PaperlayerView exact revision on arXiv

Source-linked AI summary

Sensing, Computing, and Communication for Energy Harvesting IoTs: A Survey

Dong Ma, Guohao Lan, Mahbub Hassan, Wen Hu, Sajal K. Das

arXiv:1905.03949v2eess.SPcs.NI

TL;DR

Battery maintenance limits the sustainability of large-scale IoT deployments, while EH-IoTs must operate with small, dynamic, and unpredictable power supplies. The paper systematically surveys commercialization, standards, sensing, intermittent computing, and wireless communication advances for EH-IoTs. It synthesizes existing approaches, identifies lessons and research directions, and summarizes standards activities supporting interoperability and deployment.

  • Problem

    Large-scale IoT deployments face battery maintenance limitations, while unpredictable harvested power creates challenges for reliable sensing, computing, and communication.

  • Method

    The paper systematically surveys EH-IoT products, standards, energy-efficient sensing, intermittent computing, checkpointing, timekeeping, and wireless communication techniques.

  • Results

    The survey classifies, compares, and analyzes EH-IoT solutions, identifies lessons learned, discusses future research directions, and summarizes relevant standards activities.

  • Takeaways & Limitations

    Reliable EH-IoTs require coordinated advances across sensing, computing, and communications, alongside standards that promote interoperability and deployment.

Abstract

from arXiv · show

With the growing number of deployments of Internet of Things (IoT) infrastructure for a wide variety of applications, the battery maintenance has become a major limitation for the sustainability of such infrastructure. To overcome this problem, energy harvesting offers a viable alternative to autonomously power IoT devices, resulting in a number of battery-less energy harvesting IoTs (or EH-IoTs) appearing in the market in recent years. Standards activities are also underway, which involve wireless protocol design suitable for EH-IoTs as well as testing procedures for various energy harvesting methods. Despite the early commercial and standards activities, IoT sensing, computing and communications under unpredictable power supply still face significant research challenges. This paper systematically surveys recent advances in EH-IoTs from several perspectives. First, it reviews the recent commercial developments for EH-IoT in terms of both products and services, followed by initial standards activities in this space. Then it surveys methods that enable the use of energy harvesting hardware as a proxy for conventional sensors to detect contexts in energy efficient manner. Next it reviews the advancements in efficient checkpointing and timekeeping for intermittently powered IoT devices. We also survey recent research in novel wireless communication techniques for EH-IoTs, such as the applications of reinforcement learning to optimize power allocations on-the-fly under unpredictable energy productions, and packet-less IoT communications and backscatter communication techniques for energy impoverished environments. The paper is concluded with a discussion of future research directions.

I. INTRODUCTION

EH-IoTs address battery limitations by harvesting ambient energy, but unpredictable and limited power makes sensing, computing, and communication difficult. This survey organizes developments across commercialization, standards, sensing, intermittent computing, and communications.

  • Motivation: Battery replacement or recharging is costly, inconvenient, and sometimes impossible in large-scale IoT deployments.Batteries store finite energy, creating maintenance limitations for widely deployed devices.
  • Motivation: EH-IoTs harvest ambient energy to power IoT electronics without batteries, with examples including wireless switches, smart shoes, smartwatches, and energy meters.Energy sources include kinetic, solar, thermal, and electromagnetic energy.
  • Research Challenges: Small-form-factor harvesters provide limited, dynamic, and unpredictable power, complicating continuous sensing, reliable computing, and wireless power allocation.The survey identifies smooth operation under unpredictable supply as a central EH-IoT challenge.
  • Scope and Contribution: Existing EH surveys omit recent EH-IoT advances in commercialization, standardization, context sensing, intermittent computing, and emerging communications.This survey fills that coverage gap while briefly summarizing topics already extensively reviewed elsewhere.
  • Scope and Contribution: The survey covers commercial and standards developments, energy-efficient sensing from harvesting patterns, intermittent computing, and novel wireless communications.Its communication coverage includes reinforcement-learning-based power allocation, packet-less communication, and backscatter techniques.
  • System Architecture: A generic EH-IoT architecture replaces the battery with an energy-harvesting module while retaining conventional IoT electronics.The module combines an energy-harvesting transducer with power management, including storage and conversion components.

