As governments in developing countries race to solve the global learning crisis, a key focus is on novel teaching approaches as taught in pedagogical programs. To scale, these pedagogical programs rely on government teacher training infrastructure. However, these programs face challenges in rural parts of Africa where there is a lack of advisor support, teachers are isolated and technology infrastructure is still emerging. Conversational agents have addressed some of these challenges by scaling expert knowledge and providing personalized interactions, but it is unclear how this work can translate to rural African contexts. To explore the use of such technology in this design space, we conducted two related studies. The first was a qualitative study with 20 teachers and ministry officials in rural Côte d’Ivoire to understand opportunities and challenges in technology use for these stakeholders. Second, we shared a conversational agent probe over WhatsApp to 38 teachers for 14-weeks to better understand what we learned in the survey and to uncover realistic use cases from these stakeholders. Our findings were examined through a theoretical lens of aspirations to discover sustainable design directions for conversational agents to support teachers in low infrastructure settings.

Teaching is challenging in rural areas in developing countries because of large classroom sizes, limited basic infrastructure, and lack of teacher training. Teachers' impact on educational outcomes can be improved by teacher training programs but these programs are limited by poor infrastructure in rural areas. Although infrastructural challenges can be overcome by information technology today, it is unclear how such technology can support teacher training. Therefore to explore opportunities for teacher training in rural developing contexts with technology, we conducted interviews and observations with 22 primary school teachers in two regions of rural Côte d'Ivoire. We followed an aspirations-based approach that focuses on the user's aspirations (long term desires) to help understand teachers' intrinsic desire to grow in their professional life. We found that teacher aspirations conflict with their current role but their solidarity helps them find education role models who balance this conflict. We found that teachers face challenges in teacher training but they handle issues through their solidarity and by creating solutions. We then discuss ways to measure impact on aspirations, designing for conflicting aspirations and to leverage aspirations to create a social change in teacher training using technology.

Social media messaging applications(i.e. WhatsApp, Facebook) have reached 2.3 billion users in 2019, with the majority of users emerging from developing countries. The high usage among emergent users opens the possibility of designing text-based interventions for social change but such interventions rely on experts (i.e. doctors, educators, and moderators) knowledge which is scarce in developing contexts. Expert knowledge can be scaled up using chatbots but more research is needed to support emergent users who need context-specific support such as local language interventions or may not have regular internet connectivity. Therefore to support the design of chatbot based interventions in low resource contexts, we built DIA a chatbot architecture for low resource contexts to scale expert knowledge and support localization. DIA is a human-chatbot (humbot) hybrid system that organically learns topic-specific knowledge and local language from user interactions. We built a preliminary version of DIA on WhatsApp and deployed it to mentor 38 teachers in a rural context of Côte d'Ivoire. Through our preliminary deployment, we show that DIA can help (1) build a data-set of a topic and language-specific dialogues (2) understand users' online smartphone usage through chat logs and (3) collect survey data for through conversational interaction.

We propose a multi-objective optimization algorithm to allocate tasks and plan paths for a team of UAVs. The UAVs must find a target in a bounded area and then continuously communicate the target information to the ground personnel. Our genetic algorithm approach aims to minimize the mission completion time, which includes the time to find the target (area coverage) and the time to setup a communication path (network connectivity). We evaluate strategies using a data mule, a relay chain, and a novel hybrid approach to communicate with the ground personnel. The algorithm can be tuned to prioritize coverage or connectivity, depending on the mission demands. Simulation results show reduced overall mission completion times (up to 65%), with more improvement as the UAV density increases.

In this paper, we address the impact of fading on downlink performance of wireless cellular systems. In a wireless cellular system, the mobile station selects the base station based on one set of measurements that are impacted by fading. This fading can differ at a later time when the mobile station uses the base station to communicate. Further, a base station's power distribution can depend on whether or not it is selected as a receiver. This paper formalizes these concepts in selection and evaluation stages. Selection has one set of fading factors and possibly a selection bias. Evaluation has different fading factors and distinguishes between the selected base station and other interfering base stations in the signal to interference plus noise ratio. These are studied in the context of the probability a mobile station achieves coverage in a multi-tier network using techniques from stochastic geometry. Several applications of the result are given to show the utility of the results.

