Eating disorder (ED) recovery extends beyond therapy sessions, unfolding in vulnerable moments embedded in everyday life and relationships. Yet empirical understanding of how these moments arise, how supporters contribute, and how technologies might offer timely, contextual assistance remains limited. To address this gap, we conducted a design session and two-week diary study with 27 individuals with ED and 12 social supporters. Our analysis identified diverse contexts in which patients and supporters perceived support to be needed, and the forms of support they envisioned digital tools could offer. While many needs were mutually recognized, the actual practice of support often involved mismatches, suggesting opportunities for technologies to help mediate supportive engagement. Our study contributes empirical insight into everyday support moments in ED recovery and highlights opportunities to design digital interventions that provide context-sensitive assistance, empower supporters, and extend care beyond clinical settings.

Live music provides a uniquely rich setting for studying creativity and interaction due to its spontaneous nature. The pursuit of live music agents—intelligent systems supporting real-time music performance and interaction—has captivated researchers across HCI, AI, and computer music for decades, and recent advancements in AI suggest unprecedented opportunities to evolve their design. However, the interdisciplinary nature of music has led to fragmented development across research communities, hindering effective communication and collaborative progress. In this work, we bring together perspectives from these diverse fields to map the current landscape of live music agents. Based on our analysis of 184 systems across both academic literature and video, we develop a comprehensive design space that categorizes dimensions spanning usage contexts, interactions, technologies, and ecosystems. By highlighting trends and gaps in live music agents, our design space offers researchers, designers, and musicians a structured lens to understand existing systems and shape future directions in real-time human-AI music co-creation. We release our annotated systems as a living artifact at https://live-music-agents.github.io.

We investigate how non-native English speakers (NNESs) interact with diverse information aids to assess and select AI-generated paraphrases.We develop ParaScope, an AI paraphrasing assistant that integrates diverse information aids, such as back-translation, explanations, and usage examples, and logs user interaction data.Our in-lab study with 22 NNESs reveals that user preferences for information aids vary by language proficiency, with workflows progressing from global to more detailed information.While backtranslation was the most frequently used aid, it was not a decisive factor in suggestion acceptance; users combined multiple information aids to make informed decisions.Our findings demonstrate the potential of explainable AI paraphrasing tools to enhance NNESs' confidence, autonomy, and writing efficiency, while also emphasizing the importance of thoughtful design to prevent information overload.Based on these findings, we offer design implications for explainable AI paraphrasing tools that support NNESs in making informed decisions when using AI writing systems.

Developing new machine learning applications often requires the collection of new datasets. However, existing datasets may already contain relevant information to train models for new purposes. We propose SoundCollage: a framework to discover new classes within audio datasets by incorporating (1) an audio pre-processing pipeline to decompose different sounds in audio samples, and (2) an automated model-based annotation mechanism to identify the discovered classes. Furthermore, we introduce the clarity measure to assess the coherence of the discovered classes for better training new downstream applications. Our evaluations show that the accuracy of downstream audio classifiers within discovered class samples and a held-out dataset improves over the baseline by up to 34.7% and 4.5%, respectively. These results highlight the potential of SoundCollage in making datasets reusable by labeling with newly discovered classes. To encourage further research in this area, we open-source our code at github.com/nokia-bell-labs/audio-class-discovery.

Composed Image Retrieval (CIR) retrieves relevant images based on a reference image and accompanying text describing desired modifications. However, existing CIR methods only focus on retrieving the target image and disregard the relevance of other images. This limitation arises because most methods employing contrastive learning-which treats the target image as positive and all other images in the batch as negatives-can inadvertently include false negatives. This may result in retrieving irrelevant images, reducing user satisfaction even when the target image is retrieved. To address this issue, we propose Query-Relevant Retrieval through Hard Negative Sampling (QuRe), which optimizes a reward model objective to reduce false negatives. Additionally, we introduce a hard negative sampling strategy that selects images positioned between two steep drops in relevance scores following the target image, to effectively filter false negatives. In order to evaluate CIR models on their alignment with human satisfaction, we create Human-Preference FashionIQ (HP-FashionIQ), a new dataset that explicitly captures user preferences beyond target retrieval. Extensive experiments demonstrate that QuRe achieves state-of-the-art performance on FashionIQ and CIRR datasets while exhibiting the strongest alignment with human preferences on the HP-FashionIQ dataset. The source code is available at https://github.com/jackwaky/QuRe.

Deep learning models perform poorly when domain shifts exist between training and test data. Test-time adaptation (TTA) is a paradigm to mitigate this issue by adapting pre-trained models using only unlabeled test samples. However, existing TTA methods can fail under severe domain shifts, while recent active TTA approaches requiring full-class labels are impractical due to high labeling costs. To address this issue, we introduce a new setting of TTA with binary feedback, which uses a few binary feedbacks from annotators to indicate whether model predictions are correct, thereby significantly reducing the labeling burden of annotators. Under the setting, we propose BiTTA, a novel dual-path optimization framework that leverages reinforcement learning to balance binary feedback-guided adaptation on uncertain samples with agreement-based self-adaptation on confident predictions. Experiments show BiTTA achieves substantial accuracy improvements over state-of-the-art baselines, demonstrating its effectiveness in handling severe distribution shifts with minimal labeling effort.

Test-Time Adaptation (TTA) adjusts models using unlabeled test data to handle dynamic distribution shifts. However, existing methods rely on frequent adaptation and high computational cost, making them unsuitable for resource-constrained edge environments. To address this, we propose SNAP, a sparse TTA framework that reduces adaptation frequency and data usage while preserving accuracy. SNAP maintains competitive accuracy even when adapting based on only 1\% of the incoming data stream, demonstrating its robustness under infrequent updates. Our method introduces two key components: (i) Class and Domain Representative Memory (CnDRM), which identifies and stores a small set of samples that are representative of both class and domain characteristics to support efficient adaptation with limited data; and (ii) Inference-only Batch-aware Memory Normalization (IoBMN), which dynamically adjusts normalization statistics at inference time by leveraging these representative samples, enabling efficient alignment to shifting target domains. Integrated with five state-of-the-art TTA algorithms, SNAP reduces latency by up to 93.12\%, while keeping the accuracy drop below 3.3\%, even across adaptation rates ranging from 1\% to 50\%. This demonstrates its strong potential for practical use on edge devices serving latency-sensitive applications. The source code is available at https://github.com/chahh9808/SNAP.

The rapid growth of wearable devices has opened exciting opportunities for context-aware multi-device collaboration, where multiple devices can provide enhanced user experience tailored to user needs and conditions. However, it also presents a unique challenge of reliably determining whether a set of wearables is being used by the same individual. In real-world scenarios, device sharing, exchanging, or unintended use can cause privacy risks and degraded functionality. Existing solutions primarily rely on accelerometer data to match movement patterns across devices, but they perform poorly during stationary or varied non-repetitive activities. In this paper, we introduce BioQ, a method that unobtrusively detects wearable co-location by generating and matching bio-cues. These bio-cues are generated from on-body wearable sensor data and embedded into a common latent space. Furthermore, when devices share the same sensor types, BioQ can effectively integrate multiple sensor sources to improve cue generation and matching. Experimental results show that BioQ outperforms baselines in bio-cue generation and matching and is resource-effective in model training, inference, and energy use. Our code is available at https://github.com/Nokia-Bell-Labs/contextual-biological-cues.

Self-supervised learning enables effective model pre-training on large-scale unlabeled data, which is crucial for user-specific fine-tuning in mobile sensing applications. However, pre-trained models often face significant domain shifts during fine-tuning due to user diversity, leading to performance degradation. To address this, we propose SelfReplay, an adaptive approach designed to align self-supervised models to different domains. SelfReplay consists of two stages: MetaSSL, which leverages meta-learning with self-supervised learning to pre-train domain-adaptive weights, and ReplaySSL, which further adapts the pre-trained model to each user's domain by replaying the meta-learned self-supervised task with a few user-specific samples. This produces a personalized model tailored to each user. Evaluations on mobile sensing benchmarks demonstrate that SelfReplay outperforms existing baselines, improving the F1-score by 9.4%p on average. On-device analyses on a commodity smartphone show the efficiency of SelfReplay's adaptation step, required just once after deployment, with SimCLR completing in only 10 seconds while using less than 100MB of memory.

Hate speech remains a persistent and unresolved challenge in online platforms. Content moderators, working on the front lines to review user-generated content and shield viewers from hate speech, often find themselves unprotected from the mental burden as they continuously engage with offensive language. To safeguard moderators' mental well-being, we designed HateBuffer, which anonymizes targets of hate speech, paraphrases offensive expressions into less offensive forms, and shows the original expressions when moderators opt to see them. Our user study with 80 participants consisted of a simulated hate speech moderation task set on a fictional news platform, followed by semi-structured interviews. Although participants rated the hate severity of comments lower while using HateBuffer, contrary to our expectations, they did not experience improved emotion or reduced fatigue compared with the control group. In interviews, however, participants described HateBuffer as an effective buffer against emotional contagion and the normalization of biased opinions in hate speech. Notably, HateBuffer did not compromise moderation accuracy and even contributed to a slight increase in recall. We explore possible explanations for the discrepancy between the perceived benefits of HateBuffer and its measured impact on mental well-being. We also underscore the promise of text-based content modification techniques as tools for a healthier content moderation environment.

Pre-training representations acquired via self-supervised learning could achieve high accuracy on even tasks with small training data. Unlike in vision and natural language processing domains, pre-training for IMU-based applications is challenging, as there are few public datasets with sufficient size and diversity to learn generalizable representations. To overcome this problem, we propose IMG2IMU that adapts pre-trained representation from large-scale images to diverse IMU sensing tasks. We convert the sensor data into visually interpretable spectrograms for the model to utilize the knowledge gained from vision. We further present a sensor-aware pre-training method for images that enables models to acquire particularly impactful knowledge for IMU sensing applications. This involves using contrastive learning on our augmentation set customized for the properties of sensor data. Our evaluation with four different IMU sensing tasks shows that IMG2IMU outperforms the baselines pre-trained on sensor data by an average of 9.6%p F1-score, illustrating that vision knowledge can be usefully incorporated into IMU sensing applications where only limited training data is available.

Digital food content’s popularity is underscored by recent studies revealing its addictive nature and association with disordered eating. Notably, individuals with eating disorders exhibit a positive correlation between their digital food content consumption and disordered eating behaviors. Based on these findings, we introduce FoodCensor, an intervention designed to empower individuals with eating disorders to make informed, conscious, and health-oriented digital food content consumption decisions. FoodCensor (i) monitors and hides passively exposed food content on smartphones and personal computers, and (ii) prompts reflective questions for users when they spontaneously search for food content. We deployed FoodCensor to people with binge eating disorder or bulimia (n=22) for three weeks. Our user study reveals that FoodCensor fostered self-awareness and self-reflection about unconscious digital food content consumption habits, enabling them to adopt healthier behaviors consciously. Furthermore, we discuss design implications for promoting healthier digital content consumption practices for vulnerable populations to specific content types.

Despite a rich history of investigating smartphone overuse intervention techniques, AI-based just-in-time adaptive intervention (JITAI) methods for overuse reduction are lacking. We develop Time2Stop, an intelligent, adaptive, and explainable JITAI system that leverages machine learning to identify optimal intervention timings, introduces interventions with transparent AI explanations, and collects user feedback to establish a human-AI loop and adapt the intervention model over time. We conducted an 8-week field experiment (N=71) to evaluate the effectiveness of both the adaptation and explanation aspects of Time2Stop. Our results indicate that our adaptive models significantly outperform the baseline methods on intervention accuracy (>32.8% relatively) and receptivity (>8.0%). In addition, incorporating explanations further enhances the effectiveness by 53.8% and 11.4% on accuracy and receptivity, respectively. Moreover, Time2Stop significantly reduces overuse, decreasing app visit frequency by 7.0 ∼ 8.9%. Our subjective data also echoed these quantitative measures. Participants preferred the adaptive interventions and rated the system highly on intervention time accuracy, effectiveness, and level of trust. We envision our work can inspire future research on JITAI systems with a human-AI loop to evolve with users.

