Cognitive Communications: Distributed Artificial Intelligence (DAI), Regulatory Policy and Economics, Implementation

By David Grace

This book discusses in-depth the concept of distributed artificial intelligence (DAI) and its application to cognitive communications

In this book, the authors present an overview of cognitive communications, encompassing both cognitive radio and cognitive networks, and also other application areas such as cognitive acoustics. The book also explains the specific rationale for the integration of different forms of distributed artificial intelligence into cognitive communications, something which is often neglected in many forms of technical contributions available today. Furthermore, the chapters are divided into four disciplines: wireless communications, distributed artificial intelligence, regulatory policy and economics and implementation. The book contains contributions from leading experts (academia and industry) in the field.

Key Features:

• Covers the broader field of cognitive communications as a whole, addressing application to communication systems in general (e.g. cognitive acoustics and Distributed Artificial Intelligence (DAI)

• Illustrates how different DAI based techniques can be used to self-organise the radio spectrum

• Explores the regulatory, policy and economic issues of cognitive communications in the context of secondary spectrum access

• Discusses application and implementation of cognitive communications techniques in different application areas (e.g. Cognitive Femtocell Networks (CFN)

• Written by experts in the field from both academia and industry

Cognitive Communications will be an invaluable guide for research community (PhD students, researchers) in the areas of wireless communications, and development engineers involved in the design and development of mobile, portable and fixed wireless systems., wireless network design engineer. Undergraduate and postgraduate students on elective courses in electronic engineering or computer science, and the research and engineering community will also find this book of interest.  

• Language: English
• Publisher: Wiley
• Release date: Jul 25, 2012
• ISBN: 9781118360330

Book preview

List of Figures

Figure 1.1 Example of how distributed artificial intelligence is used in a cognitive network

Figure 2.1 Cognitive radio cycle

Figure 2.2 A heterogeneous network

Figure 2.3 Cross-tier interference and intra-tier interference

Figure 2.4 Example of ABSFs in time domain techniques for heterogeneous networks

Figure 2.5 Example of lightly loaded PDCCH for heterogeneous network

Figure 2.6 Example of OFDM symbol muting in time domain techniques for heterogeneous networks

Figure 2.7 Example of consecutive subframe blanking in time domain techniques for heterogeneous networks

Figure 2.8 Example of frequency domain techniques for heterogeneous networks

Figure 2.9 Macrocell and femtocell model as cognitive radio concepts

Figure 2.10 The shaded area satisfying the first constraint

Figure 2.11 The shaded area satisfying the second constraint

Figure 2.12 The shaded area satisfying both constraints

Figure 2.13 The shaded area satisfying the first constraint (a) and the second constraint (b)

Figure 2.14 Conventional transparent relay (a) and transparent relay using cooperative strategy (b)

Figure 2.15 System model of transparent relay using cooperative strategy

Figure 2.16 Three cell model (a) and user allocation of OFDMA packet for FFR and SFR (b)

Figure 2.17 Self-organization cycle

Figure 2.18 Heterogeneous network handover example

Figure 2.19 SINR CDF of the macrocell with femtocells

Figure 2.20 SINR CDF of the macrocell only

Figure 2.21 Average throughput at one sector of centre cell for macrocell only and macrocell with femtocell (uniformly and cell edge deployed)

Figure 3.1 OFDM system block diagram

Figure 3.2 CP insertion in the OFDM symbol by copying the last part in the beginning of the symbol

Figure 3.3 Description of the waterfilling principle. Pmax is the total system power budget and SNR denotes the inverse of the sub-carriers' signal-to-noise ratio

Figure 3.4 Underlay and overlay spectrum sharing schemes

Figure 3.5 Downlink\uplink CR network

Figure 3.6 Frequency distribution of the active and non-active primary bands

Figure 3.7 Example of the SU's allocated power using PI-algorithm

Figure 3.8 Frequency distribution with two active PU bands

Figure 3.9 Achieved capacity versus allowed interference threshold: two active PU bands

Figure 3.10 Total interference introduced to the PU1 versus interference threshold

