PaperlayerView exact revision on arXiv

Source-linked AI summary

Rethinking Battery-free Sensing Communication via Wake-up Radios

Gaosheng Liu, Kasim Sinan Yildirim, Sajal K. Das, Yingxin Song

arXiv:2608.25292v1cs.NI

TL;DR

Battery-free sensors face brief, stochastic communication windows and can lose timing state when their main energy domain browns out. MagPie combines WuR-assisted discovery with a separately backed LP-RTC, then extends persisted phase into energy-gated, versioned single-hop collection. In independent censored simulations, MagPie completes 100/100 first rendezvous events in every scenario, while the study’s hardware and contention evaluations validate narrower execution and coordination claims.

  • Problem

    Battery-free sensors require infrastructure-free rendezvous despite stochastic, non-overlapping active windows and must retain acquired timing after main-domain brownouts.

  • Method

    MagPie combines a wake-up radio for first-contact opportunity with a separately backed LP-RTC, energy gating, versioned schedules, and idempotent collection-slot allocation.

  • Results

    100/100 first rendezvous events complete in every one of five trace-parameterized scenarios for MagPie, compared with 36-100/100 for Find.

  • Takeaways & Limitations

    MagPie demonstrates a hardware–protocol co-design that preserves phase across brownouts and supports static, single-hop All-to-One collection without a powered control anchor.

  • Takeaways & Limitations

    The evidence does not establish end-to-end energy efficiency, field-scale RF behavior, or multi-hop operation, and several guarantees remain conditional on coverage and system assumptions.

Abstract

from arXiv · show

Battery-free sensors expose short, stochastic communication windows and may lose timing state whenever their main energy domain browns out. MagPie combines a microampere wake-up radio (WuR) with a separately backed low-power real-time clock (LP-RTC): the WuR widens the first-contact window, while the LP-RTC preserves the acquired phase for later exchanges. Energy gates, epoch-versioned schedules, and idempotent slot allocation extend this mechanism to static, single-hop All-to-One collection without a powered control anchor. The analysis accounts for role selection and duty-cycled listening, and bounds scheduled-retry tails only under explicit conditional quantile coverage. Evaluation separates three scopes. In independent, administratively censored simulation trials, MagPie completes 100/100 first rendezvous events in each of five trace-parameterized harvesting scenarios; Find completes 36-100/100. A single-collision-domain slotted-Aloha study shows that adaptive K is necessary at high contender density and reports mean, P95, and confidence intervals through 120 components. Finally, controlled STM32WL33 experiments validate alignment, clock persistence, and six-device slot execution, including an 11.05-hour functional run. The study does not claim measured end-to-end energy, ambient-harvesting performance, or multi-hop scalability; highly variable harvesting still limits collection because coordination cannot create missing energy.

I. INTRODUCTION

Battery-free sensors have brief, stochastic active windows separated by long charging intervals, making rendezvous and timing persistence central communication problems. MagPie addresses these requirements with WuR-assisted discovery, backed-clock persistence, energy-gated schedules, and recovery-oriented collection mechanisms.

  • Motivation: Active windows last milliseconds while charging intervals are two to five orders of magnitude longer, so independently powered radios may not overlap.A main-radio exchange requires two active windows to overlap, and charging variability differs across cycles and devices.
  • Motivation: Prediction-based rendezvous and powered-coordinator schedules respectively depend on charging models or reintroduce infrastructure.The introduction identifies infrastructure-free discovery under stochastic timing and timing-state retention as separate requirements.
  • MagPie approach: MagPie widens first-contact opportunities with a wake-up radio and preserves acquired phase in a separately backed LP-RTC.Energy gates prevent radio phases from starting below their configured budget, while sender-owned epochs reconcile updates after missed notifications.
  • MagPie approach: The protocol extends pairwise timing into stationary, single-hop All-to-One collection using versioned discovery, idempotent range assignment, and recovery after state loss.The work deliberately scopes the network layer to single-hop collection rather than general-purpose routing.
  • Evaluation scope: The study separates evidence from prototype execution, trace-parameterized simulation, and collision-domain contention analysis.Reported evidence includes seed-block confidence intervals and explicitly bounded claims.

