A 1.83 mJ encrypted telemetry cycle, battery-free

What the work claims

This is a primary engineering result, demonstrated on a bench. Djidjekh, Thomas, Bernyer, Loubet, Dragomirescu and Takacs, all at LAAS-CNRS in Toulouse, present a battery-free wireless sensor node that combines four things most passive devices lack: multi-parameter sensing, on-node computation, cryptographic protection, and a communication link robust enough for a real indoor room¹. The node is powered purely by radiative wireless power transfer (WPT) at 868 MHz from a dedicated transmitter, and it answers using a 1-bit backscattering rectenna that emits its data on the orthogonal polarization to the illumination. The authors claim the complete sense-compute-encrypt-transmit cycle runs on a measured 1.83 mJ, and that a full recharge of the on-board energy buffer takes 3 minutes 11 seconds at 5 meters and 5 minutes 35 seconds at 7 meters from a 33 dBm EIRP source.

How it works

The architecture is a careful exercise in energy accounting. A dual-polarized rectenna does two jobs. Vertically polarized, it feeds a Schottky-diode rectifier (an SMS7621) whose measured RF-to-DC conversion efficiency lands between 16 and 28 percent across 864 to 880 MHz depending on input power, into a 10 kilohm load. The same aperture, switched by a 1-bit RF switch (a GRF6011, specified at 0.43 dB of insertion loss), flips to horizontal polarization to scatter the outgoing telemetry, so the power-carrying downlink and the data-carrying uplink are separated in the polarization domain. That orthogonality is the robustness trick: the reader sees the node's signal against a different polarization background than its own transmitted carrier, which suppresses clutter and self-interference.

Harvested energy is buffered by a power management chain built around a TI BQ25504 boost charger and a TPS63030 buck-boost regulator, with two 100 microfarad capacitors holding the charge between 5.2 and 2.3 volts, an available energy of roughly 2.2 millijoules. The load side is a microcontroller running at 16 MHz with its phase-locked loop disabled, talking to its sensors over I2C. Scheduling is energy-aware in a way worth noting: the VOC sensor needs about 170 milliseconds of heater warm-up at roughly 2.6 mA, so the firmware overlaps that warm-up with other work rather than idling. The payload path is a 20-byte frame (preamble, device ID, a 4-byte counter, 10 bytes of sensor data, CRC), Manchester-encoded at an effective 200 kbps because the switch toggles at 400 kHz; the whole frame lasts 0.8 milliseconds. UID, counter and data are protected with AES-128 in ECB mode, with the monotonic counter acting as an anti-replay freshness check. Correct end-to-end operation was verified by decrypting captured frames offline and recovering the sensor values¹.

Read the energy numbers together and the design constraint appears. The buffer holds about 2.2 mJ; one full cycle costs 1.83 mJ at the 3.3 volt rail. A full charge therefore buys roughly one cycle, and at 5 meters from a 33 dBm source that charge takes just over three minutes. This is a duty-cycled architecture in the strict sense: it wakes, spends almost its entire energy reserve on a single encrypted transmission, and goes back to sleep to harvest.

Where a skeptic should push

The load-bearing assumption is that the air-link budget survives contact with a real deployment environment. Every range and charging number in the paper was measured line-of-sight in a corridor and an office, at 868 MHz, with a signal generator as the illuminator. Metal benches, incubators, perfusion tubing, and especially water-heavy biological tissue all reshape that budget, and 868 MHz coupling into salty dielectric is precisely the regime where free-space numbers are most optimistic. Demonstrated versus asserted: the 1.83 mJ cycle, the 16 to 28 percent rectifier efficiency, the charging times and the end-to-end decryption are all measured and are credible. Asserted but not shown: performance in cluttered multipath, in any lossy medium, or at any data rate above 200 kbps.

The security claims deserve a colder reading. AES-128 in ECB mode is the wrong mode for the job: it is deterministic, malleable, and provides no integrity or authenticity. The anti-replay counter is genuinely useful, but a listener can still flip ciphertext bits in the unauthenticated fields, and identical plaintext blocks encrypt identically. For room-temperature telemetry this is a defensible proof of concept; for anything touching regulated biological or medical data it would not pass a serious review. The honest takeaway is narrower than the paper's framing: the achievement is fitting any cryptography at all into a battery-free energy budget, not achieving strong security.

Battery-free telemetry at neural energy budgets

For microelectrode array people, the first instinct is to ask whether this untethers the headstage, and the paper's own numbers answer: not for raw data. A modest 256-channel array digitized at 16 bits and 20 kilosamples per second produces about 82 megabits per second; this node's entire frame budget is 160 information bits every several minutes. That gap is not a criticism of the work, it is the actual finding for our field: battery-free telemetry will never carry raw MEA streams, so the acquisition chain must decide, at the electrode, what is worth a millijoule. The 10-byte payload here is the size of about forty one-byte event counts or twenty two-byte feature words. A neural version of this architecture is a spike-event or feature reporter, not a data recorder.

Three transferable mechanisms stand out. First, polarization-domain isolation: carrying power on one polarization and data on the orthogonal one is a clean analogue of the oldest problem in wireless electrophysiology, where the power or stimulation carrier pollutes the recording band; the same trick applied at a headstage or implant could relax front-end dynamic-range requirements that today drive amplifier cost. Second, task-overlapped scheduling under a hard warm-up delay is exactly the discipline a stimulation-recovery cycle needs: an array front-end that must wait out an electrode polarization window after a stimulus pulse could spend that dead time on compression instead of idling. Third, the security-per-millijoule point cuts both ways. It is an opportunity, because it shows a cryptographic floor is affordable even on harvested energy, and neural data is moving into regulatory scope. It is a threat, because the demonstrated ceiling is ECB-grade protection; vendors who treat telemetry encryption as a checklist item will ship exactly this level of weakness into devices whose data feeds closed-loop interventions.

The obsolescence angle is worth stating plainly. Tethered MEA systems are not endangered by 200 kbps backscatter links. What this work erodes is the assumption that untethered operation requires a battery, with its thermal load, its volume, its chemistry next to living tissue, and its limited cycle life. For long-baseline organoid maturation studies inside incubators, or for chronic peripheral-nerve cuffs, a duty-cycled battery-free node that wakes, reads a precomputed feature vector, encrypts and scatters it, then sleeps on harvested energy is a plausible instrumentation path. The 1.83 mJ ledger, not the radio, is the blueprint.

The bottom line

Established: a measured, end-to-end demonstration that sensing, computation, AES-128 and polarimetric backscatter fit inside a 1.83 mJ duty cycle on harvested power, with recharge times measured at meter scale. Hypothesis: that this architecture class extends to neural event reporting and to polarization-isolated wireless headstages. What would confirm it: a link budget through tissue or incubator environments, a neural payload replacing the environmental one, and a link margin study at realistic array data reduction ratios. What would break it: recharge times stretching by orders of magnitude in lossy media, or feature-level compression proving lossy enough that downstream science degrades. Either way, the energy arithmetic in this paper is now the number to beat for anyone promising untethered neural instrumentation.