COGNITIVE RADIOS IN PUBLIC SAFETY AND SPECTRUM ...

COGNITIVE RADIOS IN PUBLIC SAFETY AND SPECTRUM MANAGEMENT Thomas W. Rondeau, Charles W. Bostian, David M. Maldonado, Adam Ferguson, Sheryl Ball, Bin Le, and Scott Midkiff Center for Wireless Telecommunications, Virginia Tech 466 Whittemore Hall, Blacksburg, VA 24061 [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]

______________________________________________________________________ As wireless communications become more important to daily life, the wireless research community must create more efficient and useful radios to satisfy the rising needs for spectrum. The cognitive radio work covered in this paper addresses the spectrum and communications issues. Cognitive radios are aware, flexible, and intelligent, and they allow a radio to understand the wireless environment and make autonomous decisions to adapt themselves for efficient operation. A network of cognitive radios is a network that performs independent spectrum management to maximize all radios’ successful operation, even in the presence of legacy, or noncognitive, radios. This paper discusses work done at Virginia Tech to build a cognitive radio as well as the economic, political, and technological issues surrounding cognitive radio adoption. Specifically, this paper addresses the impact of cognitive radio technology on the public safety community and spectrum management. ____________________________________________________________________

Keywords: cognitive radio, spectrum management, public safety communications, cognition, machine intelligence, interoperability, artificial intelligence

1. INTRODUCTION To quote former Federal Communications Commission (FCC) Chairman Michael Powell, “All consumers . . . deserve a new spectrum policy paradigm that is rooted in modern-day technologies and markets. We are living in a world where demand for spectrum is driven by an explosion of wireless technology and the ever-increasing popularity of wireless services. Nevertheless, we are still living under a spectrum 'management' regime that is 90 years old. It needs a hard look, and in my opinion, a new direction” [Powell, 2002]. One answer to Powell’s concerns is the development of cognitive radios [Mitola, 1999], wireless devices that recognize user needs, act intelligently to fill those needs, and do so while working “politely” with all other radios located in the same geographic region and local regulatory limits. The radios act as autonomously as possible with minimal direction from the user. Such radios will provide many benefits to users of all kinds. In this paper we discuss applications in public safety and spectrum management. Public safety users require efficient access to spectrum and automatic quality of service (QoS) guarantees during disasters and similar events. Spectrum management has historically been a policy-based regulatory process with static allocations of frequencies for different purposes. The recent explosion of IEEE 802.11 (WiFi) wireless networking devices has shown that unlicensed spectrum is capable of providing many services and benefits that licensed spectrum cannot. Unfortunately, sharing resources, like in unlicensed spectrum, raises many problems with performance due to inefficient resource management. A radio aware of spectrum occupancy and able to readjust itself to minimize interference and maximize network efficiency would overcome most problems associated with radio resource management. That is, they can improve performance by adjusting to the spectrum, network, and application needs. This radio

concept opens the way to more and better uses of the spectrum, which introduces several interesting policy issues. Virginia Tech’s Center for Wireless Telecommunications (CWT) has developed a cognitive engine that will enable cognition in any adjustable radio. With NSF Digital Government Program support, we have built a proof-of-concept prototype that allows legacy Proxim Tsunami® 5-GHz radios to adapt to changing propagation and interference conditions. This prototype is installed at SAIC’s Public Safety Integration Center (PSIC) in McLean, Virginia. We are continuing to develop the cognitive engine and now have a variety of experimental and simulation results. This paper is focused less on the technical makeup of our cognitive engine than with the lessons learned and implications of cognitive radio techniques for the areas of public safety communications and spectrum management. For more information regarding our work, see [Rondeau, 2004] [Rieser, 2004a] [Rieser, 2004b] [Bostian, 2004]. Section 2 provides a more detailed definition of cognitive radios and their capabilities and describes a new approach to machine learning. Section 3 discusses cognitive radio use in public safety bands while Section 4 explores the technical, economic, and political issues of spectrum management with cognitive radios. Finally, Section 5 analyzes the impact of cognitive radio on other areas of government technology adoption.