II. RELATED WORK

Related work addresses intermittent sensing through computation preservation, duty cycling, wake-up radios, asynchronous discovery, prediction, or powered coordination. MagPie instead targets timing acquisition and persistence across main-domain brownouts without a powered control anchor.

  • Intermittent computing: Checkpointing and task-based runtimes preserve computation on one intermittently powered node, whereas MagPie targets shared timing across multiple nodes.This distinction separates local state recovery from acquiring and retaining communication phase.
  • Duty cycling and routing: Traditional duty cycling and energy-aware routing generally assume retained state and predetermined wake times, unlike MagPie’s main-domain brownout setting.Those approaches commonly use rechargeable batteries or large buffers and much longer scheduling intervals.
  • Wake-up radios: Wake-up-radio systems reduce idle-listening power, but Greentooth relies on a powered base station and CATO addresses a different state boundary.MagPie focuses on retaining relative time while the MCU and main radio repeatedly lose power, adding energy-gated scheduled collection.
  • Asynchronous discovery: Asynchronous discovery protocols provide deterministic or bounded rendezvous under retained-clock assumptions that fail when the timing domain browns out.MagPie borrows the duty-cycle tradeoff while first acquiring and then protecting a cross-brownout phase.
  • Prediction and coordination: Prediction-based methods avoid powered coordinators but depend on charging models or environmental timing signals, while coordinator systems recreate timing through infrastructure.MagPie treats these systems as alternatives rather than interchangeable numerical baselines.

III. MAGPIE SYSTEM DESIGN

MagPie combines energy-aware protocol phases with persistent LP-RTC state so devices can resume scheduled communication after main-domain failures. Its sizing and tuning rules remain conditional on measured hardware behavior, charging history, and supported power budgets.

  • The system model includes harvesting, capacitor storage, MCU, main and wake-up radios, NVM, and a separately powered LP-RTC, with nonparametric charging times.
  • Voltage-threshold sensing is specified using STM32WL33 comparator, DAC, ADC, and programmable-detector primitives, but is not an end-to-end integrated measurement.
  • 55 nA crystal-mode LP-RTC operation gives an ideal 75.8-hour upper bound for a 10 mF, 1.5 V backup reservoir, while leakage reduces that example to 0.82–26.9 hours.The 75.8-hour value is explicitly an ideal upper bound rather than a prototype measurement.
  • For ρ = 20ppm and Tguard = 10ms, τstale = 250 s; worst-case deployment-range oscillator error, response jitter, and quantization determine usable guard margin.Increasing the guard extends staleness tolerance but raises receiver energy in every exchange.
  • Phase-wise energy requirements include conversion loss, commit energy, quiescent draw, and measured average rail voltage, so prediction requires target-board measurements.The simulator’s 47 µF capacitor normalizes Bonito charging times, whereas the prototype uses 1000 µF.
  • Under ideal constant-current charging, stored energy and fixed-interval charge time scale linearly with capacitance, making the 1000 µF prototype buffer 21.3× slower than 47 µF.Simulated protocol comparisons therefore use equal 47 µF normalization, and the larger testbed is mechanism validation rather than an energy or latency comparison.
  • WuR listening reduces net charging current and must be duty-cycled at or below break-even; adaptive tuning selects the smallest J meeting latency while using K = 2 for pairs and K ≈m for m contenders.The controller is first-order because no end-to-end current trace calibrates E0.

B. Multi-Threshold Energy Decision Mechanism

MagPie partitions capacitor voltage into ordered energy regions that gate WuR and main-radio operations before commitment. This staged mechanism reserves energy for coordination, transactions, state persistence, and recovery without treating every handshake as a single-window task.

