Conventionally, during the knowledge distillation process (e.g., feature distillation), an additional projector is often required to perform feature transformation due to the dimension mismatch between the teacher and the student networks. Interestingly, we discovered that even if the student and the teacher have the same feature dimensions, adding a projector still helps to improve the distillation performance. In addition, projectors even improve logit distillation if we add them to the architecture too. Inspired by these surprising findings and the general lack of understanding of the projectors in the knowledge distillation process from existing literature, this paper investigates the implicit role that projectors play, but so far been overlooked. Our empirical study shows that the student with a projector 1) obtains a better trade-off between the training accuracy and the testing accuracy compared to the student without a projector when it has the same feature dimensions as the teacher, 2) better preserves its similarity to the teacher beyond shallow and numeric resemblance, from the view of Centered Kernel Alignment (CKA) (Kornblith et al., 2019), and 3) avoids being over-confident (Guo et al., 2017) as the teacher does at the testing phase. Motivated by the positive effects of projectors, we propose a projector ensemble-based feature distillation method to further improve distillation performance. Despite the simplicity of the proposed strategy, empirical results from the evaluation of classification tasks on benchmark datasets demonstrate the superior classification performance of our method on a broad range of teacher-student pairs and verify, from the aspects of CKA and model calibration that the student's features are of improved quality with the projector ensemble design.

Coreset selection methods aim to identify a representative subset of training data that preserves competitive performance. However, mainstream coreset selection approaches are model-specific and assume they already have full information about the target model when the coreset is selected. This largely restricts the usefulness of coreset selection in practice. This work aims to fill that gap by formulating and investigating the problem of cross-architecture generalizability of coresets: we develop a unified theoretical framework that analyzes the upper bound of coreset selection objective functions, extend it to scenarios involving multiple downstream architectures, and provide an empirical analysis on cross-architecture coreset performance. Based on our findings, we propose a novel ensemble scoring method that aggregates multi-source knowledge to enhance cross-architecture generalizability. Our extensive experiments across thirteen architectures and six selection ratios provide comprehensive verification of our theoretical analysis. The source code is available at https://github.com/diqichen91/CACS.git .

Coreset selection for classification often relies on assessing individual sample difficulty or importance, leading to sample-wise or range-based selection, but this can overlook the collective impact on model decision boundaries. Realizing that the representative power a coreset possesses is tightly associated with the decision boundaries a model can form on it, we propose a novel approach that directly optimizes the Decision Boundary (DB) formed by the selected coreset. Specifically, we ask: How can we collectively select samples to create a DB that is globally smoothed yet locally detailed, ensuring maximum generalizability and noise-resilience to the original dataset? To address this, we define two key objectives: (1) Global shape retention – The selected coreset should form a smoothed version of the original DB, preserving its overall structure and preventing overfitting; (2) Local detail preservation – While smoothing prevents overfitting, excessive smoothing risks losing critical nuances. Thus, the selection must also retain key points near the original DB to capture local complexities. We formulate these objectives as a convex quadratic optimization problem with linear constraints and solve it efficiently. Extensive evaluations demonstrate the consistent and substantial advantages of our method over the state-of-the-art coreset selection strategies. The source code is available at https://github.com/diqichen91/DBCore.git .

Knowledge Distillation (KD) aims to train a student model to mimic the behavior of a more powerful teacher model. In this paper, we reveal that through the lens of diffusion processes, student logits can be statistically treated as a noisy version of teacher logits, and KD helps reduce the noise level of student logits. This insight motivates us to design a framework leveraging KD to produce denoisable student logits that can be further recovered towards teacher logits via a reverse diffusion process. A key advantage of this approach is that the inference-diffusion process can occur in two physical locations and on separate devices, enabling a two-step and distributed inference process. The experimental results show that the derived denoisable student logits achieve comparable or even superior performance to standard KD’s, and the reverse diffusion process achieves a substantial improvement in accuracy, without needing the original image, thus preserving the privacy and security of the original data. Additionally, the logits can be further compressed before transmission, reducing the required bandwidth while achieving comparable overall performance.

Despite the significant success of open-vocabulary large foundation models (LFM) for segmentation in recent years, most real-life applications remain closed-vocabulary tasks and efficiency remains a critical factor for usability. Leveraging the power of open-vocabulary LFMs to create efficient, accurate closed-vocabulary segmentation models without the burden of pixel annotations provides an essential tool for employing these models effectively. This work introduces a novel Open-vocabulary to Closed-vocabulary Segmentation (O2CSeg) framework, which builds a compact, closed-vocabulary segmentation model from a large open-vocabulary LFM without any annotations. Our method capitalises on AI-assisted text prompts and learnable prompts that correspond to target class names to fully unleash the potential of open-vocabulary LFMs, eliminating the expensive annotation process. To address the challenge of noisy pseudo-labels generated by the teacher during training, we propose a confidence margin-based re-weighting knowledge distillation scheme, ensuring the student captures high-quality knowledge from the teacher. We validate our framework across diverse datasets and student configurations, demonstrating its efficacy in achieving high efficiency and accuracy for segmentation without annotations. In many cases, the derived student network outperforms its open-vocabulary teacher significantly with a higher mIoU and 40 times faster speed. Our contributions offer a promising direction for fast prototyping of efficient semantic segmentation models in scenarios where annotations are lacking, or label sets are evolving.

We propose LightLLM, a model that fine tunes pre-trained large language models (LLMs) for light-based sensing tasks. It integrates a sensor data encoder to extract key features, a contextual prompt to provide environmental information, and a fusion layer to combine these inputs into a unified representation. This combined input is then processed by the pre-trained LLM, which remains frozen while being fine-tuned through the addition of lightweight, trainable components, allowing the model to adapt to new tasks without altering its original parameters. This approach enables flexible adaptation of LLM to specialized light sensing tasks with minimal computational overhead and retraining effort. We have implemented LightLLM for three light sensing tasks: light-based localization, outdoor solar forecasting, and indoor solar estimation. Using real-world experimental datasets, we demonstrate that LightLLM significantly outperforms state-of-the-art methods, achieving 4.4x improvement in localization accuracy and 3.4x improvement in indoor solar estimation when tested in previously unseen environments. We further demonstrate that LightLLM outperforms ChatGPT-4 with direct prompting, highlighting the advantages of LightLLM's specialized architecture for sensor data fusion with textual prompts.

