WInnComm 2019 Technical Session Presentation Abstracts
~~~~~~~
Wednesday, 20 November
10:30 - 12:00
Modular Radio Workshop
TS1: Citizens Broadband Radio Service 1
14:00 - 15:30
TS3: Machine Learning in Spectrum Management
10:30 - 12:00
TS1: Citizens Broadband Radio Service 1
The role of sharing in novel platform-based ecosystemic business models in 5G
Seppo Yrjölä (Nokia, Finland)
The fifth generation, 5G, mobile communications technologies are expected to transform the telecommunications services and networks' business models and respective ecosystems. 5G is transforming networks on five streams: 1) densification by adding millimeter wave small cells in the access network to boost capacity; 2) distribution of radio and core functions, content and services on the edge clouds for pooling gains, low latency, high reliability, security and privacy; 3) decomposition of network functions to lift scalability, programmable transport mesh that interconnects the distributed datacenter infrastructure; 4) softwarization of the network with advances in analytics and machine learning enables high level of automatization in management and orchestration; and 5) virtualization and slicing, utilizing the above capabilities for new service oriented business models. With roots in economics and engineering, platform research has an intrinsically dualistic perspective to business. In the economics tradition platforms have been seen as two- or multi-sided markets connecting supply and demand, whereas in the engineering tradition they have been seen as modular technological designs for facilitating innovation. Platforms and ecosystems can be seen intertwined, as both traditions acknowledge platforms to be consisting of a complex networked/layered system of modular components and interfaces the scope and scale of which go beyond the immediate platform actors. A transformation of business models as well as entire industries from vertical or horizontal linear towards two-sided and networked is ongoing. Furthermore, with the platformization an "oblique" business models are emerging focusing on value sharing through value co-creation and co-capture, while the traditional vertical control-oriented business models have aimed at controlling value creation, and the horizontal business models controlling value capture. As an emerging field, 5G related business models have been discussed to a limited extent in the literature and sharing based platform business models in general have seldom been examined. An essential question is How to understand and capture the role of sharing in the evolution of future platform-based ecosystemic business models in 5G? This study has an intrinsically dualistic perspective to platform business utilizing the 4C ecosystemic framework and the as a Service (aaS) digital service-dominant logic business model typologies. In the 4C business model framework each of the four types of business models have varying value propositions and revenue models: connection (e.g., CBRS), content (e.g., data), context (e.g., search or location), and commerce (e.g., marketplace and platforms). From the ecosystem perspective, the typology can be interpreted as a set of nested layers, where lower layer business models are required as enablers and value levers for the higher layers to exist. On the other hand, in the digital services domain, service oriented business models like infrastructure-as-a-service (IaaS), platform-as-a-service (PaaS) and software-as-a-service (SaaS) enable a large number of digital service providers to offer a variety of cloud-based services across the cloud stack layers. This research followed a cyclical process of research-oriented action research. The study collected and utilized data from the future-oriented 6G Wireless Summit in Levi in March 2019. The findings demonstrate that in 5G modules can be defined as an add-on software subsystem that connects to the platform to add functionality to the platform defined interfaces as specifications and design rules that describe how the platform and components interact and exchange information using well-documented, and predefined standards like application programming interfaces (APIs). The novelty of this study relates to leveraging two new complementing elements of the platform architecture: data and algorithms. The 5G management and orchestration layer can incorporate an exposure function opening the assets of a network to other service providers like mobile virtual network operators, micro-operators, industry verticals, enterprises and 3rd party applications. Exposing valuable infrastructure and data assets to the developer community through a set of APIs and setting up effective partnerships will allow service providers to grow their businesses by sharing their services with these external partners. Via softwarization and virtualization of network functions and opening of interfaces, sharing economy concepts will be utilized not only at higher platform business layers but widely in network connectivity, spectrum management and radio site domains. Changes in the ownership of spectrum access rights, networks, network resources, facilities and customers will result in several different combinations depending on the situation as different sites have different requirements and infrastructures. 5G system architecture enables different levels of exposure to resources and network functions between business actors and depending on the relationships between business actors and customer there exist different levels of transparency in network resource provisioning and further, different forms of cooperation models. Demands and resources can be brought together through the matching/sharing resource configuration/orchestration roles including different kinds of operators, resource brokers (e.g., Citizens Broadband Radio Service Spectrum Access Systems (CBRS SAS) operators), and various service/application providers such as trust/security providers. Block chain or distributed ledgers technology is attracting high hopes as artificial intelligence (AI) complementing technologies. Without central authority in a distributed manner, this technology allows storing and sharing information that does not change too often giving rise to e.g., new ways of organizing spectrum, network slice and data markets, or helping to maintain trust in an inter-operator setting. Introduced network elasticity and scalability enable network and resource usage adaptation to needed capacity and service levels on demand that, in turn, improves business agility while reducing both capital and operational expenses. Traditionally, the wireless networks context has been dominated by supply side business models while our study anticipates the increase in two- or multisided service oriented business models utilizing sharing economy concepts. Moreover, the CBRS "Spectrum-as-a-Service" (SPaaS) concept was found to leverage all the elements of the extended ecosystemic platform architecture components, interfaces, data and algorithm, in particular through the introduction of SAS and spectrum license marketplace.