Test-time adaptation (TTA) has emerged as a viable solution to adapt pretrained models to domain shifts using unlabeled test data. However, TTA faces challenges of adaptation failures due to its reliance on blind adaptation to unknown test samples in dynamic scenarios. Traditional methods for out-of-distribution performance estimation are limited by unrealistic assumptions in the TTA context, such as requiring labeled data or retraining models. To address this issue, we propose AETTA, a label-free accuracy estimation algorithm for TTA. We propose the prediction disagreement as the accuracy estimate, calculated by comparing the target model prediction with dropout inferences. We then improve the prediction disagreement to extend the applicability of AETTA under adaptation failures. Our extensive evaluation with four baselines and six TTA methods demonstrates that AETTA shows an average of 19.8%p more accurate estimation compared with the baselines. We further demonstrate the effectiveness of accuracy estimation with a model recovery case study, showcasing the practicality of our model recovery based on accuracy estimation. The source code is available at https://github.com/taeckyung/AETTA.

Large language models (LLMs) have demonstrated exceptional abilities across various domains.However, utilizing LLMs for ubiquitous sensing applications remains challenging as existing text-prompt methods show significant performance degradation when handling long sensor data sequences.We propose a visual prompting approach for sensor data using multimodal LLMs (MLLMs).We design a visual prompt that directs MLLMs to utilize visualized sensor data alongside the target sensory task descriptions.Additionally, we introduce a visualization generator that automates the creation of optimal visualizations tailored to a given sensory task, eliminating the need for prior task-specific knowledge.We evaluated our approach on nine sensory tasks involving four sensing modalities, achieving an average of 10% higher accuracy than text-based prompts and reducing token costs by 15.8×.Our findings highlight the effectiveness and cost-efficiency of visual prompts with MLLMs for various sensory tasks.The source code is available at https://github. com/diamond264/ByMyEyes. ### InstructionYou are an expert in sensor data analysis.Given the sensor data, determine the correct answer from the options listed in the question.Provide the answer with the format of ANSWER , where ANSWER corresponds to one of the options listed in the question.If the answer is not in the options, choose the most possible option.The ECG data is collected from a lead II ECG sensor.The ECG data is recorded over 10 seconds.The data is normalized with the statistics of the user's data.Please refer to the provided examples and use them to answer the following question for the target data.### Examples *Example of normal*: Average heartbeat in the ECG signal (list of ['lead II']): [-0.34, -0.34, -0.35, -0.35, -0.36, … ECG_P_Peaks in the ECG signal (list of (index, value)): [(21, 0.08), (129, 0.19), (239, 0.22), … ECG_Q_Peaks in the ECG signal (list of (index, value)): [(35, -0.61), (137, -0.46), (246, -0.48), … ECG_S_Peaks in the ECG signal (list of (index, value)): [(42, -1.4), (149, -1.35), (260, -1.33), … ECG_T_Peaks in the ECG signal (list of (index, value)): [(63, 2.18), (171, 2.11), (282, 2.31), … *Example of conduction disturbance*: Average heartbeat in the ECG signal (list of ['lead II']): [-0.15, -0.24, -0.29, -0.31, -0.28, … ECG_P_Peaks in the ECG signal (list of (index, value)): [(4, 0.14), (57, 0.3), (103, 0.22), … ECG_Q_Peaks in the ECG signal (list of (index, value)): [(14, -0.05), (65, -0.05), (109, -0.07), … ECG_S_Peaks in the ECG signal (list of (index, value)): [(22, -2.28), (73, -2.1), (124, -2.35), … ECG_T_Peaks in the ECG signal (list of (index, value)): [(82, 0.03), (142, 0.64), (245, 0.25), … ### Question Average heartbeat in the ECG signal (list of ['lead II']): [-0.39, -0.39, -0.39, -0.39, -0.4,… ECG_P_Peaks in the ECG signal (list of (index, value)): [(15, 0.14), (94, -0.26), (173, -0.23), … ECG_Q_Peaks in the ECG signal (list of (index, value)): [(23, -0.27), (102, -0.81), (182, -0.55), … ECG_S_Peaks in the ECG signal (list of (index, value)): [(34, 0.15), (116, -0.51), (192, -0.45), … ECG_T_Peaks in the ECG signal (list of (index, value)): [(50, 1.39), (130, 1.1), (209, 1.31), … *Question*: When the sensor data is used for a task for classifying ECG data into 2 categories: conduction disturbance, normal, what is the most likely answer among ['conduction disturbance', 'normal']?*Answer*:

When deployed in mobile scenarios, deep learning models often suffer from performance degradation due to domain shifts. Test-Time Adaptation (TTA) offers a viable solution, but current approaches face latency issues on resource-constrained mobile devices. We propose TESLA: Time-Efficient Sparse and Lightweight Adaptation strategy for real-time mobile applications, which skips adaptation for specific batches to increase the inference sample rate. Our method balances model accuracy and inference speed by accumulating domain-informative samples from non-adapted batches and sparsely adapting them. Experiments on edge devices demonstrate competitive accuracy even with sparse adaptation rates, highlighting the effectiveness of our approach in real-time mobile applications. Our strategy can seamlessly integrate with existing lightweight adaptation and optimization algorithms, further accelerating inference across diverse mobile systems.

Federated Learning (FL) is a distributed machine learning framework that trains accurate global models while preserving clients' privacy-sensitive data. However, most FL approaches assume that clients possess labeled data, which is often not the case in practice. Federated Semi-Supervised Learning (FSSL) addresses this label deficiency problem, targeting situations where only the server has a small amount of labeled data while clients do not. However, a significant performance gap exists between Centralized Semi-Supervised Learning (SSL) and FSSL. This gap arises from confirmation bias, which is more pronounced in FSSL due to multiple local training epochs and the separation of labeled and unlabeled data. We propose $(FL)^2$, a robust training method for unlabeled clients using sharpness-aware consistency regularization. We show that regularizing the original pseudo-labeling loss is suboptimal, and hence we carefully select unlabeled samples for regularization. We further introduce client-specific adaptive thresholding and learning status-aware aggregation to adjust the training process based on the learning progress of each client. Our experiments on three benchmark datasets demonstrate that our approach significantly improves performance and bridges the gap with SSL, particularly in scenarios with scarce labeled data.

Psychiatrists diagnose mental disorders via the linguistic use of patients.Still, due to data privacy, existing passive mental health monitoring systems use alternative features such as activity, app usage, and location via mobile devices.We propose FedTherapist, a mobile mental health monitoring system that utilizes continuous speech and keyboard input in a privacy-preserving way via federated learning.We explore multiple model designs by comparing their performance and overhead for FedTherapist to overcome the complex nature of on-device language model training on smartphones.We further propose a Context-Aware Language Learning (CALL) methodology to effectively utilize smartphones' large and noisy text for mental health signal sensing.Our IRBapproved evaluation of the prediction of selfreported depression, stress, anxiety, and mood from 46 participants shows higher accuracy of FedTherapist compared with the performance with non-language features, achieving 0.15 AU-ROC improvement and 8.21% MAE reduction.

Test-time adaptation (TTA) aims to address distributional shifts between training and testing data using only unlabeled test data streams for continual model adaptation. However, most TTA methods assume benign test streams, while test samples could be unexpectedly diverse in the wild. For instance, an unseen object or noise could appear in autonomous driving. This leads to a new threat to existing TTA algorithms; we found that prior TTA algorithms suffer from those noisy test samples as they blindly adapt to incoming samples. To address this problem, we present Screening-out Test-Time Adaptation (SoTTA), a novel TTA algorithm that is robust to noisy samples. The key enabler of SoTTA is two-fold: (i) input-wise robustness via high-confidence uniform-class sampling that effectively filters out the impact of noisy samples and (ii) parameter-wise robustness via entropy-sharpness minimization that improves the robustness of model parameters against large gradients from noisy samples. Our evaluation with standard TTA benchmarks with various noisy scenarios shows that our method outperforms state-of-the-art TTA methods under the presence of noisy samples and achieves comparable accuracy to those methods without noisy samples. The source code is available at https://github.com/taeckyung/SoTTA.

Many digital wellbeing tools help users monitor and control social media use on their smartphones by tracking and setting limits on their usage time. Tracking is typically done at the granularity of phone- or app-level; however, recent social media apps provide various features such as direct messaging, comment reading/posting, and content uploading/viewing. While it is possible to track and analyze within-app feature usage, little is known about the effect of granularity on smartphone interventions. We designed and developed FinerMe to explore how the granularity of interventions (app-level vs. feature-level) affects the usage of popular social media such as Instagram and YouTube on smartphones. We conducted a field study with 56 participants over 16 days that consisted of three phases: baseline collection, self-reflection, and self-reflection with restrictive interventions. The results showed that while both app-level and feature-level interventions similarly reduced social media use, feature-level interventions enabled users to spend less time on passive app features related to content consumption (e.g., following feed on Instagram, and viewing comments on YouTube) than app-level interventions. Moreover, when self-reflection is combined with restrictive interventions at the feature-level, users were more reflective on their usage behavior than when done at the app-level.

This article investigates how the expansion of array size may improve the spatial diversity of state-of-the-art Wi-Fi system and increase its throughput. With comprehensive Wi-Fi measurement studies with augmented antennas, we identify the potential performance gain atop spatial diversity gains from existing technologies like MIMO and beamforming. We propose WINAS, a general Wi-Fi intelligent antenna selection scheme with full system implementation that can be easily integrated with commodity Wi-Fi AP. WINAS provides substantially improved throughput for downlink traffics. Our experimental evaluation suggests that WINAS improves Wi-Fi throughput up to 1.56x, and 1.47x in average, in real user-based evaluation.

Reflecting on stress-related data is critical in addressing one’s mental health. Personal Informatics (PI) systems augmented by algorithms and sensors have become popular ways to help users collect and reflect on data about stress. While prediction algorithms in the PI systems are mainly for diagnostic purposes, few studies examine how the explainability of algorithmic prediction can support user-driven self-insight. To this end, we developed MindScope, an algorithm-assisted stress management system that determines user stress levels and explains how the stress level was computed based on the user’s everyday activities captured by a smartphone. In a 25-day field study conducted with 36 college students, the prediction and explanation supported self-reflection, a process to re-establish preconceptions about stress by identifying stress patterns and recalling past stress levels and patterns that led to coping planning. We discuss the implications of exploiting prediction algorithms that facilitate user-driven retrospection in PI systems.

Various automated eating detection wearables have been proposed to monitor food intakes. While these systems overcome the forgetfulness of manual user journaling, they typically show low accuracy at outside-the-lab environments or have intrusive form-factors (e.g., headgear). Eyeglasses are emerging as a socially-acceptable eating detection wearable, but existing approaches require custom-built frames and consume large power. We propose MyDJ, an eating detection system that could be attached to any eyeglass frame. MyDJ achieves accurate and energy-efficient eating detection by capturing complementary chewing signals on a piezoelectric sensor and an accelerometer. We evaluated the accuracy and wearability of MyDJ with 30 subjects in uncontrolled environments, where six subjects attached MyDJ on their own eyeglasses for a week. Our study shows that MyDJ achieves 0.919 F1-score in eating episode coverage, with 4.03 × battery time over the state-of-the-art systems. In addition, participants reported wearing MyDJ was almost as comfortable (94.95%) as wearing regular eyeglasses.

We demonstrate StressTrendmeter, a mobile app that targets college students for anonymously sharing the source of stress via the form of hashtags, viewing stress topics based on trends, and providing social support through the empathy button and hashtag-based chat.

Federated Learning (FL) trains a machine learning model on distributed clients without exposing individual data. Unlike centralized training that is usually based on carefully-organized data, FL deals with on-device data that are often unfiltered and imbalanced. As a result, conventional FL training protocol that treats all data equally leads to a waste of local computational resources and slows down the global learning process. To this end, we propose FedBalancer, a systematic FL framework that actively selects clients' training samples. Our sample selection strategy prioritizes more "informative" data while respecting privacy and computational capabilities of clients. To better utilize the sample selection to speed up global training, we further introduce an adaptive deadline control scheme that predicts the optimal deadline for each round with varying client training data. Compared with existing FL algorithms with deadline configuration methods, our evaluation on five datasets from three different domains shows that FedBalancer improves the time-to-accuracy performance by 1.20~4.48× while improving the model accuracy by 1.1~5.0%. We also show that FedBalancer is readily applicable to other FL approaches by demonstrating that FedBalancer improves the convergence speed and accuracy when operating jointly with three different FL algorithms.