Figure 3.11 Total interference introduced to the PU2 versus interference threshold

Figure 3.12 Achieved CR versus allowed interference threshold (low): two active bands

Figure 3.13 Frequency distribution with one active PU band

Figure 3.14 Achieved capacity versus allowed interference threshold: one active PU band

Figure 3.15 Achieved capacity versus allowed interference threshold (low): One active PU band

Figure 3.16 An example of the SU's allocated power using the power allocation algorithm

Figure 3.17 Achieved capacity versus allowed interference thresholds

Figure 3.18 Outage probability versus allowed interference thresholds

Figure 3.19 Achieved capacity versus number of SU's

Figure 3.20 Achieved capacity versus per-user power

Figure 3.21 Instantaneous rates over time

Figure 4.1 Frequency response comparison between OFDM and FBMC in one sub-carrier

Figure 4.2 Typical structure of multi-carrier system

Figure 4.3 Frequency response of prototype filter

Figure 4.4 Illustration of amplitude of frequency response in M-band filter banks

Figure 4.5 Polyphase structure of M-band filter banks

Figure 4.6 Basic structure of FBMC transmitter

Figure 4.7 Basic structure of FBMC receiver

Figure 4.8 System diagram of a filter bank-based multi-carrier (FBMC) system

Figure 4.9 The scheme of proposed multi-stage DFTFB

Figure 4.10 (a) Brief structure of proposed multi-stage DFT filter banks (b) Structure of DFT filter banks with modulation component in stage L

Figure 4.11 TV bands division in IEEE 802.22 WRAN and fractional bandwidth usage

Figure 4.12 Architecture of two stage DFTFB (TS-DFTFB)

Figure 4.13 (a) Detection performance of 32 band t-DFTFB and TS-DFTFB (b) Number of multiplications of 32 band t-DFTFB and TS D TFB

Figure 4.14 Detailed power estimator module in Figure 4.12

Figure 4.15 (a) Threshold and AWGN curves, where the initial threshold is varying with respect to the actual noise power. (b) Threshold and AWGN curves, where the initial threshold is much bigger than the noise power. The values of adaptive parameters are

Figure 4.16 The usage of a spectrum band for second user

Figure 4.17 Block diagram of transform decomposition

Figure 4.18 Block diagram of decimation transform decomposition

Figure 4.19 The number of complex multiplications that CTR2-FFT, conventional TD (CTD) and our proposed method (DTD) need under the hypothesis that was mentioned previously

Figure 4.20 The number of complex multiplications that DTD needs at different distribution degrees

Figure 4.21 Symbol Error Rate versus the number of impulse noise for different auxiliary sub-carrier assignation schemes, M = 256, m = 20, SNR = 20 dB

Figure 5.1 Connectivity in a CRN composed of three primary (squares) and eight secondary (circles) nodes. Channels in use by primaries and those available to secondaries are indicated in brackets below the nodes. Solid lines indicate nodes that are connected. The dashed lines between nodes 5 and 7 and between nodes 3 and 8 indicate that even though these node pairs are within radio range of each other, they are not connected as they do not have a common available channel

Figure 5.2 An ad hoc CRN scenario with three primary (squares) and 16 secondary nodes (circles) generated using link model parameters Dp = 0.5, Ds = 0.3, κ = 1. The clustering solution is illustrated in the bottom plot. Large circles denote the clusterheads (a) Primary transmission ranges and CRN connectivity. (b) Clusters formed by distributed AP

Figure 5.3 Impact of Δ on the number of clusters formed by the distributed AP technique

Figure 5.4 Effect of the number of AP iterations on the number of clusters formed

Figure 5.5 Clustering efficiency of the distributed AP and centralized greedy techniques for single channel ad hoc networks

Figure 5.6 The node connectivities for a typical CRN scenario with 100 randomly deployed nodes in a unit square simulation area

Figure 5.7 Cooperative sensing nodes chosen by each of the various selection techniques. Reporting nodes are indicated by solid circles. Brightness denotes the probability of detection (Qd) at each location for a false alarm rate (Qf) of 1% (a) All nodes reporting (b) AP with 38 reporting nodes (c) K-means with 38 reporting nodes (d) Random with 38 reporting nodes