Dynamic spectrum sharing with other networks using optimized PHY/MAC layers & guided by Artificial Intelligence/Machine Learning
Dunmin Zheng and Santanu Dutta (Ligado Networks, USA)
Dynamic spectrum sharing with other networks using optimized PHY/MAC layers & guided by Artificial Intelligence/Machine Learning Santanu Dutta, Gary Churan, Dunmin Zheng Ligado Networks A design approach and simulation results are described for an ad hoc mesh network sharing a common band with other networks. The PHY/MAC layer is optimized for power/bandwidth efficiency over disjoint spectrum. In addition, AI/ML may be used to improve performance by spectrum occupancy prediction, aided by reinforcement learning. The network continuously senses the spectrum occupancy of the shared band and dynamically organizes itself to utilize the unoccupied segments (spectral holes) of the band. The system may be a candidate for unlicensed (GAA) networks in the CBRS band. Another application may be HF, where the band is often very congested with narrowband interference, whose spectrum occupancy changes slowly with time and distance. In addition to autonomous sensing of spectrum holes, a performance advantage may be derived by utilizing feeds from a central controller (if present), such as the Spectrum Access System (SAS) in CBRS, which allocates spectrum resources to the higher priority networks in the shared band. Furthermore, elements of AI/ML may be used to predict where and when spectral holes will appear, so that the network can better deal with rapidly changing spectrum occupancy than is possible from a purely sense-and-avoid approach. The innovations in the proposed system include: (1) Optimal utilization of disjoint spectrum at the PHY/MAC layers, using coherent power integration over frequency (unlike non-coherent integration typically used in frequency hopping systems); (2) Use of interference avoiding CDMA (CDMA-IA) as the multiple access technique, which avoids overlaying power on other networks' spectra, thereby avoiding interference to and from other networks; (3) waveform design based on application of Fountain codes in the frequency domain, which leads to implementation advantages in disjoint spectrum; (4) robust, decentralized control of the network, which reduces latency and improves network efficiency; (5) support of relay function for serving "hidden nodes"; (6) operation on a single band, with no need for paired spectrum to support FDD; (7) Potential for applying Machine Learning to predict occurrence of spectral holes, so that the subject network can better optimize its own spectrum occupancy.
Propagation Measurements in the 3.5 GHz CBRS Band
Andrew Clegg (Google, USA)
The results of extensive propagation testing in the 3.5 GHz band will be presented. The measurements were obtained in real-world urban/suburban environments, and are directly relevant to network planning and co-existence studies in the new 3.5 GHz CBRS band. The measurements will be compared to predictions using, for example, free space loss and ITM, to show that actual losses are typically much greater than predictions. Various scenarios are investigated, including indoor/outdoor losses, losses in wet/dry weather, seasonal differences, and losses during a significant snow event.
14:00 - 15:30
TS2: Citizens Broadband Radio Service 2
Coexistence management for 5G in CBRS WITHDRAWN
Heikki Kokkinen and Paweł Kryszkiewicz (Fairspectrum, Finland)
In this study, we present simulation results of GAA CBSD power and bandwidth allocation for efficient coexistence management. We compare the performance of four heuristic algorithms with a Multi-Choice algorithm, which combines the four approaches. Our study shows that individually the algorithms result a suboptimal solution close to Perato optimal curve of the solution space. The Multi-Choice algorithm delivers multiple solutions on Perato optimal curve.