Test-time adaptation (TTA) is an emerging paradigm that addresses distributional shifts between training and testing phases without additional data acquisition or labeling cost; only unlabeled test data streams are used for continual model adaptation. Previous TTA schemes assume that the test samples are independent and identically distributed (i.i.d.), even though they are often temporally correlated (non-i.i.d.) in application scenarios, e.g., autonomous driving. We discover that most existing TTA methods fail dramatically under such scenarios. Motivated by this, we present a new test-time adaptation scheme that is robust against non-i.i.d. test data streams. Our novelty is mainly two-fold: (a) Instance-Aware Batch Normalization (IABN) that corrects normalization for out-of-distribution samples, and (b) Prediction-balanced Reservoir Sampling (PBRS) that simulates i.i.d. data stream from non-i.i.d. stream in a class-balanced manner. Our evaluation with various datasets, including real-world non-i.i.d. streams, demonstrates that the proposed robust TTA not only outperforms state-of-the-art TTA algorithms in the non-i.i.d. setting, but also achieves comparable performance to those algorithms under the i.i.d. assumption. Code is available at https://github.com/TaesikGong/NOTE.

People experience various stressful events in their daily lives. Receiving social support, especially from peers who went through a similar experience, helps individuals cope with such stress. We propose StressTrendmeter, a mobile application that targets college students for anonymously sharing the source of stress via the form of hashtags, viewing stress topics based on trends, and providing social support through the empathy button and hashtag-based chat. We deployed StressTrendmeter to 222 students from two universities for five weeks. With hashtags and trending features, students found StressTrendmeter (i)helpful to spontaneously yet concisely articulate their stress topics and (ii) easy to browse through and become aware of issues around the campus. Our study reveals that social sharing with StressTrendmeter brought awareness, resonance, and accountability as students empathized and expressed support. Based on our study, we share design implications for social support systems with community awareness.

Many applications utilize sensors on mobile devices and apply deep learning for diverse applications. However, they have rarely enjoyed mainstream adoption due to many differentindividual conditionsusers encounter. Individual conditions are characterized by users’ unique behaviors and different devices they carry, which collectively make sensor inputs different. It is impractical to train countless individual conditions beforehand and we thus argue meta-learning is a great approach in solving this problem. We presentMetaSensethat leverages “seen” conditions in training data to adapt to an “unseen” condition (i.e., the target user). Specifically, we design a meta-learning framework that learns “how to adapt” to the target via iterative training sessions of adaptation. MetaSense requires very few training examples from the target (e.g., one or two) and thus requires minimal user effort. In addition, we propose asimilar condition detector(SCD) that identifies when the unseen condition has similar characteristics to seen conditions and leverages this hint to further improve the accuracy. Our evaluation with 10 different datasets shows that MetaSense improves the accuracy of state-of-the-art transfer learning and meta learning methods by 15 and 11 percent, respectively. Furthermore, our SCD achieves additional accuracy improvement (e.g., 15 percent for human activity recognition).

We present Rushmore, a system that securely displays static or animated images using TrustZone. The core functionality of Rushmore is to securely decrypt and display encrypted images (sent by a trusted party) on a mobile device. Although previous approaches have shown that it is possible to securely display encrypted images using TrustZone, they exhibit a critical limitation that significantly hampers the applicability of using TrustZone for display security. The limitation is that, when the trusted domain of TrustZone (the secure world) takes control of the display, the untrusted domain (the normal world) cannot display anything simultaneously. This limitation comes from the fact that previous approaches give the secure world exclusive access to the display hardware to preserve security. With Rushmore, we overcome this limitation by leveraging a well-known, yet overlooked hardware feature called an IPU (Image Processing Unit) that provides multiple display channels. By partitioning these channels across the normal world and the secure world, we enable the two worlds to simultaneously display pixels on the screen without sacrificing security. Furthermore, we show that with the right type of cryptographic method, we can decrypt and display encrypted animated images at 30 FPS or higher for medium-to-small images and at around 30 FPS for large images. One notable cryptographic method we adapt for Rushmore is visual cryptography, and we demonstrate that it is a light-weight alternative to other cryptographic methods for certain use cases. Our evaluation shows that in addition to providing usable frame rates, Rushmore incurs less than 5% overhead to the applications running in the normal world.

Digital intervention tools against problematic smartphone usage help users control their consumption on smartphones, for example, by setting a time limit on an app. However, today's social media apps offer a mix of quasiessential and addictive features in an app (e.g., Instagram has following feeds, recommended feeds, stories, and direct messaging features), which makes it hard to apply a uniform logic for all uses of an app without a nuanced understanding of feature-level usage behaviors. We study when and why people regret using different features of social media apps on smartphones. We examine regretful feature uses in four smartphone social media apps (Facebook, Instagram, YouTube, and KakaoTalk) by utilizing feature usage logs, ESM surveys on regretful use collected for a week, and retrospective interviews from 29 Android users. In determining whether a feature use is regretful, users considered different types of rewards they obtained from using a certain feature (i.e., social, informational, personal interests, and entertainment) as well as alternative rewards they could have gained had they not used the smartphone (e.g., productivity). Depending on the types of rewards and the way rewards are presented to users, probabilities to regret vary across features of the same app. We highlight three patterns of features with different characteristics that lead to regretful use. First, "following"-based features (e.g., Facebook's News Feed and Instagram's Following Posts and Stories) induce habitual checking and quickly deplete rewards from app use. Second, recommendation-based features situated close to actively used features (e.g., Instagram's Suggested Posts adjacent to Search) cause habitual feature tour and sidetracking from the original intention of app use. Third, recommendation-based features with bite-sized contents (e.g., Facebook's Watch Videos) induce using "just a bit more," making people fall into prolonged use. We discuss implications of our findings for how social media apps and intervention tools can be designed to reduce regretful use and how feature-level usage information can strengthen self-reflection and behavior changes.

While people primarily communicate with text in mobile chat applications, they are increasingly using visual elements such as images, emojis, and memes. Using such visual elements could help users communicate clearly and make chatting experience enjoyable. However, finding and inserting contextually appropriate images during the chat can be both tedious and distracting. We introduce MilliCat, a real-time image suggestion system that recommends images that match the chat content within a mobile chat application (i.e., autocomplete with images). MilliCat combines natural language processing (e.g., keyword extraction, dependency parsing) and mobile computing (e.g., resource and energy-efficiency) techniques to autonomously make image suggestions when users might want to use images. Through multiple user studies, we investigated the effectiveness of our design choices, the frequency and motivation of image usage by the participants, and the impact of MilliCat on mobile chat experiences. Our results indicate that MilliCat’s real-time image suggestion enables users to quickly and conveniently select and display images on mobile chat by significantly reducing the latency in the image selection process (3.19 × improvement) and consequently more frequent image usage (1.8 ×) than existing solutions. Our study participants reported that they used images more often with MilliCat as the images helped them convey information more effectively, emphasize their opinion, express emotions, and have fun chatting experience.

While mobile instant messaging (MIM) facilitates ubiquitous interpersonal communication, its constant connectivity could build the expectation of an immediate response to messages, and its notifications flood could cause interruptions at inopportune moments. We examine two design concepts for MIM-private status sharing and sender-controlled notifications-that aim to lower the pressure for an immediate reply and reduce unnecessary interruptions by untimely notifications. Private status sharing reactively reveals a customized status with a selected partner(s) only when the partner has sent a message. Sender-controlled notifications give senders the control of choosing whether to send a notification for their own messages. We built MyButler, an Android app prototype that instantiates these two concepts and integrated it with KakaoTalk, a commercial MIM app. During a two-week field study with 11 pairs (5 couples and 6 friend pairs), participants expressed themselves through a total of 210 different statuses, 64.3% of which indicated the current activity or task of the user. Participants reported that private status sharing enabled them to explain their unavailability and relieved the pressure and expectations for timely attendance. We reveal more findings on the types of privately shared statuses and their roles in MIM communication; the in-situ behaviors and patterns of using sender-controlled notifications; and the motivations of MIM users in choosing whether to alert their messages. In terms of message notifications, senders chose to send 25.4% of the messages without any notification. We found that senders' decisions to alert are affected by the receiver's status, their own status to chat, and the possibility of message content exposure to others through notifications. Based on our findings, we draw insights into how the concepts of private status sharing and sender-controlled notifications can be applied in future designs and explorations.

Energy efficiency is critical in many IoT applications with sensors that deliver data over wireless communications. Duty-cycling has been a major method for reducing energy consumption. One popular duty-cycling is MAC duty-cycling where an MCU commands the periodic or adaptive turning on and off of an RF chip. Recently, there has been an increase in the number of RF chips for IoT that are equipped with PHY duty-cycling, a new capability of autonomously switching on and off an RF chip without the use of an MCU. These two schemes working at different layers have different pros and cons in terms of operating time scale, the amount of energy saved when the RF chip is switched off, all of which depend on the characteristics of the MCU and the RF chip. In this paper, we propose a novel protocol named HD-MAC (Hierarchical Duty-cycling MAC) that hierarchically integrates duty-cycling in the MAC and physical layers. By smartly applying the new function of chip-level duty-cycling, HD-MAC is able to further reduce the amount of on-time in MAC duty-cycling; hence, energy efficiency can be improved. To optimize HD-MAC’s energy efficiency while achieving a given delay requirement, we formulate an optimization problem and solve it to obtain the optimal parameters in the cross-layer context. We implement HD-MAC on Contiki OS and perform extensive experiments using a real sensor mote Firefly with a CC1200 RF chip. We demonstrate that the energy efficiency of HD-MAC is up to 72% higher than that of existing protocols while still satisfying the delay requirement and sustaining similar reliability.

Video multicast over wireless local area network (WLAN) has been gaining attraction for applications sharing a venue-specific common video with multiple users. However, wireless multicast is limited by a receiver that has the weakest communication link to the source. Collaborative relaying could overcome this challenge by enabling selected receiver nodes to relay the packets from the source to other receivers. We propose EV-CAST, an interference and energy-aware video multicast system using collaborative relays, which entails (i) online topology management based on interference-aware link characterization, (ii) joint selection of relay nodes and transmission parameters, and (iii) polling-based relay protocol. Our proposed algorithm, the core of EV-CAST, judiciously selects the relay nodes and transmission parameters in consideration of interference, battery status, and spatial reuse. Our prototype-based experiment results demonstrate that EV-CAST enhances video multicast delivery under various network scenarios. EV-CAST enables 2x more nodes to achieve a target video packet loss ratio with 0.59x shorter airtime than the state-of-the-art video multicast scheme.

Overheating smartphones could hamper user experiences. While there have been numerous reports on smartphone overheating, a systematic measurement and user experience study on the thermal aspect of smartphones is missing. Using thermal imaging cameras, we measure and analyze the temperatures of various smartphones running diverse application workloads such as voice calling, video recording, video chatting, and 3D online gaming. Our experiments show that running popular applications such as video chat, could raise the smartphone's surface temperature to over 50$^\circ$C in only 10 minutes, which could easily cause thermal pain to users. Recent ubiquitous scenarios such as augmented reality and mobile deep learning also have considerable thermal issues. We then perform a user study to examine when the users perceive heat discomfort from the smartphones and how they react to overheating. Most of our user study participants reported considerable thermal discomfort while playing a mobile game, and that overheating disrupted interaction flows. With this in mind, we devise a smartphone surface temperature prediction model, by using only system statistics and internal sensor values. Our evaluation showed high prediction accuracy with root-mean-square errors of less than 2$^\circ$C. We discuss several insights from our findings and recommendations for user experience, OS design, and developer support for better user-thermal interactions.

We propose the design of MyButler, a sender-controlled notification management system that mitigates disruption caused by mobile instant messaging through sharing the receiver's activity information with the sender.

While the recent Wi-Fi standard 802.11ac achieves Gb/s theoretical capacity with Multi-User MIMO (MU-MIMO) technology, several studies reported that throughput of 802.11ac in practice is far from Gb/s link speed. We investigate the downlink throughput of Wi-Fi systems with commercially available 802.11ac products in multiple indoor environments to reveal the throughput of MU-MIMO system that user experiences in practice. From our experiments, Single-User MIMO (SU-MIMO) outperformed MU-MIMO at every experimental environments. We further provide analysis on our experimental results considering channel sounding overhead, user grouping, environmental impact, and transmission mode selection.

We propose Knocker, a real-time object identification technique with smartphones. Knocker leverages unique impulse signals that are generated by knocking on an object with a smartphone. Knocker does not require any special augmentation for both smartphones and objects.

