A common test for indoor solar cells, satellite IoT measured in the wild, and SIM cards as an attack path: research roundup, October 2026

This is the first monthly research roundup: six recent works on the Internet of Things, from energy harvesting to satellite connectivity and security, explained for engineers and technical managers. Each section links to the paper or its official page, and the status label on every work says whether it has been peer reviewed, is a preprint or is a funded project. This month all six are peer reviewed.

A common way to test indoor solar cells

A consensus statement coordinated by the Centre for Hybrid and Organic Solar Energy at the University of Rome Tor Vergata was published in Nature Energy on 18 August (paper). Its 73 authors come from more than 60 institutions and companies, including CNR’s Institute of Structure of Matter and ENEA. The authors propose best practices for measuring how photovoltaic cells perform under indoor lighting and a baseline for testing their stability. The procedure starts from the recommendation of a single reference artificial light source, then covers setting up the measurement system and running measurements and tests step by step, condensed into three checklists. The recommendations draw on experimental data, standards specifications, lighting databases and regulations.

Why it matters: Indoor light is one of the most practical sources for energy harvesting in a smart building, where sensors sit under the same lamps for years. Today indoor efficiency figures are hard to compare, because labs use different lamps, spectra and light levels. A shared protocol lets a hardware team compare cells on equal terms and size a harvester against a sensor’s real power budget, which decides whether a device can drop its coin cell or only extend its battery life.

Limits: This is a measurement method, not a new cell, so it says nothing about how much power a given product will deliver in a given room. According to the coordinators’ announcement, the protocol is meant as a baseline, with more complex conditions such as daylight through windows left for later work.

Glasses that power their own sensors

Researchers at Northwestern University presented GlassTENG at the ACM International Symposium on Wearable Computers, part of UbiComp/ISWC 2026, with the paper published on 10 October (paper). They built three triboelectric nanogenerators into a glasses frame: on the nose bridge over the angular artery, on an extended arm over the superficial temporal artery, and at the temple over the temporalis muscle. Each sensor converts mechanical movement into an electrical signal, and the analog front end draws 1.36 µW per sensor. The frame captures pulse waveforms, jaw activity such as clenching, tapping and eating, and upper facial movement such as blinking and raising the eyebrows. In a study with 20 participants, the system classified six activities with 93.8% accuracy and estimated heart rate with a mean absolute error of 1.82 beats per minute against a chest strap.

Why it matters: In a wearable, the battery sets the size, the weight and how often people take the device off. A sensor that generates its own signal moves almost the whole power budget to the radio and the processor.

Limits: The microwatt figure covers the analog front end, not the radio and processing needed to get data off the glasses. The results come from a 20-person study of six activities, and the authors present the work as a path toward long-term wear rather than a tested long-term deployment.

A sensor that learns on the device in 32 kB

Politecnico di Torino, with two co-authors from STMicroelectronics, described MAUPITI in IEEE Embedded Systems Letters, published in early access on 6 August (paper). The smart infrared sensor combines a 16 by 16 thermal MOSFET (TMOS) pixel array with a RISC-V microcontroller extended with low-precision SIMD instructions, and recognizes poses and gestures with less than 32 kB of on-chip memory and about 1.5 mW. To avoid the memory cost of backpropagation and replay buffers, a small convolutional encoder is trained and quantized offline, while the class prototypes of a nearest-class-mean classifier are stored and updated on the device as data arrives. On two datasets, accuracy matched a conventional classifier, and classification plus prototype updates added less than 0.29% latency. The authors have also posted the accepted version on arXiv.

Why it matters: Low-resolution thermal arrays sense people without capturing a face. Most TinyML devices only run inference with a model frozen at the factory; this design lets a sensor add a gesture or adjust to a new room without sending data off the device and without the memory that full training needs. TinyML explained covers the constraints these designs work around.

Limits: Only the prototypes adapt on the device; the encoder is fixed, so adaptation is bounded by what the offline-trained features can represent. The letter reports results on two datasets, and the abstract describes no field deployment.

