Surging data demand driven by the proliferation of smartphones is revolutionizing the wire- less scene. By 2020, the number of smartphone subscribers will grow to 9 billion and the demand in mobile data traffic will grow 10 folds. Capacity (bandwidth) and Latency are the key contributing factors to a positive user experience. For example, streaming a video from Netflix or Youtube is bandwidth limited while loading or browsing the Youtube or Net- flix website is latency limited. One can argue that by using more spectrum(wider wireless channels) can solve the capacity problem however according to Federal Communications Commission(FCC) we are already facing a spectrum deficit. Newer network technologies like 4G and LTE have improved the network performance(both capacity and latency) but 60world’s mobile subscribers will still be using the slower 3G or 2G connections for at least a decade. Another challenge that mobile devices are posing is that the networks previously dominated by the static users (or static channel conditions) with devices like laptops are increasingly becoming more heterogeneous( mix of static and mobile users with varying channel conditions).

In this dissertation, I address the issue of managing wireless resources to support various multimedia applications, in wireless networks with a high degree of user mobility. Driven by necessity for flexible wireless resource management, I design, build and eval- uate three wireless resource management schemes. First, I present JPRA, a joint power and rate adaptation scheme which adaptively distributes uneven power levels of OFDMA sub-carriers to cope with frequency selectivity in fading. Thus, improves the total network capacity. Second, I present TRINITY, a practical transmitter cooperation scheme to handle heterogeneous user profiles in wireless networks. Sophisticated MIMO-based transmission strategies, based on transmitter cooperation, have emerged recently. However, different types of users profiles (e.g., static vs mobile, stable vs dynamic channels) that make up todays enterprises, require different MIMO transmission strategies(CMSA, Network MIMO and DAS). With the wrong strategy, a user could even see a degradation in performance. TRINITY can simultaneously cater to a heterogeneous mix of users, by intelligently com- bining a plurality of MIMO transmission strategies wherein the transmitters at different nodes can cooperate to deliver significant performance gains.

Next, I present FluidNet, a scalable, light-weight framework for realizing the full potential of Cloud-based radio access networks(C-RAN). Cloud-based radio access networks (C-RAN) have been proposed as a cost-efficient way of deploying small cells. In this work, we argue that the intelligent configuration of the front-haul network between the baseband processing units(BBUs) and remote radio heads, is essential in delivering the performance and energy benefits to the RAN and the BBU pool, respectively. FluidNet deploys a logically re-configurable front-haul to apply appropriate transmission strategies in different parts of the network and hence cater effectively to both heterogeneous user profiles and dynamic traffic load patterns.

Finally, I present FlexiWeb, a network aware compaction scheme for accelerat- ing mobile web transfers. Latency is often cited as the main reason for poor mobile web performance and to reduce the page load times, middleboxes that compress page content are commonly used today. Unfortunately, this can hurt performance in many cases. Our measurements reveal that the middlebox should be used only when network conditions are bad; otherwise, most objects in the web page should be directly fetched from the source web server. Based on this observation we build FlexiWeb, a framework that supports network- aware middlebox usage and performs dynamic network-aware compression to provide further performance gains.

With the advent of personalized services and devices, like online social networks and smart- phones, a user's (otherwise private or sensitive) information is shared with these platforms. While this sharing of information undoubtedly adds value to the user's experience, it raises some pressing concerns about the user's privacy, or the lack of it.

In this dissertation, we propose three major frameworks designed and implemented specifically to protect the users' privacy on such emerging and highly personalized platforms. The three frameworks are as follows.

1. Twitsper, a wrapper for Twitter that allows user to exchange private messages without hurting Twitter's commercial interest.

2. Hermes, a decentralized social network with architecture the explicit goal of preserving a user's privacy and anonymity (including even their sharing patterns) while minimizing the cost each user incurs due to the decentralized model.

3. ZapDroid, a version of Android that is modified and enhanced specifically for the purpose of detecting and negating the impacts of zombie apps. In the context of our work, zombie apps are unused apps that still run in the background, consuming resources and leaking the users' private information. As an OS, ZapDroid is compatible with any app downloadable from the Google Play Store.

