Every warm resistor hisses. Electrons at room temperature never sit still, and their jiggle shows up as a faint, broadband voltage: thermal noise, the floor every radio engineer spends a career fighting. A team at Stanford’s Smart Sensing Systems lab has started talking in it instead. Their transmitter generates no radio carrier. It switches an antenna between two states, open circuit for one bit and short circuit for the other. It does not make a signal. It reflects the hiss that is already there.

The paper, in PNAS 123(43), 2026, is by co-first authors Geneva Ecola and Jida Zhang, both graduate students, with professor Joseph Kahn and senior author Zerina Kapetanovic, assistant professor of electrical engineering. As Ula Chrobak’s Stanford story, republished on Tech Xplore on 20 October, reports, the team built a first-principles channel model first. Tuning the hardware to that model lifted the data rate 100-fold. Kapetanovic’s Stanford profile summarises the result as kilobit-per-second links, with polar codes used to approach the capacity the model derived.

The system, drawn as a block diagram

Read it the way you would read a plant schematic:

• Input: ambient thermal noise. Free, everywhere, proportional to temperature.
• Modulator: one switch, one antenna, two impedance states.
• Channel: the reflected noise.
• Receiver: an amplifier that, in Kahn’s words, works like an “electron refrigerator”, so the reflected noise appears hotter or colder while the whole system stays at room temperature.

“The transmitter is able to modulate a signal between colder and hotter, but it does that without cooling anything down,” Kahn says. “It’s just a reflector.”

That is the essence, and it is why the energy line matters. Stanford states that a radio-frequency transmitter can use up to 1,000 times as much energy to send information as one using only thermal noise. “Generating an active signal is very power-consuming,” Zhang says. Move the expense off the node and onto the receiver, and the node can shrink to almost nothing.

The trade-off is plain: without a carrier the node spends almost nothing, but it also cannot shout, so the receiver carries the burden, and so far these links have been shown inside a densely insulated chamber that other signals could not penetrate. The team names its own next steps: linking multiple transmitters into a network, and holding the signal in noisier environments outside the box.

A century-old hiss, finally put to work

The physics belongs to two people most engineers never think about. John B. Johnson measured this noise at Bell Labs in the late 1920s, and Harry Nyquist explained it. Every receiver datasheet still leans on their work. Kapetanovic’s doctoral research at the University of Washington turned it into a link: a 2022 PNAS paper with Morales and Smith reached up to 26 bit/s at up to 7.3 m, at power comparable to an RFID tag. In 2025 a sister paper in the IEEE Journal of Microwaves (Kapetanovic, Falconer, Ecola) made a resistor plus antenna into a thermometer that “requires no power at the sensor node” and can be read underground or behind barriers. Four years took the link from 26 bit/s to kilobits, by understanding the channel instead of adding power. “We’ve built a system that others could very easily implement,” Kapetanovic says.

←TODAY: In 2026 a switched antenna reflecting room-temperature noise reaches kilobit rates, in a shielded chamber at Stanford.
→3012: Zurich’s slabs and facades report their own state for a century, and nobody climbs a scaffold to change a battery.
Fulcrum: The sensor that lasts is the one whose transmitter costs nothing to run, so the energy budget belongs on the drawing, not the invoice.

The building line (ours, not theirs)

Stanford names soil moisture, environmental sensing, and sites near electromagnetically sensitive equipment such as medical devices and observatories. Buildings are PAZ’s extrapolation, and should read as one. They are an obvious one for Swiss retrofit, where facade and slab monitoring keep hitting the same bottleneck: the sensor is cheap, and the battery swap behind ventilated cladding is not. A resistor and antenna cast into a slab, a bracket reporting its own temperature for decades: none of that exists today, and a chamber result does not prove it will. It does sharpen the question to ask of every sensor on a drawing: who powers it in year twelve? From where I sit, looking back, the sensors that survived were the ones nobody had to visit.

Atelier: An office planning a monitoring concept for a facade retrofit usually specifies sensors by what they measure, rarely by what keeps them alive. As AI-assisted analysis makes long monitoring datasets more valuable, the node that dies in year five becomes the costly gap. Monday move: add a “power” column to your sensor schedule and flag every battery-powered node that will sit behind a closed building layer.

Hack: Put a number on the hiss a passive sensor would be reflecting. The Johnson–Nyquist relation gives the open-circuit noise voltage as √(4kTRB) and the available noise power as kTB.

import math
k, T, R, B = 1.380649e-23, 293.0, 50.0, 1e6   # J/K, 20 °C in K, ohm, Hz
v = math.sqrt(4 * k * T * R * B)              # Johnson noise voltage, V rms
p = 10 * math.log10(k * T * B / 1e-3)         # available noise power, dBm
print(f"{v*1e6:.2f} uV rms, {p:.1f} dBm")

You get about 0.90 µV and −113.9 dBm. Now set T to 263.0, a winter facade at −10 °C, and the power drops by about 0.47 dB. That small, temperature-proportional shift is exactly what the 2025 Stanford thermometer reads without powering the node.

PAZ’s Grasshopper↔Archicad Library can carry a sensor attribute from the Rhino model into the Archicad schedule, so the power question survives the handover. Open your current retrofit model today and count the sensors that need a battery.

Sources & Further Reading

• Primary: Tech Xplore: Background static can carry wireless data without an actively generated radio signal