Collecting and curating data plays a crucial role in environmental surveying. In order to gather meaningful samples, it is often necessary to develop a real-time data curation system that processes data on-the-fly and enriches it with validated models from domain experts. One area where this is particularly important is underwater marine surveys, where the vastness of the sea requires human interaction in the curation process to focus on relevant areas for exploration. Additionally, ongoing surveys are susceptible to poor performance due to data drift, which hinders the ability to provide valuable feedback for guiding data collection. While recent advancements have shown promise in addressing these challenges, they often overlook the practical constraints associated with remote data collection, such as limited processing power and latency. To overcome these limitations, this paper proposes a real-time system that adapts to data drift and enables the recording of uncertain samples for further processing and analysis on shore. The results of our approach demonstrate a remarkable improvement in species recognition, achieving an almost 18% improvement compared to the best baseline method in unseen areas. Importantly, this improvement is achieved while meeting the real-time requirements of surveys and consuming only 15W of power. By effectively addressing the challenges of underwater data drift, our proposed approach provides an efficient and effective solution for environmental surveys.

We introduce LiDARSpectra, a novel approach utilizing mobile-integrated commodity Light Detection and Ranging (LiDAR) signals for synthetic indoor light spectral mapping. Our method incorporates an innovative material estimation algorithm into the LiDAR signal processing pipeline, accurately simulating reflected wavelengths from indoor surfaces. Utilizing low-resolution LiDAR scans enriched with material information, it eliminates the need for deploying dedicated spectral sensors, greatly simplifying the spectral mapping process. We validate our synthetic spectral maps against real sensor data and demonstrate their utility in applications such as indoor localization and solar energy provisioning. This presents an efficient solution for indoor spectral mapping with wide-ranging potential across fields like lighting design, indoor planting, environmental monitoring, and location-based services.

Seagrass meadows serve as critical carbon sinks, but estimating the amount of carbon they store requires knowledge of the seagrass species present. Underwater and surface vehicles equipped with machine learning algorithms can help to accurately estimate the composition and extent of seagrass meadows at scale. However, previous approaches for seagrass detection and classification have required supervision from patch-level labels. In this paper, we reframe seagrass classification as a weakly supervised coarse segmentation problem where image-level labels are used during training (25 times fewer labels compared to patch-level labeling) and patch-level outputs are obtained at inference time. To this end, we introduce SeaFeats, an architecture that uses unsupervised contrastive pre-training and feature similarity, and SeaCLIP, a model that showcases the effectiveness of large language models as a supervisory signal in domain-specific applications. We demonstrate that an ensemble of SeaFeats and SeaCLIP leads to highly robust performance. Our method outperforms previous approaches that require patch-level labels on the multi-species ‘DeepSeagrass’ dataset by 6.8% (absolute) for the class-weighted F1 score, and by 12.1% (absolute) for the seagrass presence/absence F1 score on the ‘Global Wetlands’ dataset. We also present two case studies for real-world deployment: outlier detection on the Global Wetlands dataset, and application of our method on imagery collected by the FloatyBoat autonomous surface vehicle.

Current video object detection (VOD) models often encounter issues with over-aggregation due to redundant aggregation strategies, which perform feature aggregation on every frame. This results in suboptimal performance and increased computational complexity. In this work, we propose an image-level Object Detection Difficulty (ODD) metric to quantify the difficulty of detecting objects in a given image. The derived ODD scores can be used in the VOD process to mitigate over-aggregation. Specifically, we train an ODD predictor as an auxiliary head of a still-image object detector to compute the ODD score for each image based on the discrepancies between detection results and ground-truth bounding boxes. The ODD score enhances the VOD system in two ways: 1) it enables the VOD system to select superior global reference frames, thereby improving overall accuracy; and 2) it serves as an indicator in the newly designed ODD Scheduler to eliminate the aggregation of frames that are easy to detect, thus accelerating the VOD process. Comprehensive experiments demonstrate that, when utilized for selecting global reference frames, ODD-VOD consistently enhances the accuracy of Global-frame-based VOD models. When employed for acceleration, ODD-VOD consistently improves the frames per second (FPS) by an average of 73.3% across 8 different VOD models without sacrificing accuracy. When combined, ODD-VOD attains state-of-the-art performance when competing with many VOD methods in both accuracy and speed. Our work represents a significant advancement towards making VOD more practical for real-world applications. The code will be released at https://github.com/bingqingzhang/odd-vod.

Monitoring of physiology and behavior of marine animals living undisturbed in their natural habitats can provide valuable information about their well-being and response to environmental stressors. We focus on detecting the feeding behavior in predatory fish using implantable biologgers that record and analyze electrocardiogram (ECG) signals. We propose a novel processing pipeline for resource-constrained embedded systems that can infer higher-level information, such as heart-rate and feeding events, from the ECG signals in situ. Our main contributions are in proposing efficient event detection algorithms that can reliably detect fish feeding events from noisy heart-rate data based on the unique statistical properties of feeding-induced changes in the heart-rate. We evaluate our approaches using an in-house biologger that we surgically implant in twelve coral trout fish and use to collect data during an experiment for a period of ten weeks and show that our signal processing pipeline performs well with noisy ECG signals overall. Specifically, our heart-rate estimation algorithm achieves errors of less than one beat per minute even in scenarios where popular algorithms used by domain specialists perform poorly. Furthermore, our feeding detection algorithms offer improved accuracy compared with the state-of-the-art algorithms while requiring significantly reduced computational and energy resources. We implement the proposed heart-rate estimation and feeding detection algorithms on the biologger and evaluate the associated system overhead. The results show that our proposed heart-rate estimation and feeding detection algorithms can run in-situ on the biologger as they demand rather small computational and energy resources that can conveniently be provisioned. This work is an important first step towards developing effective tools for long-term monitoring of high-level parameters pertaining to the health and behavior of marine animals in the wild.