  • The energy state machine uses five ordered voltage thresholds to gate operations across the WuR and main-radio subsystems.
  • A radio phase starts only when its assigned voltage interval can fund load, conversion loss, state commit, and quiescent draw, with margins limited to measured operating conditions.
  • The [VWuR- 1x , VWuR- 2x] region reserves one WuR operation, while [Vradio-on , VWuR- 1x] reserves a second WuR operation before main-radio engagement.
  • The [Vradio-off , Vradio-on] region sustains one complete main-radio transaction, either TX followed by RX or the reverse.
  • Failed WuR coordination falls to Vradio-on and recovers two regions, whereas completed exchange falls to Vradio-off and recovers three.This produces a listen–charge–listen pattern that favors rapid recovery and WuR availability.
  • The protocol combines role selection, WuR-based clock alignment, and persistence of the resulting alignment across power failures.

1) Role Selection Under Identical Conditions:

Identity-derived role selection prevents synchronized devices from repeatedly transmitting together, while the WuR handshake acquires relative LP-RTC phase for later scheduled overlap. Persistent clock state allows the acquired alignment to survive main-domain outages, subject to re-alignment conditions.

  • Role Selection Under Identical Conditions: A transmit-on-multiple-of-K rule can synchronize co-deployed devices into repeated collisions, so MagPie derives roles from device identity and counter state.
  • Role Selection Under Identical Conditions: Each eligible device transmits on a pseudorandom 1/K fraction of clock edges, while unique identifiers decorrelate roles even with identical energy histories.
  • Clock Alignment via WuR Handshake: During alignment, D2 sends a WuR beacon, D1 wakes after Tresp, measures phase, replies at its next LP-RTC edge, and both later open the main radio in guarded overlap.
  • Role Selection Under Identical Conditions: Role re-evaluation prevents one collision from becoming permanent, but does not randomize periodic wake phase; repeated misses require phase re-estimation.
  • Role Selection Under Identical Conditions: With m contenders in one collision domain, exactly-one-transmission probability is maximized near K = m, trading idle opportunities against collisions.
  • Clock Alignment via WuR Handshake: The reply carries D1’s counter and measured offset, allowing D2 to compute the reverse offset and confirming bidirectional synchronization.
  • Clock Alignment via WuR Handshake: After recharge, the device reloads stored offsets and computes the next rendezvous from Δij because the separately powered LP-RTC continued counting.

3) Shutdown, Recovery, and Staleness Detection:

The protocol preserves synchronization across main-domain failures and invalidates cached timing when the clock fails or drift exceeds the guard interval. Scheduled exchange then retries without repeating handshakes, while quantile-based cycle sizing explicitly accepts a miss tail.

  • Shutdown and recovery: Clock-domain survival is required for cached offsets to remain valid after shutdown.An oscillator-failure flag invalidates offsets until successful alignment clears it.
  • Staleness detection: Cached offsets are invalidated when elapsed drift exceeds τstale, with full re-alignment after at most Rmax fallback cycles.The checks run at every boot and fail safe.
  • Scheduled exchange: Once aligned, scheduled radio overlap replaces repeated handshakes.The persistent schedule remains available for later opportunities.
  • Cycle sizing: The shared cycle uses the larger empirical charging-time quantile from the two endpoints, expressed in Tclk ticks.Each device advertises its α-quantile over the last W = 32 cycles, and both adopt the resulting value.
  • Cycle sizing: At α = 0.9, the quantile deliberately accepts nominal miss probability 1 − α rather than treating past samples as a maximum.The rolling estimate does not guarantee conditional coverage under arbitrary drift.

2) Rendezvous Analysis:

MagPie bounds scheduled rendezvous success under conditional quantile coverage rather than claiming determinism. Epoch updates, observable-miss recovery, and full handshake fallback address stale schedules and missed payloads.

  • Rendezvous guarantee: Conditional quantile coverage bounds the probability that both endpoints are ready at a scheduled boundary.The analysis explicitly excludes unobserved regime shifts that make the selected cycle systematically too short.
  • Rendezvous guarantee: Under conditional independence, the joint readiness bound strengthens to p_k ≥ α^2.
  • Rendezvous guarantee: Proposition 1 bounds the number of cycles until the first successful exchange under the stated coverage conditions.
  • Rendezvous guarantee: At α = 0.9 and p_0 = 0.8, expected attempts are at most 1.25 and five consecutive misses have probability at most 3.2 × 10^-4.
  • Recovery: When coverage fails, no tail guarantee follows, although surviving clock phase keeps the next opportunity computable.A sender-owned epoch and Rmax = 3 observable misses trigger schedule refresh and full bidirectional re-alignment.