B. Commercial and Academic EH-IoTs

Self-powered IoTs span smart buildings, wearables, transportation, industrial systems, and implantable medical devices, using environmental, kinetic, thermal, RF, and bioenergy sources. Commercial products and research prototypes demonstrate battery-free sensing and communication across these domains.

  • Self-powered IoTs use diverse energy sources, including indoor light, RF, airflow, kinetic motion, body heat, and bioenergy.
  • Smart building: Smart-building prototypes and products support airflow sensing, power monitoring, activity recognition, wireless switching, and RF-powered air-pollution monitoring.
  • Wearable devices: Wearable systems harvest energy from walking, body heat, or sunlight to support activity recognition, sensing, and Bluetooth communication.
  • Industry/transportation: Transportation and industrial deployments use motion harvesters for process control, asset monitoring, tracking, and railway condition monitoring.
  • Implantable medical devices: Battery-powered implantable medical devices have limited lifespans and may require surgical replacement, motivating energy-harvesting alternatives.
  • The survey concludes that kinetic energy is especially popular for wearables, industry, and implantable medical devices, while smart buildings can use many energy sources.

C. Standardization Activities for EH-IoTs

EH-IoT standardization targets both wireless communication and energy-harvester testing. Initial activities address extremely limited and inconsistent energy supplies while supporting interoperability and deployment.

  • ISO and IEC standardization covers wireless communication protocols and testing methods for energy-harvesting transducers.
  • Wireless protocols: The wireless short packet protocol targets EH devices with extremely small, inconsistent energy supplies and carries short smart-home control messages.
  • Wireless protocols: WSP uses random access instead of conventional carrier sensing or listen-before-talk, allowing transmitter-only devices to participate.
  • Wireless protocols: WSP proposes amplitude and frequency modulation to balance energy efficiency and mobility, recommending FM for mobile devices.
  • Transducer testing: IEC has released nine standards covering testing methods for piezoelectric, electromagnetic, thermoelectric, and triboelectric harvesters.
  • The survey presents standardization as an initial step, while further optimization of EH-IoT sensing, computing, and communications remains necessary.

III. CONTEXT SENSING FROM EH PATTERNS

EH patterns can serve as proxies for conventional sensors, enabling context sensing from kinetic, thermoelectric, solar, and RF harvesters. Instantaneous signals provide richer detection, whereas accumulated energy saves power with coarser sensing.

  • Energy-harvesting patterns have been used to detect diverse contexts without continuously powering specialized sensors during energy-starving periods.
  • Current EH-IoTs power sensors, MCUs, and radios for context monitoring, while novel designs reuse harvesting signals for sensing.
  • Instantaneous-power analysis supports richer context detection but requires more frequent sampling of fluctuating power values.
  • Accumulated-energy analysis samples stored energy less often, reducing power use at the expense of coarser-grained context sensing.
  • Kinetic harvesting signals encode external motion and vibration patterns that can be processed for context recognition.
  • Human activity recognition: Walking and running produce distinct instantaneous voltage patterns and capacitor-charging slopes, enabling activity recognition without specialized sensors.
  • Human activity recognition: A piezoelectric harvester classified six activities with over 90% accuracy using waveform peaks, time lengths, and slopes.
  • Human activity recognition: Hand-worn KEH classified five activities with 80% accuracy versus 95% for an accelerometer, showing a recognition gap in that setting.

2) Transportation Mode Detection:

Energy-harvesting signals support transportation-mode detection and other sensing tasks, including calorie estimation, step counting, gait recognition, hotword detection, and airflow sensing. Reported results show useful accuracy but also setting-dependent limitations.