In this paper, a comprehensive study of the downlink performance in a heterogeneous cellular network (or HetNet) is conducted via stochastic geometry. A general HetNet model is considered consisting of an arbitrary number of open-access and closed-access tiers of base stations (BSs) arranged according to independent homogeneous Poisson point processes. The BSs within each tier have a constant transmission power, random fading factors with an arbitrary distribution and arbitrary path-loss exponent of the power-law path-loss model. For such a system, analytical characterizations for the coverage probability are derived for the max-SINR connectivity and nearest-BS connectivity models. Using stochastic ordering, interesting properties and simplifications for the HetNet downlink performance are derived by relating these two connectivity models to the maximum instantaneous received power (MIRP) connectivity model and the maximum biased received power (MBRP) connectivity models, providing good insights about HetNets and their downlink performance in these complex networks. Furthermore, the results also demonstrate the effectiveness and analytical tractability of the stochastic geometric approach to study the HetNet performance.

A key challenge in planning is how to schedule under uncertainty. We consider scheduling communications to dispersed wireless clients sharing a hub over a period of time in a dynamic environment. Scheduling under uncertainty research has produced either complex or unsatisfactory solutions for this problem. This paper develops a simple receding horizon linear program to allocate times for client specific traffic in order to realistically maximize client throughput. Given future uncertainty it considers clients' (estimated) future rates and loads and aggregates them utilizing various permutations. For the scenario we present, the simplest receding horizon approach using only two time intervals can perform within 3.4 percentage points of the optimal whereas a greedy approach can only achieve within 13 percentage points of the optimal. The aggregation over many periods minimizes the variations due to uncertainty and greatly simplifies the problem while providing good performance. This result can be extended to many other scheduling with uncertainty type scenarios.

In this paper, a comprehensive study of the downlink performance in a heterogeneous cellular network (or hetnet) is conducted. A general hetnet model is considered consisting of an arbitrary number of open-access and closed-access tier of base stations (BSs) arranged according to independent homogeneous Poisson point processes. The BSs of each tier have a constant transmission power, random fading coefficient with an arbitrary distribution and arbitrary path-loss exponent of the power-law path-loss model. For such a system, analytical characterizations for the coverage probability and average rate at an arbitrary mobile-station (MS), and average per-tier load are derived for the max-SINR connectivity model. Using stochastic ordering, interesting properties and simplifications for the hetnet downlink performance are derived by relating this connectivity model to the maximum instantaneous received power (MIRP) connectivity model, providing good insights about the hetnets and the downlink performance in these complex networks. Furthermore, the results also demonstrate the effectiveness and analytical tractability of the stochastic geometric approach to study the hetnet performance.

In a cognitive radio (CR) network, the CR devices opportunistically communicate in the frequency bands occupied by the primary users in order to improve the spectral efficiency in these bands. By sensing the primary users, e.g., the television transmitter-receiver pairs and the wireless microphone systems, each CR device determines whether or not to operate in the band. The CR devices, due to erroneous sensing, either fail to detect the primary user, causing excessive interference at the primary users, or have a false-alarm, causing it to remain silent in a white-space band leading to poor spectrum utility. The impact of these imperfections on the primary user operations in terms of the coverage probability is characterized where the primary users and the CR devices are distributed according to independent homogeneous Poisson point processes on the plane.

In this paper, we analyze the signal-to-interference-plus-noise ratio (SINR) performance at a mobile station (MS) in a random cellular network. The cellular network is formed by base stations (BSs) placed in a one-, two-, or three-dimensional space according to a possibly non-homogeneous Poisson point process, which is a generalization of the so-called shotgun cellular system. We develop a sequence of equivalence relations for the SCSs and use them to derive semi-analytical expressions for the coverage probability at the MS when the transmissions from each BS may be affected by random fading with arbitrary distributions as well as attenuation following arbitrary path-loss models. For homogeneous Poisson point processes in the interference-limited case with power-law path-loss model, we show that the SINR distribution is the same for all fading distributions and is not a function of the base station density. In addition, the influence of random transmission power, power control, and multiple channel reuse groups on the downlink performance is also discussed. The techniques developed for the analysis of SINR have applications beyond cellular networks and can be used in similar studies for cognitive radio networks, femtocell networks, and other heterogeneous and multi-tier networks.