[TOP]
TS3: Machine Learning in Spectrum Management
Interference-aware power coordination algorithm for 5G Ultra-Dense Networks (UDN)
Ángel G. Andrade (Universidad Autónoma de Baja California & Facultad de Ingeniería, Mexico); Adolfo Reyna-Orta and Alexis Anzaldo (Universidad Autónoma de Baja California, Mexico)
For accomplish 5G proposes, network densifications emerge as a fundamental strategy to improve spectrum use, data rate, offloading the macrocells, and increasing the resources reused on the small base stations. However, the interference generated due to the massive deployment of small-base stations affects the channel propagation conditions allocated to users, causing inefficient resource allocation. In this work, the problem of power allocation is discussed to reduce the levels of interference on downlink transmissions for Ultra-Dense mobile network (UDN). The resource allocation problem is divided into two stages; channel and power allocation. For channel allocation, an algorithm that considers proportional fair among users (PF) was used. Power allocation coordination through a Genetic Algorithm (GA) is presented and evaluated concerning sharing and partitioning spectrum approach. The UDN capacity and the number of attended users are maximized. Also, it is analyzed under which conditions further densification is beneficial and improves the system capacity. Results show that when interference coordination is efficiently applied, more users can be attended, and the degradation of network capacity was minimal.
Machine Learning as Enabler for Cross-Layer Resource Allocation: Opportunities and Challenges with Deep Reinforcement Learning
Andres Kwasinski and Fatemeh Shah-Mohammadi (Rochester Institute of Technology, USA)
The vision for ubiquitous mobile computing is indivisibly linked with pervasive connectivity. As such, the realization of this vision requires devices to be able to interconnect in different wireless environments where the traditional layered design approach is inefficient for many of the applications scenarios and does not adapt well to a network infrastructure based on intra and inter-network heterogeneity. While a layered network protocol design is advantageous in simplifying the design problem into compartmentalized modules, it also underperforms in wireless networking settings where some examples are well known (e.g. TCP over a wireless link). Cross-layer techniques are known to improve performance for a majority of the types of traffic carried in today's networks. Yet, commercial wireless devices architecture remains layered, because the development of the complete networking stack is divided into different teams, each specialized in implementing one layer or sub-layer in a specific processor (e.g. main CPU or baseband radio processor). This results in software modules with very limited capabilities to exchange the data and the commands needed to implement cross-layer solutions. Nevertheless, Cognitive Radio technology provides the means to enable seamless modular implementation of cross-layer technology through the introduction of a software component tasked with collecting data from the different layers and using the derived information to intelligently adapt operating parameters across multiple layers. This software component, known as a "cognitive engine", is implemented through a Machine Learning technique. In this presentation, we will explain how a cognitive engine for cross-layer resource allocation is implemented using a multi-agent Deep Q-Network, an emerging class of Reinforcement Learning algorithms which combine the process of reinforcement learning with a class of artificial neural network known as Deep Neural Networks to approximate the Q action-value function from Q-Learning. Single-agent Deep Q-Networks are the state of the art in reinforcement learning, yielding remarkable performance in video game playing experiments, autonomous vehicles control, and other applications. However, the wireless environment of a cognitive radio is a multi-agent scenario that, while still showing the performance gains seen with single-agent Deep Q-Networks, also introduces new challenges. These challenges, that include an expanded state and action space and a potentially non-stationary environment, lead to increased time for the Deep Q-Network to learn the best decision to make when allocating resources. Fortunately, the multi-agent environment does not only introduce challenges but it also presents opportunities to address them. In this presentation we will explain the challenges presented by multi-agent Deep Q-Learning in a wireless environment and how these challenges can be addressed through the use of "transfer learning", where an experienced cognitive radio shares experience with other cognitive radios that need to run the Deep Q-Network algorithm to learn their best parameter setting.
Spectrum Awareness Under Co-Channel Usage via Deep Temporal Convolutional Networks
Amir Ghasemi (Communications Research Centre Canada, Canada); Chaitanya Parekh (Communications Research Centre, Canada); Paul Guinand (Communications Research Centre Canada, Canada)
Modulation recognition of radio signals is a key component of spectrum environment awareness enabling radios to share the spectrum more effectively and allowing the regulators to monitor compliance and identify rogue users. Modulation recognition typically relies on signal features derived analytically by domain experts. Recently, it has been shown that deep learning can achieve state-of-the-art performance by directly processing raw I/Q signals without requiring any manual feature engineering. This ability to learn useful representations of radio signals directly from raw data enables a paradigm shift where algorithms could be fine-tuned to specific propagation environments or radio impairments, which might otherwise be difficult to model analytically, by simply collecting sufficient raw data. In this paper we consider the problem of modulation recognition in circumstances where there is a co-channel signal. Both the problem where the type of the co-channel signal is known and the problem of modulation recognition for both of two unknown interfering signals are addressed. We use a temporal convolutional neural network which, for the case of a known co-channel signal type achieves a classification accuracy in excess of 80% at a signal-to-interference ratio (SIR) of 0dB. The modulation recognition of co-channel interfering signals in noise is of particular interest for spectrum regulation and compliance. We also show the efficacy of this architecture for the classical case of modulation recognition of a single signal. In this case, we can achieve 100% classification accuracy for signal-to-noise ratio (SNR) levels greater than 0dB.