  • Transportation mode detection: Machine learning applied to wearable piezoelectric KEH voltage identified car, bus, or train travel with 85% accuracy and could distinguish train routes.
  • Calorie estimation: A wearable piezoelectric transducer’s voltage output can support calorie estimation using anthropometric features and AC voltage signals.
  • Calorie estimation: KEH-based calorie estimates were very close to estimates from a 3-axial accelerometer for most subjects during walking and running.
  • Step counting: A capacitor-voltage waveform can count walking steps because each stair-like waveform segment corresponds to one step.
  • Step counting: Capacitor voltage can also trigger Bluetooth beacon transmissions once a threshold is reached, allowing a nearby smartphone to count steps.
  • Step counting: Peak detection of KEH power generation identified steps with 96% accuracy across four subjects and 570 steps under varied walking scenarios.
  • Airflow sensing: Piezoelectric-harvester voltage inferred airflow speed with 0.2m/s error while sampling the transducer every 100 ms.

8) Acoustic Communication:

Energy harvesters can also support sensing and communication by exploiting environmental signals, including thermal gradients, solar illumination, and acoustic modulation. These approaches enable batteryless devices to detect contexts, estimate locations, and receive data under constrained power conditions.

  • Acoustic Communication: Acoustic modulation lets a nearby laptop speaker transmit to a piezoelectric harvester at 5 bps with a target bit error rate below 1% over 80 cm.A later modulation scheme hides the transmission within background music, addressing the audibility of the original approach.
  • Energy-Harvesting-Based Context Sensing: Thermoelectric harvesters can detect temperature-related contexts such as water flow, heating appliances, and chemical reactions.They convert temperature differences into usable electricity while serving as sensing elements.
  • Energy-Harvesting-Based Context Sensing: DoubleDip harvests a pipe’s thermal gradient to wake a water-flow sensor only when hot water flows, reducing energy use compared with fixed duty cycling.The harvested energy both wakes the sensor and compensates battery expenditure.
  • Solar-Based Localization: Solar cells support localization by measuring received light strength and exploiting distinct responses across locations and cell materials.LuxTrace estimates distance to light sources, while multi-cell sensing recognizes places from illumination and wavelength responses.
  • Solar-Based Localization: LuxTrace achieves 21 cm distance estimation accuracy at the 80% quantile using a wearable solar-cell indoor-positioning prototype.The system attaches solar cells to the user’s shoulder and uses received light strength with a trained model.
  • Solar-Based Localization: Solar-cell sensing recognizes places with 86.2% accuracy using two cell types trained on data from five cell types across nine locations.The method uses illumination-related energy generation and material-dependent wavelength responses.

2) Gesture Recognition:

Energy-harvesting hardware can act as a proxy sensor for gestures and other contexts, particularly by interpreting changes in harvested kinetic, solar, or radio-frequency signals. This can reduce power consumption, although accuracy may remain below that of specialized sensors.

  • Gesture Recognition: Solar panels can recognize gestures because nearby hand motion changes the harvested solar energy.This enables gesture recognition in solar-powered IoT devices without adding a conventional motion sensor.
  • Gesture Recognition: Transparent solar cells extend gesture recognition to device screens, despite weaker visible-light responsiveness caused by lower absorption efficiency.Their transparency creates opportunities to turn smartwatch displays and other screens into harvesters.
  • Gesture Recognition: Transparent solar cells recognize some gestures almost as well as opaque cells, whose average accuracy is 95%.The comparison indicates that transparency need not eliminate gesture-recognition capability.
  • Gesture Recognition: SolarGest provides a simulator that generates photocurrent waveforms for arbitrary solar cells and hand gestures, supporting research when physical cells are inaccessible.The simulator was developed because many newer cell types remain confined to research laboratories.
  • Lessons Learned: Across energy-harvesting sources, kinetic harvesting has received the most attention for context detection and is considered promising for replacing motion sensors in active scenarios.Solar and RF harvesting remain less mature but may support applications because light and radio conditions change rapidly with context.
  • Lessons Learned: Reusing harvesters as sensors can significantly reduce system power consumption and prolong device operation time.Accumulated-energy sensing may save more power but is expected to be less accurate than instantaneous signals.
  • Lessons Learned: Energy harvesters provide cruder measurements than specialized sensors, and recent studies report inferior context-detection performance.Further research is required before harvesters can replace specialized sensing hardware.