Figure 5.8 Detection performance of the various sensor selection techniques

Figure 5.9 Effect of the number of reporting nodes on the detection performance

Figure 6.1 Typical neural network structure [12]

Figure 6.2 The inserted data sample x affects its BMU and its neighbourhood. The black and the grey dots represent state of the map before and after the input of the data sample, respectively, while the arrows stand for the direction and the intensity (length of the arrow) of the adjustment during the training [30]

Figure 7.1 Cognitive wireless mesh networking (CogMesh) scenarios

Figure 7.2 Cluster-based network formation in CogMesh

Figure 7.3 Reinforcement learning

Figure 7.4 Performance, when : Impact of the temperature to expected rewards achieved by SU 1

Figure 7.5 Performance, when : Impact of the temperature to expected rewards achieved by SU 1

Figure 7.6 Performance comparison between the proposed algorithm and the system's optimum

Figure 7.7 The expected rewards of the SU's versus the PU's behavior factor

Figure 7.8 Channel model of the primary users

Figure 7.9 Cognitive radio network with N = 4 and M = 5 at time slot k. Collision occurs when more than one secondary user transmits over the same free channel

Figure 7.10 Strategy dynamics of Algorithms 1 and 2

Figure 7.11 Strategy dynamics of Algorithms 1 and 2 with different initial values of and

Figure 7.12 Strategy dynamics of Algorithms 1 and 2 with the same belief parameter

Figure 7.13 Comparison of the accumulated utilities corresponding to different OSA schemes

Figure 7.14 Comparison of the achieved fairness index of different OSA schemes

Figure 8.1 The reinforcement learning model in a cognitive radio scenario

Figure 8.2 Reinforcement learning-based spectrum sharing algorithm

Figure 8.3 Point-to-point architecture

Figure 8.4 Channel usage at (1) Event 50, (2) Event 100, (3) Event 500, (4) Event 1000

Figure 8.5 Cumulative distribution function of system blocking probability at discrete points over the service area

Figure 8.6 Cumulative distribution function of system dropping probability at discrete points over the service area

Figure 8.7 Algorithm flowchart

Figure 8.8 Cumulative distribution function of system blocking probability of transmitter and receiver pairs

Figure 8.9 Cumulative distribution function of system dropping probability of transmitter and receiver pairs

Figure 8.10 Average values of Ustd through thousands of events

Figure 8.11 Average blocking probability with different preferred channel weight thresholds

Figure 8.12 Average dropping probability with different preferred channel weight thresholds

Figure 8.13 Percentage of activation with different preferred channel weight thresholds

Figure 8.14 Average blocking probability with different size of preferred channel set

Figure 8.15 Average dropping probability with different size of preferred channel set

Figure 8.16 System blocking probability of uniform random exploration at different offered traffic levels

Figure 8.17 System dropping probability of uniform random exploration at different offered traffic levels

Figure 8.18 Exploration costs (number of trials required per task) for a learning agent

Figure 8.19 System blocking probability at different offered traffic levels

Figure 8.20 System dropping probability at different offered traffic levels

Figure 8.21 Percentage of activation in exploitation at different offered traffic levels

Figure 9.1 CPT of i-th configuration

Figure 9.2 Test Case 1: Scenario 1: No prior knowledge of the system capacity under the specific configuration [1]

Figure 9.3 Test Case 2: Scenario 2: With prior knowledge of the system capacity of being 6 Mbps under the specific configuration [1]

Figure 9.4 Generalized scheme of the under question NN-based pattern [4]

Figure 9.5 Performance of the ‘winning’ scheme with respect to the training (known) data set [4]

Figure 9.6 Performance of the ‘winning’ scheme with respect to validation (unknown) data set [4]

Figure 9.7 MATLAB Data File: the number of the first line refers to the number of the input variables, here equal to 5 (RSSI, Input PacKeTS, Output PacKeTS, Input BYTES, Output BYTES), and the last column refers to the bit rate (used only for labelling reasons). Each Line is a data sample and each column is a different input variable [9]