2. COGNITIVE RADIO OVERVIEW A cognitive radio is a flexible and intelligent radio capable of creating any waveform and protocol supported by the radio hardware and software. Waveforms consist of all of the parameters that define the way in which the radio transmits information including transmitter power, operating frequency, modulation, pulse shape, symbol rate, coding, etc. Protocols are the

rules by which network nodes transfer information. A cognitive radio develops waveforms and chooses protocols in real-time using artificial intelligence. These actions require three components: 1. Perception: Sensors that collect data on both external factors (channel conditions, other radios, regulations, user needs) and internal factors

(waveform capabilities,

computational power available, battery life). 2. Conception: An intelligent core that learns and understands how to combine knowledge from the sensing mechanism to instruct the adaptation mechanism. 3. Execution: An optimization and adaptation mechanism that alters the radio’s behavior.

Figure 1 presents a very generic architecture for a cognitive radio. The cognitive engine is a separate system within the total solution, which relies on information from the user, radio, and policy domains for instructions on how

to

best

control

the

communication system, which is itself shown as a simplified protocol stack. This structure works well as a generalized architecture as it makes no claims to how the cognitive engine (and therefore the rest of the cognitive radio) should behave while still mapping the interactions of the rest of the systems.

Figure 1. Generic cognitive radio architecture.

In Figure 1, there are three input domains that are considered by the cognitive engine. The user domain tells the cognitive engine what services and performances are required by the user’s applications. The external environment and RF channel provide environmental context to the radio’s transmission and reception behavior. Finally, the policy domain restricts the system to work within the boundaries and limitations set by the regulatory bodies as interpreted by the policy engine. Having a highly flexible communication system such as a software defined radio (SDR) that can be programmed and reprogrammed quickly with a wide range of possible waveforms and protocols is of extreme importance to the cognitive radio operation. The development of such devices is an open research problem. To be a bit more specific about the operation of the cognitive engine in our work, we go back to Joseph Mitola, the originator of the cognitive radio idea, who defined a cognition cycle that all cognitive radios should follow [1999]. Mitola’s work existed mostly in the application layer of communications devices. We have extended this concept to the lower physical and medium access layers, which are responsible for controlling the radio frequency (RF) channel to transmit bits and coordinating communication among connected hosts to transfer blocks of data. Through our work on the bottom two layers (see Figure 2), we have developed a slightly modified cognition cycle that better fits our needs, as shown in Figure 3. The cognition cycle shows the information flow from inputted information to outputted actions. The radio hardware collects information from the external world to provide the cognitive system with the environmental context, while other environmental observations occur to read information about the user needs, policy demands, and hardware capabilities. All the information is then processed into an understandable model of the current state and passed to the machine learning core, which uses a combination of case-based decision making and optimization

techniques to learn and adapt proper behavioral patterns. The radio is then instructed as to the new behavior and will adapt itself accordingly. The optimization procedure discovers the best method of adaptation using artificial intelligence techniques, such as in our case, genetic algorithms. The system is aided by the presence of the knowledge base and the case-based decision maker, which keep track of progress and behavior over time to help the optimization process find better solutions faster. As the radio adapts to the new configuration, the cycle is repeated to ensure that the adaptation was appropriate and that no other environmental changes require further adaptation. The following section takes a closer look at how the optimization and learning processes work. The International Standards Organization’s (ISO) Open Systems Interconnect (OSI) standard defines interconnections and responsibilities of different layers of the communication system Each layer is independent of the others, allowing for many different types of systems to stand in for each other depending on the platform. Often, the Presentation and Session layers are now removed from the protocol stack and their responsibilities are absorbed by the Application and Transport layers. Figure 2. The ISO’s OSI Protocol Layers [NCS, 1996].

Figure 3. Cognition Loop of CWT’s Cognitive Engine.