Power Line Communication (PLC) based WiFi extenders can improve WiFi coverage in homes and enterprises. Unlike in traditional WiFi networks which use an underlying high data rate Ethernet backhaul, a PLC backhaul may not support high data rates. Specifically, our measurements show that the end-to-end throughput for the concatenated PLC-WiFi link is determined by the bottleneck segment of the aggregated WiFi-PLC link.

The main objective of this dissertation is to maximize the aggregate of WiFi-PLC networks. The main challenge we address is that PLC capacity could be of a lower capacity than the WiFi link. Thus, users that are connected via a good WiFi link could suffer from the poor PLC link associated with the PLC extender they are attached to. Therefore, we develop WOLT, a system that connects each user to a signal extender for the sake of maximizing the aggregate throughput. It does so while accounting for the PLC link capacities and the users' WiFi link qualities. The results obtained from real testbed experiments and high-fidelity simulations and they show that WOLT is capable of increasing the aggregate throughput by more than 2.5x compared to a greedy user association baseline.

Moreover, users that are occluded from the main router could suffer from well-known WiFi impairments such as deep fading and shadowing. To overcome this issue, Distributed Antenna Systems (DAS) have been shown to improve the robustness and stability of the WiFi link. We approach this problem by two solutions. First, we develop PLC-DAS, a system that finds the correct DAS set of extenders for each user in home settings. We compare PLC-DAS against three baselines and across three fairness models. The results from our simulations show an improvement up to 4.5x compared to blindly using all WiFi-PLC extenders to form a DAS transmitter, while maintaining a fairness Jain’s index value of at least 0.97 with proportional and max-min fairness models.

Secondly, we develop Priza, a scalable system that clusters WiFi-PLC extenders into cells and assigns frequencies to each cell. Priza is evaluated via real testbed experiments and high-fidelity simulations and it can increase the aggregate throughput by more than 3x compared to the baseline that creates as large DAS cells as possible.

The open and shared nature of the wireless medium makes it easy for adversaries to launch simple, yet effective, denial of service attacks (DoS attacks).

As an example, jamming attacks, involve the uncoordinated transmission of electromagnetic energy on the medium.

In a carrier sensing network (e.g., 802.11), this attack strategy increases the number of collisions at the receiver side and/or blocks the medium access to legitimate nodes at the transmitting side.

Both of the above effects degrade the wireless network performance significantly.

Frequency hopping (FH) has been traditionally used to overcome jamming attacks.

However, we analytically and experimentally show that FH is inadequate to efficiently cope with jamming in today's networks.

Later we propose a suite of systems that aim at coping with jamming attacks at various levels (i.e., detection, localization and prevention).

We first identify two intelligent and effective jamming attacks that can be launched in 802.11 WLANs and we provide robust detection systems.

In particular, we design and implement

(i) CMD, a system to detect active jamming attacks that exploit the carrier sensing functionality of 802.11 networks and

(ii) FIJI, a cross-layer system for detecting (and mitigating) jamming attacks that exploit the performance anomaly of 802.11 WLANs.

Furthermore, given the importance of locating the jamming device in many deployment scenarios (e.g., battlefield), we propose a lightweight jamming localization scheme.

Our system utilizies ideas borrowed from the gradient descent optimization method.

The system's evaluations, show the potentials and applicability of our localization strategy.

The final step for coping with jamming attacks is jamming prevention.

Based on our initial measurement driven analysis, we do not rely on a FH scheme, that tries to simply avoid the jammer.

On the contrary, we design, implement and evaluate a prevention system, called ARES (Anti-jamming REinforcement System), to fight against the saboteur.

ARES is applicable to carrier sensing networks and tunes the parameters of rate adaptation and power control to improve the performance under the presence of an attack while ensuring that operations under benign conditions are unaffected.

Our extensive evaluations, show that ARES improves the network throughput across all scenarios by up to 150%.