Figure 9.8 SOM visualizations: (a) only the label with the most instances appear in the cells, (b) all labels that have at least one instance appear in the cell and (c) SOM of (b) is supplemented with the number of instances that each label has in the cell [9,10]

Figure 9.9 Comparative diagram of the predicted (solid line) and measured (dotted line) values of bitrate [10]

Figure 9.10 Inference of user preferences

Figure 9.11 View of CTMS implementation used for the derivation of results: (a) Retrieval of profile information; (b) Collection of user feedback [16]

Figure 9.12 User feedback for professional user role and high, medium and low QoS

Figure 9.13 Adapted conditional probabilities for Utility Volume in professional context given (a) high, (b) medium and (c) Low QoS

Figure 9.14 Network topology which was used during the simulation

Figure 9.15 SOM depicting the congestion levels (0 in blue labels when the link can serve all the traffic, 1 in lighter labels when some packets drop but yet is not treated as a congested link and 2 in darker labels when the link is expected to become congested) of the link under question

Figure 10.1 An illustration of collaborative spectrum sensing and coalition

Figure 10.2 Illustration of channel selection

Figure 10.3 Spectrum access success probabilities for different P_rec

Figure 10.4 Performance gain of adaptive branching probability with a bandit algorithm

Figure 10.5 An illustration of the evolution of default channel

Figure 10.6 Three realizations of user proportion evolution

Figure 10.7 The evolution of user proportion with different parameters

Figure 10.8 Upper bound of user proportion

Figure 11.1 Spectrum allocation in the United Kingdom prior to digital switchover

Figure 11.2 Metrological radar stations in Europe [18]

Figure 11.3 Time/frequency (right) and spatial opportunity for interweaving secondary transmissions in primary spectrum [1]

Figure 11.4 A typical interference margin/temperature at primary receiver creates spectrum opportunities for underlay sharing by secondary systems [38]

Figure 11.5 An illustration of the overlay approach for secondary spectrum sharing where cognition of primary signals at secondary transmitter enables interference cancellation at primary receiver [38]

Figure 11.6 The concept of a spectrum quasi-continuum consisting of elementary sub-channels that could be dynamically pooled by cognitive radio in response to user requirements

Figure 12.1 UK UHF spectrum after the completion of the digital switchover (courtesy Neul)

Figure 12.2 TVWS potential range due to lower frequency and higher power in comparison with WiFi. Tx power = 4W EIRP, frequency = 700 MHz, Tx antenna 25 m, Rx antenna 4 m

Figure 12.3 Typical output of a geolocation database (BT) showing free channels at a given location

Figure 12.4 Hidden node problem of cognitive radio [8]

Figure 12.5 Probability of detection of a DVB-T signal is plotted against the signal-to-noise ratio for several sensing algorithms. Arrow marks the SNR ratio that corresponds to Ofcom's requirement [11]

Figure 12.6 Coverage map of DTT transmitter located in Guildford, Surrey [27]

Figure 12.7 Aggregate interference levels at the edge of DTTV coverage area plotted as a function of total service area for different deployment densities. The keep out distance is 30 km. Conservative and liberal regulatory caps to interference are shown as thick dark lines

Figure 12.8 Aggregate interference levels at the edge of DTTV coverage area plotted as a function of total service area for different deployment densities. The keep out radius is 70 km. Conservative and liberal regulatory caps to interference are shown as thick dark lines

Figure 12.9 White space roadmap (courtesy Cambridge Consultants, April 2010)

Figure 12.10 Usage example of the IEEE 802.19af in TVWS frequencies [30]

Figure 12.11 UHF channels availability map for cognitive access to TVWS in Germany (left panel) and Sweden, computed for WSD with 20 dBm transmit power and 1.5 Tx height [38]

Figure 12.12 Left panel shows UHF channels availability map for secondary spectrum access to TV white spaces in the UK [41] Results are calculated using Ofcom's database of transmitters, Dark: < 50 MHz, Light > 150 MHz. Right panel shows population-weighted cumulative distribution