2.1 Cognitive Machines Successful cognitive radios must be aware, learn, and take action for any situation that might arise. Applications range from voice communications under low power conditions, communications in high interference zones, to more complex and critical, hostile military networks of interoperating vehicles and soldiers with many different network needs. A radio must respond to any of these scenarios and adapt its many different parameters that define the radio’s waveform and protocols. These radios do not only require learning; instead, they need highly sophisticated learning and decision making capabilities. Machine learning has been well documented with both criticisms [Hawkins, 2004] as well as successes [Negnavitsky, 2002]. Successful applications of artificial intelligence (AI) tend to occur specifically in narrowly defined, well-bounded applications. While the radio application is still bounded, the technical demands for intelligence in a radio exceed those normally associated with successful applications of classic artificial intelligence techniques such as expert systems or neural networks (see Figure 4). Our work requires a machine to have stronger reasoning capabilities and potential for creating and testing new design solutions. These needs are not met by the conventional machine

• • • • • •

Neural Networks Fuzzy Systems Expert Systems Bayesian Networks Evolutionary Algorithms Case-Based Systems

learning techniques and have prompted our new Figure 4. Typical AI techniques.

developments in machine learning. The novelty of the new method lies in the learning and adaptation processes. These processes are augmented by a feedback system that compares the radio’s actions to its performance to help direct future adaptations. Our cognitive engine works with evolutionary

techniques that provide powerful methods of analyzing, predicting, and discovering new solutions and then using the feedback information to adjust itself intelligently for more accurate analysis and predictions that lead to improved solutions. The system uses a tiered approach to learning and optimization that mixes aspects of different AI techniques together to make a more robust computational intelligence. The machine learning techniques exploit the major benefits of evolution through random searching. The techniques also make skilled use of prior knowledge to direct the adaptation towards areas most likely to contain solutions. The general system topology is a two-layered evolutionary algorithm: the bottom layer discovers new solutions through trial, and the top layer monitors the behavior of the bottom layer as well as the feedback information from the environment. In the case of the cognitive radio, the solutions are new radio parameters. The feedback information consists of performance measurements such as bit error rate, frame error rate, throughput, etc. (see Table 1 for more complete details). The feedback information is then compared to the performance predicted by the first layer algorithm. As the observed performance deviates from the predictions, the top layer genetic algorithm adjusts the bottom layer to compensate for areas with large errors. Such directed adaptation has been studied and previously used in evolutionary algorithm work through techniques like chromosome tagging and templates or in knowledge-based evolutionary techniques [Goldberg, 1989]. We are now applying this technique to a meta-level control over another algorithm to create a highly directed and powerful learning machine. The cognitive machine relies on a sophisticated input/output system; the more inputs received, the more knowledge the radio has, and therefore, the more accurate its modeling and predictions are. The next section discusses a particular sensing method in more detail.

Combining the sensor input and the learning machine with a knowledge base improves the cognitive machine even more, thus mixing traditional AI techniques together to extract the benefits of each one. With a case-base of previously observed inputs and successful combinations of solutions to inputs, the cognitive machine increases its performance and abilities.

Table 1. Sample tabulation of knobs and meters by layer (adapted with permission from Prof. Huseyin Arslan) Layer Meters Knobs (observable parameters) (writable parameters) Source coding MAC Frame error rate Data rate Channel coding rate and type Frame size and type Interleaving details Channel/slot/code allocation Duplexing Multiple access Encryption Transmitter power PHY Bit error rate SINR Spreading type and code Received signal power Modulation type Noise power Modulation index Interference power Pulse shaping Power consumption Symbol rate Fading statistics Carrier frequency Doppler spread Dynamic range Delay spread Equalization Angle of Arrival Antenna directivity Other Computational power Battery Life

CPU Frequency scaling

3. COGNITIVE RADIOS IN PUBLIC SAFETY Our work in cognitive radios comes mostly from the area of public safety since we were originally developing broadband systems for emergency response applications. As the work progressed, we recognized that emergency response teams require fast, reliable, and hands-off

radio equipment. A well-documented public safety concern is that different emergency response teams, even within the same jurisdiction, may have incompatible radio systems impeding coordination of response efforts. Both the issues of reliability and interoperability are addressed here through the advantages provided by a cognitive radio.

3.1 Reliable, Robust, and Hands-off Radios Radio communication is absolutely essential to disaster response and emergency relief efforts: it provides the means to coordinate and disseminate information to a large group effort. In order to focus on their primary task during an emergency, even with an established communications infrastructure, the emergency responders are unable and unwilling to adjust their radios. Our work has been aimed at providing the autonomous and reliable communications required in any situation, including the extreme ones encountered in emergencies. Our initial efforts to apply cognitive radio techniques to help emergency responders culminated with the integration of a broadband channel sounder into commercial Proxim Tsunami® radios. The channel sounder uses radio techniques to sound, or “see,” the radio propagation channel [Cox, 1972]. With these radio eyes, the cognitive engine senses and models the radio channel. This provides it with an understanding of the operating environment that is then useful in determining the proper response to compensate for any deficiencies in the channel. Figure 3 illustrates the behavior of this radio system currently operating at SAIC’s PSIC facility.

