Cloud radio access networks (CRANs) have become a key enabling technique for the next generation wireless communications. Resource allocation in CRANs still needs to be further improved to reach the objective of minimizing power consumption and meeting demands of wireless users over a long period. Inspired by the success of Deep Reinforcement Learning (DRL) on solving complicated control problems, we present a novel framework,ReCARL, for power-efficient resource allocation in CRANs with deep reinforcement learning. Specifically, we define the state space, action space and reward function for the DRL agent, apply a deep neural network (DNN) to approximating the action-value function, and formally formulate the resource allocation problem (in each decision epoch) as a convex optimization problem. Under ReCARL, we propose two different DRL agents: one has a regular DNN structure trained with the basic deep Q-learning method (ReCARL-Basic); while the other has a context-aware DNN structure trained with a hybrid deep Q-learning method (ReCARL-Hybrid). We evaluated the performance of ReCARL along with the two DRL agents by comparing them with two widely-used baselines via extensive simulation. The simulation results show that ReCARL achieves significant power savings while meeting user demands, and it can well handle highly dynamic cases.

Cloud Radio Access Networks (RANs) have become a key enabling technique for the next generation (5G) wireless communications, which can meet requirements of massively growing wireless data traffic. However, resource allocation in cloud RANs still needs to be further improved in order to reach the objective of minimizing power consumption and meeting demands of wireless users over a long operational period. Inspired by the success of Deep Reinforcement Learning (DRL) on solving complicated control problems, we present a novel DRL-based framework for power-efficient resource allocation in cloud RANs. Specifically, we define the state space, action space and reward function for the DRL agent, apply a Deep Neural Network (DNN) to approximate the action-value function, and formally formulate the resource allocation problem (in each decision epoch) as a convex optimization problem. We evaluate the performance of the proposed framework by comparing it with two widely-used baselines via simulation. The simulation results show it can achieve significant power savings while meeting user demands, and it can well handle highly dynamic cases.

In this paper, we propose to leverage the emerging deep learning techniques for spatiotemporal modeling and prediction in cellular networks, based on big system data. First, we perform a preliminary analysis for a big dataset from China Mobile, and use traffic load as an example to show non-zero temporal autocorrelation and non-zero spatial correlation among neighboring Base Stations (BSs), which motivate us to discover both temporal and spatial dependencies in our study. Then we present a hybrid deep learning model for spatiotemporal prediction, which includes a novel autoencoder-based deep model for spatial modeling and Long Short-Term Memory units (LSTMs) for temporal modeling. The autoencoder-based model consists of a Global Stacked AutoEncoder (GSAE) and multiple Local SAEs (LSAEs), which can offer good representations for input data, reduced model size, and support for parallel and application-aware training. Moreover, we present a new algorithm for training the proposed spatial model. We conducted extensive experiments to evaluate the performance of the proposed model using the China Mobile dataset. The results show that the proposed deep model significantly improves prediction accuracy compared to two commonly used baseline methods, ARIMA and SVR. We also present some results to justify effectiveness of the autoencoder-based spatial model.

We envision that in the near future, just as Infrastructure-as-a-Service, radios, and radio resources in a wireless network can also be provisioned as a service to mobile virtual network operators (MVNOs), which we refer to as Radio-as-a-Service (RaaS). A major obstacle for wide adoption of RaaS is the lack of incentives and fairness for allocating radio resources among MVNOs. In this paper, we present a novel auction-based model to enable fair pricing and fair resource allocation according to real-time needs of MVNOs for RaaS. Based on the proposed model, we study the auction mechanism design with the objective of maximizing social welfare. First, we present an integer linear programming and Vickrey-Clarke-Groves-based auction mechanism for obtaining optimal social welfare. To reduce time complexity, we present a polynomial-time greedy mechanism for the RaaS auction. Both methods have been formally shown to be truthful and individually rational. Extensive simulation results show that the proposed greedy auction mechanism can quickly produce close-to-optimal solutions. Furthermore, to prevent winning bidders from making 0 payment, we introduce reserve prices, and present auction mechanisms with reserve prices, which are shown to be truthful and individually rational too.