Figure 12.13 TVWS channels available for low-power cognitive access in Central London [14]

Figure 12.14 Home distribution using a TVWS system

Figure 12.15 TVWS systems could be used for micro-/metrocell backhaul

Figure 12.16 Rural not-spot coverage with TVWS

Figure 12.17 TVWS for rural broadband: home equipment

Figure 12.18 BT trial on the Isle of Bute

Figure 12.19 Terminal to terminal ‘hopping’ with TVWS (e.g. different frequencies)

Figure 12.20 A 1 km² area of London (Bayswater). The shading shows the coverage possible when 20% of premises have an indoor transmitter for WiFi or LTE in TVWS spectrum [46]

Figure 13.1 Illustrative example showing the data rate requirement (dotted line) and available throughput due to received signal to interference and noise ratio (SINR) (solid line), between indoor and outdoor scenarios for a cellular base station

Figure 13.2 A joint macro-femtocell deployment architecture

Figure 13.3 Macrocell to femtocell interference variations with FAP distance for different BS transmission powers

Figure 13.4 Femtocell to femtocell interference variations together with safety distance

Figure 13.5 Impact of wall penetration loss on received signal

Figure 13.6 Example showing various femtocell deployments: (a) overlapped, (b) overlapped but not interfering, and (c) non-overlapped

Figure 13.7 Interference scenario in joint macro-femto deployments

Figure 13.8 FFR-based resource sharing in joint macro femto deployments

Figure 13.9 An example of the graph colouring problem for 5 FAP

Figure 13.10 Logical diagram showing a virtual clustered femtocell network system

Figure 13.11 FAP deployment scenarios: (a) before cluster formation, (b) after clustering (applying VCF), and (c) the non-clustered system (NCS) where the shading represents channels of a VCC

Figure 13.12 Performance comparison between clustered and non-clustered network for various FAP deployments

Figure 13.13 SINR performance comparison for clustered and non-clustered systems at different deployment densities

Figure 13.14 Spectral efficiency performance comparison for clustered and non-clustered system at different deployment densities

Figure 13.15 Coverage optimization to minimize the interference for two co-located femtocells: (a) before and (b) after optimization

Figure 13.16 (a) Before load balancing, (b) after load balancing, and (c) joint load balancing and coverage optimization

Figure 13.17 Interference scenario for SISO-based omni-directional and MIMO-based directional transmission

Figure 13.18 The generalized Enhanced FFR (EFFR) scheme

Figure 14.1 Frequency-dependent attenuation and noise level for different transmission distances (spreading factor k = 1.5)

Figure 14.2 The optimal carrier frequency and the corresponding product of attenuation and noise versus the propagation distance

Figure 14.3 A simple scenario

Figure 14.4 Bandwidth of underwater acoustic channel

Figure 14.5 Number of collisions versus number of nodes

Figure 14.6 Energy consumption versus time slot length

Figure 14.7 Number of retransmissions versus time slot length

Figure 14.8 Delivery delay versus time slot length

Figure 14.9 Throughputs versus time slot length

Figure 15.1 The LNA is a crucial component of receivers, as it should provide gain and have a low NF to keep receiver NF low enough, while at the same time it should be very linear. The spectra are drawn on a dB-scale, while the time-signals are drawn on a linear scale. (a) The LNA mitigates the effect of noise added by the following stages of the receiver. (b) Nonlinearity in the LNA distorts the spectrum, and hence increases BER

Figure 15.2 State-of-the-art ADC-performance (a) Currently, no ADC achieves a DR of 100 dB and a BWof 6 GHz. (b) A 2 times higher bandwidth-resolution product requires roughly twice the power. (from [4] which is regularly updated)

Figure 15.3 Our mathematical abstraction of a transmitter and receiver

Figure 15.4 A BPF transfer characteristic and terminology

Figure 15.5 Example transfer of a SAW-filter for the 850 MHz GSM-band

Figure 15.6 The goal of a receiver is to amplify the weak signal to be demodulated and to suppress other signals

Figure 15.7 A sub-sampling receiver performs frequency conversion and sampling in one step, but requires a dedicated high-Q filter for each band. It suffers severely from noise folding