Deep mobile sensing applications are suffering from various individual conditions in the wild. We propose a meta-learned adaptation technique to adapt to a target condition with a few labeled data. We evaluate our system on a public dataset and it outperforms baselines.

We present Reeboc that combines machine learning and k-means clustering to analyze the conversation of a chat, extract different emotions or topics of the conversation, and recommend emojis that represent various contexts to the user. Instead of simply analyzing a single input sentence, we consider recent sentences exchanged in a conversation. we performed a user study with 17 participants in 8 groups in a realistic mobile chat environment. Participants spent the least amount of time in identifying and selecting the emojis of their choice with Reeboc (38% faster than without emoji recommendation).

Gas monitoring, often as a part of safety and health systems, is widely used in a variety of sectors such as manufacturing, auto- mobiles, medical, households, food and beverages, environments, and HVACs. Existing gas monitoring approaches such as the use of catalytic and infrared sensors and electrochemical and metal oxide semiconductor technologies, typically require expensive hardware or are power hungry, and thus not suitable for long-term and large scale deployments. We propose a gas intensity measuring system with cost-effective commercial off-the-shelf (COTS) RFID devices. Our key intuition is that the changes in phase and strength of the signal result from not only the signal propagation but also from the hardware (i.e., tag's circuit). To this end, we propose a method to in- tegrate RFID with a chemiresistor, called Carbon Nanotubes (CNTs), whose electrical property varies with the nearby gas concentration. Our system shows the potential of low-cost, easy-to-make, and wireless sensing based gas monitoring systems.

We present ProDroid, a framework that provides Android app developers an ability to quickly produce functional prototypes. With ProDroid, developers can create a new app that imports various kinds of functionality provided by other existing Android apps. Our evaluation shows that with the help of ProDroid, a developer was able to import a function from an existing Android app into a new prototype with only 55 lines of Java code, while the function itself requires 10,334 lines of Java code to implement.

While there are many research proposals for wearable Automatic Dietary Monitoring (ADM) systems that detect eating of a user, it is difficult to notice real-world users wearing such devices in public. We propose a new wearable ADM system that could be used daily by real-world users. It is designed in a form of necklace, providing natural and firm contact of sensor on user's skin to accurately capture eating activities. At our preliminary experiments, our wearable ADM system detected eating of a user with 86.1% accuracy.

Recent improvements in deep learning and hardware support offer a new breakthrough in mobile sensing; we could enjoy context-aware services and mobile healthcare on a mobile device powered by artificial intelligence. However, most related studies perform well only with a certain level of similarity between trained and target data distribution, while in practice, a specific user's behaviors and device make sensor inputs different. Consequently, the performance of such applications might suffer in diverse user and device conditions as training deep models in such countless conditions is infeasible. To mitigate the issue, we propose MetaSense, an adaptive deep mobile sensing system utilizing only a few (e.g., one or two) data instances from the target user. MetaSense employs meta learning that learns how to adapt to the target user's condition, by rehearsing multiple similar tasks generated from our unique task generation strategies in offline training. The trained model has the ability to rapidly adapt to the target user's condition when a few data are available. Our evaluation with real-world traces of motion and audio sensors shows that MetaSense not only outperforms the state-of-the-art transfer learning by 18% and meta learning based approaches by 15% in terms of accuracy, but also requires significantly less adaptation time for the target user.

We present X-Droid, a framework that provides Android app developers an ability to quickly and easily produce functional prototypes. Our work is motivated by the need for such ability and the lack of tools that provide it. Developers want to produce a functional prototype rapidly to test out potential features in real-life situations. However, current prototyping tools for mobile apps are limited to creating non-functional UI mockups that do not demonstrate actual features. With X-Droid, developers can create a new app that imports various kinds of functionality provided by other existing Android apps. In doing so, developers do not need to understand how other Android apps are implemented or need access to their source code. X-Droid provides a developer tool that enables developers to use the UIs of other Android apps and import desired functions into their prototypes. X-Droid also provides a run-time system that executes other apps' functionality in the background on off-the-shelf Android devices for seamless integration. Our evaluation shows that with the help of X-Droid, a developer imported a function from an existing Android app into a new prototype with only 51 lines of Java code, while the function itself requires 10,334 lines of Java code to implement (i.e., 200× improvement).

A cellular network is a closed system, and each network operator has built a unique “walled garden” for their network by combining different operation policies, network configurations, and implementation optimizations. Unfortunately, some of these combinations can induce performance degradation due to misconfiguration or unnecessary procedures. To detect such degradation, a thorough understanding of even the minor details of the standards and operator-specific implementations is important. However, it is difficult to detect such problems, as the control plane is complicated by numerous procedures. This paper introduces a simple yet powerful method that diagnoses these problems by exploiting the operator-specific implementations of cellular networks. We develop a signaling collection and analysis tool that collects control plane messages from operators and finds problems through comparative analysis. The analysis process consists of three different control plane comparison procedures that can find such problems effectively. These individual procedures use a time threshold, control flow sequence, and signaling failure as the basis for comparison. To this end, we collect approximately 3.1 million control-plane messages from 13 major cellular operators worldwide. As a case study, we analyze the circuit-switched fallback technology that triggers generation crossover between third generation and long-term evolution technologies.

We introduce Zaturi, a system enabling parents to create an audio book for their babies by utilizing micro spare time at work. We define micro spare time at work as tiny fragments of time with low cognitive loads that frequently occur at work, such as waiting for an elevator. We show that putting together micro spare time at work helps a working parent (1) build a tangible symbol conveying his/her thoughts to the beloved baby and (2) develop his/her own feelings of parental achievement without compromising regular working hours. Zaturi lets the parent immediately be aware of micro spare time and provides a crafted interface to seamlessly record the book piece by piece, so that the baby can enjoy listening to the book recorded in the parent's own voice. Through an extensive design process, we characterize the notion of micro spare time and build a working prototype of Zaturi. We also report parents' perceptions and family reactions after a two-week deployment.

Smartphone notifications provide application-specific information in real-time, but could distract users from in-person social interactions when delivered at inopportune moments. We explore breakpoint-based notification management, in which the smartphone defers notifications until an opportune moment. With a video survey where participants selected appropriate moments for notifications from a video-recorded social interaction, we identify four breakpoint types: long silence, a user leaving the table, others using smartphones, and a user left alone. We introduce a Social Context-Aware smartphone Notification system, SCAN, that uses build-in sensors to detect social context and identifies breakpoints to defer smartphone notifications until a breakpoint. We conducted a controlled study with ten friend groups who had SCAN installed on their smartphones while dining at a restaurant. Results show that SCAN accurately detects breakpoints (precision=92.0%, recall=82.5%), and reduces notification interruptions by 54.1%. Most participants reported that SCAN helped them to focus better on in-person social interaction and found selected breakpoints appropriate.

While a certain sound serves a purpose to someone, for others, the same sound could be noise. Specifically, an alarm sound would be necessary for someone to wake up and start the day, while it would be an unwanted sound for people sharing the same room, who need not wake up as early. Noise cancellation is useful in this scenario, but most existing techniques require costly equipments (e.g., high quality speakers or microphones) or devices that are uncomfortable to wear during sleep. We thus explore the possibility of using only commodity smartphones to achieve active noise control and present Virtual Earplugs. As most people own a smartphone that include a microphone and speakers, and has an alarm clock feature, we believe our system could be easily used in practice. We highlight the technical challenges in realizing our vision and demonstrate the feasibility of our approach using our preliminary prototype. Our results indicate that we can reduce the alarm sound by up to 19~dB in only 2.1~seconds of processing delay.

Multi-rate forward erasure correction (FEC)-applied wireless multicast enables reliable and efficient video multicast with intelligent selection of physical (PHY) layer data rate and FEC rate. The optimal PHY/FEC rates depend on the cause of the packet losses. However, previous approaches select the PHY/FEC rates by considering only channel errors even when interference is also a major source of packet losses. We propose InFRA, an interference-aware PHY/FEC rate adaptation framework that (i) infers the cause of the packet losses based on received signal strength indicator (RSSI) and cyclic redundancy check (CRC) error notifications, and (ii) determines the PHY/FEC rates based on the cause of packet losses. Our prototype implementation with off-the-shelf chipsets demonstrates that InFRA enhances the multicast delivery under various network scenarios. InFRA enables 2.3x and 1.8x more nodes to achieve a target video packet loss rate with a contention interferer and a hidden interferer, respectively, compared with the state-of- the-art PHY/FEC rate adaptation scheme. To our best knowledge, InFRA is the first work to take the impact of interference into account for the PHY/FEC rate adaptation.

Malicious activities on the Web are increasingly threatening users in the Internet. Home networks are one of the prime targets of the attackers to host malware, commonly exploited as a stepping stone to further launch a variety of attacks. Due to diversification, existing security solutions often fail to detect malicious activities that remain hidden and pose threats to users' security and privacy. Characterizing behavioral patterns of known malware can help to improve the classification accuracy of threats. More importantly, as different malware might share commonalities, studying the behavior of known malware could help the detection of previously unknown malicious activities. We pose the research question if it is possible to characterize such behavioral patterns analyzing the traffic from known infected clients. We present our quest to discover such characterizations. Results show that commonalities arise but their identification may require some ingenuity. We also present our discovery of malicious activities that were left undetected by commercial IDS.

Malware is a major threat to security and privacy of network users. A huge variety of malware typically spreads over the Internet, evolving every day, and challenging the research community and security practitioners to improve the effectiveness of countermeasures. In this paper, we present a system that automatically extracts patterns of network activity related to a specific malicious event, i.e., a seed. Our system is based on a methodology that correlates network events of hosts normally connected to the Internet over (i) time (i.e., analyzing different samples of traffic from the same host), (ii) space (i.e., correlating patterns across different hosts), and (iii) network layers (e.g., HTTP, DNS, etc.). The result is a Network Connectivity Graph that captures the overall "network behavior" of the seed. That is a focused and enriched representation of the malicious pattern infected hosts exhibit, purified from ordinary network activities and background traffic. We applied our approach on a large dataset collected in a real commercial ISP where the aggregated traffic produced by more than 20,000 households has been monitored. A commercial IDS has been used to complement network data with alerts related to malicious activities. We use such alerts to trigger our processing system. Results shows that the richness of the Network Connectivity Graph provides a much more detailed picture of malicious activities, considerably enhancing our understanding.

HTTP is a popular channel for malware to communicate with malicious servers (e.g., Command & Control, drive-by download, drop-zone), as well as to attack benign servers. By utilizing HTTP requests, malware easily disguises itself under a large amount of benign HTTP traffic. Thus, identifying malicious HTTP activities is challenging. We leverage an insight that cyber criminals are increasingly using dynamic malicious infrastructures with multiple servers to be efficient and anonymous in (i) malware distribution (using redirectors and exploit servers), (ii) control (using C&C servers) and (iii) monetization (using payment servers), and (iv) being robust against server takedowns (using multiple backups for each type of servers). Instead of focusing on detecting individual malicious domains, we propose a complementary approach to identify a group of closely related servers that are potentially involved in the same malware campaign, which we term as Associated Server Herd (ASH). Our solution, SMASH (Systematic Mining of Associated Server Herds), utilizes an unsupervised framework to infer malware ASHs by systematically mining the relations among all servers from multiple dimensions. We build a prototype system of SMASH and evaluate it with traces from a large ISP. The result shows that SMASH successfully infers a large number of previously undetected malicious servers and possible zero-day attacks, with low false positives. We believe the inferred ASHs provide a better global view of the attack campaign that may not be easily captured by detecting only individual servers.

A Multi-User MIMO (MU-MIMO) Access Point (AP) can obtain a capacity gain by simultaneously transmitting to multiple clients. This technique requires Channel State Information (CSI) at the transmitting AP to set antenna gains and phases to enable simultaneous reception through beamforming. The AP must also select both the mode (number of transmit and collective receive antennas) and the user set prior to transmission. While the ideal mode and user selection is a function of CSI, CSI must be estimated with an overhead intensive channel sounding process. We design, implement, and evaluate Pre-sounding User and Mode selection Algorithm (PUMA), a method for mode and user selection prior to channel sounding. We show that even without CSI, PUMA (i) exploits theoretical properties of MU-MIMO system scaling with respect to mode, (ii) characterizes the relative cost of each potential mode, and (iii) estimates per-stream transmission rate and aggregate throughput in each mode for a potential user set, all without CSI. Once PUMA has selected the appropriate mode and user group, the chosen protocol's channel sounding method is used on the intended user subset to carry out the transmission. We show that, on average, PUMA selects the mode and group that achieves an aggregate rate within 3% of the saturation throughput of what would have been achieved by sounding all users (which would require significant additional overhead). Moreover, we show that PUMA obtains 30% higher aggregate throughput compared to the best fixed-mode policy that uses the maximum number of available transmit and receive antennas.

