Press Enter to Search or ESC to Close
Imagine a place where the air feels perpetually heavy, the ground visibly cracks with thirst, and the anticipation of rain is as tangible as the dusty horizon. This was the reality for Takalar, a region that had been engulfed in an unyielding drought, causing both the land and its people to yearn for reprieve. […]
Late one evening at our workbench in Makassar, four inexpensive USB software-defined radio dongles sat taped to a circular plastic mount, their coaxial pigtails running to four directional antennas pointed North, East, South, and West. On the laptop screen in front of me, a synthetic test signal was bouncing off the concrete walls of the room, and a standard neural network was losing its mind—guessing 358 degrees one second, 2 degrees the next, and occasionally pointing straight into a brick wall where no transmitter existed. That budget gap sparked the question behind GARUD-A (Gelombang Analisis Radio Untuk Deteksi Arah), our flagship prototype for the Digiwave 2026 AI Innovation Competition: what if we stopped trying to force cheap $25 RTL-SDR dongles to measure phase—something their independent crystal oscillators simply cannot do without hardware modifications—and instead taught a deep neural network the physical laws of electromagnetic amplitude patterns?
Whenever I show our GARUD-A prototype to seasoned RF engineers, the first thing they do is peer behind the antenna mast looking for the phase-matching cables and the master clock distribution box. When they see four ordinary RTL-SDR dongles plugged into a standard USB hub with no clock modification at all, their eyebrows shoot up: ‘Wait—how are you doing direction finding without phase coherence?’ We wanted an array that any regional technician in Indonesia could assemble from standard off-the-shelf parts in an afternoon without touching a soldering iron to a surface-mount crystal. That meant abandoning phase interferometry altogether and returning to a much older, rock-solid physical principle: directional amplitude comparison.
Have you ever watched someone’s eyes glaze over when you show a slide full of partial differential equations and loss-function Greek letters? While preparing our GARUD-A package for Digiwave 2026, I realized that telling judges and fellow engineers about Physics-Informed Neural Networks wasn’t enough—they needed to be able to play with the radio waves themselves. So I asked: can we pack the complete mathematical engine of GARUD-A—the 4-sector antenna gain physics, urban multipath reflection generator, RTL-SDR noise floor simulator, and real-time PINN inference—into a single, lightning-fast HTML5 web app hosted right on zulhaidir.com/garuda/?
Staring at a stopwatch that says ’07:00′ is a humbling experience for any researcher. When we prepared our final submission for the Digiwave 2026 AI Innovation Competition, we had a comprehensive technical report, a risk assessment document, a live web simulator, and months of field experiments—and we had exactly seven minutes of video to make our case. Remembering a lesson from the storytelling workshop I attended at the Young ICT Leaders Forum in Busan back in 2018—’Be authentic, know your story, and filter every idea through the citizen’s need’—I sat down to write a scene-by-scene storyboard and teleprompter script that treated our seven minutes like a documentary short.
Imagine telling a neighborhood association that you are installing an ‘AI-powered radio listening post’ on a pole down the street. Even if your only goal is to protect their cell phone reception and local airport radar from illegal interference, the very phrase ‘radio listening post’ sounds like something out of a spy movie. When we authored the formal AI Risk Assessment & Governance Document for GARUD-A in October 2026, we decided to solve the privacy problem at the hardware-software boundary itself: what if our AI direction-finding node was architecturally blind to the *content* of a transmission?
When you stand on top of a hill with a portable spectrum analyzer for two hours on a Wednesday morning, your screen gives you a snapshot of the radio spectrum at that exact spot and time. But what happens at 9:00 PM when nighttime traffic peaks, or five kilometers away behind a ridge where your hilltop antenna is shadowed? For years, that blind spot bothered me. For our research accepted at the 2026 Photonics & Electromagnetics Research Symposium (PIERS) in Suzhou, China, my UNHAS doctoral research team and I asked: what happens if we deploy a high-density mesh of 20 low-cost IoT-SDR monitoring stations operating 24/7 to automate ITU-R SM.1880-2 occupancy measurements across a region?
Walk into any spectrum management meeting and you will hear someone say, ‘The VHF band in this city is 90% occupied—we can’t issue any more licenses!’ Then an engineer drives out to the city with a spectrum analyzer, watches the band for an hour, and sees half the channels sitting dead quiet. Who is right? While writing our PIERS 2026 paper and mentoring students at UNHAS, I put together a plain-English field guide to ITU-R SM.1880-2—breaking down how revisit time, integration windows, and thresholding turn noisy waterfall plots into trusted international metrics.
Every time I drive past a high-power commercial FM broadcast tower in Makassar or Jakarta, I think about all the kilowatts of electromagnetic energy radiating out into the sky 24 hours a day. Less than a tiny fraction of one percent of that energy actually hits a car radio antenna—the other 99.99% simply dissipates into the environment. Meanwhile, on the ground below, engineers are struggling to figure out how to power thousands of tiny Internet of Things (IoT) sensors without changing lithium batteries every year. That insight inspired our second paper accepted for PIERS 2026 in Suzhou: can we repurpose our low-cost SDR spectrum monitoring network and a hybrid 1D Convolutional Neural Network (1D-CNN) to geolocate VHF FM broadcast stations (87.5–108 MHz) and map optimal Green IoT RF energy harvesting zones?
When I clicked ‘Export to PDF’ on my final Formulir Aplikasi Insinyur Profesional (FAIP) in May 2026, the page counter in the corner of my screen read 602 pages. Sitting at my desk in Makassar, I scrolled from Page 1 all the way down to Page 602, watching twenty-six years of my life flash by in tables, certificates, schematics, and project reports. Enrolling in the PPI Past Learning Recognition (Rekognisi Pembelajaran Lampau / RPL) program at Universitas Negeri Makassar (UNM) changed my perspective completely. The RPL process doesn’t just ask *what* job titles you held; it forces you to dissect *how* you thought as an engineer across every project since graduation.