Figure 15.8 A heterodyne receiver performs a frequency conversion on the signal to be demodulated in order to facilitate further processing

Figure 15.9 A block schematic showing the possible DSP-steps in a heterodyne receiver to obtain

Figure 15.10 The position of flo with respect to fc determines how well the image and interference close to the desired signal can be suppressed

Figure 15.11 The zero-IF receiver rejects the image by using a complex frequency translation. For zero-IF, fif = 0 and the image is the signal itself

Figure 15.12 Some possible implementations for creating I and Q baseband signals

Figure 15.13 Two main architectures exist to combine the I and Q signals to a single real analogue output signal where the image is rejected. (a) Hartley architecture (b) Weaver architecture

Figure 15.14 Image frequency suppression as a function of IQ-mismatch. The phase error is ϕ and the gain error is 10 log10(1 + ε)

Figure 15.15 Wideband matching can be obtained with different methods (a) Using a resistor. (b) Using feedback. (c) Using a common-gate amplifier

Figure 15.16 The noise-cancelling LNA of [19]. The signal is amplified, and the noise from the transistor (modelled as a current source) is cancelled by proper choice of the parallel amplifier gain A

Figure 15.17 Using a good SA, the effect of the receiver linearity on each vacant channel can be calculated, allowing the selection of a channel with achievable requirements. In the scenario shown here, with three large primary signals, only channels 3, 4, 8, 9, 16, and 17 will be usable

Figure 15.18 High linearity can be obtained by keeping voltage swings low as long as possible

Figure 15.19 Harmonic downmixing is a fundamental problem when RF-filtering is lacking

Figure 15.20 Appropriate weighting of different square wave LO-phases yields a closer approximation to a sine wave, effectively removing tthird, fifth, eleventh, thirteenth, (and so on) harmonics, leaving the seventh and ninth harmonics as the first uncancelled ones

Figure 15.21 Beamforming provides a means for spatial filtering to suppress interferers and lowers the NF by providing passive gain

Figure 15.22 Applying complex weight to signals can be implemented in several ways

Figure 15.23 The use of a rational function to approximate the sine function allows complex weights to be easily generated in the analogue domain, thus reducing DR-requirements further on in the analogue receiver [31]

Figure 15.24 A tuneable BPF can be implemented as the cascade of a downconversion mixer, LPF, and upconversion mixer, with a surprisingly simple circuit implementation. (a) BPF implemented as LPF with down/upconversion (b) Straightforward implementation (c) Using a shared resistor and removing redundant switches

Figure 15.25 Measurements of the 65 nm CMOS implementation of [34] (the circuit shown in Figure 15.24c)

Figure 15.26 The quad-band receiver of Broadcom [13] extensively uses tuneable BPFs to implement a SAW-less receiver

Figure 15.27 An LC-oscillator occupies a significant portion of chip area. (a) Typical circuit schematic. (b) Circuit layout

Figure 15.28 The bimodal LC-oscillator of [1] and frequency coverage. (a) Schematic. (b) Frequency coverage

Figure 15.29 Block diagram of a standard transmitter

Figure 15.30 The Kahn transmitter separates the phase and envelope of the baseband signal to allow the use of a high-efficiency nonlinear PA

Figure 15.31 Predistortion is a widely applied technique to linearize transmitters

Figure 15.32 The DDRF-architecture as proposed by [41] combines most of the analogue functionality of a direct-conversion transmitter in a single block (1) Architecture. (2) Implementation of DRFC-block

List of Tables

Table 2.1 Six different scenarios internetworking between 3GPP and WLANs

Table 2.2 Transmission power for different cell types

Table 2.3 Stand-alone channel sensing

Table 2.4 Cooperative channel sensing

Table 2.5 Timing for the transparent relay in IEEE802.16j

Table 2.6 Timing for the transparent relay with cooperative strategy

Table 2.7 Simulation configuration

Table 3.1 Complexity comparison

Table 4.1 Coefficients of the prototype filter

Table 6.1 Organization of the basic information elements (for arbitrary network i) on which the cognitive mechanisms are based