Survivability is usually defined as the capability of a network to deliver data successfully in a timely manner, even in the presence of attacks. Security as well as reliability are included in survivability considerations. Due to the lossy and open nature of wireless links, design of solutions to make the network survive against attacks as well as wireless idiosyncrasies is difficult. In this dissertation, we approach several issues that have to be considered in order to ensure survivability in various contexts. Firstly, we need to understand the causes for why the network is failing or does not deliver data efficiently as supposed. Network forensics could help with this matter. However, provisioning forensic evidence could impact the network performance in terms of throughput and delay. Secondly, we need to make the network reliable when switching from commonly used bands to another, because conventional network protocols may be affected by the characteristics of that particular frequency. Finally, we need to make sure there is security for the communication relying on the online social networks against active seekers of information.

For the first issue, we examine the problem of monitoring packet level transmissions and provisioning forensic evidence. We provide an analytical framework that computes the likelihood that forensic evidence exists with respects to packet transmissions. We validate our analytical framework via simulations and real-world experiments on two different wireless testbeds. Then we use the analytical framework as a basis for a protocol within a forensic analyzer to determine the likelihood that nodes drop packets. Our assessments are shown to be close to the ground truth, with an average deviation of 2.3%. Furthermore, we quantify the trade-offs between provisioning forensic evidence and achieving high performance in wireless networks. We find that the performance remains unaffected up to a certain evidence availability requirement. Beyond this, the throughput could degrade and the delay could increase by as much as 30%.

Then, this dissertation presents our work in evaluating neighbor discovery in 60 GHz band. Neighbor discovery is a fundamental process in the self-configuration of ad-hoc wireless networks. However, the unique physical layer characteristics of 60 GHz make neighbor discovery different from that in traditional 2.4 and 5 GHz bands. In particular, neighbors can be discovered via both direct and reflected beams. We analyze two neighbor discovery approaches in 60 GHz band. We examine the impact of system parameters on the discovery efficiency.

Finally, this dissertation includes our efforts in doing covert communication on public photo-sharing sites. Steganography can be used to embed secret messages in shared images. However, the image processing performed by photo sharing sites may destroy the embedded messages. We provide an in-depth measurement study of the feasibility of hiding data on four popular photo sharing sites and identify the challenges in communicating covertly on these sites. We propose an approach of embedding secret messages while ensuring their integrity is preserved. We also present an approach for bootstrapping the communication without an out-of-band channel.

In recent years we have witnessed an almost exponential growth in traffic that is transferred over the Internet. This traffic growth is a result of various innovations in network technology. However, this growth has also introduced various problems with regards to performance and security in modern networks. In this thesis we develop various frameworks that address some of those problems. The thesis is divided in two parts, the first deals with security issues and the second deals with performance issues.

In the first part, we analyze how modern day IDS systems are unable to efficiently cope with both the volume of traffic and the amount of information that can be collected from modern day systems. To deal with information overload, we develop and automated framework for feature selection that specifies the optimal set of features that a given system should monitor. This allows the IDS to focus only on what is important. To address the scalability problem in IDS systems, we develop Jaal, a framework that enables intrusion detection at ISP scales. The key idea in Jaal is to monitor traffic and construct in-network packet summaries. The summaries are then processed centrally to detect attacks with high accuracy.

In the second part, we first highlight how battery technology is often the perfor-

mance bottleneck in smartphones. Consequently we develop a mathematical framework that accurately predicts how much energy a given high definition video will consume on a mobile device. The predictive framework empowers user by pre-calculating energy con- sumption and letting the user decide whether a certain video download is worth the energy budget available. Finally, we develop NEST, a novel transport framework for delivering ex- tractive summaries of a dataset distributed across multiple producers over NDN. The goal is to exploit diversity in network conditions between a consumer and different producers towards delivering the consumer-specified summary while minimizing latency.

One of the major problems experienced by wireless multi-hop networks is the intermittent network connectivity, which is a consequence of fluctuating link quality due to signal fading.

