Edge vision systems are critical for real-time applications like autonomous navigation and remote monitoring, requiring reliable low-latency data transmission. Wireless networks face challenges, including dynamic bandwidth, unpredictable packet loss, and congestion that severely impact quality-of-service. This doctoral research introduces a content-aware multipath transmission framework combining reinforcement learning-based routing with deep learning reconstruction for optimizing real-time image delivery over heterogeneous wireless networks. Our system uses YOLO object detection and ResNet feature extraction to identify semantically critical image regions, employing a Deep Q-Network agent for intelligent routing decisions across multiple wireless interfaces. Compared to existing approaches treating all data uniformly, our content-aware scheduler achieves 48% lower network latency compared to single-path transmission while maintaining superior reconstruction quality.

Remote supervision of autonomous vehicles requires real-time video streaming under severe bandwidth constraints. Existing adaptive streaming protocols treat all spatial regions uniformly, failing to account for the varying criticality of objects based on proximity. I propose a depth-aware adaptive streaming system that leverages stereo depth estimation to intelligently allocate bandwidth by prioritizing encoding quality for nearby safety-critical objects while reducing bitrate for distant regions. My preliminary work validates this approach through improved far-object detection and substantial bandwidth reduction. This doctoral research develops a unified framework that dynamically adjusts spatial quality allocation based on depth maps, scene complexity, and network conditions.

As robots (edge-devices, agents) find uses in an increasing number of settings and edge-cloud resources become pervasive, wireless networks will often be shared by flows of data traffic that result from communication between agents and their corresponding edge-cloud nodes (cloud compute or data resource accessed by an agent). In such a setting, any agent communicating with the edge-cloud is unaware of the state of the network resource, which evolves in response to not just the agent’s own communication at any given time but also to communication by the other agents, which stays unknown to the agent. We address the challenge of an agent learning a policy that allows it to decide whether or not to communicate with its cloud node, using limited feedback it obtains from its own attempts to communicate, with the goal of optimizing its utility. The policy must generalize well to any number of other agents sharing the network and must not be trained for any particular network configuration. Our proposed policy is a deep reinforcement learning model Query Net (QNet) that we train using a proposed simulation-to-real framework. Our simulation model has just one parameter and is agnostic to specific configurations of any wireless network. It however allows training an agent’s policy over a wide range of outcomes that an agent’s communication with its edge-cloud node may face when using a shared network, by suitably randomizing the simulation parameter. We propose a learning algorithm that addresses the challenges we observe in training QNet. We validate our simulation-to-real driven approach through experiments conducted on real wireless networks including WiFi and cellular. We compare QNet with other policies to demonstrate its efficacy. Our WiFi experiments involved as few as five agents, resulting in barely any contention for the network, to as many as 50 agents, resulting in severe contention. The cellular experiments spanned a broad range of network conditions, with baseline network round-trip times ranging from a low of 0.07 s to a high of 0.83 s.

The growing popularity of live online classes, even in remote areas, stresses the need for a good and seamless quality of experience to enhance learning. However, these bandwidth-hungry applications challenge the current cellular networks to maintain consistent bandwidth and latency. In this work, we, therefore, propose using the collaboration of multiple devices with their individual cellular networks to support such live video streaming. We design a content-aware system Compact that splits video into foreground and background using video tiles (independently encoded spatial blocks) and streams them over different paths. Compact depends on its scheduler, which exhaustively searches for the best quality based on the network estimates. We extensively evaluate our system using network traces while walking and traveling on the bus or car. Compared to the single path, Compact manages to reduce the median stall and E2E lag by 70.6% and 28.57%, and the tail stall and lag by 83.9% and ≈ 80% on a bus trace. Furthermore, we performed a live experiment to test Compact on the actual cellular network.