We propose a novel Visible Light Positioning (VLP) method, called Iris, that uses light spectral information (LSI) to localize humans completely passively in the sense that it neither requires the user to carry any device, nor does it require any modifications to existing lighting infrastructure. Iris localizes a user based on the interference they produce on the LSI recorded at an array of spectral sensors embedded in the environment. We design a deep neural network that can effectively learn location fingerprints directly from the sensor LSI data and predict locations accurately under varying lighting conditions. We prototype Iris using a commercial-off-the-shelf light spectral sensor, AS7265x, which can measure light intensity over 18 different wavelength channels. We benchmark Iris against the state-of-the-art passive VLPs that rely on conventional photo-sensors capable of measuring only a single light intensity value aggregated over the entire visible spectrum. Our evaluations over two typical indoor environments, a 25 m2 one-bedroom apartment and a 13m × 8m office space, demonstrate that Iris can significantly reduce both the localization errors and the number of required sensors, while increasing robustness against changes in environmental lighting.

Machine learning (ML) algorithms have shown great potential in edge-computing environments, however, the literature mainly focuses on model inference only. We investigate how ML can be operationalized and how in-situ curation can improve the quality of edge applications, in the context of ML-assisted environmental surveys. We show that camera-enabled ML systems deployed on edge devices can enable scientists to perform real-time monitoring of species of interest or characterization of natural habitats. However, the benefit of this new technology is only as good as the quality and accuracy of the edge ML model inferences. In this demonstration, we show that with small additional time investment, domain scientists can manually curate ML model outputs and thus obtain highly reliable scientific insights, leading to more effective and scalable environmental surveys.

Crown-of-Thorn Starfish (COTS) outbreaks are a major cause of coral loss on the Great Barrier Reef (GBR) and substantial surveillance and control programs are ongoing to manage COTS populations to ecologically sustainable levels. In this paper, we present a comprehensive real-time machine learning-based underwater data collection and curation system on edge devices for COTS monitoring. In particular, we leverage the power of deep learning-based object detection techniques, and propose a resource-efficient COTS detector that performs detection inferences on the edge device to assist marine experts with COTS identification during the data collection phase. The preliminary results show that several strategies for improving computational efficiency (e.g., batch-wise processing, frame skipping, model input size) can be combined to run the proposed detection model on edge hardware with low resource consumption and low information loss.

In this paper, we investigate the impacts of location on the spectral distribution of received light, i.e., the intensity of light for different wavelengths, in indoor environments. Our findings show that, even when using the same light source, different locations exhibit slightly different spectral distribution due to reflections from their localised environment containing different materials or colours. Based on this observation, we present Spectral-Loc, a novel indoor localization method that employs light spectrum information to detect the device's position. Because spectrum sensors are increasingly being used in new products and applications, such as white balance in smartphone photography, Spectral-Loc can be quickly implemented without the need for extra hardware or infrastructure. We used a commercially available light spectrum sensor, the AS7265x, to prototype Spectral-Loc, which can measure light intensity over 18 different wavelength sub-bands. We benchmark the localization accuracy of Spectral-Loc against the conventional light intensity sensors that provide only a single intensity value. Our evaluations in two indoor areas, a meeting room and a large office, show that using light spectral information considerably decreases the localization error for different percentiles.

Communication among batteryless devices is key to their success in replacing traditional battery-supported systems. However, low and unpredictable availability of ambient energy combined with limited energy storage capacity of the devices make efficient communication challenging. As a stepping stone toward addressing this challenge, we propose to leverage common patterns in harvested energy across the devices. In this abstract, we explore one possible approach that exploits a property of many fluorescent light sources used worldwide: their brightness changes with double the power line frequency. We design a circuit that transforms the corresponding changes in energy harvested with a solar panel into a digital signal that is frequency- and phase-synchronized across multiple devices. Based on our design, we build a novel batteryless node, called Flync. Using two Flync nodes, we demonstrate that the synchronized signal can be generated with less than 1 µA and a maximum measured node to node jitter of 363.24 µs.

Each year, hundreds of thousands of people contract Healthcare Associated Infections (HAI). Poor hand hygiene compliance among healthcare workers is thought to be the leading cause of HAIs and methods were developed to measure compliance. Surprisingly, human observation is still considered the gold standard for measuring compliance by World Health Organization (WHO). Moreover, no automated solutions exist for monitoring hand hygiene techniques, such as "how to hand rub" technique by WHO. In this work, we introduce RFWash; the first radio-based device-free system for monitoring Hand Hygiene (HH) technique. On the technical level, HH gestures are performed back-to-back in a continuous sequence and pose a significant challenge to conventional two-stage gesture detection and recognition approaches. We propose a deep model that can be trained on unsegmented naturally-performed HH gesture sequences. RFWash evaluation demonstrates promising results for tracking HH gestures, achieving gesture error rate of≈67% compared to fully supervised approach. The work is a step towards practical RF sensing that can reliably operate inside future healthcare facilities.

Welcome to the 19th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN 2020), a premiere event on embedded sensing and networked systems that brings together researchers from academia, industry, and government. We are proud to see the continuing trend of increased interest in IPSN in recent years. A substantial 30% year-to-year increase in the number of submissions allowed us to share with you a particularly strong program this year. However, it is with mixed feelings that we write this message. We have all seen the world come to an abrupt stop in the last few weeks due to the spread of COVID-19 virus and will sadly not see you in the beautiful Darling Harbour in Sydney. Instead, the conference will happen in an entirely virtual format that will be split equally across three major world regions. This will inevitably harm the traditional cross-collaborative spirit of the Cyber- Physical Systems week, an event that brings together researchers across diverse fields, including Embedded, Hybrid, and Real-Time Systems. The inability to travel and exchange ideas face-to-face is a challenge, but also a wonderful opportunity to reach a wider global audience and we are determined to capitalize on the advantages and make the virtual IPSN a success. Testimony to the high quality of our program is the incredible work of our authors and organizers. Following the tradition in IPSN, our technical program committee brought together 27 distinguished experts covering the wide breadth of IPSN topics. The committee has collectively reviewed 124 submissions, providing a minimum of 3 high-quality reviews to each author. The top 50 of these submissions received additional 2 reviews and were discussed in person at the program committee meeting in St. Louis. We accepted 27 papers and followed a robust shepherding process to address the reviews prior to publication. We are proud of the authors and would like to thank them for striving to achieve the highest quality. We want to thank the program committee for their many hours of reviewing and discussion that made the program come together.

