Information-centric networking (ICN) design is motivated to be developed in the future Internet for improved delivery efficiency, content scalability and availability. ICN architectures are based on named content, which is radically different from the traditional host-centric paradigm based on named hosts. With the deployment of in-network storage for caching in the access points (AP), it is efficient to offload the tremendous increasing amount of content in this new ICN architecture. To allocate the resources efficiently, the content provider (CP) is able to cache the popular data objects with the cooperation of the cache-enabled APs, by offering appreciable economic incentives. In the past ICN research, Zipf discrete distribution is normally be utilized to represent the content popularity in Internet. The universal Zipf distribution, however, cannot perfectly capture the statistical features of content popularity in various geographical locations in ICN. We employed data-driven methodology to predict the content popularity from the collected data of local CP users without the premise on the fixed content popularity distribution. In addition, since the users' privacy may be compromised by semi-honest CP (e.g., the users may receive aimed spam/fraud emails or phone calls according to the disclosure of the data). We exploited optimal local hashing (OLH) protocol to protect each individual user’s content preference information, while capture the characteristics of whole users' data. We studied the system revenue maximization problem by taking users content preferences into account to enhance the effectiveness, and sought practical an effective means to process the uncertainty of noisy data, and reduce the privacy risks of users' information.
With the advanced technologies and equipment on computation, communication, automation, controls and sensing, the traditional electric infrastructure is motivated to be modernized into smart grid, which enables two-way communication including electricity and information. To overcome the power outage/interruptions due to the supply and demand mismatch, smart metering is a promising solution to forecast and monitor electricity consumption of consumers. The smart meters are installed in each consumer's end (household, company, factory, etc.). The amount of electricity a customer used is measured and saved in an energy profiles, which will be sent to the utility provider at a requested time interval (the frequency can be as few as 1-5 minutes). The provider utility can predict the user's demand accurately, optimize the operation of all distribution resources, and improve the efficiency of the energy network. However, the energy profiles will be a potential target for well-motivated adversaries to compromise the customer's privacy. In this nearly real-time delivery of energy consumption profiles, the attackers can exactly observe the consumer's behavior, by comparing the differences between consumption profiles. There are some research efforts trying to address security and privacy concerns while meeting the requirements in smart grid. Nevertheless, the cryptographic solutions can only keep data protected during transmissions, but not for the cases that the adversary compromises the utility providers' servers, or the utility providers themselves are not trustworthy. Under the assumption that the utility provider is semi-honest, i.e., honest-but-curious, I proposed to allow customers to add distributed differential noises to the measured data before the smart meters send it to the utility provider. Based on the aggregated "noisy" but statistically correct data, we let the utility provider employ data-driven approach to characterize the uncertainty of customers' forecasting power demand, match the demand with the supply, and try to minimize the cost of energy generation. We show that the proposed scheme can effectively reduce the power generation cost of the utility provider while preserving the customers' differential privacy in smart grid.
Cognitive radio networks play an important role in CPS as a typical communication network. Cognitive radio (CR) is a promising technology to improve spectrum utilization, which enables secondary users (SUs) to access the licensed spectrum. Due to high economic values of spectrum resources, CR technology will potentially initiate spectrum trading, which benefits PUs with monetary gains and SUs with accessing opportunities to satisfy their service demands. In spectrum market, although many spectrum trading architecture help to capture spectrum accessing opportunities, they ignore the SUs' traffic demand uncertainty, which may have negative impact on primary service provider’s (PSP) revenue maximization. The past spectrum trading architectures are not able to represent SUs' traffic demands in real-time manner. Jointly considering SUs' real-time traffic demand and the privacy of SUs' data, I proposed a new entity, secondary traffic estimator and database (STED), which is responsible for estimating the SUs’ traffic demands in real-time manner and answering PSP's queries about SUs’ traffic demands. Considering the large population of SUs in the PSP's coverage boundary, it is not efficient to crowdsource SUs’ traffic demands by collecting each SU’s demands in terms of time consumption and communication overhead. Thus, we propose to let the STED employ data-driven approach to collect sampled SUs’ demands, construct reference demand distribution from sampled demands, and leverage reference distribution to estimate the real demand distribution of all SUs. By exploiting the stochastic nature of capacity/throughput of SUs', I studied the probability distribution of SUs' traffic demand and different distribution metrics to quantify the uncertain at certain confidence levels. From the aspect of privacy preservation, by exploiting the differential privacy technique in the scheme, we ensure that the output from STED keeps the characteristic of the whole data set, while preserves information privacy of each individual data simultaneously. Have capturing the feature of traffic demand uncertainty, under the proposed spectrum trading architecture and differential privacy technique scheme, I solve the revenue maximization problem by building the risk-averse two-stage stochastic model under multiple constraints (privacy guaranty, spectrum availability, traffic delivery satisfaction), and convert it into a traditional two-stage solution, and verify the effectiveness of the proposed scheme.
Compared to common commodities, spectrum has a very special feature, i.e., its spatial reusability, which has promoted many research works on the centralized design of spectrum trading. Although the centralized spectrum trading design takes spectrum reuse into account and guarantees economic properties, it may not capture instantaneous accessing opportunities, and have scalability issues, when the network size of SUs increases. When the network grows too fast, i.e, there are too many users involved in the spectrum trading, the centralized system may need advanced processor to handle it. Beyond the centralized spectrum trading design, distributed designs can also provide quick responses to some emergency situations, e.g., unfortunate attacks of 9/11, Hurricane Katrina, etc., where the centralized infrastructure deployment may be destroyed. However, most existing distributed spectrum trading have little concern of spatial reuse and frequency reuse. I proposed a novel matching based spectrum trading scheme, which jointly considers spectrum reuse features, and allows spectrum trading between PUs and SUs in a distributed manner in CRNs. I thoroughly investigated the network performance optimization problems under multiple constrains (i.e., the spectrum availability, channel interference), to maximize total revenue of PUs. Due to the NP-Completeness of solving some formulated problems, I provided approximation algorithms via matching method, which is originally from economics, and corresponding feasible solutions as well. With the advent of my proposed scheme, I not only improve the economic revenue of spectrum trading systems while involving more special characteristics (e.g., frequency reuse) of CRNs, but also extended the design into multi-radio multi-channel spectrum trading market, which significantly affect the performance of throughput in the network.