Cloud-RANs (Radio Access Networks) assume the existence of a high-capacity, low-delay/latency fronthaul to support cooperative transmission schemes such as CoMP (Coordinated Multi-Point) and coordinated beamforming. However, building such hierarchical wired fronthauls is challenging as the typical I/Q data stream is non-elastic - I/Q data over the wired fronthaul has little tolerance for delay jitters and zero tolerance for losses. Any distortion to the I/Q data stream will make the resulting wireless transmission completely unintelligible. We propose Spiro, a mechanism that efficiently transports RF signals over a wired fronthaul network. The primary goal of Spiro is to make I/Q data streams elastic and resilient to unexpected network condition changes. This is accomplished through a novel combination of compression and data prioritization of I/Q data on the wired fronthaul. For a given wireless throughput, Spiro can reduce the bandwidth demand of the fronthaul data stream by up to 50% without any noticeable degradation in the wireless reception quality. Further bandwidth reduction via compression and frame losses only have a limited impact on the wireless throughput.

Fingerprinting the Operating System (OS) running on a device based on its traffic has several applications, such as NAT detection, policy enforcement in enterprise networks, and billing for shared access in mobile networks. In this paper, we propose to utilize several features in TCP/IP headers for OS identification, and use real traffic traces to evaluate the accuracy of fingerprinting. Our trace-driven study shows that several techniques that successfully fingerprint desktop OSes are not effective for fingerprinting mobile devices. Therefore, we propose new features for fingerprinting OSes on mobile devices. We also consider NAT/tethering detection, an important application of OS fingerprinting. We use the presence of multiple OSes from the same IP address along with TCP timestamp, clock frequency, and boot time to detect tethering. Evaluation shows that our approach effectively detects tethering and outperforms existing schemes.

Spectrum management and device coordination for dynamic spectrum access (DSA) networks have received significant research attention. However, current wireless devices have yet to fully embrace DSA networks due to the difficulties in realizing spectrum-agile communications. We address the practical hurdles and present solutions toward implementing DSA devices, answering an important question “what is a simple practical extension to current wireless devices that makes them spectrum-agile?” To this end, we propose RODIN, a general per-frame spectrum-shaping protocol that has the following features to support DSA in commercial off-the-shelf (COTS) wireless devices: direct manipulation of passband signals from COTS devices, fast FPGA-based spectrum shaping, and a novel preamble design for spectrum agreement. RODIN uses an FPGA-based spectrum shaper together with a preamble I-FOP to achieve per-frame spectrum shaping with a delay of under 10 μs.

The IEEE 802.11n standard defines channel bonding that allows wireless devices to operate on 40 MHz channels by doubling their bandwidth from standard 20 MHz channels. Increasing channel width increases capacity, but it comes at the cost of decreased transmission range and greater susceptibility to interference. However, with the incorporation of Multiple-Input Multiple-Output (MIMO) technology in 802.11n, devices can now exploit the increased transmission rates from wider channels with minimal sacrifice to signal quality and range. The goal of our work is to identify the network factors that influence the performance of channel bonding in 802.11n networks and make intelligent channel bonding decisions. We discover that channel width selection should consider not only a link's signal quality, but also the strength of neighboring links, their physical rates, and interferer load. We use our findings to design and implement a network detector that successfully identifies interference conditions that affect channel bonding decisions in 100% of our test cases. Our detector can form the foundation for more robust and accurate algorithms that can adapt bandwidth to variations in channel conditions. Our findings allows us to predict the impact of network conditions on performance and make channel bonding decisions that maximize throughput.

The growing demand for real-time streaming video on portable devices has increased the importance of multimedia multicast in mobile wireless networks. A defining characteristic of such multicast networks is its heterogeneity in both the channel states and the MIMO capabilities of its clients. However, current wireless multicast schemes adapt poorly to such heterogeneity. We introduce Procrustes, a multimedia multicast scheme that is built upon a novel PHY-layer rateless code. Unlike bit-level rateless codes (such as Raptor [14] codes), Procrustes clients automatically adjust the PSNR of the received multicast video stream to match both the instantaneous channel state and the number of active receive antennas. We demonstrate the performance of Procrustes in a simulated environment.

Cooperative packet recovery has been widely investigated in wireless networks, where corrupt copies of a packet are combined to recover the original packet. While previous work such as MRD (Multi Radio Diversity) and Soft apply combining to bits and bit-confidences, combining at the symbol level has been avoided. The reason is rooted in the prohibitive overhead of sharing raw symbol information between different APs of an enterprise WLAN. We present Epicenter that overcomes this constraint, and combines multiple copies of incorrectly received “symbols” to infer the actual transmitted symbol. Our core finding is that symbols need not be represented in full fidelity - coarse representation of symbols can preserve most of their diversity, while substantially lowering the overhead. We then develop a rate estimation algorithm that actually exploits symbol level combining. Our USRP/GNURadio testbed confirms the viability of our ideas, yielding 40% throughput gain over Soft, and 25-90% over 802.11. While the gains are modest, we believe that they are realistic, and available with minimal modifications to today's EWLAN systems.

Wireless network scheduling and control techniques (e.g., opportunistic scheduling) rely heavily on access to Channel State Information (CSI). However, obtaining this information is costly in terms of bandwidth, time, and power, and could result in large overhead. Therefore, a critical question is how to optimally manage network resources in the absence of such information. To that end, we develop a cross-layer solution for downlink cellular systems with imperfect (and possibly no) CSI at the transmitter. We use rateless codes to resolve channel uncertainty. To keep the decoding complexity low, we explicitly incorporate time-average block-size constraints, and aim to maximize the system utility. The block-size of a rateless code is determined by both the network control decisions and the unknown CSI of many time slots. Therefore, unlike standard utility maximization problems, this problem can be viewed as a constrained partial observed Markov decision problem (CPOMDP), which is known to be hard due to the “curse of dimensionality.” However, by using a modified Lyapunov drift method, we develop a dynamic network control scheme, which yields a total network utility within O(1/Lav) of utility-optimal point achieved by infinite block-size channel codes, where Lavis the enforced value of the time-average block-size of rateless codes. This opens the door of being able to trade complexity/delay for performance gains in the absence of accurate CSI. Our simulation results show that the proposed scheme improves the network throughput by up to 68% over schemes that use fixed-rate codes.

The emergence of MIMO antennas and channel bonding in 802.11n wireless networks has resulted in a huge leap in capacity compared with legacy 802.11 systems. This leap, however, adds complexity to selecting the right transmission rate. Not only does the appropriate data rate need to be selected, but also the MIMO transmission technique (e.g., Spatial Diversity or Spatial Multiplexing), the number of streams, and the channel width. Incorporating these features into a rate adaptation (RA) solution requires a new set of rules to accurately evaluate channel conditions and select the appropriate transmission setting with minimal overhead. To address these challenges, we propose ARAMIS (Agile Rate Adaptation for MIMO Systems), a standard-compliant, closed-loop RA solution that jointly adapts rate and bandwidth. ARAMIS adapts transmission rates on a per-packet basis; we believe it is the first 802.11n RA algorithm that simultaneously adapts rate and channel width. We have implemented ARAMIS on Atheros-based devices and deployed it on our 15-node testbed. Our experiments show that ARAMIS accurately adapts to a wide variety of channel conditions with negligible overhead. Furthermore, ARAMIS outperforms existing RA algorithms in 802.11n environments with up to a 10 fold increase in throughput.

Light field is a large set of spatially correlated images of the same static scene captured using a 2D array of closely spaced cameras. Interactive light field streaming is the application where a client continuously requests successive light field images along a view trajectory of his choosing, and in response the server transmits appropriate data for the client to correctly reconstruct desired images. The technical challenge is how to encode captured light field images into a reasonably sized frame structure a priori (without knowing eventual clients' view trajectories), so that at stream time, expected server transmission rate can be minimized, while satisfying client's view-switch requests. In this paper, using I-frames, redundant P-frames and distributed source coding (DSC) frames as building blocks, we design coding structures to optimally trade off storage size of the frame structure with expected server transmission rate. The key novelty is to facilitate the use of “landmarks” in the structure-popular reference frames cached in the decoder buffer-so that the probability of having at least one useful predictor frame available in the buffer for disparity compensation is greatly increased. We first derive recursive equations to find the optimal caching strategy for a given coding structure. We then formulate the structure design problem as a Lagrangian minimization, and propose fast heuristics to find near-optimal solutions. Experimental results show that the expected server streaming rate can be reduced by up to 93.6% compared to an I-frame-only structure, at twice the storage required.

We present the design and experimental evaluation of Simultaneous TRansmission with Orthogonally Blinded Eavesdroppers (STROBE). STROBE is a cross-layer approach that exploits the multi-stream capabilities of existing technologies such as 802.11n and the upcoming 802.11ac standard where multi-antenna APs can construct simultaneous data streams using Zero-Forcing Beamforming (ZFBF). Instead of using this technique for simultaneous data stream generation, STROBE utilizes ZFBF by allowing an AP to use one stream to communicate with an intended user and the remaining streams to orthogonally “blind” (actively interfere with) any potential eavesdropper thereby preventing eavesdroppers from decoding nearby transmissions. Through extensive experimental evaluation, we show that STROBE consistently outperforms Omnidirectional, Single-User Beamforming (SUBF), and directional antenna based transmission methods by keeping the transmitted signal at the intended receiver and shielded from eavesdroppers. In an indoor Wireless LAN environment, STROBE consistently serves an intended user with an SINR 15 dB greater than an eavesdropper.

One of the key features of high speed WLAN such as 802.11n is the use of MIMO (Multiple Input Multiple Output) antenna technology. The MIMO channel is described with fine granularity by Channel State Information (CSI) that can be utilized in many ways to improve network performance. Many complex parameters of a MIMO system require numerous samples to obtain CSI for all possible channel configurations. As a result, measuring the complete CSI space requires excessive sampling overhead and thus degrades network performance. We propose CSI-SF (CSI Sampling & Fusion), a method for estimating CSI for every MIMO configuration by sampling a small number of frames transmitted with different settings and extrapolating data for the remaining settings. For instance, we predict CSI of multi-stream settings using CSI obtained only from single stream packets. We evaluate the effectiveness of CSI-SF in various scenarios using our 802.11n testbed and show that CSI-SF provides an accurate, complete knowledge of the MIMO channel with reduced overhead from traditional sampling. We also show that CSI-SF can be applied to network algorithms such as rate adaptation, antenna selection and association control to significantly improve their performance and efficiency.

Spectrum management and device coordination for Dynamic Spectrum Access (DSA) networks have received significant research attention. However, current wireless devices have yet to fully embrace DSA networks due to the difficulties in realizing spectrum-agile communications. We address the practical hurdles and present solutions towards implementing DSA devices, answering an important question "what is a simple practical extension to current wireless devices that makes them spectrum-agile?" To this end, we propose RODIN, a general per-frame spectrum-shaping protocol that has the following features to support DSA in commercial off-the-shelf (COTS) wireless devices: (a) direct manipulation of passband signals from COTS devices, (b) fast FPGA-based spectrum shaping, and (c) a novel preamble design for spectrum agreement. RODIN uses an FPGA-based spectrum shaper together with a preamble I-FOP to achieve per-frame spectrum shaping with a delay of under 10 μ s.

The IEEE 802.11n standard allows wireless devices to operate on 40MHz-width channels by doubling their channel width from standard 20MHz channels, a concept called channel bonding. Increasing channel width should increase bandwidth, but it comes at the cost of decreased transmission range and greater susceptibility to interference. However, with the incorporation of MIMO (Multiple-Input Multiple-Output) technology in 802.11n, devices can now exploit the increased transmission rates from wider channels at a reduced sacrifice to signal quality and range. The goal of our work is to understand the characteristics of channel bonding in 802.11n networks and the factors that influence that behavior to ultimately be able to predict behavior so that network performance is maximized. We discuss the impact of channel bonding choices as well as the effects of both co-channel and adjacent channel interference on network performance. We discover that intelligent channel bonding decisions rely not only on a link's signal quality, but also on the strength of neighboring links and their physical rates.

