Workshops 1B & 2B: Big RF and Context Aware Cognitive Radio
Tuesday, 24 March 2015 (Fung Auditorium)

The performance of any cognitive radio algorithm is dependent on numerous external factors so that, for instance, security measures are appropriate to the deployment, transmitted waveforms account for local regulatory variances, radios make the best use of the services of available networks and devices, and the QoS sensitive algorithms respond to the user, application, and operating conditions. All of these external factors external that influence the relative performance and acceptability of cognitive radio (CR) performance are part of the CR’s operating context.

In the WinnF Cognitive Radio Work Group (CRWG) defined a CR’s context as all relevant information about the CR’s operation. Thus a CR’s context includes the metrics that measure the CR’s communications performance – the so-called “meters” of a CR – such as BER, jitter, end-to-end delay, and signal-to-noise ratio – commonly encountered in any waveform stack as well as relevant information beyond waveform stack metrics, such as CR location, information from or about other radios in the area, mission objectives, user identity, and the specific application(s) being supported. But understanding the context of a CR often requires a CR to have insight far beyond what it can directly observe and reasoning capabilities beyond what a consumer device could be expected to provide.

Instead, CR implementers turned to networked databases, such as the TV White Space databases, to implement the reasoning processes and handle the various policy implications. But as CR turns to the 3.5 GHz band where incumbent properties change much more quickly and as cellular service providers begin to trying to synthesize insights about their network (i.e., network intelligence) with increasingly complexity due to femtocells and self-organizing networks, it is clear that being able to quickly process and extract meaning from the sea of data will be increasingly important.

Because of the issues with the volume, velocity, variety, and veracity of the data in these envisioned RF databases, we believe that, like with Big Data, traditional relational database techniques will be insufficient for the purpose of gaining meaningful insights about the RF environment in a way that can be used by CRs, regulators, and network managers. To refer to this problem space, the Cognitive Radio Work Group (CRWG) in the Wireless Innovation Forum (WInnF) has adopted the term “Big RF” to reflect the similarities between the IT and RF problem domains. Big RF conditions apply across all network domains from relatively small local networks to larger nationwide deployments. Big Data resources, techniques and concepts could be applied and extended to RF data analysis problems once we account for the unique properties of radios (e.g., data converter and sensor limitations, component nonlinearities) and the radio environment (e.g., propagation effects, lossy channels, interference, and bandwidth constraints).

A notional diagram of major components and tools in Big RF based on a canonical Big Data diagram is shown in Figure 6. While many blocks are largely the same though with more specificity to the RF domain (e.g., REMS and the TVWS databases as potential data sources), there are the following major differences between Big Data and Big RF applications.

• Multiple analysis consumers – While the only consumers of the results of the Big Data analysis are humans, Big RF is a tool to aid the RF understanding of human and CR users. This means that there will need to be tools developed that put the Big RF analysis results in a format that is meaningful for CRs (CR Presentation).
• Looping data flow – By being both a data source and a potential consumer of the analyses, the system can have looping behavior, which means some care should be taken to consider system stability.
• Self-similarity – at the scale of a single cognitive radio and at the scale of a Big RF system, all of the basic processes are the same – sensing and collecting data (observation), streaming and other sorts of analysis (orientation), model building (learning), and customized RF-specific logic that informs decisions and actions.

The CRWG is currently undertaking a study into enabling context aware cognitive radio with a strong emphasis on enabling technologies based on Big Data. The following are some of the ongoing discussions and problems that the CRWG, in particular, is trying to address.

• Scope and definition – Is there a clear distinction between what constitutes a “traditional” cognitive radio problem and what is a Big RF problem, or is it simply a matter of degree? A related question to be addressed is - What sorts of problems would be best handled by a Big RF approach and what could be better addressed by more traditional CR techniques?
• Implementation - What deployment architectures are needed and how can they fit into the existing and emerging cognitive radio ecosystem? How can these be automated to provide feedback and information to cognitive radios in addition to people? Is there a way to adapt the multi-core scalable distributed Big Data tools such as Hadoop to opportunistically leverage the computational resources of thousands of software radios or are Big RF specific tools required? How can we formalize system descriptions for subsequent analysis, e.g., describing side lobes and antenna patterns and user location trends? Would SSRF be sufficient for that descriptive language? Would another language surveyed in the CRWG’s previous work suffice? Or does a new standard need to be defined? Should the possible languages be collected into an ontology to simplify processing?