Monitoring of physiology and behavior of marine animals living undisturbed in their natural habitats can provide valuable data on their well-being and response to environmental stressors. We focus on detection of feeding of predatory fish using implantable biologgers that record electrocardiogram (ECG) signals. We propose a novel processing pipeline for resource-constrained embedded systems that can infer higher-level information, such as heart-rate and feeding events, from the ECG signals. Our main contribution is a lightweight change-detection algorithm, that can reliably detect fish feeding in noisy heart-rate data based on unique statistical properties of feeding-induced changes in heart-rate. We evaluate our approach using an in-house biologger that we surgically implant in twelve coral trouts over a period of ten weeks. We show that our signal processing pipeline performs well with noisy ECG signals overall. Specifically, our heart-rate estimation algorithm achieves errors of less than one beat per minute even in scenarios where popular algorithms used by domain scientists perform poorly. Furthermore, our feeding detection algorithm achieves good accuracy and matches the performance of state-of-the-art algorithms while requiring significantly less memory and computational resources. This work is an important first step towards long-term monitoring of high-level condition and health of marine animals in the wild.

Conventional systems for motion context detection rely on batteries to provide the energy required for sampling a motion sensor. Batteries, however, have limited capacity and, once depleted, have to be replaced or recharged. Kinetic Energy Harvesting (KEH) allows to convert ambient motion and vibration into usable electricity and can enable batteryless, maintenance free operation of motion sensors. The signal from a KEH transducer correlates with the underlying motion and may thus directly be used for context detection, saving space, cost and energy by omitting the accelerometer. Previous work uses the open circuit or the capacitor voltage for sensing without using the harvested energy to power a load. In this paper, we propose to use other sensing points in the KEH circuit that offer information-rich sensing signals while the energy from the harvester is used to power a load. We systematically analyze multiple sensing signals available in different KEH architectures and compare their performance in a transport mode detection case study. To this end, we develop four hardware prototypes, conduct an extensive measurement campaign and use the data to train and evaluate different classifiers. We show that sensing the harvesting current signal from a transducer can be energy positive, delivering up to ten times as much power as it consumes for signal acquisition, while offering comparable detection accuracy to the accelerometer signal for most of the considered transport modes.

Each year, hundreds of thousands of people contract Healthcare Associated Infections (HAIs). Poor hand hygiene compliance among healthcare workers is thought to be the leading cause of HAIs and methods were developed to measure compliance. Surprisingly, human observation is still considered the gold standard for measuring compliance by World Health Organization (WHO). Moreover, no automated solutions exist for monitoring hand hygiene techniques, such as "how to hand rub" technique by WHO. In this paper, we introduce RFWash; the first radio-based device-free system for monitoring Hand Hygiene (HH) technique. On the technical level, HH gestures are performed back-to-back in a continuous sequence and pose a significant challenge to conventional two-stage gesture detection and recognition approaches. We propose a deep model that can be trained on unsegmented naturally-performed HH gesture sequences. RFWash evaluation demonstrates promising results for tracking HH gestures, achieving gesture error rate of < 8% when trained on 10-second segments, which reduces manual labelling overhead by ≈ 67% compared to fully supervised approach. The work is a step towards practical RF sensing that can reliably operate inside future healthcare facilities.

Energy-efficient location tracking with battery-powered devices using energy harvesting necessitates duty-cycling of GPS to prolong the system lifetime. We propose an energy and mobility-aware scheduling framework that adapts to real-world dynamics to achieve optimal long-term tracking performance. To forecast energy, the framework uses an exponentially weighted moving average filter to compute a virtual energy budget for the remainder of the forecast period. The virtual energy budget is then used as input for our proposed information-based GPS sampling approach, which estimates the current tracking error through dead-reckoning and schedules a new GPS sample when the error exceeds a given threshold. In order to improve the long-term tracking performance, the threshold is adapted based on the current energy and movement trends to balance the expected information gain from a new GPS sample with its energy cost. We evaluate our approach on empirical traces from wild flying foxes and compare it to strategies that sample GPS using fixed and adaptive duty cycles and by using dead-reckoning with a fixed threshold. Our analysis shows that the proposed information-based GPS sampling strategy reduces the mean tracking error compared to existing methods and approaches the performance of the optimal offline sampling strategy.

Careful energy management is a prerequisite for long-term, unattended operation of solar-harvesting sensing systems. We observe that in many applications the utility of sensed data varies over time, but current energy-management algorithms do not exploit prior knowledge of these variations for making better decisions. This paper presents PreAct, the first energy-management algorithm that exploits time-varying utility to optimize application performance. PreAct's design combines strategic long-term planning of future energy utilization with feedback control to compensate for deviations from the expected conditions. We implement PreAct on a low-power microcontroller and compare it against the state of the art on multiple years of real-world data. Our results demonstrate that PreAct is up to 53 % more effective in utilizing harvested solar energy and significantly more robust to uncertainties and inefficiencies of practical systems. These gains translate into an improvement of 28% in the end-to-end performance of a real-world application we investigate when using PreAct.

We consider the problem of data collection from a network of energy harvesting sensors, applied to tracking mobile assets in rural environments. Our application constraints favor a fair and energy-aware solution, with heavily duty-cycled sensor nodes communicating with powered base stations. We study a novel scheduling optimization problem for energy harvesting mobile sensor network, that maximizes the amount of collected data under the constraints of radio link quality and energy harvesting efficiency, while ensuring a fair data reception. We show that the problem is NP-complete and propose a heuristic algorithm to approximate the optimal scheduling solution in polynomial time. Moreover, our algorithm is flexible in handling progressive energy harvesting events, such as with solar panels, or opportunistic and bursty events, such as with Wireless Power Transfer. We use empirical link quality data, solar energy, and WPT efficiency to evaluate the proposed algorithm in extensive simulations and compare its performance to state-of-the-art. We show that our algorithm achieves high data reception rates, under different fairness and node lifetime constraints.

Indoor localization using radio signals is challenging. A recently proposed algorithm based on WiFi channel state information is an effective solution. However, it relies on a computationally expensive grid search. We propose a new algorithm based on a modified matrix pencil method that reduces the computational complexity by two orders of magnitude without any loss of accuracy.

This work presents a variation of a data collection problem referred as TSP-Data Collection (TSP-DC). Previously we have demonstrated that a two-stage algorithm using Linear-Programming Optimization and Gradient-Descent Optimization (LPO-GDO) is able to solve TSP-DC for minimum tour time. This poster abstract shows that LPO-GDO is also able to solve TSP-DC for an objective function of minimum energy, giving different solutions for different relative energy costs of data transmission and sink movement, and for high or low data loads at sensor nodes.