Light field is a large set of spatially correlated images of the same static scene captured using a 2D array of closely spaced cameras. Interactive light field streaming is the application where a client continuously requests successive light field images along a view trajectory of her choosing, and in response the server transmits appropriate data for the client to correctly reconstruct desired images. The technical challenge is how to encode captured light field images into a reasonably sized frame structure a priori (without knowing eventual clients' view trajectories), so that during streaming session, expected server transmission rate can be minimized, while satisfying client's view requests. In this paper, we design efficient frame structures, using I-frames, redundant P-frames and distributed source coding (DSC) frames as building blocks, to optimally trade off storage size of the frame structure with expected server transmission rate. The key novelty is to optimize structures in such a way that decoded images in caches of neighboring cooperative peers, connected together via a secondary network such as ad hoc WLAN for content sharing, can be reused to further decrease the server-to-client transmission rate. We formulate the structure design problem as a Lagrangian minimization, and propose fast heuristics to find near-optimal solutions. Experimental results show that the expected server streaming rate can be reduced by up to 83% compared to an I-frame-only structure, at less than twice the storage required.

We explore the design of a high capacity multi-radio wireless network using commercial 802.11n hardware. We first use extensive real-life experiments to evaluate the performance of closely located 802.11n radios. We discover that even when tuned to orthogonal channels, co-located 802.11n radios interfere with each other and achieve significantly less throughput than expected. Our analysis reveals that the throughput degradation is caused by three link-layer effects: (i) triggering of carrier sensing, (ii) out of band collisions and (iii) unintended frequency adaptation. Using physical layer statistics, we observe that these effects are caused by fundamental limitations of co-located radios in achieving signal isolation. We then consider the use of beamforming antennas, shielding and antenna separation distance to achieve better signal isolation and to mitigate these problems. Our work profiles the gains of different physical isolation approaches and provides insights to network designers to realize high-performance wireless networks without requiring synchronization or protocol modifications.

We present an experimental performance evaluation study of WiFi links in an open-space outdoor environment. We consider a large scale wireless sensor network scenario of seismic data collection from sensors that are buried in ground and a set of access points (APs) form the hierarchical aggregation layer and the backbone of the network. We conduct two different link characterization studies. First, we evaluate the links between the sensor nodes and a wireless AP using IEEE 802.11a/b/g. We construct the path loss model and investigate the reachability distance of this link for different protocols and different sensor node antenna heights. We then characterize the long distance wireless backhaul links between the APs. We use 802.11n and high gain directional antenna for high throughput and long distance. We evaluate how different PHY and MAC layer enhancements of 802.11n impacts its performance in an open outdoor environment. We observed up to 148 Mb/s throughput at 800 meter line-of-sight links without sophisticated tuning of antenna orientation. We believe our findings can be a benchmark for WiFi based outdoor network deployment, especially for high throughput long distance links.

Recently, there has been a tremendous growth in the number of installed distributed computing platforms such as those for content distribution networks, cloud computing infrastructures, and distributed data centers. Such distributed platforms need a scalable end-to-end (e2e) network monitoring component to provide Quality of Service (QoS) guarantees to the services and improve the overall performance. An important challenge for a network monitoring infrastructure is the periodicity of the measurements as this aspect trades off the monitoring overheads with staleness of the results. In the Network Genome project, we explore the relationships between different e2e network metrics with the aim of leveraging such relationships for reducing monitoring costs while maintaining measurement accuracy. We perform our analysis using long range network measurements from PlanetLab, where we have been collecting e2e network data (route, number of hops, capacity bandwidth and available bandwidth) as part of the S3 system since January 2006. In this paper, we focus on the correlation between the Capacity and Available Bandwidth metrics between host pairs in the PlanetLab testbed. Our analysis shows that the ranking of hosts with respect to their Capacity to/from a set of nodes is a good indicator of the ranking of hosts with respect to their Available Bandwidth to/from the same set of nodes.

In Wireless Local Area Networks (WLANs), channel management is important in achieving reliable data communications and satisfying QoS requirements. The key aspects of wireless channel management are monitoring the channel quality and adapting quickly to the network conditions by switching to a better channel. We propose a wireless channel monitoring system with co-located monitoring surrogates. Our system works on multi-radio Access Points (APs) where a co-located surrogate radio monitors the condition of various channels while the master radio serves the clients for data communication. Although we have designed our system for generic WLANs, we believe it will be most useful for IEEE 802.11n networks where there are a large number of channels and dynamic frequency selection is required. Our system enables intelligent, fast channel adaptation, reduces service disruption time, and consequently helps realize the performance potential of 802.11n. We present our multi-radio co-located wireless channel monitoring surrogate system and evaluate its effectiveness on our IEEE 802.11n network testbed. We also perform case studies to demonstrate the benefit our system brings compared against the existing schemes.

Peak Signal-to-Noise Ratio (PSNR) is the simplest and the most widely used video quality evaluation methodology. However, traditional PSNR calculations do not take the packet loss into account. This shortcoming, which is amplified in wireless networks, contributes to the inaccuracy in evaluating video streaming quality in wireless communications. Such inaccuracy in PSNR calculations adversely affects the development of video communications in wireless networks. This paper proposes a novel video quality evaluation methodology. As it not only considers the PSNR of a video, but also with modifications to handle the packet loss issue, we name this evaluation method MPSNR. MPSNR rectifies the inaccuracies in traditional PSNR computation, and helps us to approximate subjective video quality, Mean Opinion Score (MOS), more accurately. Using PSNR values calculated from MPSNR and simple network measurements, we apply linear regression techniques to derive two specific objective video quality metrics, PSNR-based Objective MOS (POMOS) and Rates-based Objective MOS (ROMOS). Through extensive experiments and human subjective tests, we show that the two metrics demonstrate high correlation with MOS. POMOS takes the averaged PSNR value of a video calculated from MPSNR as the only input. Despite its simplicity, it has a Pearson correlation of 0.8664 with the MOS. By adding a few other simple network measurements, such as the proportion of distorted frames in a video, ROMOS achieves an even higher Pearson correlation (0.9350) with the MOS. Compared with the PSNR metric from the traditional PSNR calculations, our metrics evaluate video streaming quality in wireless networks with a much higher accuracy while retaining the simplicity of PSNR calculation.

We develop a new wireless link quality metric, ECOT (Estimated Channel Occupancy Time) that enables a high throughput route setup in wireless mesh networks. The key feature of ECOT is being applicable to diverse mesh network environments where IEEE 802.11 MAC (Medium Access Control) variants are used. We take into account the detailed operational features of various 802.11 MAC protocols, such as 802.11 DCF (Distributed Coordination Function), 802.11e EDCA (Enhanced Distributed Channel Access) with BACK (Block Acknowledgment), and 802.11n A-MPDU (Aggregate MAC Protocol Data Unit), and derive an integrated link metric that enables finding maximum throughput end-to-end routes. Through simulations in randomized topological environments, we evaluate the performance of the proposed link metric and routing strategy to demonstrate that our proposed schemes can achieve up to 354.4% throughput gain over existing ones.

We provide a comparative analysis of various routing strategies that affect the end-to-end performance in wireless mesh networks. We first improve well-known link quality metrics and routing algorithms to better operate in wireless mesh environments. We then investigate the route optimality and its impact on the network performance by comparing the achieved end-to-end performance with the optimal offline routing. Various network topologies, number of concurrent flows, and interference types are considered in our evaluation and we reveal that a nonoptimal route is easily established because of routing protocol's misbehavior, interflow interference, and their interplay, thus affecting the end-to-end performance.

Interference among concurrent transmissions complicates QoS provisioning for multimedia applications in wireless mesh networks. In this paper we propose MARIA (mesh admission control and QoS routing with interference awareness), a scheme towards enhancing QoS support for multimedia in wireless mesh networks. We characterize interference in wireless networks using a conflict graph based model. Nodes exchange their flow information periodically and compute their available residual bandwidth based on the local maximal clique constraints. Admission decision is made based on the residual bandwidth at each node. We implement an on-demand routing scheme that explicitly incorporates the interference model in the route discovery process. It directs routing message propagations and avoids "hot-spots" with severe interference. Simulation results demonstrate that by taking interference into account MARIA outperforms the conventional approach. It finds routes with less interference and enhances the performance significantly. We use video as an example application and MARIA improves the quality of delivered videos, with up to 7.3 dB average PSNR gain.

In the absence of end-to-end quality of service (QoS), overlay routing has been used as an alternative to the default best effort Internet routing. Using end-to-end network measurement, the problematic parts of the path can be bypassed, resulting in improving the resiliency and robustness to failures. Studies have shown that overlay paths can give better latency, loss rate, and TCP throughput. Overlay routing also offers flexibility as different routes can be used based on application needs. There have been very few proposals of using bandwidth as the main metric of interest, which is of great concern in media applications. We introduce our scheme BARON (Bandwidth-Aware Routing in Overlay Networks) that utilizes capacity between the end hosts to identify viable overlay paths and measures available bandwidth to select the best route. We propose our path selection approaches, and using the measurements between 174 PlanetLab nodes and over 13,189 paths, we evaluate the usefulness of overlay routes in terms of bandwidth gain. Our results show that among 658,526 overlay paths, 25% have larger bandwidth than their native IP routes, and over 86% of (source, destination) pairs have at least one overlay route with larger bandwidth than the default IP routes. We also present the effectiveness of BARON in preserving the bandwidth requirement over time for a few selected Internet paths.

The performance of an IEEE 802.11 station heavily depends on the selection of an AP (Access Point) that the station is associated with to access the Internet. The conventional approach to the AP selection is based on the received signal strength called RSSI (Received Signal Strength Indication) from APs within the transmission range. This approach however, might yield unbalanced traffic load among APs as the station chooses an AP only based on the signal strength, instead of considering the AP load and the level of contention on medium access. Accordingly, the station that is associated with the highest-RSSI AP might suffer from poor network performance. In this paper, we propose a new association metric, EVA (Estimated aVailable bAndwidth) with which a station can find the AP that provides the maximum achievable throughput among scanned APs. EVA is designed to estimate the available bandwidth on a channel with respect to a station that is to join a WLAN (Wireless Local Area Network). A station equipped with EVA observes a channel state in a per-slot basis, and yet does not request any external information from nearby APs or neighbor stations. Our estimation mechanism is non-intrusive, fully distributed, and independent of the infrastructure. Through simulation study, we evaluate the accuracy of the estimation and show that EVA-based association yields enhanced throughput performance compared with the legacy scheme.

In simulating wireless networks, modeling of the physical layer behavior is an important yet difficult task. Modeling and estimating wireless interference is receiving great research attention, and is crucial in a wireless network performance study. The implementation of physical layer capture, preamble detection, and carrier sense threshold plays an important role in successful frame reception in the presence of interference. We showed in our previous testbed study that the operations of the frame reception and the capture effect in real IEEE 802.11a systems differ from those of popular research simulators. We present our modifications of the IEEE 802.11a PHY models to the current simulators. The modifications can be summarized as follows. (i) The current simulators' frame reception is based only on the received signal strength. However, the real 802.11 systems can start the frame reception only when the Signal-to-Interference Ratio (SIR) is high enough to detect the preamble. (ii) Different chipset vendors implement the frame reception and capture algorithms differently, resulting in different operations for the same event. We provide different simulation models for several popular chipset vendors and show the performance differences between the models. (iii) The current simulators set the carrier sense threshold equal to the receiver sensitivity. The standard however states that it should be 20 dB higher than the receiver sensitivity. We implement our modifications to the QualNet simulator and conduct a wireless network performance study to evaluate the impact of PHY model implementation.

Supporting multimedia services in wireless mesh networks is receiving more attention from the research community. While wired networks have mature infrastructure and protocols providing QoS for multimedia, supporting multimedia in multihop wireless mesh networks faces greater technical challenges. The unreliable nature and shared media of multihop communications make the deployment of multimedia applications in wireless mesh networks a difficult task. To identify and understand the issues and problems of providing multimedia in multihop wireless mesh networks, we take video streaming as an example, setting up a real testbed to conduct extensive experiments in various scenarios and analyze its performance. In contrast to simulation or network-layer statistics based studies, our investigation is directly focused on video quality in multihop scenarios. The results better represent real networks and reveal interesting aspects of video performance in multihop wireless mesh networks, which we believe is helpful in designing efficient QoS solutions for multimedia services in the wireless mesh networks.