E. Networking Protocol

The networking protocol builds a versioned merge tree, then distributes unique collection slots through two passes. Persistent epochs, ranges, and assignments make replay idempotent across brownouts in a static, single-hop All-to-One setting.

  • Networking scope: The network layer targets N battery-free reporters and one logically distinguished sink in a static, single-hop domain.Collection begins only when the sink has enough energy to receive the round.
  • Protocol phases: Component discovery creates a versioned timing tree, followed by slot distribution that assigns every reporter a unique index.The two phases are component discovery and slot distribution.
  • Slot distribution: The slot-distribution passes partition contiguous identifier-ordered ranges from the smallest-identifier root.Each node takes the first index in its range and partitions the remainder by subtree size.
  • Replay safety: Epoch-and-range persistence makes replay idempotent: equal assignments reproduce the same slot, while older epochs are discarded.Disjoint sibling ranges prevent concurrent branches from issuing duplicate indices.
  • Overhead: A completed merge tree uses N − 1 accepted alignments, while each collection round contains N reporter transmissions.Per-node storage is O(d_i), and total topology storage is O(N).
  • Cycle coordination: Rounding the synchronization cycle to a distribution-cycle multiple adds less than one distribution cycle of waiting.The resulting cycle is bounded by C_sync + C_dis and at most 2max(C_sync, C_dis).

F. Overhead Analysis

The overhead analysis combines modeled primitive costs with separate simulation and prototype scopes. It quantifies communication and listening costs, while explicitly withholding end-to-end energy and cross-artifact numerical claims.

  • Limitations: The primitive table supports only a first-order design study, not an energy-efficiency claim or production hardware sizing.End-to-end shunt measurements would need to include regulator, threshold, false-wake, NVM, and backup-domain costs.
  • Primitive costs: Modeled active cost is 79.2 µJ at the sender and 165 µJ at the receiver for one scheduled data transfer.These values exclude WuR listening, conversion loss, and state commits.
  • Primitive costs: A maximum 30 s WuR listening window draws 360 µJ at 12 µW from the harvester.Listening appears as a longer charge interval in the model.
  • Trade-offs: The receiver’s 10 ms guard costs 150 µJ before payload reception at the configured 15 mW.Wider guards extend offset lifetime but are paid every cycle.
  • Evaluation design: The study reports six evaluation questions spanning rendezvous, sustained exchange, contention, scaling, sensitivity, and hardware execution.
  • Prototype: The STM32WL33 prototype uses an approximately 4 µA WuR and a separately backed AM1805 32.768 kHz time base.Controlled 915 MHz energy supplies the main buffer, with harvesting and communication time-multiplexed.
  • Evidence scope: Numerical results from the pairwise simulator, six-device firmware, and separate contention study are not interchangeable.The artifacts use different settings, including K = 2, K = 32, and density-adaptive K.
  • Simulation scope: Pairwise synchronization uses 100 independent first-rendezvous trials per harvesting setting with administrative censoring at τ = 5000 s.Results use seeded synthetic traces parameterized from five harvesting settings rather than direct HDF5 replay.

B. Pairwise Synchronization

MagPie improves first-contact and sustained exchange performance in the disclosed harvesting simulations, while network delivery remains constrained when reporters lack radio-grade energy.