Satellite IoT measured on a live network

A team from The Hong Kong Polytechnic University, Shanghai Jiao Tong University and the Hong Kong University of Science and Technology presented at ACM SIGCOMM 2026, published on 11 August (paper), what it calls the first large-scale measurement of direct-to-satellite IoT in the field, using one of the largest operational satellite IoT networks. The authors report that capacity is not the pressing issue. Instead, low throughput, long-tail latency and poor energy sustainability hold the service back, and they trace these to architecture and protocol choices. They propose three drop-in changes: reliability driven by negative acknowledgements, flow control to cut long delays, and finer-grained sleep management. In testbed and live production tests, the redesigned protocol delivered 2.1 times the throughput, 52% fewer long-tail latencies and 38% less energy.

Why it matters: Satellite IoT is being sold as the way to reach assets beyond terrestrial LPWAN coverage. For a battery-powered tracker, the energy spent waiting for a pass and retrying matters more than how many devices a satellite can serve, and the tail of the latency distribution decides whether an alarm arrives in time. Protocol fixes like these could reach devices through software updates rather than new hardware. Our connectivity overview puts satellite next to the terrestrial options.

Limits: The measurements come from one operational network, which the abstract does not name, so the findings may not transfer to every satellite IoT architecture. The improvements need the operator and device makers to adopt the changes.

Using body tissue as the network

Researchers at the Georgia Institute of Technology, with colleagues at MIT and Emory University School of Medicine, reported in Science on 24 September (paper) a communication platform that uses tissue as a conductive medium between implants and wearables. Devices emit transient electric fields, and a receiving device activates only when it gets pulses that switch on its specific transistor circuit. Signals crossed the skin, the space under it, the abdominal cavity and the gut. The implants can be injected with a syringe, draw negligible power while listening, and the authors report more than ten times the tissue coverage of Bluetooth. In rats, networks of sensors and neural interfaces enabled wireless control of two limbs; according to Georgia Tech’s announcement, a sensor detecting front-paw movement triggered stimulation of a hind-leg muscle.

Why it matters: This is a wake-up network with almost no standby cost, the same pattern IoT designers use when a node sleeps until a radio trigger arrives. For medical IoT, it points to closed-loop therapies where a sensor can fire an actuator elsewhere in the body without a phone or a Bluetooth Low Energy link in the loop. IoT in healthcare covers where such devices would fit.

Limits: The in-vivo results are in rats, and Georgia Tech states that therapeutic use in humans has not been demonstrated. The system is designed for short trigger messages, not for streaming data.

When the SIM card attacks the modem

Researchers from the University of Birmingham and the security company Fuzzware presented CATana at the USENIX WOOT Conference on Offensive Technologies, held on 10 and 11 August in Baltimore alongside USENIX Security (paper). They studied RUN AT, a proactive command that lets a SIM ask the device’s modem to execute an AT command. Using their CATana toolkit, they surveyed 26 devices, 8 IoT devices and 18 smartphones, and found that 9 expose this SIM-to-AT interface, which led to 4 vulnerabilities. Case studies show command execution, arbitrary file reads, forced downgrades to 2G and denial of service. According to the university’s announcement, the modules tested include ones used in EV chargers, the team informed the GSMA and affected vendors, and key manufacturers have released updates and hardened configurations.

Why it matters: In cellular IoT, the SIM is usually treated as a trusted part of the device. This work shows it belongs in the attack surface, especially as eSIM remote management and long supply chains add more ways for a hostile profile or card to reach a fleet. Teams using an IoT module can ask the vendor whether RUN AT is enabled and keep module firmware updated. See IoT security fundamentals for the wider threat model.

Limits: An attacker first needs a malicious SIM or control of the SIM’s software, which takes physical access, a compromised operator or tampering in the supply chain. The survey covers 26 devices, and the abstract does not say which products were affected or fixed.

How we pick and label research

We select recent work on the Internet of Things from journals, conferences, preprint servers and EU-funded projects, summarize it from the paper or its abstract and official announcements, and link the source. Preprints are labelled as not yet peer reviewed, and we do not peer review any work ourselves. To report an error, see our editorial policy, which describes how corrections are made.

Topics: research, energy harvesting, TinyML, satellite IoT, IoT security, wearables