Mobile phones are becoming a new platform to offer health services throughout the world, particularly in developing countries where the health sector experiences serious shortages. Diabetes is a metabolic disorder that refers to the condition created by the body's inability to regulate glucose levels. It has been labeled the "silent epidemic" for its insidious and chronic nature. The medical profession has postulated that patients could benefit from a system providing continuous glucose readings and that the predicted explosion of diabetes in the developing parts of the world could be averted using mobile health solutions. In remote areas where healthcare workers are scarce and patients have high travel cost it might be the only way to monitor the condition. We define a model that evaluates a mobile health solution to monitor diabetes and explore the influence of two factors on the model: the discount rate and the cost of a telecom service. We test the model on data from three countries with a considerable health-care burden associated with the disease: the USA, Brazil and India.

In this paper, we consider the downlink signal-to-interference-plus-noise ratio (SINR) analysis in a heterogeneous cellular network with K tiers. Each tier is characterized by a base-station (BS) arrangement according to a homogeneous Poisson point process with certain BS density, transmission power, random shadow fading factors with arbitrary distribution, arbitrary path-loss exponent and a certain bias towards admitting the mobile-station (MS). The MS associates with the BS that has the maximum instantaneous biased received power under the open access cell association scheme. For such a general setting, we provide an analytical characterization of the coverage probability at the MS.

Given multiple widespread stationary data sources such as ground-based sensors, an unmanned aircraft can fly over the sensors and gather the data via a wireless link. When sensors have limited energy resources, network lifetime can be extended by reducing the power that the sensors use for communication with the aircraft. Complex vehicle and communication dynamics and imperfect knowledge of the environment make accurate system models difficult to acquire and maintain, so we present a reinforcement learning approach that allows the data-ferrying aircraft to optimize data collection trajectories and sensor power use in situ, obviating the need for system identification. By allowing the ferry aircraft to fly longer trajectories, we learn energy-conserving radio transmission policies that are significantly better than hand-coded heuristics. Furthermore, we introduce a meta-level reinforcement learner that makes energy policy learning faster and more robust, and that transfers knowledge acquired in earlier tasks to new data-ferrying tasks.

This paper analyzes a scenario where a cognitive radio (CR) moves through a field of primary users (PU). As the CR moves, it goes through periods when a given channel is available and when the channel is unavailable (i.e. an outage) due to the CR avoiding a primary user. In addition, there are periods when the channel is deemed available but, in fact, CR communication could interfere with the primary user. The paper derives the distribution of these different periods under different detection and interference scenarios. The results provide insights into how useful a channel is to a CR and the impact of the CR on the PU. The model applies to a scenario of unmanned aircraft access to TV whitespace.

This paper studies the carrier-to-interference ratio (CIR) and carrier-to-interference-plus-noise ratio (CINR) performance at the mobile station (MS) within a multi-tier network composed of M tiers of wireless networks, with each tier modeled as the homogeneous n-dimensional (n-D, n=1,2, and 3) shotgun cellular system, where the base station (BS) distribution is given by the homogeneous Poisson point process in n-D. The CIR and CINR at the MS in a single tier network are thoroughly analyzed to simplify the analysis of the multi-tier network. For the multi-tier network with given system parameters, the following are the main results of this paper: (1) semi-analytical expressions for the tail probabilities of CIR and CINR; (2) a closed form expression for the tail probability of CIR in the range [1,infinity); (3) a closed form expression for the tail probability of an approximation to CINR in the entire range [0,infinity); (4) a lookup table based approach for obtaining the tail probability of CINR, and (5) the study of the effect of shadow fading and BSs with ideal sectorized antennas on the CIR and CINR. Based on these results, it is shown that, in a practical cellular system, the installation of additional wireless networks (microcells, picocells and femtocells) with low power BSs over the already existing macrocell network will always improve the CINR performance at the MS.