While we are treating Big RF as a subset of Big Data, there are some identifiable differences with other Big Data implementations, such as the following.

• Security – Many Big RF solutions will require accessing data sources that are not directly under the control of the analyst, which introduces several angles of attack and may necessitate the addition of routines to correlate data to detect poisoning attacks. And because of the noisy nature of the wireless medium, even without malicious intent, data in a Big RF system will likely still have a greater likelihood to be corrupted or erroneous. Similarly, if Big RF results are made available to cognitive radio subscribers, then it will be challenging to preserve privacy from attacks on quasi-identifiers returned from Big RF systems much as it is challenging to obfuscate primary user location in the TVWS geolocation databases.
• Dynamics – With more sensors and shorter coherence times for some phenomena, quicker responses will frequently be needed in a Big RF system. However, it should be noted that any Big RF system should accommodate multiple time scales as adaptations could be needed in less than a second (e.g., link adaptation or primary user avoidance), on the order of hours to days (e.g., adjusting network configurations), or on the order of months to years (e.g., accommodating regulatory changes).

To broaden the range of contribution sources and to help popularize these concepts, the Forum intends to host a workshop WInnComm 2015, focusing on contextual reasoning in cognitive radios and Big RF. While the efforts to date have been mostly industry driven, we believe this workshop would greatly benefit from academic research input. The workshop will include a call for contribution in the following areas:

• Applications – 5G small cell (interference mitigation), context aware applications, Internet of Things, MHealth, Smart Grid, Public safety network disaster management, SON and public safety, Health Care, Home Energy management, Retail spaces
• Data sources and appropriate interface methods– spectrum databases, drones with sensors, phones / devices, FSMS, FSAS
• Dynamics – stability of the system; system complexity and implications for system engineering
• Security / privacy – securing the communications and network that enables context CR and Big RF; sanitizing quasi-identifiers while still allowing actionable insights; privacy with deep-packet inspection
• Implementation methods, studies on the tradeoffs for batch and stream processing, complex system modeling, system architectures
• Proposed standards for presentation layers, data formats, schema, ontologies, languages

 Workshop Agenda

Workshop 1B
Tuesday, 24 March, 10:30-12:00 (Fung Auditorium)

Invited Speaker: Mark Majernik, K&M Systems

Presentations:

• “Socially Aware Routing Algorithms for Context Aware Cognitive Radio,” James Neel, Cognitive Radio Technologies

• Distributive Sensing Techniques for Context Aware Spectrum Mapping

  Deborah J Walter and Kurt Bryan (Rose-Hulman Institute of Technology, USA)

  Cognitive radio performance is dependent on the real-time awareness of the radio spectrum use across frequency, space, and time. Today's congested spectrum, complex RF environments, and dynamic mobile emitters, require spectrum managers to develop automated, context-aware techniques for characterizing the RF environment. We propose the use of distributive sensors working in cooperation to efficiently map the RF environment. Conventional spectral sensing involves the determination of frequency bands that are not heavily occupied by licensed users. More generally, cognitive radio systems require the ability to measure, sense, learn and to be aware of the context of the spectral white spaces. Therefore, cognitive systems require real-time information across multiple dimensions such as time, space, frequency and code. Although we recognize that the RF environment is in general congested, the sensing problem can be recast into a multi-dimensional sensing environment which can be described as a sparse problem. In this paper we introduce a strategy for simultaneously locating the source of RF signals and detecting their spectral characteristics using a distributive network of passive sensors that are mobile and reconfigurable. We consider the use of compressed sensing techniques in this setting. Questions of particular interest are: what is the optimal arrangement of sensors for locating RF emitters? How many sensors are needed for locating a given number of RF emitters? How does such a system perform in the presence of noise? We will also validate our approach with data collected using a physical testbed designed by a team of Rose-Hulman seniors, in response to an Air Force Research Laboratory Student Challenge.

  
  

Workshop 2B
Tuesday, 24 March, 14:00-15:30 (Fung Auditorium)