Figure 3. Pulse (1) is sent by transmitter (2), corrupted by the channel conditions (3), and received by the sounder (4) for processing.

The channel sounder first transmits a pulse from the base station. As the pulse travels between the radios, it is corrupted. For example, Figure 3 shows a corrupted pulse with many repetitions, which indicates multipath, or multiple reflections of the transmitted signal from objects between the transmitter and receiver. The sounder then samples the incoming signal [Rieser, 2001] and stores the information. The sounder model says a lot about the channel and can help direct adaptation of the radio because of the relationship between the channel and the waveform. As the cognitive radio learns these relationships, it can improve the adaptation speed and quality in future channels with similar properties. A cognitive radio with the channel sounder sensing mechanism can provide reliable communications for emergency responders by recognizing the changing environments and intelligently adjusting its waveform and protocols to correct for the channel conditions. With

communications playing a critical role in the work of emergency responders, cognitive radios offer significant benefits.

3.2 Interoperable Radios Public safety agencies have shown the need for universally interoperable voice and data communications that can support coordinated responses to incidents including a major terrorist attack such as 9/11/2001, the 1993 standoff in Waco, Texas, the 1996 rural Montana arrest of the Unabomber, and local fire and law enforcement operations. In an ideal world, all public safety personnel would carry radios that are compatible in waveforms, protocols, and frequency coverage, and their operators could use them to form and interconnect tactical networks as required in any situation. Spectrum assignments would be flexible, making unused channels available for revenue-earning commercial leases but still allow public safety agencies to reclaim them instantly for emergency use. Meeting these needs under real world economic, political, and commercial constraints requires the development of affordable radios with (a) sufficient frequency and waveform agility to interoperate with any of the existing standards, (b) the capability to serve as a gateway between two or more networks using incompatible standards, and (c) the intelligence to find spectrum, configure themselves, and begin operation in response to simple operator commands. Software defined radios offer (a) and (b), since their operating frequencies and waveforms are software controlled, and switching between standards simply requires running different code. Cognitive radios add the intelligence needed in (c) to SDR. Cognitive radios perform intelligent adaptation with minimal operator involvement and will require little or no retraining of personnel. They offer a powerful but incremental solution to the problems of incompatible and aging communications equipment, multiple equipment

standards, and limited and fragmented budget cycles, funding, planning, frequency coordination, and available spectrum. Ideally, a properly designed cognitive radio can interoperate with everything else that is out there – now. It can interconnect mutually incompatible networks, allowing full interoperability through a system of systems approach.

4. COGNITIVE RADIOS AND SPECTRUM MANAGEMENT Spectrum management is an area of extreme importance and requires innovative technology to help reinvent spectrum policy. Changes in spectrum management are not only strongly suggested, as evident from the President’s Spectrum Policy Task Force [FCC, 2002], but are being actively pursued by the FCC in proceedings like Interference Temperature [FCC, 2003a] and Cognitive Radio [FCC, 2003b]. Currently, spectrum is poorly allocated with much of it wasted and underused while other parts are overcrowded and congested. Cognitive radios provide important benefits realizable through aware, intelligent, and adaptive radios. We now analyze the cognitive radio impact with respect to economics and policy.