In wireless networks where communications are made over a shared medium, interference and collisions are the primary causes of packet drops. In multi-hop networks such as wireless mesh networks, due to the hidden terminal problem, limiting the effects of collisions and interference is a key in achieving high performance. Typical wireless medium access control protocols perform carrier sensing to avoid collisions. Most research efforts so far has used the binary model when studying carrier sensing and interference; a node is either carrier sensed or not, and a link is either interfered or not. In reality however, there exists a gray zone. Carrier sensing and interference should be represented in continuous values. Using the measurement data from our 802.11a wireless mesh network test-bed, we propose metrics that represent the levels of carrier sensing and interference. Using our metrics, we also propose methods to estimate broadcast throughput and goodput. We evaluate the accuracy of our methods by comparing our models with the measured data. In addition, we investigate the impact of the capture effect on interference.

Multihomed, mobile wireless computing and communication devices can spontaneously form communities to logically combine and share the bandwidth of each other's wide-area communication links using inverse multiplexing. But, membership in such a community can be highly dynamic, as devices and their associated WWAN links randomly join and leave the community. We identify the issues and trade-offs faced in designing a decentralized inverse multiplexing system in this challenging setting and determine precisely how heterogeneous WWAN links should be characterized and when they should be added to, or deleted from, the shared pool. We then propose methods of choosing the appropriate channels on which to assign newly arriving application flows. Using video traffic as a motivating example, we demonstrate how significant performance gains can be realized by adapting allocation of the shared WWAN channels to specific application requirements. Our simulation and experimentation results show that collaborative bandwidth aggregation systems are, indeed, a practical and compelling means of achieving high-speed Internet access for groups of wireless computing devices beyond the reach of public or private access points

Modern mobile devices are multi-homed with WLAN and WWAN communication interfaces. In a community of nodes with such multi-homed devices-locally inter-connected via high-speed WLAN but each globally connected to larger networks via low-speed WWAN, striping high-volume traffic from remote large networks over a bundle of low speed WWAN links can overcome the bandwidth mismatch problem between WLAN and WWAN. In our previous work, we showed that a packet striping system for such multi-homed devices-a mapping of delay-sensitive packets by an intermediate gateway to multiple channels using combination of retransmissions (ARQ) and forward error corrections (FEC)-can dramatically enhance the overall performance. In this paper, we improve upon a previous algorithm in two respects. First, by introducing two-tier dynamic programming tables to memoize computed solutions, packet striping decisions translate to simple table lookup operations given stationary network statistics. Doing so drastically reduces striping operation complexity. Second, new weighting functions are introduced into the hybrid ARQ/FEC algorithm to drive the long-term striping system evolution away from pathological local minima that are far from the global optimum. Results show the new algorithm performs efficiently and gives improved performance by avoiding local minima compared to the previous algorithm

Abstract — Estimation of network proximity among nodes is an important building block in several applications like service selection and composition, multicast tree formation, and overlay construction. Recently, scalable techniques have been proposed to estimate inter-node latencies, including network coordinate systems like GNP and Vivaldi. However, existing mechanisms for querying such information do not scale well to a very large number of nodes, when one wants to accurately find a set of nodes globally closest to a given node. In this paper we are concerned with distributing the position data among a set of infrastructure nodes, and propose ways of partitioning and querying this data. The trade-offs between accuracy and overhead in this distributed infrastructure are explored. We evaluate our solution through simulations with real and synthetic network measurement data. I.

Abstract — As the academic world moves away from physical journals and proceedings towards online document repositories, the ability to efficiently locate work of interest among the torrent of newly-generated papers will become increasingly important. To aid in this endeavor, we designed SmartSeer, a system that allows users to register personalized continuous queries over the CiteSeer database of technical documents. Users are then alerted whenever papers that match their queries are put online. SmartSeer has two main design requirements. First, to allow effective information retrieval, it should support rich continuous queries (as opposed to simple keyword searches). Second, to make effective use of donated infrastructure, it should be capable of running on a loosely maintained group of unreliable machines spread across multiple organizations (as opposed to assuming a reliable and tightly coupled distributed system). Existing work on distributed continuous query systems fails at least one of these requirements. Our design for SmartSeer is based on Distributed Hash Tables (DHTs), and thereby leverages previous work on DHT-based query systems. A prototype of SmartSeer has been implemented and evaluated on Planetlab. Though we evaluate our design only for the SmartSeer application, we believe it also provides useful insights into other distributed and rich continuous query systems (web alerts, news alerts etc). I.

Efficient resource discovery based on dynamic attributes such as CPU utilization and available bandwidth is a crucial problem in the deployment of computing grids. Existing solutions are either centralized or unable to answer advanced resource queries (e.g., range queries) efficiently. We present the design of NodeWiz, a grid information service (CIS) that allows multi-attribute range queries to be performed efficiently in a distributed manner. This is obtained by aggregating the directory services of individual organizations in a peer-to-peer information service.

Distributed computing has been widely embraced as a cost-effective means of performing compute-intensive tasks by pooling the computational resources of collaborating systems. We envision the emergence of an analogous approach to communication resource sharing which we call distributed communication. Distributed communication enables sharing a set of relatively low-speed WAN channels emanating from communities of multi-homed devices interconnected with a highspeed wireless LAN. We envisage opportunities to aggregate cellular links from spontaneously formed ad hoc groups of mobile devices, as well as broadband access links (e.g., DSL) from neighboring residences. But, will individuals be willing to share bandwidth as easily as they share bits? A prototype system that we have constructed convinces us that the technical challenges of distributed communication can be overcome. And there appears to be no other means of satisfying the growing demand for access bandwidth as quickly and as cheaply.

Multi-homed mobile devices have multiple wireless communication interfaces, each connecting to the Internet via a low speed and bursty WAN link such as a cellular link. We propose a packet striping system for such multi-homed devices — a mapping of packets by agateway to multiple channels, such that the overall performance is enhanced. We model and analyze the striping of delay-sensitive packets over multiple burst-loss channels. We derive the expected packet loss ratio when FEC (Forward Error Correction) and retransmissions are applied for error protection over multiple channels. We next model and analyze the case when the channels are bandwidth-limited. We develop a dynamic programming based algorithm that solves the optimal striping problem for the ARQ, the FEC, and the hybrid FEC/ARQ case.

Estimation of network proximity among nodes is an important building block in several applications like service selection and composition, multicast tree formation, and overlay construction. Recently, scalable techniques have been proposed to estimate inter-node latencies, including network coordinate systems like GNP and Vivaldi. However, existing mechanisms for querying such information do not scale well to a very large number of nodes, when one wants to accurately find a set of nodes globally closest to a given node. In this paper we are concerned with distributing the position data among a set of infrastructure nodes, and propose ways of partitioning and querying this data. The trade-offs between accuracy and overhead in this distributed infrastructure are explored. We evaluate our solution through simulations with real and synthetic network measurement data.

Multi-homed mobile devices have multiple wireless communication interfaces, each connecting to the Internet via a long range but low speed and bursty WAN link such as a cellular link. We propose a packet striping system for such multi-homed devices - a mapping of delay-sensitive packets by an intermediate gateway to multiple channels, such that the overall performance is enhanced. In particular, we model and analyze the striping of delay-sensitive packets over multiple burst-loss channels with random delays. We first derive the expected packet loss ratio when forward error correction (FEC) is applied for error protection over multiple channels. We next model and analyze the case when the channels are bandwidth-limited with shifted-gamma-distributed transmission delays. We develop a dynamic programming-based algorithm that solves the optimal striping problem for the ARQ, the FEC, and the hybrid FEC/ARQ case.

Multi-homed, mobile wireless computing and communication devices can spontaneously form communities to logically combine and share the bandwidth of each other's wide-area communication links using inverse multiplexing. But membership in such a community can be highly dynamic, as devices and their associated WAN links randomly join and leave the community. We identify the issues and tradeoffs faced in designing a decentralized inverse multiplexing system in this challenging setting, and determine precisely how heterogeneous WAN links should be characterized, and when they should be added to, or deleted from, the shared pool. We then propose methods of choosing the appropriate channels on which to assign newly-arriving application flows. Using video traffic as a motivating example, we demonstrate how significant performance gains can be realized by adapting allocation of the shared WAN channels to specific application requirements. Our simulation and experimentation results show that collaborative bandwidth aggregation systems are, indeed, a practical and compelling means of achieving high-speed Internet access for groups of wireless computing devices beyond the reach of public or private access points.

Gateway nodes are important elements in a sensor network since they provide the ability to establish long-range reach-back communication in order to retrieve critical data to remote locations. The gateways are however prone to failures just like any sensor nodes, and they consume significantly more energy since they transmit over longer distances compared with sensor-to-sensor links. We introduce an adaptive and fault-tolerant method for gateway assignment in sensor networks. Our approach is fully distributed and achieves the following objectives: (i) it allows surviving gateways to recover for other failed gateways; (ii) it distributes energy usage and traffic load between several gateway nodes within the sensor network. Each gateway adaptively controls its region of influence based on local conditions such as remaining energy level and traffic load. Our methodology was evaluated via simulation using a network model containing 400 sensor nodes with virtual targets. The simulation results indicate that our scheme is robust to gateway failures. Moreover, our scheme successfully balances the energy consumption and traffic load among the gateways.

With the wide availability of high-speed network access, we are experiencing high quality streaming media delivery over the Internet. The emergence of ubiquitous computing enables mobile users to access the Internet with their laptops, PDAs, or even cell phones. When nomadic users connect to the network via wireless links or phone lines, high quality video transfer can be problematic due to long delay or size mismatch between the application display and the screen. Our proposed solution to this problem is to enable network proxies with the transcoding capability, and hence provide different, appropriate video quality to different network environment. The proxies in our transcoding-enabled caching (TeC) system perform transcoding as well as caching for efficient rich media delivery to heterogeneous network users. This design choice allows us to perform content adaptation at the network edges. We propose three different TeC caching strategies. We describe each algorithm and discuss its merits and shortcomings. We also study how the user access pattern affects the performance of TeC caching algorithms and compare them with other approaches. We evaluate TeC performance by conducting two types of simulation. Our first experiment uses synthesized traces while the other uses real traces derived from an enterprise media server logs. The results indicate that compared with the traditional network caches, with marginal transcoding load, TeC improves the cache effectiveness, decreases the user-perceived latency, and reduces the traffic between the proxy and the content origin server.

Typical applications of mobile ad hoc networks (MANET) require group-oriented services. Digital battlefields and disaster relief operations make data dissemination and teleconferences a key application domain. Network-supported multicast is hence critical for efficient any-to-many communications. However, very little work has been done on "reliable" transport multicast. We propose and evaluate reliable adaptive lightweight multicast (RALM). The design choices of RALM are motivated by lessons we learned from evaluating the performance of traditional wired reliable multicast transport protocols (in particular, SRM) in ad hoc networks. We argue the two components, reliability and congestion control, are essential in designing a reliable multicast transport protocol for MANETs. RALM addresses both reliability and congestion control. It achieves reliability by guaranteeing data delivery to troubled receivers in a round-robin fashion. RALM's send-and-wait congestion control uses NACK feedback to adjust to congestion experienced by receivers. We show through simulations that RALM achieves perfect reliability while exhibiting low end-to-end delay and minimal control overhead compared against other protocols.

Application-level multicast networks overlaid on unicast IP networks are increasingly gaining in importance. While there have been several proposals for overlay multicast networks, very few of them focus on the stringent requirements of real-time applications such as streaming media. We propose RITA (Receiver Initiated Timely Adaptation) framework for an efficient overlay multicast infrastructure. RITA is based on a combination of landmark clustering and RTT measurements, and is particularly suitable for multimedia real-time applications. Our goal is to balance the network-oriented goals of building an efficient multicast tree with the application-oriented goals of providing good QoS with minimal disruptions. Using accurate global soft state information tables, our approach promptly constructs and reconfigures high quality trees. A distinguishing feature of our approach is that the tree reconfiguration is initiated just-in-time by the application client at the receiver when the media quality falls below a specific threshold. The goal is to achieve dynamic tree reconfiguration with very low switching delay such that end users do not perceive any application performance degradation.