Optimizing tail latency while efficiently managing computational resources is crucial for delivering high-performance, latency-sensitive services in edge computing. Emerging applications, such as augmented reality, require low-latency computing services with high reliability on user devices, which often have limited computational capabilities. Consequently, these devices depend on nearby edge servers for processing. However, inherent uncertainties in network and computation latencies—stemming from variability in wireless networks and fluctuating server loads—make service delivery on time challenging. Existing approaches often focus on optimizing median latency but fall short of addressing the specific challenges of tail latency in edge environments, particularly under uncertain network and computational conditions. Although some methods do address tail latency, they typically rely on fixed or excessive redundancy and lack adaptability to dynamic network conditions, often being designed for cloud environments rather than the unique demands of edge computing. In this paper, we introduce SafeTail, a framework that meets both median and tail response time targets, with tail latency defined as latency beyond the percentile threshold. SafeTail addresses this challenge by selectively replicating services across multiple edge servers to meet target latencies. SafeTail employs a reward-based deep learning framework to learn optimal placement strategies, balancing the need to achieve target latencies with minimizing additional resource usage. Through trace-driven simulations, SafeTail demonstrated near-optimal performance and outperformed most baseline strategies across three diverse services.

Storing large volumes of traffic video content in cloud storage is an expensive undertaking, given the limited capacity of cloud storage and its inability to store data beyond a few weeks. To address this issue, this article introduces CLOUD-CODEC , a novel video encoding approach tailored specifically for traffic monitoring video. CLOUD-CODEC offers three key advantages: (i) real-time encoding without any delay, (ii) near-perfect video quality upon decoding, and (iii) one-fifth the storage size of traditional encoding methods. CLOUD-CODEC is generally applicable to traffic cameras under various weather and lighting conditions. The encoding algorithm is a lightweight DNN-based object detection and box-shaped segmentation approach. The method can uniquely detect and segment cars, pedestrians, and moving objects with the marginal box-shaped contours. Periodic object detection makes it possible for CLOUD-CODEC to operate in real-time and estimate the movement of objects between predictions. Proof-of-concept evaluations using a massive dataset indicate that CLOUD-CODEC reduces video size by 80%—surpassing AV1 (34.9%), CloudSeg (58.4%), Detection (76.9%), Segmentation (73.1%), and Segm&Sort (69.5%). It achieves a frame rate of 95.8 when encoding and a VMAF score of 72.54 after decoding, with a storage size that is one-fifth of traditional methods. Field-testing of CLOUD-CODEC on metropolitan traffic cameras demonstrates its ability to extend storage time by 74.92%.

Smartphone app usage has steeply risen in India in the past decade. But limited efforts in the past assess the privacy aspects of these smartphone apps. Many of these are used for common utilities and handle sensitive user data. Such sensitive data leaks can have a wide variety of consequences when exposed to untrusted players (e.g., repressive governments, data/content hosting companies, and other third parties). These could range from mere embar-rassment to personal targeting and surveillance. This paper presents a measurement study on the data collection and privacy considerations of some of the most popular apps on the Indian Google Play Store. We selected 24 apps, and analyzed their data collection behavior on phones as well as the security of the servers to whom they send the data. We also obfuscated the data being collected and sent to the backend servers. Interestingly, for the non-government apps we found that extensive (mostly personally identifiable information), often “unnecessary”, data collection is being performed. In other words, we observed that a lot of these apps work fine even without these pieces of sensitive information. We then classified the data collected as necessary or not necessary based on this information. Furthermore, we found that often such sensitive data may be available in plaintext to the intermediate players managing/deploying the hosting infrastructure. We also found that while the governmental services-based apps collect fewer such unnecessary data points, they often store the data on web-fronted back-end databases with little to no user authentication mechanisms enabled. We expect our study to enable better understanding among both users and app developers about the privacy implications of these data collection practices.

The proliferation of video analytics in applications like autonomous driving, traffic surveillance, and teleoperated vehicles requires on-premise (on edge) execution of deep learning models to meet latency requirements and curb bandwidth usage by limiting frequent offloading of inference tasks. However, constrained by the compute and power availability on the edge, a cheaper model is typically deployed. These shallower models have two major associated problems: 1) using the same model for all cameras/vehicles gives inconsistent accuracy, and 2) trained models are prone to data drift.