Table 8.1 Weighting factor values

Table 8.2 Simulation parameters

Table 8.3 Simulation parameters

Table 8.4 Simulation parameters

Table 9.1 Possible values of the under investigation parameters [4]

Table 9.2 Values of the predefined parameters

Table 9.3 ‘Winning’ test case

Table 9.4 Values of the batch training algorithm for the test case with the best performance [10]

Table 9.5 Instance of the monitoring procedure for learning user preferences

Table 9.6 Variables that were/could be used for the tests

Table 10.1 Comparison between cognitive radio and electronic commerce

Table 10.2 Differences between the recommendation propagation and epidemic propagation

Table 11.1 Four different scenarios for dynamic spectrum access

Table 11.2 Necessary conditions for secondary spectrum access in various regulatory regimes

Table 12.1 Ofcom's proposed parameters for licence-exempt access to TVWS using sensing and geolocation database methods

Table 12.2 Possible applications for TVWS spectrum

Table 13.1 The network environment parameters used in all simulations

Table 13.2 Comparison of the three different access mechanisms

Table 13.3 Summary of the various interference scenarios in a joint macro-femtocell deployment

Table 14.1 The feasible bandwidth Btx(d) corresponding to propagation distance d

Table 15.1 CR requirements set by different authorities assuming mobile devices that rely on spectrum sensing

About the Editors

David Grace is Head of Communications Research Group and a Senior Research Fellow within the Department of Electronics at the University of York. He is also a Co-Director of the York-Zhejiang Lab on Cognitive Radio and Green Communications, and a Guest Professor at Zhejiang University. He received his PhD from University of York in 1999, the subject of his thesis being Distributed Dynamic Channel Assignment for the Wireless Environment. Current research interests include cognitive communications, including cognitive radio and cognitive networks, specifically applying distributed artificial intelligence to resource and topology management to improve overall capacity; cognitive green radio; architectures for beyond 4G wireless networks; dynamic spectrum access and interference management. He is currently a co-investigator of the FP7 BuNGee project dealing with broadband next generation access, and recently he was the principal investigator of a UK MOD project on Cognitive Routing for Tactical Ad Hoc Networks.

In 2000, he jointly founded SkyLARC Technologies Ltd, and was one of its directors. From 2003–2007 he was the technical lead for the 14-partner FP6 CAPANINA project. He is an author of over 160 papers, and a co-author on Broadband Communications via High Altitude Platforms, also published by John Wiley & Sons, Ltd. From 2005–2009 he was COST 297 WG1 Chair which dealt with radio communications for high altitude platforms. He currently chairs the Worldwide Universities Network Cognitive Communications Consortium (WUN CogCom), which has members from more than 90 organizations worldwide, and is a member of COST IC0902. He is the WUN CogCom Liaison Chair for IEEE Committee on Cognitive Networks, and is a founding member of the new IEEE Technical Sub-Committee on Green Communications and Computing (GCC). In 2013, he will be an IEEE ICC Symposium Co-Chair: Cognitive Networks Track.

Honggang Zhang is a Full Professor of Department of Information Science and Electronic Engineering as well as the Co-Director of York-Zhejiang Lab for Cognitive Radio and Green Communications at the Zhejiang University, China. He is an Honorary Visiting Professor of the University of York, UK. He received the PhD degree in Electrical Engineering from Kagoshima University, Japan, in March 1999. From October 1999 to March 2002, he was with the Telecommunications Advancement Organization (TAO) of Japan, as a TAO Research Fellow. From April 2002 to November 2002, he joined the TOYOTA IT Centre. From December 2002 to August 2004, he was with the UWB Research Consortium, the Communications Research Laboratory (CRL) and the National Institute of Information and Communications Technology (NICT) of Japan. He was the principle author and contributor for proposing DS-UWB in IEEE 802.15 WPAN standardization task group. From September 2004 to February 2008, he has been with CREATE-NET (Italy), where he lead its wireless teams in exploring Cognitive Radio (CR) and UWB technologies while participated the European FP6/FP7 projects (EUWB, PULSERS 2). Dr. Zhang serves as the Chair of Technical Committee on Cognitive Networks (TCCN) of the IEEE Communications Society (ComSoc). He was the founding TPC Co-Chair of CrownCom 2006 as well as the Steering Committee Member of CrownCom 2006–2009. He was the Co-Chair of IEEE Globecom 2008 Symposium. In the area of green communications, Dr. Honggang Zhang was the Lead Guest Editor of the IEEE Communications Magazine special issues on ‘Green Communications’. He was the General Chair of IEEE/ACM GreenCom 2010 (2010 IEEE/ACM International Conference on Green Computing and Communications) and the Co-Chair of the IEEE International Workshop on Green Communications (GreenComm 2010–2011) in conjunction with IEEE ICC/Globecom. He is the co-author/editor of the book Green Communications: Theoretical Fundamentals, Algorithms and Applications (CRC Press).