4.1. Economic Implications 4.1.1 History of Spectrum Problems Radio spectrum has been allocated by the FCC since the agency was created in 1934. From the beginning, the Commission’s decisions have been criticized as inefficient. The original administrative fiat mechanism awarded frequency rights on a first-come, first-served basis, unless there was more than one applicant. In these cases, comparative hearings were conducted to determine which potential licensee should prevail. The merit of the process was criticized for being based on irrelevant criteria and purely political motivations [Hazlett, 1993]. In 1959,

economist Ronald Coase proposed private rights and market allocation of radio spectrum to the FCC [Coase, 1959]. The FCC began to auction spectrum in 1993, a successful move when evaluated on the criteria of economic efficiency because auctions award spectrum rights to the user who values them most. As new technologies increase spectrum value, spectrum markets can allow ownership to move to the most valuable use. Spectrum allocation inefficiencies are particularly obvious in unlicensed bands where loose property rights invite excessive use and promote interference, a problem known as Tragedy of the Commons. At the same time, restrictive property rights in currently underutilized licensed spectrum create transaction costs that are prohibitive for entrepreneurs with technologies capable of increasing utilization. In these cases, the burden falls on entrepreneurs to prove to the FCC that their technology will increase consumer welfare. This is complicated by the difficulty of measuring increases in consumer welfare. For example, a tradeoff between spectrum usage and interference occurs in block spectrum allocation. The regulator can increase usage by increasing the number of transmitting agents; however, the subsequent interference increases are undesirable. There is no conclusive way of determining the collective optimal level of usage relative to interference.

4.1.2 How Cognitive Radio Resolves these Problems Cognitive radio is a technology with the potential of providing a solution to the problem of efficient spectrum allocation. It addresses both the problem of interference in unlicensed spectrum bands as well as under-use in licensed spectrum. These benefits are the result of the cognitive engine’s ability to perceive external users and adjust its behavior accordingly. It can manage unlicensed spectrum so that additional users are blocked once interference compromises

service quality. The cognitive engine can also be programmed to utilize licensed bands when available without inconveniencing the licensed user. Users can also be assigned priorities so that high value users have priority over low value users, or those whose broadcasts benefit the public good (such as emergency services) can displace other users as needed. Current research attempts to determine a way of increasing utilization in TV broadcast spectrum, where usage is sparse relative to its neighboring wireless bands [Hazlett, 2001]. Prior research has emphasized the value lost by allowing TV broadcast spectrum to remain underutilized. The goal of this research has been to affect policy-making into reallocating the spectrum in a more efficient manner. Cognitive radio has the ability to utilize licensed spectrum without reallocating property rights which gives it a unique position relative to policy.

5. FUTURE BENEFITS FROM COGNITIVE RADIOS This paper presented an overview of cognitive radio impacts on public safety and spectrum management which begins to address the most practical problem confronting cognitive radio: FCC certification and acceptance. The cognitive engine we are building will demonstrate how a prototype cognitive radio can interconnect between different kinds of networks (a digital Project 25 radio system to an analog FM radio system, for example), and support both day-today public safety activities and large-scale incidents and events. Likewise, the cognitive radios provide significant improvements in spectrum utilization. Through both technical and economic analysis, we have presented methods that address the spectrum allocation concerns recognized throughout all levels of government and society. The technical solutions offered here only begin to address the problem of regulatory control. Through both documents [FCC, 2005] and personal communication with the FCC, we

have learned that the commission is excited and supportive of SDR and cognitive radio efforts, but they have yet to find an acceptable and useful way of regulating them. While we hope that one day we will have radios intelligent and capable enough to work with each other without regulatory restrictions, that is a dream for another time. Meanwhile, we have presented a system architecture that includes a policy engine that prevents illegal waveforms from being used on the cognitive radio. The approach is to employ computer engineering techniques of verification to test the waveforms against regulatory policy to ensure compliance [Chaki, 2003] [Dwyer, 2001]. This area in particular holds great promise for technology adaptation as well as significant research possibilities. The benefits of the cognitive radio research presented extend beyond issues in wireless communications. The analysis of spectrum as a resource leads to other areas of economic resource analysis. In public safety, the cognitive radio tools will change and enhance emergency rescue efforts. And finally, new methods and theories in machine learning are leading us to major advancements in computational intelligence for improved understanding and decision making abilities.