Thursday, 21 November
10:30 - 12:00
TS4: Platforms for Software Defined Systems
14:00 - 15:30
TS5: Implementations of Testing of 5G Networks
TS6: Drone Swarm Networking
10:30 - 12:00
TS4: Platforms for Software Defined Systems
Software Programmable AND Hardware Adaptable: Can You Have Your Cake and Eat It, Too?
Manuel Uhm (Xilinx Inc. & Wireless Innovation Forum, USA)
Heterogeneous processing systems are in vogue for most embedded systems, typically leveraging some combination of general purpose processors, digital signal processors, graphics processors, hardware accelerators and/or FPGA fabric. While this makes sense from a hardware perspective as there is no single processor type that is ideally suited for all algorithms and applications, it can make for an extreme software development challenge as different tools, languages and debug methodologies are often required to program each processor type with no easy way to do system level testing and debugging. Higher levels of abstraction have been touted as a solution to this complex problem, however, the industry has not converged on a single methodology as a typical embedded system could require programming in multiple languages or APIs such as C, C++, Python, VHDL, Verilog, CUDA and/or OpenCL, just to name a few of the more common choices. Since it seems highly unlikely that industry will achieve convergence on this thorny issue, a more logical approach is to embrace multiple levels of abstraction through a unified platform that simplifies development and debugging. This presentation will outline just such an approach using a new category of heterogeneous processor, the Adaptive Compute Acceleration Platform (ACAP). Xilinx has launched the first ACAP with a portfolio of heterogeneous processors known as Versal.
CERTIF: Conformity tests on software defined radio platforms
Olivier Kirsch (KEREVAL, France)
In order to facilitate interoperability and portability of software defined radio components, the conformance to SDR standards (including APIs and behavior specifications) is mandatory. Due to the huge number of requirements and to ensure reproducibility of the compliance assessment, a testing methodology has been developed and implemented into the bench CERTIF. Firstly this presentation will summarize briefly the test methodology applied to verify the conformance of a software radio platform to the SDR standards. In a second time we will show all kind of non conformance issues detected by the bench CERTIF on the basis of concrete examples and we will show the reproducibility of the test results. All these cases come from experience feedbacks on the bench. The presentation will be made by Olivier Kirsch, Test project Manager of KEREVAL. KEREVAL is a French testing Lab.
14:00 - 15:30
TS5: Implementation of Testing of 5G Networks
Proposal of an Efficient Multiplexing Scheme based on OFDMA and Massive MIMO Beamforming
Yoshimi Fujii and Kensuke Tsuda (Kozo Keikaku Engineering Inc., Japan)
The 5G base station uses Massive MIMO technology in full-fledged manner, and a base station provided with tens to hundreds of antenna elements forms beams for individual terminals in the area to improve power efficiency or to reduce interferences among adjacent base stations. By doing them, installation of higher-density base stations becomes possible, and the transmission capacity per unit area is dramatically improved. However, there is a limit to improve beam accuracy, because digitally controlled analog phase shifters are used for a wide band area reaching 100 MHz to 400 MHz in the case of a millimeter wave band. In order to realize full digital beamforming system, which is essential for performing phase control on OFDM subcarrier basis to achieve much finer beams, the amount of arithmetic processing of the base station is massively increased, and therefore, it is still at the examination stage. In this research, we propose an efficient base station architecture with full digital beamforming. By performing phase control in each OFDM subcarrier unit by baseband digital signal processing for each antenna element, more precise beam forming is enabled, compared to the conventional analog method. In addition to that, we also propose a method of beamforming by resource block units. Linking with resource block allocation of OFDMA, this could be used to maximize the potential of full digital beamforming and dramatically improve space efficiency.
National Advanced Spectrum and Communications Test Network (NASCTN)
Melissa Midzor (National Institute of Standards and Technology, USA)
The Federal government is required to operate in a compressed spectrum ranges due to FCC auctions. This presents a risk that a variety of commercial and federal operators will harmfully interfere with each other. The National Advanced Spectrum and Communications Test Network (NASCTN) is a multi-agency-chartered organization that supports the growing need for accurate, reliable, and unbiased measurements and analyses to support increased spectrum sharing by both federal agencies and non-federal spectrum users. This partnership between DoD, NIST, NTIA, NASA, NOAA, and NSF provides unique capabilities and a neutral test organization to tackle difficult and contested measurement problems in shared spectrum. Includes development of new test methodologies, providing trusted validated data in determining levels of interference between Commercial and Federal assets, access to key test facilities, and the ability to operation as a trusted agent to protect proprietary and classified information. Prior projects include LTE impacts on GPS L1 (data accepted by all stakeholders to inform technical discussions of policy), and Aggregate AWS-3 LTE Emissions (informs interference models used by DoD for expedited and expanded entry of commercial deployments into the 1755-1780 MHz band). NASCTN's new 5G Laboratory Testbed is extensible & adaptable to take on today's challenges in Spectrum Sharing, Coexistence, and Interference testing. Capabilities include a complete network infrastructure that hosts a standalone IT network and carrier-grade base stations, support of 5G FR1 andFR2 in addition to 4G LTE, wifi, and GPS, and access to anechoic chambers for OTA measurements.