Sensor networks increasingly rely on harvesting energy from the environment to sense, process, and transmit data. Online energy availability forecasting and energy management are critical to ensure long-term energy-neutral operation of battery-powered energy-harvesting sensor nodes. Existing methods focus on applications with time-invariant utility and custom-tailored hardware platforms, which limits their effectiveness across diverse application domains, different platforms, and in the face of aging hardware components. To address these limitations, we formulate an optimisation problem with respect to time-varying utility under the given hardware constraints. We also present PREACT, an online energy-management algorithm that approximates the optimal solution to the optimisation problem by incorporating long-term energy forecasting.

Heart rate is one of the most important vital signals for personal health tracking. A number of approaches were proposed to monitor heart rate, ranging from wearables to device-less systems. While WiFi has been shown to track heart rate accurately, existing solutions rely on directional antennas to improve the signal quality and ultimately the accuracy of heart rate estimation. Special hardware used in these approaches limits their applicability and truly device-less and ubiquitous heart rate monitoring is yet to be achieved.

We consider the problem of self-localization by a resource-constrained mobile node given perturbed anchor position information and distance estimates from the anchor nodes. We consider normally-distributed noise in anchor position information. The distance estimates are based on the log-normal shadowing path-loss model for the RSSI measurements. The available solutions to this problem are based on complex and iterative optimization techniques such as semidefinite programming or second-order cone programming, which are not suitable for resource-constrained environments. In this paper, we propose a closed-form weighted least-squares solution. We calculate the weights by taking into account the statistical properties of the perturbations in both RSSI and anchor position information. We also estimate the bias of the proposed solution and subtract it from the proposed solution. We evaluate the performance of the proposed algorithm considering a set of arbitrary network topologies in comparison to an existing algorithm that is based on a similar approach but only accounts for perturbations in the RSSI measurements. We also compare the results with the corresponding Cramer-Rao lower bound. Our experimental evaluation shows that the proposed algorithm can substantially improve the localization performance in terms of both root mean square error and bias.

The prediction of periodical time-series remains challenging due to various types of scaling, misalignments and distortion effects. Here, we propose a novel model called Temporal embedding-enhanced convolutional neural Network (TeNet) to learn repeatedly-occurring-yet-hidden structural elements in periodical time-series, called abstract snippet detectors, to predict future changes. Our model effectively learns a new feature space for a time-series dataset. In the new feature space, distorted time-series that have implicit similarity but substantial differences in value and sequence to regular patterns are re-aligned to the regular patterns in the dataset, and subsequently contribute to a robust prediction mode. The model is robust to various types of distortions and misalignments and demonstrates strong prediction power for periodical time-series. We conduct extensive experiments and discover that the proposed model shows significant and consistent advantages over existing methods on a variety of data modalities ranging from human mobility to household power consumption records, when evaluated under four metrics. The model is also robust to various factors such as number of samples, variance of data, numerical ranges of data etc. The experiments verify that the intuition behind the model can be generalized to multiple data types and applications and promises significant improvement in prediction performance across the datasets studied.

State-of-the-art trajectory compression methods usually involve high space-time complexity or yield unsatisfactory compression rates, leading to rapid exhaustion of memory, computation, storage, and energy resources. Their ability is commonly limited when operating in a resource-constrained environment especially when the data volume (even when compressed) far exceeds the storage limit. Hence, we propose a novel online framework for error-bounded trajectory compression and ageing called the Amnesic Bounded Quadrant System (ABQS), whose core is the Bounded Quadrant System (BQS) algorithm family that includes a normal version (BQS), Fast version (FBQS), and a Progressive version (PBQS). ABQS intelligently manages a given storage and compresses the trajectories with different error tolerances subject to their ages. In the experiments, we conduct comprehensive evaluations for the BQS algorithm family and the ABQS framework. Using empirical GPS traces from flying foxes and cars, and synthetic data from simulation, we demonstrate the effectiveness of the standalone BQS algorithms in significantly reducing the time and space complexity of trajectory compression, while greatly improving the compression rates of the state-of-the-art algorithms (up to 45 percent). We also show that the operational time of the target resource-constrained hardware platform can be prolonged by up to 41 percent. We then verify that with ABQS, given data volumes that are far greater than storage space, ABQS is able to achieve 15 to 400 times smaller errors than the baselines. We also show that the algorithm is robust to extreme trajectory shapes.

Long-term location tracking, where trajectory compression is commonly used, has gained high interest for many applications in transport, ecology, and wearable computing. However, state-of-the-art compression methods involve high space-time complexity or achieve unsatisfactory compression rate, leading to rapid exhaustion of memory, computation, storage and energy resources. We propose a novel online algorithm for error-bounded trajectory compression called the Bounded Quadrant System (BQS), which compresses trajectories with extremely small costs in space and time using convex-hulls. In this algorithm, we build a virtual coordinate system centered at a start point, and establish a rectangular bounding box as well as two bounding lines in each of its quadrants. In each quadrant, the points to be assessed are bounded by the convex-hull formed by the box and lines. Various compression error-bounds are therefore derived to quickly draw compression decisions without expensive error computations. In addition, we also propose a light version of the BQS version that achieves O(1) complexity in both time and space for processing each point to suit the most constrained computation environments. Furthermore, we briefly demonstrate how this algorithm can be naturally extended to the 3-D case. Using empirical GPS traces from flying foxes, cars and simulation, we demonstrate the effectiveness of our algorithm in significantly reducing the time and space complexity of trajectory compression, while greatly improving the compression rates of the state-of-the-art algorithms (up to 47%). We then show that with this algorithm, the operational time of the target resource-constrained hardware platform can be prolonged by up to 41%.

Increasing human population, economical development, and industrialization of our society lead to disturbance of natural ecosystems. To prevent long-term damage of the environment, the affected ecosystems need to be rehabilitated to a sustainable form after industrial operations cease. Environment rehabilitation is a long-term process that is complex and costly as it includes restoration of soil, water bodies, and reintroduction of plant, insect, and animal species. We present a wireless sensor network for evidence-based rehabilitation of disturbed natural ecosystems. Drawing on our case study of an open-cut surface mine in rural Australia, we show that microclimate data can provide insights into the efficiency of specific rehabilitation processes, disentangle the impact of microclimate on the rehabilitation success, and provide early indicators into potential rehabilitation problems. We worked with ecologists to provide domain-based advice on addressing a range of rehabilitation problems and developed a system that periodically generates reports on the rehabilitation status of areas of interest. The report incorporates data from expert surveys from the field and microclimate data from sensors, and provides recommendations to improve rehabilitation in under-performing areas. Such evidence-based assessment of rehabilitation is an important step towards ensuring compliance with set rehabilitation objectives, potentially leading to both more successful and less costly environment rehabilitation.