REFERENCES C. W. Bostian, S. F. Midkiff, T. M. Gallagher, C. J. Rieser, and T. W. Rondeau, “Rapidly Deployable Broadband Communications for Disaster Response,” Sixth International Symposium on Advanced Radio Technologies (SAFECOM Session), Boulder, CO, Mar. 2004, pp. 87 – 92. S. Chaki, E. M. Clarke, A. Groce, S. Jha, and H. Veith, "Modular Verification of Software Components in C,” Proc. Intl Conf. Software Engineering, 2003, pp. 385 - 395. R. Coase, “The Federal Communications Commission,” Journal of Law & Economics, vol. 2, pp. 1 - 40, 1959. D. C. Cox, “Delay Doppler Characteristics of Multipath Delay Spread and Average Excess Delay for 910 MHz Urban Mobile Radio Paths,” IEEE Trans on Antennas and Propagation, vol. 20, no. 5, pp. 625 - 635, 1972.

M. B. Dwyer, J. Hatcliff, R. Joehanes, S. Laubach, C. S. Pasareanu, H. Zheng, and W. Visser, "Tool-supported program abstraction for finite-state verification,” Proc. IEEE Intl Conf. Software Engineering, 2001, pp. 177 - 187. FCC, “Spectrum Policy Task Force Report,” FCC Document ET Docket No. 02-135, Nov. 2002. FCC, “Establishment of an Interference Temperature Metric to Quantify and Manage Interference and to Expand Available Unlicensed Operation in Certain Fixed, Mobile and Satellite Frequency Bands: Notice of Inquiry,” FCC Documents ET Docket No. 03-237, Nov. 2003. FCC, “Facilitating Opportunities for Flexible, Efficient, and Reliable Spectrum Use Employing Cognitive Radio Technologies: Notice of Proposed Rule Making and Order,” FCC Document ET Docket No. 03-108, Dec. 2003. FCC, “Facilitating Opportunities for Flexible, Efficient, and Reliable Spectrum Use Employing Cognitive Radio Technologies: Report and Order,” FCC Document ET Docket No. 05-57, Mar. 2005. D. E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning, AddisonWesley: Reading, MA, 1989. J. Hawkins, On Intelligence, Times Books: New York, 2004. T. W. Hazlett, “The Political Economy of Radio Spectrum Auctions.” Institute of Governmental Affairs, University of California: Davis, Program on Telecommunications Policy, Jun. 1993, pp. 25 - 28. T. W. Hazlett, “The Wireless Craze, the Unlimited Bandwidth Myth, The Spectrum Auction Faux Pas, and the Punchline to Ronald Coase’s ‘Big Joke’: an Essay on Airwave Allocation Policy,” Harvard Journal of Law and Technology, vol. 14, no. 2, pp. 337 - 469, 2001. J. Mitola, “Cognitive Radio for Flexible Mobile Multimedia Communications,” IEEE International Workshop on Mobile Multimedia Communications, 1999, pp. 3 - 10. National Communications System (NCS), “Open Systems Interconnection – Reference Model (OSI-RM),” FED-STD-1037C: Telecommunications: Glossary of Telecommunication Terms, Aug. 1996. M. Negnavitsky, Artificial Intelligence: A Guide to Intelligent Systems, Addison-Wesley: Harlow, England, 2002. M. K. Powell, “Broadband Migration III: New Directions in Wireless Policy,” Remarks at the Silicon Flatirons Telecommunications Program, Boulder, CO, Oct. 2002.

C. J. Rieser, “Design and Implementation of a Sampling Swept Time Delay Short Pulse (SSTDSP) Wireless Channel Sounder for LMDS,” Masters Thesis, Virginia Tech, Jul. 2001. C. J. Rieser, T. W. Rondeau, C. W. Bostian, and T. M. Gallagher, “Cognitive Radio Testbed: Further Details and Testing of a Distributed Genetic Algorithm Based Cognitive Engine for Programmable Radios,” Proc of MILCOM, vol. 3, Oct. 2004, pp. 1437 - 1443. C. J. Rieser, “Biologically Inspired Cognitive Radio Distributed Genetic Algorithms for Secure and Robust and Networking,” Ph.D. Dissertation, Virginia Tech, Aug. 2004.

Engine Model Utilizing Wireless Communications

T. W. Rondeau, C. J. Rieser, B. Le, and C. W. Bostian, “Cognitive Radios with Genetic Algorithms: Intelligent Control of Software Defined Radios,” Proc of Software Defined Radios, Nov. 2004, pp. C-3 - C-8.

ACKNOWLEDGMENT This work was supported by the National Science Foundation under awards 9983463, DGE9987586, and CNS-0519959.