IV. INTERMITTENT COMPUTING FOR BATTERYLESS EH-IOTS

Intermittent power failures disrupt execution because volatile state is lost and devices repeatedly restart from checkpoints. Checkpointing preserves progress but introduces energy, memory, and recomputation costs, potentially creating repeated execution loops.

  • Intermittent Execution: EH-IoT power failures can occur several times per second, causing volatile variables and registers to be lost during execution.Checkpointing is therefore needed to periodically save volatile state in non-volatile memory.
  • Checkpointing Costs: Checkpointing requires copying state to non-volatile memory and restoring it after power returns, creating potentially significant energy overhead.The overhead varies because different program stages contain different numbers of state variables.
  • Sisyphean Execution: Insufficient harvested energy between checkpoints can force repeated small advances followed by rollbacks, preventing smooth program execution.This failure pattern can cause the program to revisit the same execution region repeatedly.
  • Sisyphean Execution: Sisyphean execution occurs when power fails after progress, restarting the program from an earlier checkpoint and repeatedly undoing later instructions.The illustrated example reaches instruction 8 before rolling back to instruction 4, forming a Sisyphean loop.
  • Intermittent Execution: The RSA glucose-encryption example illustrates how intermittent execution can preserve non-volatile variables while losing a volatile loop index after power failure.The example encrypts glucose value 2 as 2^3 mod 15 = 8 before an interruption.

3) State inconsistency:

Batteryless EH-IoTs face state and timing inconsistency when power failures interrupt updates across volatile and non-volatile storage. Research therefore combines checkpoint placement and activation strategies with mechanisms for preserving coherent execution across interruptions.

  • State Inconsistency: Checkpointing guarantees complete state preservation only when variables are maintained exclusively in volatile memory or exclusively in non-volatile memory.Splitting variables across both memory types can prevent rollback to a consistent global state.
  • Timing Inconsistency: A wearable monitoring multiple physiological signals can assign inconsistent timestamps when power fails between sensor readings.Different sensor sets may then be collected and reported at different times.
  • Timing Inconsistency: Persistent real-time clocks are difficult to support in EH-IoTs because frequent power failures interrupt their power supply.This prevents reliable timing across outages and can produce inaccurate temporal relationships among sensor data.
  • Checkpoint Placement and Activation: Checkpoint placement must balance checkpointing overhead against residual energy, but unpredictable failures make optimal placement difficult and burden programmers.Compile-time insertion creates potential checkpoints, while runtime activation determines whether to execute them.
  • Checkpoint Placement and Activation: Mementos uses programmer-inserted trigger points and Control Flow Graph strategies, while Idetic additionally considers recomputation energy through a Control Data Flow Graph.Pure CFG placement can waste energy by re-executing an entire function after a failure near its return.
  • Checkpoint Placement and Activation: Energy-aware runtime activation skips a checkpoint when available energy exceeds a threshold sufficient for reaching the next checkpoint.Threshold selection is challenging because computation costs differ between successive checkpoints and may require offline emulation.
  • Checkpointing at Run Time: Run-time checkpointing removes compile-time placement burdens but may save large volatile states too late and exceed the decoupling capacitor’s available energy.It also increases store-and-restore energy and non-volatile memory demands when many variables must be preserved.

3) Task-based Checkpointing:

Task-based execution models divide intermittently powered programs into atomic, often idempotent tasks to preserve consistency across power failures. Checkpointing, task design, watchdog timers, and timekeeping address execution progress and temporal correctness under unpredictable energy.