Smart factories have data packets with a mix of stringent and non-stringent deadlines with varying levels of importance that need to be delivered via a wireless network. However, the scheduling of packets in the wireless network is crucial to satisfy the deadlines. In this work, we propose a technique of utilizing IEEE 802.11ax, popularly known as WiFi 6, for such applications. IEEE 802.11ax has a few unique characteristics, such as specific configurations of dividing the channels into resource units (RU) for packet transmission and synchronized parallel transmissions. We model the problem of scheduling packets by assigning profit to each packet and then maximizing the sum of profits. We first show that this problem is strongly NP-Hard, and then propose an approximation algorithm with a 12-approximate algorithm. Our approximation algorithm uses a variant of local search to associate the right RU configuration to each packet and identify the duration of each parallel transmission. Finally, we extensively simulate different scenarios to show that our algorithm works better than other benchmarks.

In recent years, LiFi has become increasingly pop-ular as an indoor communication technology that utilizes the unlicensed visible light and infra-red spectrum to transmit data. A major challenge of utilizing LiFi is that its area of coverage is limited. Thus, a large number of LiFi access points (APs) is often complemented by deploying a WiFi AP to form a hybrid WiFilLiFi network. However, such deployment does not lead to any additional improvement in the performance of WiFi or LiFi APs. Recent WiFi APs are known to have high overhead due to the requirement of channel state information (CSI), which is essential for utilizing spatial multiplexing. Thus, in this work, we propose a system called WiLiConnect (WiFi-LiFi Connectivity with CSI), which communicates the CSI requirement of the WiFi channel through the LiFi APs, thereby reducing the overhead of WiFi APs. We formulate this problem of load-balancing the overhead of CSI sharing across the LiFi APs, and show that the general problem is NP-Hard. We then propose a round-robin algorithm to solve a special case of the problem, where all the users are assumed to have a single antenna. We further utilize extensive simulation to show that WiLiConnect significantly reduces the overhead of sending CSI. Specifically, WiLiConnect incurs only 0.06% overhead on a WiFi AP having 8 antennas on the total sum rate.

We address the problem of localizing an (unauthorized) transmitter using a distributed set of sensors. Our focus is on developing techniques that perform the transmitter localization in an efficient manner, wherein the efficiency is defined in terms of the number of sensors used to localize. Localization of unauthorized transmitters is an important problem which arises in many important applications, e.g., in patrolling of shared spectrum systems for any unauthorized users. Localization of transmitters is generally done based on observations from a deployed set of sensors with limited resources, thus it is imperative to design techniques that minimize the sensors’ energy resources. In this paper, we design greedy approximation algorithms for the optimization problem of selecting a given number of sensors in order to maximize an appropriately defined objective function of localization accuracy. The obvious greedy algorithm delivers a constant-factor approximation only for the special case of two hypotheses (potential locations). For the general case of multiple hypotheses, we design a greedy algorithm based on an appropriate auxiliary objective function—and show that it delivers a provably approximate solution for the general case. We develop techniques to significantly reduce the time complexity of the designed algorithms by incorporating certain observations and reasonable assumptions. We evaluate our techniques over multiple simulation platforms, including an indoor as well as an outdoor testbed, and demonstrate the effectiveness of our designed techniques—our techniques easily outperform prior and other approaches by up to 50-60% in large-scale simulations and up to 16% in small-scale testbeds.

In recent times, the need for low latency has made it necessary to deploy application services physically and logically close to the users rather than using the cloud for hosting services. This paradigm of computing, known as edge or fog computing, is becoming increasingly popular. An edge user allocation policy determines how to allocate service requests from mobile users to MEC servers. Current state-of-the-art techniques assume that the total resource utilization on an edge server is equal to the sum of the individual resource utilizations of services provisioned from the edge server. However, the relationship between resources utilized on an edge server with the number of service requests served from there is usually highly non-linear, hence, mathematically modelling the resource utilization is challenging. This is especially true in case of an environment with CPU-GPU co-execution, as commonly observed in modern edge computing. In this work, we provide an on-device Deep Reinforcement Learning (DRL) framework to predict the resource utilization of incoming service requests from users, thereby estimating the number of users an edge server can accommodate for a given latency threshold. We further propose an algorithm to obtain the user allocation policy. We compare the performance of the proposed DRL framework with traditional allocation approaches and show that the DRL framework outperforms deterministic approaches by at least 10% in terms of the number of users allocated.