Given multiple widespread stationary data sources such as ground-based sensors, an unmanned aircraft can fly over the sensors and gather the data via a wireless link. To minimize delays and system resources, the aircraft should collect the data at each sensor node via the shortest trajectory. Trajectory planning is hampered by the complex vehicle and communication dynamics and by uncertainty in the locations of sensors, so we develop a technique based on model-free learning. Previous work showed that model-free stochastic optimization can find good trajectories quickly enough for use in the field, but scaled poorly as the number of sensors increased, requiring roughly O(n) flights for n sensors. Here we modify the gradient computation, combining the global optimization criterion with multiple overlapping local ones, introduce data-mule--specific credit assignment, and use observed behavior to redistribute global rewards to local regions in the trajectory. This improves scalability of the initial trajectory learning phase nearly to O(1). We target a scenario in which sensors are known to lie somewhere near a known trajectory, for example after having been parachuted out of a deployment aircraft.

Cognitive radios (CR) must ensure that their transmissions do not cause harmful interference to the primary users that operate in the same band. Primary users with short communication range transmitters, such as wireless microphone systems, are considered. The CR network is distributed according to a Poisson process. The CRs detect the primary user either through primary transmitter signals or through beacons collocated with the primary receiver. Three levels of cooperative sensing by CRs are analyzed, namely, no cooperation, full cooperation and partial cooperation. The characteristic function of the interference at the primary receiver for all combinations of primary user detection and cooperative sensing schemes are derived for the system modeled along a line (1-D) and about a plane (2-D). The analysis in 1-D corresponds to a highway scenario and 2-D corresponds to a planar system. The carrier-to-interference ratio (C/I) at the primary receiver is evaluated using the characteristic function. The analysis clearly shows the need for cooperative spectrum sensing when the density of the CRs is high. It is shown that the effect of shadowing, small scale fading, and power control at the CRs can be captured into the density of the CR network. The results hold for the CRs that are wide-band devices as well as narrow-band devices.

Wireless LANs suffer from performance problems caused by insufficient medium access opportunity given to the access point. Consequently, the downlink buffer fills up, which often leads to packet losses. We propose to address this problem by using a size-based scheduling approach, which is known to favor short flows and the start up of new ones-a very appealing property from the user's perspective as interactive applications and new flows are serviced quickly. Still, size-based scheduling policies have a well-known Achilles heel: large flows can block each other for long periods of time and low rate multimedia transfers may end up with a low priority when their accumulated transferred volume becomes large. To solve the above deficiencies, we propose a new packet scheduling scheme called Least Attained Recent Service (LARS) that applies a temporal decay to the volume of data associated with each flow. In this way, its priority depends more on what has happened recently. With this strategy, LARS can bound the impact of a new arriving flow on ongoing flows, thus limiting lock out durations. It can also efficiently protect low rate multimedia transfers irrespectively of the load conditions.

This paper analyzes the carrier-to-interference ratio of the so-called shotgun cellular system (SCS). In the SCS, base-stations are placed randomly according to a two-dimensional Poisson point process. Such a system can model a dense cellular or wireless data network deployment, where the base station locations end up being close to random due to constraints other than optimal coverage. The SCS is a simple cellular system where we can introduce several variations and design scenarios such as shadow fading, power control features, and multiple channel reuse groups, and assess their impact on the performance. We first derive an analytical expression for the characteristic function of the inverse of the carrier-to-interference ratio. Using this result, we show that the carrier-to-interference ratio is independent of the base station density and further, we derive a semi-analytical expression for the tail-probability. These results enable a complete characterization of the cellular performance of the SCS. Next, we incorporate shadow fading into the SCS and demonstrate that it merely scales the base station density by a constant. Hence, the cellular performance of the SCS is independent of shadow fading. These results are further used to analyze dense cellular scenarios.

Mobility in ad hoc networks causes routes to break disrupting traffic flows and triggering additional routing overhead. We address this problem with a novel hybrid routing protocol, Aisle Routing. Aisle Routing introduces a route refresh mechanism to reactive source routing which proactively and adaptively refreshes routes through an aisle-like area and is designed so the route-control overhead is relatively small. This paper presents the Aisle Routing mechanism and describes how it can be integrated with existing reactive source routing protocols. Simulation results demonstrate that Aisle Routing has high packet delivery ratio and low delay in various mobility scenarios.