Antenna array technology has been proposed to alleviate the problem of signal fading,

and it provides significant performance increase on a single link. However, translating this link-level performance increase to an end-to-end gain in multi-hop networks is not straightforward; a cross-layer design is necessary to efficiently facilitate this translation. In this dissertation, we present cross-layer design approaches for providing end-to-end performance increase in multi-hop networks using antenna arrays. Each approach is designed to utilize a special capability with antenna arrays.

Using antenna arrays, nodes can increase the signal strength in a specific direction;

i.e., perform directional communications. Using directional communications in a multi-hop network requires nodes to periodically update the directions of their neighbors, which introduces an overhead. We propose topology control algorithms that enable the use of directional communications in multi-hop networks with bounded overhead. The bounds provided by

our Low Degree Spanner (LDS) and Distributed LDS (D-LDS) algorithms are near-optimal.

Space-Time Block Coding (STBC) with antenna arrays (referred to as MIMO-STBC) offers significant robustness to fading without an overhead at the higher layers.

Robust MIMO-STBC links can also provide performance improvements at the higher layers

by the design of proper protocols. Such a design necessitates an accurate representation of the MIMO-STBC link behavior. To date, simplistic representations have been used. We design an accurate representation of MIMO-STBC communications, which we show to have a high fidelity to the MIMO-STBC communications in practice.

Antenna arrays also facilitate the spatial multiplexing of signals, allowing a node to transmit and receive multiple signals simultaneously. In a multi-hop network, spatial multiplexing enables receptions from multiple concurrent transmitters. However, such a reception is successful only if

both the number and the strength of concurrent transmissions is controlled by a higher-layer mechanism. We design topology control algorithms for activating a maximal number of communications simultaneously, while ensuring that every communication is successful with high probability.

There are trade-offs between security, reliability and performance. In this dissertation, we consider these aspects of wireless operational networks in different settings. Specifically, with respect to security we look at replay attacks and functional reliability, with respect to reliability and performance we look at video transmission and visible light communications.

Packet replay attack is a type of denial-of-service(DOS) attack, wherein an attacker replays overheard packets in the wireless network. Our experiments indicate that even a single attacker can degrade the route throughput by up to 61%. We design a lightweight detection and prevention system, COPS (for Copycat Online Prevention System), that intelligently uses a combination of digital signatures and Bloom filters to cope with the attack. We implement COPS on real hardware and perform experiments on our 42 node wireless testbed. Our measurements indicate that COPS achieves its objective; it can efficiently contain the effects of replayed packets to a local neighborhood without incurring high resource consumption penalties.

Later, we look into the functional reliability (FR) of the nodes in wireless networks. The FR is typically assessed based on evidence collected by nodes with regards to other nodes in the network. However, such evidence is often affected by factors such as channel induced effects and interference. We design a framework for collaborative assessment of the FR of nodes, with respect to different types of functions; our framework accounts for the above factors that influence evidence collection. We also design a module that drastically reduces the overhead at the expense of slightly increased uncertainty in the assessed FR values. We implement our framework on an indoor/outdoor wireless testbed. We show that with our framework, each node is able to determine the FR for every other node in the network with high accuracy.

Next, we research on the performance of video transmission in wireless networks. The end-user experience in viewing a video depends on the distortion; however, also of importance is the delay experienced by the packets of the video flow since it impacts the timeliness of the information contained and the playback rate at the receiver. Unfortunately, these performance metrics are in conflict with each other in a wireless network. We investigate this trade-off between distortion and delay for video. We validate our analysis via extensive simulations. Surprisingly, we find that the trade-off depends on the specific features of the video flow: it is better to trade-off high delay for low distortion with fast motion video, but not with slow motion video. Our simulation results further quantify the trade-offs in various scenarios.

