Cloud Based Regional Spectrum Services
The importance of dramatically improving spectrum efficiency has become particularly urgent in light of the emergence of smartphones as the primary computing and communication platform and the corresponding exponential growth in mobile data. In order to fulfill these growing communication requirements, there have been proposals to overload unlicensed frequency bands, by granting access to these bands to new radio technologies. The operation of various wireless technologies in the same frequency band (e.g., Wi-Fi and LTE-U, TV white space and the 3.5-3.7 GHz bands) brings enormous challenges due to incompatibility of these technologies to co-exist while maintaining different goals, like fairness, maximizing throughput, QoE, etc. This project aims to address these challenges by designing and validating a novel decentralized dynamic spectrum management architecture, which will enable effective radio technology neutral approaches to spectrum assignment. While local spectrum access decisions such as Listen-Before-Talk (LBT) at individual wireless devices might have made sense in the past due to complexity and scalability concerns they will not suffice for next generation of wireless networks. Moreover, we now have significantly more computation and near-universal Internet connectivity enabling realization of advanced data-driven spectrum management regimes based on actual radio resource usage parameters reported by radio devices, access points, and base stations. The proposed architecture enables data-driven spectrum management approaches with regional visibility of wireless networks and their spectrum use. These approaches are expected to outperform local radio based methods by a significant margin in dense environments. This is because the proposed data-driven framework can accommodate logically centralized spectrum resource allocation algorithms, which can effectively deal with hidden node/network problems and achieve high throughput and fair sharing between multiple systems including those using different radio access technologies.
The proposed spectrum management architecture is built on a set of standardized APIs and protocol interfaces for wireless devices or networks to report and/or set radio resource parameters over a common control plane for spectrum. The architecture will support exchange of radio usage and control parameters between multiple autonomous wireless domains operating in the same geographic region, thus providing a level of local visibility that enables a range of distributed (and possibly competitive) algorithms for mitigating interference. Further, given the right incentives, networks in a given region can actively collaborate by participating in a cooperative optimization algorithm that sets operating radio parameters over multiple networks to achieve efficiency and fairness while adhering to both global and local policies. The architecture also provides a well-defined interface to higher-level cloud services including regional spectrum aggregators/brokers and global spectrum databases and policy repositories. The main challenges in the design of the architecture include the design of radio-technology neutral abstractions for exchanging spectrum use information, design of the spectrum control plane and identification of key protocol interfaces, design and evaluation of distributed algorithms with desirable convergence properties.
We have developed an initial system concept for the control plane based on the notion of a 3-tier hierarchy corresponding to radio-device level, regional network level, and global cloud-services level distribution of spectrum information – see Fig. 1. At the lowest level, there is an interface for each radio device to express its current spectrum usage and set its radio operating parameters. The spectrum use information can be made available to neighboring radios via beacons or dedicated wireless control channels where available or can be propagated upward to radio network operators who use a second control interface to express aggregate radio system parameters (also called “radio MAP”). Radio MAP information, from individual wireless networks, is further aggregated at the domain controller level (typically one per autonomous wireless domain (WD), but other mappings such as one per region or group of WDs are also possible). The WD controller is envisioned as a software-defined entity similar in spirit to wired network SDN controllers but with syntax extensions for handling radio-specific parameters. A protocol interface for exchange of radio MAP information between WD controllers and for inter-network policy negotiation is also provided at the network peering level as shown. Finally, an additional protocol interface is provided to the cloud-service level, which serves roles such as regional spectrum aggregator/broker across multiple domains or as a repository for regional and global policies. As shown in the figure, the information exchanged over these protocol interfaces can be used to run spectrum coordination algorithms either at the individual spectrum controllers or at the cloud servers. The radio resource control settings from these algorithms are fed back to individual wireless networks for coordinated operation.
Results To Date & Future Work Plan
We have deployed various components of the architecture in ORBIT testbed, using nodes (general purpose computers) equipped with USRP radios and Wi-Fi cards. In order to demonstrate the feasibility of deploying the architecture and the gains achieved due to cooperative spectrum management, a simple Wi-Fi LTE-U co-existence experiment as shown in the figure below is conducted. In this benchmark scenario, it is shown how frequency allocation algorithm will cause severe interference between LTE base station (BS) and Wi-Fi access point (AP), if each WD controller conducts it individually. In fact the result of our experiment shows that the Wi-Fi AP in the edge gets zero throughput (is completely shut down) before the WD controllers coordinate.
This is due to the lack of a sensing mechanism in LTE for other transmissions, whereas Wi-Fi has been designed to co-exist with other technologies and as a result it inherently performs Listen Before Talk (LBT) and carrier sensing before transmission. This performance degradation for the “Before Coordination” scenario is a result of WD controllers’ lack of information about the existence of a Wi-Fi AP and LTE BS operating on the same channel in their overlapping wireless coverage region. In “After Coordination” scenario, an optimized channel assignment scenario is shown, in which the controllers have exchanged BS/AP location information and made the channel assignment decision accordingly. The result of our experiment shows that the average throughput does not change for 2 scenarios, whereas Jain’s fairness index nearly doubles (0.47 to 0.89) after cooperation, thus confirming that fairness improves significantly when inter-network cooperation enabled by the architecture is used. Further work is planned on distributed algorithm design and comprehensive deployment and evaluation of the architecture in larger scale real-world scenarios.
Prof. Dipankar Raychaudhuri
ray (AT) winlab (DOT) rutgers (DOT) edu