021). Music-based biofeedback to reduce tibial shock in over-ground running: A proof-of-concept study. Scientific reports
As is already clear from the title of the second paper: originally the idea was to use music-based biofeedback as a way to reduce impact. Unfortunately, this feature is not - yet? - present in the commercial project which focuses on the measurement and clearly reporting a proxy to mechanical load. This does make the message focused and is probably a good commercial move. I bought one of the sensors and already tested it out during a 5k-run. I was pleasantly surprised with the smooth on-boarding and the slick, well thought-out, user-friendly app.
Become part of the ‘The Future of injury-free running’ and go and get OnTracx!
The product launch
The OnTracx CEO during the Flanders Technology and Innovation festival
There is something about surprising interfaces: clapping to switch on lights is more fun than a flipping a switch. Pressing a panic-button to order a pizza is more fun than ordering via an app. Recently I came across this surprising interface: a flute controlled mouse cursor for a first person shooter. I recognize a good idea when I see one, and immediately wanted replicate the idea and make it freely available. So I got to work:
Vid: a microcontroller controlling mouse movements based on pitch detection.
What do we need for flute-based mouse? First we need a way to determine if a note is being played and if a note is produced, we need to be able to determine which note is being played by the musician. Next, we need to hijack and control a cursor via the detected note and trigger a click event when a specific note is played. Finally we need to play a flute, preferably a recorder, to move the mouse cursor in an obviously superior and relaxed fashion. It is not strictly required to use a recorder but a recorder is very much advised.
The note determination can be done by a fundamental frequency detector. A detector returns a frequency in Hertz and a confidence score which tells you how reliable the detection is. With some filtering, this is exactly what we need. If the frequency is close enough to a configured value, a note is detected. The confidence score tells us to either accept or ignore the detection. With this info it is possible to connect a note-detection to an action - like moving a cursor left or right, up or down.
Finally we need to move the mouse cursor. There are a few ways to do this.
Fig: Flute-based web-browsing as envisioned by its developer.
A portable way to move a mouse cursor is to let a micro-controller impersonate as a standard mouse, a ‘USB Human Interface Device’. Once the micro-controller is attached via USB it registers as a mouse and allows to move the cursor and register click events. To build a flute-based mouse, the micro-controller then needs a microphone and a pitch estimator to finally send cursor events.
I based my project on an RP2040 - a micro-controller chip designed by Raspberry Pi - since it offers a simple way to present itself to an operating system as a mouse. Just include PluggableUSBHID.h and USBMouse.h and use the Mouse API to control the mouse. For me it only behaved as a standard mouse if Serial is not used at the same time: in other words the dual USB profile does not seem to work reliably.
Sending mouse events from your code looks, for example, like ` Mouse.move(-4, 7)` to move the mouse minus four units in the horizontal and seven units in the vertical direction. Click events have a similarly straightforward API. The RP2040 also has a built-in microphone, which makes it ideal for audio applications, or so it seems.
Unfortunately, the RP2040 chip performs poorly for computationally heavy audio processing workloads. Such applications need to perform many floating point operations per second, but the RP2040 lacks a hardware floating point unit (FPU) which makes it relatively slow. When attempting to run a pitch-detection algorithm, the RP2040 was too slow to run the algorithm in real-time. After profiling the pitch estimation algorithm there was a clear place where most float operations occurred. Replacing those with much quicker fixed point operations makes the algorithm faster than real-time and usable on the RP2040.lt
To give a sens of the difference in speed between fixed point and floating point operations on the RP2040: with the default arduino build process, a million floating point operations take over 883 000 microseconds, a million fixed point operations take 8 microseconds. Fixed point operations are around 5 orders of magnitude faster!
The hardware based solution works reliably but, evidently, it needs a piece of hardware. To make sure everybody can enjoy a solution in software is provided in this section in the form of a chrome browser extension.
Moving a cursor is not possible in a browser: if a pointer location could be modified it would open a whole range of possibilities for abuse. A surprisingly easy workaround, however, is to hide the actual cursor and show a replacement cursor-like icon. This fake cursor can be moved programmatically. With the position of this fake cursor known, a click event can be triggered and result in, for example, following a link.
To take this idea to its logical next step, I implemented a chrome browser plug-in for flute-based web-browsing. I also relased this on GitHub under the Pitch perfect Pointer Positioning or PiPePoPo brand. Check the installation instructions in the PiPePoPo repository. Perhaps most of interest is how audio processing is handled by a Web Audio API Audio Worklet.
Vid: Controlling a cursor via a browser extension.
Join the flute-based web-browsing revolution today and experience web browsing like never before and install PiPePoPo.
I am not sure how but PiPePoPo was also featured on HackADay and the official Arduino Blog.
UGent
Power banks have become a staple for charging smartphones, tablets, and other devices on the go. They seem ideal to power small microcontroller projects but, they often pose a problem for low-current applications. Most modern power banks include an auto-shutdown feature to conserve energy when they detect a current draw below a specific threshold, often around 50–200mA. The idea being that the power bank can shut off after charging a smartphone. However, if you rely on power banks to power DIY electronics projects or remote applications with low current draw, this auto-off feature can be a significant inconvenience.
birdhouse-static.jpg, birdsong_arduino.ino, birhouse-example.mp4, and birhouse-static.webp
In my house, I have a few smart home features: to control ventilation, to open and close solar screens, and to switch a few smart sockets. Up until a couple of days ago, the ventilation and screen controllers operated using custom software running on an ESP32. However, configuring, maintaining, upgrading, and integrating with this custom software gradually became a headache.