Communication in delay tolerant networks can be facilitated by the use of dedicated mobile ldquoferriesrdquo which physically transport data packets between network nodes. The goal is for the ferry to autonomously find routes which minimize the average packet delay in the network. We prove that paths which visit all nodes in a round-trip fashion, i.e., solutions to the traveling salesman problem, do not yield the lowest average packet delay. We propose two novel ferry path planning algorithms based on stochastic modeling and machine learning. We model the path planning task as a Markov decision process with the ferry acting as an independent agent. We apply reinforcement learning to enable the ferry to make optimal decisions. Simulation experiments show the resulting routes have lower average packet delay than solutions known to date.

This paper explores the role of meshed airborne communication networks in the operational performance of small unmanned aircraft systems. Small unmanned aircraft systems have the potential to create new applications and markets in civil domains, enable many disruptive technologies, and put considerable stress on air traffic control systems. We argue that of the existing networked communication architectures, only meshed ad hoc networking can meet the communication demands for the large number of small aircraft expected to be deployed in future. Experimental results using the heterogeneous unmanned aircraft system are presented to show that meshed airborne communication is feasible, that it extends the operational envelope of small unmanned aircraft at the expense of increased communication variability, and that net-centric operation of multiple cooperating aircraft is possible. Additionally, the ability of airborne networks of small unmanned aircraft to exploit controlled mobility to improve performance is discussed.

Multihop packet relaying has been shown to extend the achievable communication space in sparsely connected ad hoc networks, in sensor networks, and over long distance links. This paper investigates the properties of relay-enabled networks as a function of the number of relays and whether packets are sent one at a time or whether multiple packets are simultaneously sent along the relay chain. When packets are sent one at a time, an optimal number of relays is found that maximizes the throughput and minimizes delay. The optimal number is in terms of a standard hop, which enables the number of relays,maximum throughput, and minimum delay to be determined for any communication within the environment. When multiple packets are sent at a time, a maximum throughput is found that is independent of distance and noise. The performance depends on the separation in hops between simultaneous transmissions in the relay chain. The optimal separation is found in the limit of many relays to be 5 hops across many different radio environments. These results allow us to define feasible delay-throughput-distance performance regions and provide inputs to better multi-hop wireless network design.

Widely distributed sensors must discover paths back to data collection points possibly through sparsely connected and mobile networks. Current addressing and service discovery schemes in mobile networks are not well-suited to multihop disconnected networks. This paper describes an architecture and protocol for sensor data collection through highly mobile ad-hoc network (MANET) that may never experience end-to-end connectivity. Special gateway nodes are described which are responsible for intelligently routing messages to their intended destination(s). These gateway nodes qualify their links and announce their status to the MANET, a simple approach to service discovery that is effective in this implementation. The protocol is implemented and tested in a laboratory and outdoor environment.

Public safety and commercial communications have distinct performance objectives. The former requires a high availability (i.e., low blocking), both in normal times and during a disaster, whereas the latter requires a high network revenue performance in normal times. Therefore, public safety and commercial communications traditionally use separate networks, which results in either spectrum inefficiency or unsatisfactory performance. In this paper, we propose a public safety and commercial spectrum sharing strategy via intelligent network pricing and call admission control. We provide a precise characterization of the performance objectives of such a sharing strategy, which is to maximize the commercial revenue while guaranteeing a low blocking probability to public safety calls. We demonstrate through analysis and simulation that our sharing strategy simultaneously satisfies the objectives of both types of communications

This paper considers the problem of an attacker disrupting an encrypted victim wireless ad hoc network through jamming. Jamming is broken down into layers and this paper focuses on jamming at the Transport/Network layer. Jamming at this layer exploits AODV and TCP protocols and is shown to be very effective in simulated and real networks when it can sense victim packet types, but the encryption is assumed to mask the entire header and contents of the packet so that only packet size, timing, and sequence is available to the attacker for sensing. A sensor is developed and tested on live data. The classification is found to be highly reliable for many packet types. The relative roles of size, timing, and sequence are discussed along with the implications for making networks more secure.

In this paper, we compare the dynamic source routing (DSR) protocol and our energy aware dynamic source routing protocol (EADSR). The implementation of the routing protocols is carried out using the click modular router infrastructure on laptops with wireless Ethernet cards running Linux. We demonstrate the working of both ad hoc routing protocols through our experiments and highlight the energy efficient behavior of EADSR as compared to DSR.