TS6: Drone Swarm Networking
AS Testbed Architecture for 3D Mobility Research using Advanced Wireless Technology
Vuk Marojevic (Mississippi State University, USA); Ismail Güvenç, Rudra Dutta and Mihail Sichitiu (North Carolina State University, USA); Jeffrey Reed (Virginia Tech, USA)
Enabling ubiquitous mobility and connectivity in three dimensions unleashes new applications and capabilities for mission-critical, commercial and other applications. The emergence of advanced wireless technologies and systems allows new wireless connectivity and networking principles to be explored and new protocols to be designed, developed and tested in production environments. In addition to ground-based mobility services, such as vehicle-to-vehicle communications, virtual and augmented reality, and massive Internet of things (IoT), new use cases for advanced wireless technologies are emerging in the unmanned aerial systems (UAS) spaces with unprecedented opportunities in the commercial, government, civil, and military sectors. R&D has proposed a variety of new use cases for UAS, but only a fraction of those have been verified by reproducible experiments in production-like setups. The safe integration of UAS into the national airspace, and the emergence of advanced UAS missions, with dynamic mobility patterns and tighter coordination, require new experiences and data. For this reason, we envision and architect a facility called AERPAW, an aerial experimentation and research platform for advanced wireless that provides a blueprint for designing and operating a community R&D testbed that combines commercial wireless networking technologies with open-source software radios in production-like 3D mobility environments. We discuss the requirements, deployment challenges and solution alternatives, and the important role of the wireless R&D and user communities for building and maintaining a wireless testbed in which ground and unmanned aerial vehicles interact with one another.
SwarmSense: Effective and Resilient Drone Swarm and Search for Disaster Response and Management Application
Bo Ryu (EpiSys Science, USA); Cheolmin Jeon, Jeongsoo Ha, Hanbum Ko and Byeongman Lee (EpiSys Science, Korea)
In this paper, we present SwarmSense, a novel collaborative navigation algorithm for a group of drones to effectively coordinate and share information for disaster response and management applications such as wildfires. Specifically, SwarmSense is aimed for efficient and resilient ways to plan complex unmanned aerial systems search and rescue missions by mapping wildfire zones and detecting displaced survivors in need with the ability to react to dynamic changes so as to maximize the potential to minimize damage and save lives. In other words, SwarmSense is designed to find areas where the disaster hotspots are, map and track the hotspots precisely in real time, and find and rescue the survivors spread across the disaster area. Any drone swarm and search algorithm must consider the following major challenges: (i) limited resources; (ii) limited information availability; (iii) extremely large area with highly challenging navigation conditions. First, the algorithm must take into the limited resources in terms of the number of drones available and short battery life. Second, the algorithm must address the lack of information about the disaster in terms of the location and size of fire zones, continuously changing weather conditions, and the location of any survivors in need of the rescue. Finally, the algorithm must be able to adapt with robustness for challenging terrain conditions in combination with the dynamic weather conditions. Underestimating the terrain and weather effects lead to explosion/crash of the drones, while the conservative approach will lead to poor disaster response effectiveness. SwarmSense addresses these challenges with high performance measured in terms of the three metrics: (i) firezone detection ratio (percentage of the firezones detected by the drones with no prior knowledge of their locations); (ii) firezone mapping precision under changing constantly due to dynamic weather conditions; and (iii) drone mission completion ratio (the number of drones completing the entire scenario without being destroyed by fires or terrain). We used the AMASE simulator developed by Air Force Research Laboratories (AFRL), available at https://github.com/afrl-rq/OpenAMASE, and the ten 60-minute scenarios provided by the AMASE during the competition held at AFRL during March 29-31, 2019 (https://fire-hack.devpost.com/), and show that SwarmSense detects near 100% within the first 40 minutes for all ten scenarios, achieves a high degree of precision (provided by the AMASE scoring mechanism proportional to the mapping precision), and maintains as high as 83% of the drones despite extremely challenging weather (wind) and terrain conditions.
|