  • Task decomposition and idempotency: Long-running programs can be decomposed into short atomic tasks, with idempotency preventing repeated execution from producing inconsistent results.A write-after-read to non-volatile memory is an idempotency violation that can cause inconsistency after a power failure.
  • Checkpoint-based execution: Checkpoint-connected systems launch checkpoints at task boundaries, but compile-time idempotency analysis can impose heavy programmer burdens.DINO and Ratchet use static code analysis, while CleanCut automates part of the decomposition process.
  • Task-based execution: Chain and Alpaca replace line-based coding with task-granularity programming, connecting tasks through a control-flow graph without checkpointing between tasks.Their task-based model is intended to support execution across intermittent power failures.
  • Task data exchange: Chain assigns non-volatile channels to task pairs, whereas Alpaca addresses the resulting memory inefficiency from multiple task input and output versions.Both approaches rely on task inputs stored in non-volatile memory and idempotent task execution.
  • Execution progress: Watchdog timers address the ‘Sysiphean’ problem when available energy cannot complete the instructions or task before the next checkpoint.The same noncompletion problem can affect checkpointing and task-based execution when the energy budget is insufficient.
  • Timekeeping: SRAM or capacitor remanence decay can estimate elapsed time across power failures, but current techniques track elapsed time only within one minute.TARDIS uses SRAM decay for coarse-grained tracking, while CusTARD uses capacitor-voltage decay.

V. COMMUNICATIONS FOR EH-IOTS

EH-IoT communications must operate with tiny, unpredictable harvested energy. The survey organizes work around adaptive generative-radio optimization, packet-less signaling, and reflective radio, including reinforcement learning and IRS-based techniques.

  • Communication taxonomy: Energy-harvesting communication research spans generative radios, which actively create RF waves, and reflective radios, which modulate and reflect incident waves.This distinction structures the survey’s taxonomy of EH-IoT communications.
  • Generative-radio optimization: EH-optimized transmission schedules packet timing and power using channel and energy states, with offline methods requiring prior system knowledge and online methods using current observations.Online optimization is more practical when channel and energy-arrival statistics are difficult to obtain before deployment.
  • Reinforcement learning: The survey reviews reinforcement learning for communication optimization because it can support practical EH-IoT scenarios with limited prior knowledge of channel states and energy arrivals.Applications include transmission power allocation, transmission policies, user scheduling, and multiple networking contexts.
  • Packet-less communication: Packet-less communication conveys small event messages with a single pulse, avoiding packet headers and payloads to reduce transmission energy.The survey identifies packet-less protocols as an emerging approach for energy-constrained event detection.
  • Reflective radio: Reflective radio reduces transmission energy by modulating incident RF signals rather than generating new signals, but depends on available impinging waves.The survey covers backscatter evolution from RFID and ambient backscatter to IRS-based techniques.
  • Representative results: Reinforcement learning studies also reported a 0.3 prediction error ratio for solar energy arrival and a 0.99 coverage ratio for sensor-state scheduling.These results come from separate wireless sensor network studies.
  • Representative results: Q-learning reduced packet loss rate by 60% compared with random selection in one wireless information and power transfer study.The study used reinforcement learning to select the receiver’s data transmission rate after energy transfer.

C. Packet-less Communication

Packet-less communication uses minimal signaling for event detection, often transmitting a single pulse instead of a conventional packet. The approach is attractive for nanoscale IoT but faces scalability and practical deployment constraints.

  • Single-pulse signaling: Event-detection systems can transmit a single pulse when the message is binary or drawn from a small set of event types.This avoids packet overhead such as node addresses, preambles, and payloads.
  • Event notification: A structural-monitoring design used an ultrasonic pulse through an airplane wing’s metal substrate to report fault detection after vibration exceeded a threshold.Packet-less designs must still address event localization and event-type identification.
  • Nanoscale IoT: 1 pJ per pulse can account for 68% of total device power in IoNT, making pulse minimization central to energy balance.A single pulse can encode event type and location when harvested event energy determines pulse features such as width or amplitude.
  • Scalability: Amplitude-based classification does not scale to many sensor nodes, while direction-of-arrival hardware can separate location detection from event-type classification.The latter approach reduces the number of classes that pulse features must distinguish.

1) Reflective Radio Evolution:

Reflective radio evolves from reader-dependent RFID to ambient backscatter and programmable intelligent reflective surfaces. These techniques can reduce transmission power, but IRS research remains early and is dominated by simulation.