Users are surrounded by sensors that are available through various devices beyond their smartphones. However, these sensors are not fully utilized by current end-user applications. A key reason sensor use is so limited is that application developers must exactly identify how the sensor data can be used by smartphone apps. To mitigate this problem, we present SenseWear, a sensor-sharing platform that extends the functionality of a smartphone to use remote sensors with limited additional developer effort. Sensor sharing has several uses, including augmenting the hardware in smartphones, creating new gestural interactions with smartphone applications, and improving application's Quality of Experience via higher-quality sensors from other devices, such as wearables. We developed and present six use cases that use remote sensors in various smartphone applications. Each extension requires adding fewer than 20 lines of code on average. Furthermore, using remote sensors did not introduce a perceptible increase in latency, and creates more convenient interaction options for smartphone apps.

For bandwidth-efficient streaming of 360-degree videos, the streaming technique must adapt both to the changing viewport of the user and variations of the available network bandwidth. The state-of-the-art streaming techniques for this problem attempt to solve an optimization using simplified rules that do not adapt very well to the uncertainties related to the viewport or network. We adopt a 3D-Convolutional Neural Networks (3DCNN) model to extract spatio-temporal features of videos and predict the viewport. Given the sequential decision-making nature of such streaming technique, we then apply a Reinforcement Learning (RL) based adaptive streaming approach. We address the challenges of using RL in this scenario, such as large action space and delayed reward evaluation. Comprehensive evaluations with real network traces show that the proposed method outperforms three tile-based streaming techniques for 360-degree videos. Compared to the tile-based streaming techniques, the average user-perceived bitrate of the proposed method is 1.3-1.7 times higher and the average quality of experience of the proposed method is also 1.6-3.4 times higher. Subjective user studies further confirm the superiority of the proposed approach.

Conventional streaming solutions for streaming 360-degree panoramic videos are inefficient in that they download the entire 360-degree panoramic scene, while the user views only a small sub-part of the scene called the viewport. This can waste over 80% of the network bandwidth. We develop a comprehensive approach called Mosaic that combines a powerful neural network-based viewport prediction with a rate control mechanism that assigns rates to different tiles in the 360-degree frame such that the video quality of experience is optimized subject to a given network capacity. We model the optimization as a multi-choice knapsack problem and solve it using a greedy approach. We also develop an end-to-end testbed using standards-compliant components and provide a comprehensive performance evaluation of Mosaic along with five other streaming techniques - two for conventional adaptive video streaming and three for 360-degree tile-based video streaming. Mosaic outperforms the best of the competitions by as much as 47-191% in terms of average video quality of experience. Simulation-based evaluation as well as subjective user studies further confirm the superiority of the proposed approach.

We address the problem of localizing an (illegal) transmitter using a distributed set of sensors. Our focus is on developing techniques that perform the transmitter localization in an efficient manner, wherein the efficiency is defined in terms of the number of sensors used to localize. Localization of illegal transmitters is an important problem which arises in many important applications, e.g., in patrolling of shared spectrum systems for any unauthorized users. Localization of transmitters is generally done based on observations from a deployed set of sensors with limited resources, thus it is imperative to design techniques that minimize the sensors' energy resources. In this paper, we design greedy approximation algorithms for the optimization problem of selecting a given number of sensors in order to maximize an appropriately defined objective function of localization accuracy. The obvious greedy algorithm delivers a constant-factor approximation only for the special case of two hypotheses (potential locations). For the general case of multiple hypotheses, we design a greedy algorithm based on an appropriate auxiliary objective function - and show that it delivers a provably approximate solution for the general case. We develop techniques to significantly reduce the time complexity of the designed algorithms, by incorporating certain observations and reasonable assumptions. We evaluate our techniques over multiple simulation platforms, including an indoor as well as an outdoor testbed, and demonstrate the effectiveness of our designed techniques - our techniques easily outperform prior and other approaches by up to 50-60% in large-scale simulations.