In this paper we demonstrate a wireless 802.11 testbed. This testbed allows for direct comparison between mobile wireless routing protocols without affecting their routing or MAC layer protocols or inter-layer interaction. We have built a low cost environment to facilitate such wireless network research. The core idea is "compressing" the network and emulating mobility without actually moving the nodes. We successfully emulate the RF effects along with mobility. The viability of this test bed is checked with the help of a click implementation of the dynamic source routing protocol. Our experiences show that such a testbed provides valuable feedback not available through simulation.

This paper proposes a new power aware routing objective for an ad hoc network of battery-limited wireless nodes---the maximum flow-life curve --- that maximizes the traffic flow utility over time. The objective improves upon related objectives such as minimizing the total power or maximizing the time to network partition. To find a routing that maximizes the flow-life curve, we prove an equivalence with a simpler problem and present an algorithm based on linear programming. The efficiency and fairness of the objective are demonstrated on several examples

In packet switches, packets queue at switch inputs and contend for out- puts. The contention arbitration policy directly affects switch perfor- mance. The best policy depends on the current state of the switch and current traffic patterns. This problem is hard because the state space, possible transitions, and set of actions all grow exponentially with the size of the switch. We present a reinforcement learning formulation of the problem that decomposes the value function into many small inde- pendent value functions and enables an efficient action selection.

We classify an input space according to the outputs of a real-valued function. The function is not given, but rather examples of the function. We contribute a consistent classifier that avoids the un(cid:173) necessary complexity of estimating the function.

This paper considers two-dimensional interference-limited cellular radio systems. It introduces the shotgun cellular system that places base stations randomly and assigns channels randomly. Such systems are shown to provide lower bounds to cellular performance that are easy to compute, independent of shadow fading, and apply to a number of design scenarios. Traditional hexagonal systems provide an upper performance bound. The difference between upper and lower bounds is small under operating conditions typical in modern TDMA and CDMA cellular systems. Furthermore, in the strong shadow fading limit, the bounds converge. To give insights into the design of practical systems, several variations are explored including mobile access methods, sectorizing, channel assignments, and placement with deviations. Together these results indicate cellular performance is very robust and little is lost in making rapid minimally planned deployments.

We solve the adaptive call admission control (CAC) problem in multimedia networks via reinforcement learning (RL). The problem requires that network revenue be maximized while simultaneously meeting quality of service (QoS) constraints that forbid entry into certain states and use of certain actions. We show that RL provides a solution to this constrained semi-Markov decision problem and is able to earn significantly higher revenues than alternative heuristics. Unlike other model-based algorithms, RL does not require the explicit state transition models to solve the decision problems. This feature is very important if one considers large integrated service networks supporting a number of different service types, where the number of states is so large that model-based optimization algorithms are infeasible. Both packet-level and call-level QoS constraints are addressed, and both conservative and aggressive approaches to the QoS constraints are considered. Results are demonstrated on a single link and extended to routing on a multilink network.

Tasks such as admission control in ATM and predicting overload conditions in telephone networks require a function that specifies what conditions will result in loss rates exceeding a threshold, p/sup */. This paper considers the formal task of deriving such a classification function based on samples at different conditions. When the size of these samples is small relative to 1/p/sup */, previously proposed methods incorrectly classify conditions that surpass the threshold by orders of magnitude. This paper derives general conditions for consistent and robust classifiers and presents specific methods that meet these conditions. The paper analyzes the methods with respect to asymptotic and finite sample behaviour and the results are confirmed using simulated data.

This paper contributes a distributed packet controller which reduces queueing to a single stage in two-stage packet switches. Software and neural network based controllers are described. Simulations under a range of traffic conditions for a 1024/spl times/1024 switch size shows the simplest architecture has the best performance.

The algorithm for choosing nonblocking sets of data cells from the queues can significantly affect the throughput and queuing behavior. The authors present an algorithm that is shown to have maximum throughput. This algorithm is reduced on a banyan network to a constraint satisfaction problem by using an equivalence approach. To gain the required computational speed, the massive parallelism of neural networks is used. A neural network design using multiple overlapping winner-take-all circuits is defined. This is shown to be stable and to result only in nonblocking sets of data cells. An efficient interface between the neural network and the queue is also defined. The performance of the banyan with a neural network controller is compared to a noninternal-blocking switch with various controllers. The banyan is within a factor of two of the nonblocking switch.>