Delay Tolerant Networks (DTN) are characterized by the absence of continuous connectivity resulting in high delivery delays that may exceed the acceptable limit for practical applications. In this paper, we address this issue by introducing Augur. Augur is a new routing protocol for DTNs targeted to minimize delays of message delivery. The routing scheme benefits from the spatiotemporal history data of the nodes to route message s only through gateways having less expected delay to deliver a message to its destination. We demonstrate through a comparative evaluation that Augur outperforms the state of the art DTN protocols in terms of delivery probability, overhead ratio and latency. We found that at low traffic rates Augur reduces the overhead ratio by up to 94%, and by up to 88% at high traffic. We also observed that the improvement in latency was reduced by up to half over the existing protocols in both traffic rates while still improving the delivery probability of messages.

Wireless sensor network platforms share the wireless communication channels with Wi-Fi and Bluetooth based networks, resulting in heavy use of these bands. As a consequence, platforms in wireless sensor networks need to carefully consider external interference to achieve reliable communication. In this paper, we present an experimental analysis of wireless channels for wireless sensor network operating on dual frequency bands. Specifically, we designed a set of detailed experiments aiming to find out correlation patterns in 900 MHz and 2.4 GHz ISM bands. We conducted our experiments on two testbeds and investigated the band correlation between two distinctive radio transceivers in two different office-space environments. From our data samples, we quantified frequency channel and band correlations in parallel experiments that eliminate artifacts stemming from different external activity on the test site. We found that network formed in 900 MHz band has 15% more connectivity than network formed in 2.4 GHz band, even on radio channels that minimize overlap with Wi-Fi networks.

Researchers have proposed many multi-channel and dual-band communication systems to address the limitations of single-channel hardware and software. The most common dual-band communication for sensor network applications use 900 MHz and 2.4 GHz radios. There are now some testbeds such as Flocklab and Twonet that allow experimentation with dual-radio systems. However, there is no widely accepted efficient and comprehensive mechanism to survey all the channels of the dual band systems during networking experiments. In this work, we evaluate two mechanisms that can survey the RF environment in multiple channels in both the bands concurrently. We implement these two mechanisms on Twonet and evaluate them. We find that the channel scanning methodology is generally sound but sometimes static channel assignment may be attractive due to its simplicity.

Radio connectivity in wireless sensor networks is highly intermittent due to unpredictable and time-varying noise and interference patterns in the environment. Because link qualities are not predictable prior to deployment, current deterministic solutions to unreliable links, such as increasing network density or transmission power, require overprovisioning of network resources and do not always improve reliability. We propose a new dual-radio network architecture to improve communication reliability in wireless sensor networks. Specifically, we show that radio transceivers operating at well-separated frequencies and spatially separated antennas offer robust communication, high link diversity, and better interference mitigation. We derive the optimal parameters for the dual-transceiver setup from frequency and space diversity in theory. We observe that frequency diversity holds the most benefits as long as the antennas are sufficiently separated to prevent coupling. Our experiments on an indoor/outdoor testbed confirm the theoretical predictions and show that radio diversity can significantly improve end-to-end delivery rates and network stability at only a small increase in energy cost over a single radio. Simulation experiments further validate the improvements in multiple topology configurations, but also reveal that the benefits of radio diversity are coupled to the number of available routing paths to the destination.

Long-term outdoor localisation with battery-powered devices remains an unsolved challenge, mainly due to the high energy consumption of GPS modules. The use of inertial sensors and short-range radio can reduce reliance on GPS to prolong the operational lifetime of tracking devices, but they only provide coarse-grained control over GPS activity. In this paper, we introduce our feature-rich lightweight Camazotz platform as an enabler of Multimodal Activity-based Localisation~(MAL), which detects activities of interest by combining multiple sensor streams for fine-grained control of GPS sampling times. Using the case study of long-term flying fox tracking, we characterise the tracking, connectivity, energy, and activity recognition performance of our module under both static and 3-D mobile scenarios. We use Camazotz to collect empirical flying fox data and illustrate the utility of individual and composite sensor modalities in classifying activity. We evaluate MAL for flying foxes through simulations based on retrospective empirical data. The results show that multimodal activity-based localisation reduces the power consumption over periodic GPS and single sensor-triggered GPS by up to 77% and 14% respectively, and provides a richer event type dissociation for fine-grained control of GPS sampling.

Limited visibility into the global network state renders testing and debugging sensor network applications a challenging task. Existing debugging methods are often non-intrusive and require modifications of the binary image. Hardware based debugging instrumentation such as JTAG has not been widely used beyond a single node, mainly due to its relatively high cost and lack of software support for distributed debugging. This demonstration presents a novel architecture for distributed debugging of wireless sensor networks using a low-cost extension board to access the on-chip debug module of the node's processor. Connecting several of those debug boards using a backbone network provides distributed control and monitoring of the sensor network in test.

We investigate the problem of scheduling sensor node up-times to maximize the utility of the data they collect while operating within their resource constraints. We show that the optimal scheduling algorithm can improve data utility by more than 70% compared to naive schedules. We consider a suite of sensors with different capabilities and resource demands and represent their subsets as virtual sensors. For each virtual sensor, we calculate its optimal data fusion parameters and evaluate the sensors' performance in a given environment. The selection of virtual sensors best suited to collect data in a given environment can be modeled as an Integer Linear programming problem, and we study three different algorithms to solve the problem efficiently. We evaluate the performance of virtual sensor scheduling algorithms by extensive simulation. We show that even though the naive greedy scheduling approaches work well in some scenarios, none of them are able to match our best scheduling algorithm consistently, under varying environmental conditions and sensor resources.