In this paper, we propose to learn Lifestyles of mobile users via mobile Phone Sensing (LIPS), and we develop a system and algorithms to realize this idea. First, we present the workflow and architecture of our system, LIPS. Combining both unsupervised and supervised learning, we propose a hybrid scheme for lifestyle learning, which consists of two parts: characterization and prediction. Specifically, we present a two-stage algorithm to characterize the lifestyle of a mobile user using Places of Interest (PoIs), which leverages two different algorithms for coarse-grained and fine-grained clustering in two stages respectively. Based on discovered PoIs, we present a method to build a model to predict his/her future activities using a supervised classification algorithm. In addition, we present an adaptive sampling algorithm for improving energy efficiency, which leverages both the discovered PoIs and the lifestyle model for adaptively controlling the sampling rate. We implemented the proposed system and algorithms based on the Android platform. We have validated and evaluated LIPS via extensive field tests carried out for over 1.5 months in 6 cities of USA. The experimental results show that LIPS can 1) well discover PoIs of mobile users, 2) precisely predict their future activities, and 3) achieve significant energy savings (compared to periodic sampling).

Limited research efforts have been made for Mobile CrowdSensing (MCS) to address quality of the recruited crowd, i.e., quality of services/data each individual mobile user and the whole crowd are potentially capable of providing, which is the main focus of the paper. Moreover, to improve flexibility and effectiveness, we consider fine-grained MCS, in which each sensing task is divided into multiple subtasks and a mobile user may make contributions to multiple subtasks. In this paper, we first introduce mathematical models for characterizing the quality of a recruited crowd for different sensing applications. Based on these models, we present a novel auction formulation for quality-aware and fine-grained MCS, which minimizes the expected expenditure subject to the quality requirement of each subtask. Then we discuss how to achieve the optimal expected expenditure, and present a practical incentive mechanism to solve the auction problem, which is shown to have the desirable properties of truthfulness, individual rationality and computational efficiency. We conducted trace-driven simulation using the mobility dataset of San Francisco taxies. Extensive simulation results show the proposed incentive mechanism achieves noticeable expenditure savings compared to two well-designed baseline methods, and moreover, it produces close-to-optimal solutions.

In a mobile crowd sensing system, a smartphone undertakes many different sensing tasks that demand data from various sensors. In this paper, we consider the problem of scheduling different sensing tasks assigned to a smartphone with the objective of minimizing sensing energy consumption while ensuring Quality of SenSing (QoSS). First, we consider a simple case in which each sensing task only requests data from a single sensor. We formally define the corresponding problem as the Minimum Energy Single-sensor task Scheduling (MESS) problem and present a polynomial-time optimal algorithm to solve it. Furthermore, we address a more general case in which some sensing tasks request multiple sensors to report their measurements simultaneously. We present an Integer Linear Programming (ILP) formulation as well as an effective polynomial-time heuristic algorithm, for the corresponding Minimum Energy Multi-sensor task Scheduling (MEMS) problem. Extensive simulation results show that the proposed algorithms achieve over 79% energy savings on average compared to a widely-used baseline approach, and moreover, the proposed heuristic algorithm produces close-to-optimal solutions.

We envision that in the near future, just as Infrastructure-as-a-Service (IaaS), radios and radio resources in a wireless network can also be provisioned as a service to Mobile Virtual Network Operators (MVNOs), which we refer to as Radio-as-a-Service (RaaS). In this paper, we present a novel auction-based model to enable fair pricing and fair resource allocation according to real-time needs of MVNOs for RaaS. Based on the proposed model, we study the auction mechanism design with the objective of maximizing social welfare. First, we present an Integer Linear Programming (ILP) based auction mechanism for obtaining optimal social welfare. To reduce time complexity, we present a polynomial-time greedy mechanism for the RaaS auction. Both methods have been formally shown to be truthful and individually rational. Extensive simulation results show that the proposed greedy auction mechanism can quickly produce close-to-optimal solutions.

Currently, a few online review and recommendation systems (such as Yelp and Trip Advisor) have attracted millions of users and are gaining increasing popularity. They usually rate and rank places and attractions based on subjective ratings provided by users. In this paper, we present design, implementation and evaluation of a mobile phone Sensing based Objective Ranking (SOR) system, which ranks a target place based on data collected via mobile phone sensing. Our system has the following desirable features: 1) it is easy to use, 2) its architecture is so scalable that various embedded and external sensors can be easily integrated into it, 3) an online scheduling algorithm is proposed and used to schedule sensing activities for coverage maximization, which has a constant approximation ratio of 1/2, 4) a personalizable ranking algorithm is developed and used to rank target places based on various sensor readings and user preferences. We validate and evaluate SOR via both field tests (using real hiking trails and coffee shops in Syracuse, NY as target places) and simulation. The field-testing results show that data collected and processed by SOR can well capture characteristics of target places, and personalizable rankings produced by SOR can well match user preferences. In addition, simulation results well justify effectiveness of the proposed scheduling algorithm.