  • Reflective-radio principle: Reflective radios transmit by modulating and reflecting incident RF signals, avoiding active RF generation used by conventional radio modules.This design targets the wireless-communication component that typically dominates IoT sensor power consumption.
  • RFID limitations: RFID requires a nearby carrier emitter and limits tags to communication with the reader at its operating frequency.These requirements constrain pervasive deployment.
  • Ambient backscatter: Ambient backscatter reflects surrounding signals from sources such as TV towers, cellular stations, and Wi-Fi access points, enabling direct node-to-node communication.It can also piggyback IoT data on licensed spectrum.
  • Intelligent reflective surfaces: IRS uses electronically controlled two-dimensional elements whose reflection properties can be adjusted independently and in real time.This programmability supports controlled reflection for EH-IoT communication scenarios.
  • Lessons learned: Reflective radio, especially meta-surface-based designs, has potential to cut EH-IoT transmission power, while IRS technology remains in its infancy.The survey identifies substantial future research opportunities.
  • IRS applications: IRS research covers analytical modeling, low-complexity MIMO, and ambient backscatter with advanced modulations including QPSK and 8PSK.Reported systems support reflected-signal video transmission at high data rates.
  • Research maturity: IRS-enabled wireless communication studies have predominantly used simulations, whereas actual prototypes and experiments remain rare.This limits the current empirical validation of proposed IRS techniques.

VI. FUTURE RESEARCH DIRECTIONS

Future EH-IoT research should improve context sensing accuracy and privacy while jointly designing harvesting hardware and detection algorithms. It should also exploit multimodal and RF harvesting signals for richer, more efficient sensing.

  • Energy harvesters can replace specialized sensors, saving significant power but reducing context-detection performance.Improving sensing therefore requires advances in both algorithms and hardware.
  • Algorithm: Deep learning has not yet been explored for context detection from energy-harvesting signals, but deployment data or simulations could support training.
  • Hardware: Basic kinetic harvesters and solar cells have limitations for context recognition, including narrow, application-tuned frequency responses in many kinetic harvesters.
  • Hardware: Advanced harvesters could improve detection by capturing multiple axes, elements, modes, or wider optical-frequency ranges.
  • Hardware: Jointly optimizing harvesting capacity and context-detection performance is a proposed multidisciplinary direction.
  • Multimodal sensing: Combining kinetic and solar harvesting could provide richer signals for more accurate context detection, requiring algorithms that fuse multiple EH signals.
  • RF EH sensing: RF EH-based sensing remains rare, despite the potential to reuse Wi-Fi energy patterns for applications such as activity, heartbeat, fall, occupancy, and gait sensing.
  • Privacy: Energy-harvesting data can reveal private contexts, motivating future protections such as access control, authentication, and privacy-preserving learning.

D. Hardware Assistance for Intermittent Computing

EH-IoT deployments must operate reliably despite low, unstable energy, creating challenges for debugging, security, networking, communication, and system integration. The survey therefore points toward energy-aware development tools and holistic hardware, protocol, and deployment designs.

  • Conventional debugging can miss energy-harvesting bugs because it supplies constant power and extra monitoring code consumes energy.Energy-aware debugging tools remain limited, although one approach manipulates delivered energy to account for debugging overhead.
  • Low and unpredictable energy can expose EH-IoTs to attacks when devices suspend during secure protocols, including denial-of-service attacks against gateways.
  • EH duty cycling requires nodes to wait for accumulated energy, complicating MAC synchronization and wake-up design.
  • Energy desynchrony in multi-hop EH networks makes relay selection and routing overhead relevant to energy utility, delay, and throughput.
  • IRS ambient backscatter has reached data rates on the order of several Mbps, but current experiments do not exploit independently programmable elements.
  • Current EH-IoT architectures mainly replace batteries with EH modules, but holistic integration could reduce energy loss across harvesting, storage, processing, and radio components.
  • RF EH communication still needs transceiver designs that jointly optimize data reception and energy harvesting because conversion efficiency is low at small harvested-energy levels.
  • Large-scale EH-IoT deployment remains challenging because most research relies on simulations or single prototypes, while practical systems require compact, autonomous, robust, long-lived designs.
Loading 1905.03949v2…