360° videos provide an immersive experience to users, but require considerably more bandwidth to stream compared to regular videos. State-of-the-art 360° video streaming systems use viewport prediction to reduce bandwidth requirement, that involves predicting which part of the video the user will view and only fetching that content. However, viewport prediction is error prone resulting in poor user Quality of Experience (QoE). We design PARSEC, a 360° video streaming system that reduces bandwidth requirement while improving video quality. PARSEC trades off bandwidth for additional client-side computation to achieve its goals. PARSEC uses an approach based on super-resolution, where the video is significantly compressed at the server and the client runs a deep learning model to enhance the video to a much higher quality. PARSEC addresses a set of challenges associated with using super-resolution for 360° video streaming: large deep learning models, slow inference rate, and variance in the quality of the enhanced videos. To this end, PAR-SEC trains small micro-models over shorter video segments, and then combines traditional video encoding with super-resolution techniques to overcome the challenges. We evaluate PARSEC on a real WiFi network, over a broadband network trace released by FCC, and over a 4G/LTE network trace. PARSEC significantly outperforms the state-of-art 360° video streaming systems while reducing the bandwidth requirement.

We address the problem of localizing multiple intruders (unauthorized transmitters) using a distributed set of sensors in the context of a shared spectrum system. In contrast to single transmitter localization, multiple transmitter localization (MTL) has not been thoroughly studied. In shared spectrum systems, it is important to be able to localize simultaneously present multiple intruders to effectively protect a shared spectrum from malware-based, jamming, or other multi-device unauthorized-usage attacks. The key challenge in solving the MTL problem comes from the need to "separate" an aggregated signal received from multiple intruders into separate signals from individual intruders. Furthermore, in a shared spectrum paradigm, presence of an evolving set of authorized users (e.g., primary and secondary users) adds to the challenge.In this paper, we propose an efficient algorithm for the MTL problem based on the hypothesis-based Bayesian approach called MAP. Direct application of the MAP approach to the MTL problem incurs prohibitive computational and training cost. In this work, we develop optimized techniques based on MAP with significantly improved computational and training costs. In particular, we develop a novel interpolation method, ILDW, which helps minimize the training cost. We generalize our techniques via online-learning to the setting wherein there may be a set of dynamically-changing authorized users present in the background. We evaluate our developed techniques on large-scale simulations as well as on small-scale indoor and outdoor testbeds. Our experiments demonstrate that our technique outperforms the prior approaches by significant margins, i.e., error up to 74% less in large-scale simulations and 30% less in real-world testbeds.

GNSS/GPS is a positioning system widely used nowadays in our lives for real-time localization in Earth. This technology is highly vulnerable to spoofing/jamming attacks caused by malicious intruders. In the recent years, commodity and low-cost radio-frequency hardware have been used to interfere with the legitimate GPS signal. Existing spoofing detection solutions use costly receivers and computationally expensive algorithms which limit the large-scale deployment. In this work we propose a GNSS spoofing detection system that can run on spectrum sensors with Software-Defined Radio (SDR) capabilities and cost in the order of 20 euros. Our approach exploits the predictability of the Doppler characteristics of the received GPS signals to determine the presence of anomalies or malicious attackers. We propose an artificial recurrent neural network (RNN) based on Long short-term memory (LSTM) for anomaly detection. We use data received by low-cost SDR receivers that are processed locally by low-cost embedded machines such as Nvidia Jetson Nano to provide inference capabilities. We show that our solution predicts very accurately the Doppler shift of GNSS signals and can determine the presence of a spoofing transmitter.

