The first contact
In simulations of battery-free sensors, MagPie completed every first contact across five charging scenarios, while the comparison protocol Find did not complete every trial in two of them. MagPie finished 100 of 100 first-rendezvous trials in each scenario. Find completed all 100 trials in Office, Stairs and Washer, but finished 93 in Jogging and 36 in Cars.
First rendezvous here means the first successful contact and synchronization between two devices. The paper asks whether a wake-up radio, or WuR, can widen that initial contact opportunity while a separately backed low-power real-time clock preserves timing for later exchanges. MagPie combines energy gating, schedules marked with an epoch version and slot allocation designed to be safe to repeat. Its collection design is static and single-hop, sends reports to one sink and does not use a powered control anchor.
What the comparison measured
The evidence came in three forms: trace-parameterized simulation, a slotted-Aloha study with one collision domain and controlled experiments on STM32WL33 hardware. For pairwise synchronization, the simulation used 100 independent first-rendezvous trials per setting, divided into ten seed blocks, with administrative censoring at 5,000 seconds. The timing comparison used restricted mean synchronization time, an average that caps each trial at that cutoff.
Across the five parameterized scenarios, Find's restricted mean synchronization time was 82.7 to 755.8 times MagPie's. Because the trials were censored at 5,000 seconds, this measure includes the time limit when a run had not synchronized by then.
After discovery, the study measured the interval between scheduled exchanges. In the first four scenarios, both protocols completed 100 exchanges. MagPie's mean sustained interval was 1.8 to 3.6 times shorter than Bonito's, while Bonito did not discover a peer in Cars before the stated horizon.
The paper also gives a retry bound, but only under a stated coverage premise. In its alpha = 0.9 example, the lower bound for joint readiness is p0 = 0.8, corresponding to at most 1.25 expected attempts and no more than 3.2 × 10^-4 probability of five consecutive misses. If the coverage premise fails, the bound does not apply.
When more devices share the channel
A separate MAC test examined role selection when many contenders shared one collision domain. At N = 120, adaptive K = m required 318.0 mean opportunities, with a 95% confidence interval of 316.1 to 320.0. Its 95th-percentile result was 359.0 opportunities, with a bootstrap interval of 353.1 to 362.1, compared with 725 opportunities for fixed K = 32. The model omitted energy readiness, capture, hidden terminals and fading.
In the network energy model, receiver-side delivery varied by scenario. It reached 99.7% to 99.8% in Stairs and Office, 89.0% in Washer and approximately 70% in Jogging and Cars. With N = 120 reporters, collection took 3.57 seconds in Office, Stairs and Washer, 4.32 seconds in Jogging and 11.46 seconds in Cars. Mean delivery at that size was 99.8%, 99.7%, 88.3%, 70.4% and 70.2% in the same order.
These are receiver-side model results under an error-free channel, idealized sink readiness and idealized energy readiness, not measurements of end-to-end energy performance.
To test resilience, the protocol was run with a forced clock-loss probability on each cycle. With p = 0, modeled delivery was 99.8% and there were no re-alignments. With p = 0.2, delivery was 80.6% at N = 5, 79.2% at N = 10, 80.2% at N = 20 and 79.5% at N = 40. This was a stress test, not a measured AM1805 failure distribution.
A controlled hardware check
Controlled pairwise hardware traces produced 459 complete MagPie latency pulses in 3,459 seconds, compared with 40 Find pulses in 3,315 seconds. The rates were 478 versus 43 completed alignments per hour, an observed 11-fold difference. Charging was controlled from 500 to 5,000 milliseconds, so this was not an ambient-harvesting deployment.
In the All-to-One testbed, five reporting boards and one sink ran for 11.05 hours. Formation took 945 seconds, the sink logged 2,998 packets and mean source-to-sink latency was 13.2 seconds. The log lacked a denominator for scheduled transmission opportunities, so it cannot yield a packet-delivery ratio.
A narrower claim
The result is a specific protocol evaluation, not evidence of measured end-to-end energy, ambient-harvesting performance, field-scale RF behavior or multi-hop scalability. The tested evidence consisted of trace-parameterized simulation, a slotted-Aloha model with one collision domain and controlled STM32WL33 experiments. Variable harvesting can still limit collection.
Paper data and sources
Original title: Rethinking Battery-free Sensing Communication via Wake-up Radios
Authors: Gaosheng Liu, Kasim Sinan Yildirim, Sajal K. Das, Yingxin Song
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text