The high popularity of wireless sensor networks has led to novel applications with diverse, and sometimes demanding, data communication requirements, for example, streaming camera images in surveillance applications. In response bulk-data transfer protocols were proposed that provide low latency and high throughput communication over multiple hops. However, due to typical hardware platforms only providing a single radio, which implies that forwarding nodes need to serialize send and receive actions, the maximum end-to-end throughput is limited to 1/2 the radio capacity. To bridge this performance gap we present Fast Forward, a connection-oriented multi-hop bulk-data transfer protocol optimized for dual-radio platforms, data packets are sent across a path of alternating radio and frequency channels to exploit parallel transfers and avoid intra-path interference. We implemented Fast Forward in TinyOS to run on the Opal platform equipped with two IEEE 802.15.4 radios. In this paper we show that, with some minor tweaking of the original protocol stack to streamline internal access to the SPI bus, Fast Forward is capable of operating both radios in parallel so packets can be forwarded at full speed. We have evaluated Fast Forward on a 12-node testbed in an office environment. The sustained throughput peaks around 23.7 kBps, or 76 % of the radio capacity while the best single-radio protocol flattens out at 19 %. When introducing artificial packet loss the built-in link-level acknowledgements ensure that Fast Forward manages to deliver packets with high yield (close to 100 %) at the sink across 11 hops.

We present Twonet, a large-scale sensor network testbed with dual-radio nodes. Twonet has 100 Opal nodes with low-power 32-bit ARM CPU and 2.4 GHz and 900 MHz radios. These nodes are managed by a network of 20 Raspberry Pi nodes at tier 2 and a PC server at tier 1. These nodes together provide a robust testbed for public access. Twonet represents a major addition to the collection of wireless sensor network testbeds that are publicly available. We hope Twonet's availability will foster sensor network research based on a modern 32-bit sensor node architecture and multi-channel wireless networking.

Development of wireless sensor network applications remains a challenge, due to lack of visibility into the global network state. Debugging instrumentation using printf-like instructions affects the execution timing and non-intrusive approaches, such as JTAG, have not been used beyond a single node due to their high cost.

Location sensing provides endless opportunities for a wide range of applications in GPS-obstructed environments, where, typically, there is a need for a higher degree of accuracy. In this article, we focus on robust range estimation , an important prerequisite for fine-grained localization. Motivated by the promise of acoustic in delivering high ranging accuracy, we present the design, implementation, and evaluation of acoustic (both ultrasound and audible) ranging systems. We distill the limitations of acoustic ranging and present efficient signal designs and detection algorithms to overcome the challenges of coverage, range, accuracy/resolution, tolerance to Doppler's effect, and audible intensity. We evaluate our proposed techniques experimentally on TWEET, a low-power platform purpose-built for acoustic ranging applications. Our experiments demonstrate an operational range of 20m (outdoor) and an average accuracy ≈2cm in the ultrasound domain. Finally, we present the design of an audible-range acoustic tracking service that encompasses the benefits of a near-inaudible acoustic broadband chirp and approximately two times increase in Doppler tolerance to achieve better performance.

Low power listening (LPL) has been widely adopted to save energy in wireless sensor networks. However, LPL is ineffective in adapting to dynamic networks with asymmetric traffic patterns, as it sets a network-wide check interval. As a result, nodes with low data traffic waste significant energy resources doing idle listening. This problem is particularly exacerbated in multi-radio networks where majority of data comes through the most reliable radio and the duty cycles of other radios could be reduced. We address this issue in AutoSync, a protocol that combines synchronous LPL with automatic selection of check intervals to reduce energy consumption in both single and multi-radio networks. We first present the justification for AutoSync's design, and we then discuss our implementation of AutoSync in TinyOS. We compare AutoSync against existing protocols in both simulations and empirical experiments. Results show that AutoSync attains a substantial increase in the operational lifetime and mean power consumption over existing protocols in single radio networks and even more in dual radio networks.

Robot platforms have been proposed to address challenges of wireless sensor network deployments in dynamic environments. Enabling a fraction of nodes to be mobile can lead to improvements in sensing coverage, network lifetime, and network connectivity. We present the design of ArmBot, a Cortex-M3 based robot platform, that natively supports TinyOS applications and system services. ArmBot is cost-efficient as it is built from off-the-shelf hardware components and versatile due to its power-efficient ARM processor core. We evaluate the platform in the context of core sensor network services, such as time-synchronization, and analyze its energy efficiency under varying mobility and computation loads.

We consider the problem of distributing time-sensitive information from a collection of sources to mobile users traversing a wireless mesh network. Our strategy is to distributively select a set of well-placed nodes (warehouses) to act as intermediaries between the information sources and clusters of users. Warehouses are selected via the distributed construction of Hierarchical Well-Separated Trees (HSTs), which are sparse structures that induce a natural spatial clustering of the network. Unlike many traditional multicast protocols, our approach is not data driven. Rather, it is agnostic to the number and position of sources as well as to the mobility patterns of users. Whereas source-rooted tree multicast algorithms construct a separate routing infrastructure to support each source, our sparse and flexible infrastructure is precomputed and efficiently reused by sources and users, its cost amortized over time. Moreover, the route acquisition delay inherent in on-demand wireless ad hoc network protocols is avoided by exploiting the HST addressing scheme. Our algorithm ensures with high probability a guaranteed stretch bound for the information delivery path, and is robust to lossy links and node failure by providing alternative HST-induced routes. Nearby users are clustered and their requests aggregated, further reducing communication overhead.

In this paper, we present a routing scheme that exploits knowledge about the behavior of mobile sinks within a network of data sources to minimize energy consumption and network congestion. For delay-tolerant network applications, we propose to route data not to the sink directly, but to send it instead to a relay node along an announced or predicted path of the mobile node that is close to the data source. The relay node will stash the information until the mobile node passes by and picks up the data. We use linear programming to find optimal relay nodes that minimize the number of necessary transmissions while guaranteeing robustness against link and node failures, as well as trajectory uncertainty.

We present the Whirlpool Routing Protocol (WARP), which efficiently routes data to a node moving within a static mesh. The key insight in WARP's design is that data traffic can use an existing routing gradient to efficiently probe the topology, repair the routing gradient, and communicate these repairs to nearby nodes.