Consumer IoT networks are characterized by heterogeneous devices with diverse functionality and programming interfaces. This lack of homogeneity makes the integration and secure management of IoT infrastructures a daunting task for users and administrators. In this paper, we introduce VISCR, a Vendor-Independent policy Specification and Conflict Resolution engine that enables intent-based conflict-free policy specification and enforcement in IoT environments. VISCR converts the topology of the IoT infrastructure into a tree-based abstraction and translates existing policies from heterogeneous vendor-specific programming languages, such as Groovy-based SmartThings, OpenHAB, IFTTT-based templates, and MUD-based profiles, into a vendor-independent graph-based specification. These are then used to automatically detect rogue policies, policy conflicts, and automation bugs. We evaluated VISCR using a dataset of 907 IoT apps, programmed using heterogeneous automation specifications, in a simulated smart-building IoT infrastructure. In our experiments, among 907 IoT apps, VISCR exposed 342 of IoT apps as exhibiting one or more violations, while also running 14.2x faster than the state-of-the-art tool (Soteria). VISCR detected 100% of violations reported by Soteria while also detecting new types of violations in 266 additional apps.

Conventional streaming solutions for streaming 360-degree panoramic videos are inefficient in that they download the entire 360-degree panoramic scene, while the user views only a small sub-part of the scene called the viewport. This can waste over 80% of the network bandwidth. We develop a comprehensive approach called Mosaic that combines a powerful neural network-based viewport prediction with a rate control mechanism that assigns rates to different tiles in the 360-degree frame such that the video quality of experience is optimized subject to a given network capacity. We model the optimization as a multi-choice knapsack problem and solve it using a greedy approach. We also develop an end-to-end testbed using standards-compliant components and provide a comprehensive performance evaluation of Mosaic along with four other streaming techniques - two for conventional adaptive video streaming and two for 360-degree tile-based video streaming. Mosaic outperforms the best of the competition by as much as 50% in terms of median video quality.

We use a crowdsourcing approach for RF spectrum patrolling, where heterogeneous, low-cost spectrum sensors are deployed widely and are tasked with detecting unauthorized transmissions in a collaborative fashion while consuming only a limited amount of resources. We pose this as a collaborative signal detection problem where the individual sensor's detection performance may vary widely based on their respective hardware or software configurations, but are hard to model using traditional approaches. Still an optimal subset of sensors and their configurations must be chosen to maximize the overall detection performance subject to given resource (cost) limitations. We present the challenges of this problem in crowdsourced settings and present a set of methods to address them. The proposed methods use data-driven approaches to model individual sensors and develops mechanisms for sensor selection and fusion while accounting for their correlated nature. We present performance results using examples of commodity-based spectrum sensors and show significant improvements relative to baseline approaches.

Computation offloading frameworks partition an application's execution between a cloud server and the mobile device to minimize its completion time on the mobile device. An important component of an offloading framework is the partitioning algorithm that decides which tasks to execute on mobile device or cloud server. The partitioning algorithm schedules tasks of a mobile application for execution either on mobile device or cloud server to minimize the application finish time. Most offloading frameworks partition parallel applications devices using an optimization solver which takes a lot of time. We show that by allowing duplicate execution of selected tasks on both the mobile device and the remote cloud server, a polynomial algorithm exists to determine a schedule that minimizes the completion time. We use simulation on both random data and traces to show the savings in both finish time and scheduling time over existing approaches. Our trace-driven simulation on benchmark applications shows that our algorithm reduces the scheduling time by 8 times compared to a standard optimization solver while guaranteeing minimum makespan.

Mobile cloud computing is increasingly being used in recent times to offload parts of an application to the cloud to reduce its finish time. However, quality of offloading decisions depend on network conditions and hence many offloading solutions assume that MAC layer retransmissions will tackle transient frame errors. This can lead to suboptimal solutions, as well as, degrade service level guarantee of reducing finish time compared to execution without offloading. In this work, we propose an error-aware solution that uses run-time channel conditions to adapt the offloading decisions. We guarantee that given a failure rate bound (ϵ), offloading decisions will achieve application execution in less time than that of local execution with a probability of (1-ϵ) while operating in networks with unpredictable error characteristics. Simulation results show that at channel error rate of 20%, our heuristic provides 90% guarantee of better performance than on-device computation and reduces the mean finish time by 18% compared to execution without any offloading.