  • First rendezvous: 100/100 first-rendezvous trials completed in every scenario for MagPie, while Find completed 36–100/100.Find completed all trials in Office, Stairs, and Washer, 93 in Jogging, and 36 in Cars.
  • First rendezvous: 82.7–755.8× RMST ratios separated Find from MagPie across the five parameterized scenarios.These ratios describe the disclosed simulator rather than a universal hardware speedup.
  • Reference comparison: 1.3× faster than MagPie, the powered reference was faster in stable Stairs, while MagPie had lower RMST in the other four settings.Because both coordinator and WuR factors changed, this is a system comparison rather than component attribution.
  • Sustained connection: 1.8–3.6× lower mean sustained intervals were achieved by MagPie after discovery, with both protocols completing 100 exchanges in the first four scenarios.MagPie retains the shared interval with its backed clock, whereas a miss triggers bounded retry rather than immediate rediscovery.
  • Collision-domain formation: At N = 120, adaptive K = m required mean 318.0 opportunities and P95 359.0, while fixed K = 32 raised P95 to 725 opportunities.These are lower bounds on wall-clock formation because energy readiness is omitted.
  • Network execution: 99.7–99.8% receiver-side delivery occurred in stable Stairs and Office, while Jogging and Cars fell to approximately 70% under the error-free channel.The losses were energy-related by construction because reporters often lacked radio-grade energy at their assigned instant.

E. Sensitivity and Assumption Checks

Sensitivity checks show that quantile underestimation, drift, clock failures, and contender density can materially affect readiness, retries, and formation time; controlled hardware confirms protocol execution.

  • Cycle estimation: 25% underestimation reduced observed joint readiness to 25.4–74.5%, whereas 25% overestimation raised it to 91.4–99.9% but lengthened every scheduled cycle by approximately 25%.Without artificial error, readiness was 84.7–94.1%.
  • Drift and retries: At ρ = 20ppm, 5, 10, and 20 ms guards remained valid for 125, 250, and 500 s; at 100 ppm, those ages fell to 25, 50, and 100 s.Increasing the guard extends validity but raises receiver energy in every exchange.
  • Drift and retries: Rmax = 3 capped the needless-fallback bound at 0.8% and disconnection detection delay at three cycles.The retry bound does not assume independent cycles.
  • Hardware validation: 478 versus 43 completed alignments per hour produced an 11.0× increase for MagPie over Find in controlled logic traces.Mean synchronization latency was 3678 versus 24,012 ms, with maxima of 34,405 versus 222,942 ms.
  • Network operation: A five-reporter, one-sink firmware run lasted 11.05 hours, completed formation in 945 s, and logged 2998 received data packets.The log lacked the denominator of scheduled opportunities, so it cannot yield a packet-delivery ratio.

V. DISCUSSION AND LIMITATIONS

The discussion limits MagPie’s guarantees to conditional timing and stationary, single-hop collection, with current evidence establishing execution and modeled costs rather than measured end-to-end energy efficiency.

  • Assumptions: The guarantees remain conditional because persistent beat frequencies can defeat uniform-phase assumptions and regime shifts can invalidate quantile coverage.The firmware fixes parameters; online joint adaptation is not evaluated.
  • Clock persistence: LP-RTC persistence can last only while its backup domain survives and drift remains within the guard; shared faults could create more severe recovery bursts.Leakage reduced a nominal 3.2-day ideal to hours, and 100-ppm sensitivity shortened a 10-ms guard to 50 s.
  • Energy accounting: The 1000 µF testbed buffer takes 21.3× the ideal charge time of the simulator’s 47 µF reference over the same voltage interval.The results establish timing behavior and modeled primitive costs, not measured end-to-end energy efficiency.
  • Reference comparison: The powered reference is a system comparison because it changes both control infrastructure and WuR factors, preventing attribution to either component.An iso-storage ablation grid is needed before assigning gains to individual primitives.
  • Collection boundary: Approximately 70% receiver-side delivery in the two most variable settings shows that collection remains limited by energy even with fixed slots and an error-free channel.Energy-aware scheduling could help but would consume control energy and risk hidden reassignments after brownouts.
  • Evaluation scope: The evidence excludes ambient-trace replay, measured opportunity-normalized delivery, and topology effects beyond the modeled single-hop setting.The controlled run validates execution, while the energy model assumes an available sink and error-free assigned slots.
Loading 2608.25292v1…