Mobile wireless sensors require position updates for tracking and navigation. We present a localization technique that uses the Doppler shift in radio transmission frequency observed by stationary sensors. We consider two scenarios. In the first, the mobile node is carried by a person. In the second, the mobile node controls a robot. In both approaches the mobile node transmits an RF signal, and infrastructure nodes measure the Doppler-shifted frequency. Such measurements enable us to calculate the position and velocity of the mobile transmitter. Our experimental results demonstrate that this technique is viable and accurate for resource-constrained mobile sensor tracking and navigation.

We present a method to extract topology information from detection events of mobile entities moving through a network of binary sensors. We extract the topological structure of possible paths in the network by analyzing the time correlation of events at different sensors. The histograms of time delays between any two sensors contain the necessary information to reconstruct the network topology. This data is heavily corrupted by noise due to multiple agents in the network. We therefore use a mixture model of multiple Gaussian and a uniform distribution to explicitly isolate the noise. Our algorithm yields a graph representing the topology of our sensor network along with average travel time between nodes.

Location-awareness is an important requirement for many mobile wireless applications today. When GPS is not applicable because of the required precision and/or the resource constraints on the hardware platform, radio interferometric ranging may offer an alternative. In this paper, we present a technique that enables the precise tracking of multiple wireless nodes simultaneously. It relies on multiple infrastructure nodes deployed at known locations measuring the position of tracked mobile nodes using radio interferometry. In addition to location information, the approach also provides node velocity estimates by measuring the Doppler shift of the interference signal. The performance of the technique is evaluated using a prototype implementation on mote-class wireless sensor nodes. Finally, a possible application scenario of dirty bomb detection in a football stadium is briefly described.

In this paper, we address the problem of tracking cooperative mobile nodes in wireless sensor networks. Aiming at a resource efficient solution, we advocate the use of sensors that maintain their location information and rely on the tracking service only when their locations change. In the proposed approach, the tracked node transmits a signal and infrastructure nodes measure the Doppler shifts of the transmitted signal. We show that Mica2 motes can measure RF Doppler shifts with 0.2 Hz accuracy corresponding to a 0.14 m/s error in relative speed estimates using radio inter-ferometric technique.

This paper presents an enhanced version of a novel radio interferometric positioning technique for node localization in wireless sensor networks that provides both high accuracy and long range simultaneously. The ranging method utilizes two transmitters emitting radio signals at almost the same frequencies. The relative location is estimated by measuring the relative phase offset of the generated interference signal at two receivers. Here, we analyze how the selection of carrier frequencies affects the precision and maximum range. Furthermore, we describe how the interplay of RF multipath and ground reflections degrades the ranging accuracy. To address these problems, we introduce a technique that continuously refines the range estimates as it converges to the localization solution. Finally, we present the results of a field experiment where our prototype achieved 4~cm average localization accuracy for a quasi-random deployment of 16 COTS motes covering the area of two football fields. The maximum range measured was 170~m, four times the observed communication range. Consequently, node deployment density is no longer constrained by the localization technique, but rather by the communication range.

Experiences developing a sensor network-based acoustic shooter localization system are presented. The system is able to localize the position of a shooter and the trajectory of the projectile using observed acoustic events, such as the muzzle blast and the ballistic shockwave. The network consists of a large number of cheap sensors communicating through an ad-hoc wireless network, which enables the system to resolve multiple simultaneous acoustic sources, eliminate multipath effects, tolerate multiple sensor failures while providing good coverage and high accuracy, even in such challenging environment as urban terrain. The paper describes the hardware and software platform developed for this application and summarizes the lessons learned during the development of the system.

We present a novel radio interference based sensor localization method for wireless sensor networks. The technique relies on a pair of nodes emitting radio waves simultaneously at slightly different frequencies. The carrier frequency of the composite signal is between the two frequencies, but has a very low frequency envelope. Neighboring nodes can measure the energy of the envelope signal as the signal strength. The relative phase offset of this signal measured at two receivers is a function of the distances between the four nodes involved and the carrier frequency. By making multiple measurements in an at least 8-node network, it is possible to reconstruct the relative location of the nodes in 3D. Our prototype implementation on the MICA2 platform yields an average localization error as small as 3 cm and a range of up to 160 meters. In addition to this high precision and long range, the other main advantage of the Radio Interferometric Positioning System (RIPS) is the fact that it does not require any sensors other than the radio used for wireless communication.

An ad-hoc wireless sensor network-based system is presented that detects and accurately locates shooters even in urban environments. The localization accuracy of the system in open terrain is competitive with that of existing centralized countersniper systems. However, the presented sensor network-based solution surpasses the traditional approach because it can mitigate acoustic multipath effects prevalent in urban areas and it can also resolve multiple simultaneous shots. These unique characteristics of the system are made possible by employing novel sensor fusion techniques that utilize the spatial and temporal diversity of multiple detections. In this article, in addition to the overall system architecture, the middleware services and the unique sensor fusion algorithms are described. An analysis of the experimental data gathered during field trials at US military facilities is also presented.

Wireless sensor network applications, similarly to other distributed systems, often require a scalable time synchronization service enabling data consistency and coordination. This paper describes the Flooding Time Synchronization Protocol (FTSP), especially tailored for applications requiring stringent precision on resource limited wireless platforms. The proposed time synchronization protocol uses low communication bandwidth and it is robust against node and link failures. The FTSP achieves its robustness by utilizing periodic flooding of synchronization messages, and implicit dynamic topology update. The unique high precision performance is reached by utilizing MAC-layer time-stamping and comprehensive error compensation including clock skew estimation. The sources of delays and uncertainties in message transmission are analyzed in detail and techniques are presented to mitigate their effects. The FTSP was implemented on the Berkeley Mica2 platform and evaluated in a 60-node, multi-hop setup. The average per-hop synchronization error was in the one microsecond range, which is markedly better than that of the existing RBS and TPSN algorithms.

An ad-hoc wireless sensor network-based system is presented that detects and accurately locates shooters even in urban environments. The system consists of a large number of cheap sensors communicating through an ad-hoc wireless network, thus it is capable of tolerating multiple sensor failures, provides good coverage and high accuracy, and is capable of overcoming multipath effects. The performance of the proposed system is superior to that of centralized countersniper systems in such challenging environment as dense urban terrain. In this paper, in addition to the overall system architecture, the acoustic signal detection, the most important middleware services and the unique sensor fusion algorithm are also presented. The system performance is analyzed using real measurement data obtained at a US Army MOUT (Military Operations in Urban Terrain) facility.