The use of contention-based MAC protocols combined with hidden terminal problems make multi-hop wireless ad hoc networks much more sensitive to load and congestion than wired networks or even wireless cellular networks. In such an environment, we argue that multicast reliability cannot be achieved solely by retransmission of lost packets as is typically done in wired networks with protocols such as SRM. We contend that in order to achieve reliable multicast delivery in such networks, besides reliability mechanisms, we must also consider jointly two components: reliability and congestion control. In this paper, we propose CALM, a congestion controlled, adaptive, lightweight multicast transport protocol and show that congestion control alone can significantly improve reliable packet delivery in ad hoc networks when compared to traditional "wired" reliable multicast protocols.

Tertio millennio ineunte (at the dawn of the third millennium) it is clear that wireless communications are the natural form of communication among people and even things. Virtually every appliance is equipped with a small, cheap, lightweight wireless interface and the possibility of communicating anytime, from anywhere, to anybody/anything, in the world and beyond, is imminent. In the field of wireless communications and mobile computing, mobile ad hoc networks have received great attention recently. Research into ad hoc networks began in the early 1980s as DARPA packet radio networks. Ad hoc networks operate without any central administration (i.e., base stations and mobile switching centers). Because of its independence from fixed infrastructure, ad hoc networking is considered as the most promising network architecture to enable the ‘many any communications.’ The lack of infrastructure in ad hoc networks, which differentiates ad hoc networks from cellular networks, raises several research challenges. Each network node must act as a router and packet forwarder. Each node communicates via wireless radios that have limited transmission radius. If a node wishes to communicate with another node that is not within its transmission range, it must build a multihop route and rely on intermediate nodes to forward the packet. In addition, each node is mobile and hence the network topology constantly changes in an unpredictable manner. These challenges of ad hoc networks in combination with those of traditional wireless networks (e.g., limited bandwidth, power constraints, security, limited storage) make ad hoc networking research interesting and challenging. Ad hoc networks are deployed in places where it is impossible or difficult to build an infrastructure because of cost, security, and timeliness. Examples of these situations are mostly emergency and military scenarios such as search and rescue, battlefield, and disaster recovery. In recent years, new ad hoc networking technologies, Bluetooth for instance, have emerged and enable personal area networks and home networks to be new application areas of wireless mobile ad hoc networks. With this special issue we bring together seven papers that represent state-of-the-art contributions. The papers published in this issue range from papers that thoroughly survey advanced technologies to papers that report cutting-edge research to papers that describe industrial experiences. The first paper is authored by one of the pioneers in ad hoc networking research, Anthony Ephremides. His paper looks back at the history of ad hoc networks and the growth of this research field. Current research problems are also highlighted in this article. The paper by Bisdikian et al. describes recent mobile communications projects by IBM T.J. Watson Research Center. Their work on Bluetooth technology, BlueDrekar software stack based on Bluetooth specification, and WebSplitter middleware that enables web browsing in multiple devices are introduced. Their interesting vision of tomorrow's web in connected devices is also shared in this paper. The paper by Charles Perkins et al. reports connecting the AODV (Ad hoc On-demand Distance Vector) protocol to the IPv6 Internet. Several technologies are required to achieve this: general Internet gateway connectivity, address auto-configuration, mobile IPv6, changes to router advertisement, changes to treatment of default routes, and multi-gateway operations. These innovations are illustrated in this paper. The following two papers are survey papers. Camp et al. review the mobility models used in ad hoc networking simulations. They categorize the mobility models into entity models where each node's movement is independent of other nodes and group models where a number of nodes in a group share similar mobility patterns. The impact of the selection of mobility models on ad hoc networks performance is studied through extensive simulations. Achieving QoS in ad hoc networks is a difficult task. Perkins and Hughes survey the recent work in this area. They examine the research area in three components: routing, resource reservation, and MAC. The authors also give insights and suggestions for future work for QoS research in ad hoc networking. The paper by Belding-Royer proposes a new hierarchical routing scheme for ad hoc networks called Adaptive Routing along Clustering (ARC). This protocol increases routing flexibility, robustness, and scalability. ARC uses limited broadcasting to minimize the control message overhead. The simulation results show that when combined with AODV, ARC performs favorably over other clustering algorithms. Finally, Kong et al. describe an adaptive security framework for military wireless networks with Unmanned Aerial Vehicles (UAVs). When mobile backbone infrastructure is available, UAVs perform authentication for security services. When the infrastructure is absent, their system localizes the security services at each node for ad hoc communications mode. The paper reports results from their testbed and simulation implementations. We thank the WCMC Editor-in-Chief Mohsen Guizani for his support and the staff at Wiley (Mark Hammond, Laura Kempster, and Claire Bailey). Our deep gratitude also goes to over sixty reviewers for their excellent and detailed reviews. We also thank the authors of all the submitted papers. Enjoy the issue!

We present our experiences in implementing and validating the on-demand multicast routing protocol (ODMRP) in a real wireless ad hoc network testbed. ODMRP maintains a mesh for each multicast group to provide multiple alternate paths. Redundancy created by the mesh helps overcome frequent topology changes resulting from node mobility, channel fading, and interferences. The protocol does not maintain permanent route tables with full topological views. Instead, multicast senders reactively and dynamically discover routes and obtain multicast group information on demand. ODMRP is implemented in our testbed network consisting of six hosts using the kernel level multicast support option built into the Linux operating system. We describe the key design and implementation features of our protocol and report preliminary testbed experiment results of ODMRP and DVMRP (distance vector multicast routing protocol), a traditional tree based scheme.

Higher layer protocols in wireless networks need to dynamically adapt to observed network response. A common approach is for each protocol to employ an end-to-end monitoring and measuring mechanism and estimate quantities of interest, usually related to delay, latency or bandwidth. A less conventional approach is to employ network layer feedback mechanisms in place or in aid of end-to-end efforts. In this paper we use an 802.11-based link throughput and permissible throughput measurement. In our experiments these source-destination pair permissible throughput measurements are propagated in a multi-hop network using the AODV algorithm. In this setting, we are investigating how source rate adaptive multimedia applications can make use of this network feedback. We find that the investigated network feedback approach provides a strong network control feature, an accurate and timely available measurement at the sources, while the overhead of supporting and maintaining it is minimal. Finally, we argue that network based feedback approaches are particularly promising for higher layer protocol support and soft-QoS support in wireless networks.

In recent years, routing has been the most focused area in ad hoc networks research. On-demand routing in particular, is widely developed in bandwidth constrained mobile wireless ad hoc networks because of its effectiveness and efficiency. Most proposed on-demand routing protocols however, build and rely on a single route for each data session. Whenever there is a link disconnection on the active route, the routing protocol must perform a route recovery process. In QoS routing for wired networks, multiple path routing is popularly used. Multiple routes are however, constructed using link-state or distance vector algorithms which are not well-suited for ad hoc networks. We propose an on-demand routing scheme called split multipath routing (SMR) that establishes and utilizes multiple routes of maximally disjoint paths. Providing multiple routes helps minimizing route recovery process and control message overhead. Our protocol uses a per-packet allocation scheme to distribute data packets into multiple paths of active sessions. This traffic distribution efficiently utilizes available network resources and prevents nodes of the route from being congested in heavily loaded traffic situations. We evaluate the performance of our scheme using extensive simulation.

Ad hoc networks are deployed in situations where no base station is available and a network has to be built impromptu. Since there is no wired backbone, each host is a router and a packet forwarder. Each node may be mobile, and topology changes frequently and unpredictably. Routing protocol development has received much attention because mobility management and efficient bandwidth and power usage are critical in ad hoc networks. No existing protocol however, considers the load as the main route selection criteria. This routing philosophy can lead to network congestion and create bottlenecks. We present dynamic load-aware routing (DLAR) protocol that considers intermediate node routing loads as the primary route selection metric. The protocol also monitors the congestion status of active routes and reconstructs the path when nodes of the route have their interface queue overloaded. We describe three DLAR algorithms and show their effectiveness by presenting and comparing simulation results with an ad hoc routing protocol that uses the shortest paths.

The on-demand multicast routing protocol (ODMRP) is an effective and efficient routing protocol designed for mobile wireless ad hoc networks. One of the major strengths of ODMRP is its capability to operate both as a unicast and a multicast routing protocol. This versatility of ODMRP can increase network efficiency as the network can handle both unicast and multicast traffic with one protocol. We describe the unicast functionality of ODMRP and analyze the protocol performance in a real ad hoc network testbed of seven laptop computers in an indoor environment. Both static and dynamic networks are deployed. We generate various topological scenarios in our wireless testbed by applying mobility to network hosts and study their impacts on our protocol performance. We believe that the performance study in a testbed network can help us analyze the protocol in a realistic way and indicate the direction for future research.

In this paper we investigate the performance of multicast routing protocols in wireless mobile ad hoc networks. An ad hoc network is composed of mobile nodes without the presence of a wired support infrastructure. In this environment, routing/multicasting protocols are faced with the challenge of producing multihop routes under host mobility and bandwidth constraints. In recent years, a number of new multicast protocols of different styles have been proposed for ad hoc networks. However, systematic performance evaluations and comparative analysis of these protocols in a common realistic environment has not yet been performed. In this study, we simulate a set of representative wireless ad hoc multicast protocols and evaluate them in various network scenarios. The relative strengths, weaknesses, and applicability of each multicast protocol to diverse situations are studied and discussed.

As mobile computing gains popularity, the need for ad hoc routing protocols will continue to grow. There have been numerous simulations comparing the performance of these protocols under varying conditions and constraints. One question that arises is whether the choice of MAC protocol affects the relative performance of the routing protocols being studied. This paper investigates the answer to that question by simulating the performance of three ad hoc routing protocols when run over different MAC protocols. It is determined that the choice of MAC layer protocol does, in fact, affect the relative performance of the routing protocols.

Nodes in mobile ad hoc networks communicate with one another via packet radios on wireless multihop links. Because of node mobility and power limitations, the network topology changes frequently. Routing protocols therefore play an important role in mobile multihop network communications. A trend in ad hoc network routing is the reactive on-demand philosophy where routes are established only when required. Most of the protocols in this category, however, use a single route and do not utilize multiple alternate paths. We propose a scheme to improve existing on-demand routing protocols by creating a mesh and providing multiple alternate routes. Our algorithm establishes the mesh and multipaths without transmitting any extra control message. We apply our scheme to the Ad-hoc On-Demand Distance Vector (AODV) protocol and evaluate the performance improvements by simulation.

An ad hoc wireless network is composed of mobile hosts without any wired infrastructure support. In mobile ad hoc networks, unicast and multicast routing protocols are faced with the challenge of producing multihop routes because of limited radio propagation range. In addition, routing protocols must manage mobility and be bandwidth and power efficient. The on-demand multicast routing protocol (ODMRP) is a protocol designed for ad hoc networks with multicast purposes. Two unique features of ODMRP are its unicast capability and its utilization of a mobility prediction scheme to perform rerouting in anticipation of route disconnection. We describe ODMRP unicast routing functionality and assess the mobility prediction effectiveness and efficiency. We evaluate the ODMRP performance via detailed simulation and compare it with other ad hoc routing schemes.

An ad hoc wireless network is an infrastructureless network composed of mobile hosts. The primary concerns in ad hoc networks are bandwidth limitations and unpredictable topology changes. Thus, efficient utilization of routing packets and immediate recovery of route breaks are critical in routing and multicasting protocols. A multicast scheme, on-demand multicast routing protocol (ODMRP), has been recently proposed for mobile ad hoc networks. ODMRP is a reactive (on-demand) protocol that delivers packets to destinations on a mesh topology using scoped flooding of data. A number of enhancements can be applied to improve the performance of ODMRP. In this paper, we propose a mobility prediction scheme to help select stable routes and to perform rerouting in anticipation of topology changes. We also introduce techniques to improve transmission reliability and eliminate route acquisition latency. The impact of our improvements is evaluated via simulation.

This paper presents a novel multicast routing protocol for mobile ad hoc wireless networks. The protocol, termed ODMRP (on-demand multicast routing protocol), is a mesh-based, rather than a conventional tree-based multicast scheme and uses a forwarding group concept (only a subset of nodes forwards the multicast packets via scoped flooding). It applies on-demand procedures to dynamically build routes and maintain multicast group membership. ODMRP is well suited for ad hoc wireless networks with mobile hosts where bandwidth is limited, topology changes frequently, and power is constrained. We evaluate ODMRP's scalability and performance via simulation.