Finally, we look into a visible light system in two rooms with a door open in between. Two emitters tx1 and tx2 are located in room 1 and room 2 respectively. We use BPPM, vary pulse width within the slot to provide different dimming levels. We propose a modfied ray-tracing algorithm to calculate the channel impulse response. We also provide a BER analysis considering different combinations of system parameters. Our results show that if tx2 is just illuminating, it does not impact the performance of communications in room 1. However, if both tx1 and tx2 are transmitting, the performance in room 1 is degraded to different levels depending on the position of the receivers. To improve the performance of communications in room 1 in this case, we can increase the dimming level of tx1. Moreover, when dimming level is limited, our results show that reducing the bit rate of tx1 improves the performance in room 1 dramatically. For example, when the bit rate of tx2 is 8Mb/s, reducing the bit rate of tx1 from 8Mb/s to 4Mb/s makes the BER drop from 10^-3 to 10^-13 .

Even though wireless and mobile devices have evolved with richer capabilities, resources in such devices are still limited, and thus in many cases, are insufficient to accomplish the tasks entrusted in them. The goal of my work is to improve the efficiencies of applications on wireless and mobile devices, focusing on resource constrained settings.

First, we develop TIDE, a user-centric framework that helps to identify high energy consuming applications on users’ smartphones. TIDE identifies energy hungry applications by looking at the correlation between applications’ activities and high drainage periods on the smartphones. Experiments on Android smartphones show that TIDE is able to identify correctly ≈ 90% of the energy hungry applications, while imposing reasonably low energy overheads.

Subsequently, we develop a framework to identify redundant images uploaded from multiple wireless devices in bandwidth constrained networks, e.g., the destructed networks at natural disaster scenes. Our framework intelligently combines state-of-the-art vision techniques to identify redundant images uploaded to a server. Suppressing the transfer

of redundant contents significantly lowers network load, so that the delay in transferring unique and important contents in such critical scenarios is reduced up to ≈ 44%.

We then design ACTION, a framework for accurate and timely object (e.g., human) detection in bandwidth constrained settings. In ACTION, the objects of interest are effectively detected at individual camera sensors. Metadata of detected objects is then aggregated at a designated fusion node to improve the detection accuracy. Most accurate information of each detected object is then chosen to upload to a central controller, while adhering to the bandwidth constraints. We show that ACTION helps reduce up to three folds the amount of transferred data, while still delivering important information to the central node.

Finally, we design EECS, a framework for adaptive detection algorithm selection in multi-camera settings. In EECS, only a subset of camera sensors is chosen to detect objects; further, the most energy efficient algorithm is assigned to each camera to reduce energy consumption, while still ensuring a desired accuracy. We show that EECS can be tuned to achieve the right trade-offs between energy efficiency and desired accuracy.

This dissertation has two main objectives -- solving power and latency issues in mobile augmented reality. For power, we showcase the power drain due to the two heaviest components -- simultaneous localization and mapping (SLAM) and deep convolutional neural networks (DNNs) and design solutions to reduce the power consumption on mobile devices. Our single-user solution is to use DNNs as needed, to detect new objects or recapture objects that significantly change in appearance, and otherwise depend on low-power object tracking. For multi-user solutions, we use peer-to-peer communications to exchange key information among devices, and finally assign roles to each of them -- primary or secondary. A primary device continuously tracks target objects and shares their information to slaves. Secondary devices do not need SLAM or DNN but leverage the shared information from the master and other lightweight methods to keep track of the objects with high precision, and thus significantly reduce power consumption. In addition, we can rotate the master functionality across participants in order to distribute energy expenditures among them and increase the longevity of the AR experience. For latency, we perform a first-of-its-kind measurement study on both public LTE and industry LTE testbed for two popular multi-user AR applications, yielding several insights such as: (1) The radio access network (RAN) accounts for a significant fraction of the end-to-end latency (31.2\%, or 3.9 s median); (2) AR network traffic is characterized by large intermittent spikes on a single uplink TCP connection, resulting in frequent TCP slow starts that can increase user-perceived latency; (3) Applying a common traffic management mechanism of cellular operators, QoS Class Identifiers (QCI), can help by reducing AR latency by 33\% but impacts non-AR users. Based on these insights, we propose AR solutions to intelligently adapt IP packet sizes and periodically provide information on uplink data availability, respectively. Our solutions help ramp up network performance, improving the end-to-end AR latency and goodput by ~40-70\%.