Preface

Cognitive Communications promises to revolutionize the way wireless communication devices and networks behave through ‘intelligent’ assignment of communication resources and operation. Much of the discussion within the research community today is on the narrower subject of cognitive radio, but what we hope to demonstrate with this book is a wider perspective.

Cognitive communications has its history in the early adaptive/dynamic channel assignment schemes that were used to assign, allocated radio spectrum to different devices, which were particularly popular in the early to mid 1990s. These schemes, especially in distributed form, exhibited many of the features we see in cognitive radio schemes put forward today, namely the ability to sense or be aware of the radio spectrum environment, and based on the outcome of the this sense select the most appropriate spectrum (or channel) to use. Such techniques are now widely used in short range systems, for example DECT (Digital Enhanced Cordless Telecommunications) and IEEE 802.11 (WiFi). Parallels with these early technologies are not often drawn, with many researchers instead choosing to specify the origin of the field with cognitive radio, a phrase coined by Dr Joseph Mitola III in 1999. His real contribution to the field was the incorporation of Distributed Artificial Intelligence (DAI), which he used as a way of learning about the radio environment and then acting on the findings, thereby giving devices even more flexibility and autonomy.

We now see cognitive communications, especially in the form of cognitive radio, applied to the distributed selection of the radio spectrum, which is put forward as a way of overcoming spectrum shortages seen by many, due to command and control regulation. Such regulation permits a primary user to have sole right to an allocation of spectrum within a specific geographical area (often on a country or at least region basis). Today, some radio regulators such as Ofcom in the UK and FCC in USA are ‘cognitive friendly’, with the understanding that by allowing more flexibility in how radio spectrum is assigned, coupled with intelligence or at least spectrum awareness and the ability to act and react, could potentially significantly increase the efficient utilization of spectrum. Studies have shown up to 90% of the radio spectrum might be unused at a particular time and geographical area, with conventional techniques. Early suggestions for use include the TV white space spectrum, where cognitive secondary devices share the radio spectrum with the primary TV systems, and also more efficient use of certain unlicensed spectrum bands.

Over the next few years one can expect to see the field grow even further, spurred on by various practical use cases, including the use of TV White Spaces in particular. We can also see the field widening to include application of cognition to other areas of communications, for example cognitive networks and cognitive acoustics, even its application to control of the propagation environment in smart buildings. One can also see cognition being applied to ‘green’ radio for energy efficiency improvement. The ability to be ‘smart’ should deliver significant energy savings. This especially includes the development of power efficient spectrum assignment, instead of the pursuance of ever higher spectral efficiencies, achieved through high order modulation schemes, where transmissions are artificially constrained in bandwidth, requiring higher power transmissions. Instead cognitive devices will have the ability to exploit excess bandwidth available locally to operate with much more power efficient low order modulation. Such techniques are likely to be readily exploited alongside cognitive topology management, where traffic is rerouted to optimize the power consumption of devices and networks, allowing underused and hence often, energy inefficient devices, to sleep.

This book has emerged out of the activities of the WUN Cognitive Communications Consortium (WUN CogCom) – www.wun-cogcom.org. A research discussion forum designed to bring together researchers from the different disciplines of wireless communications, artificial intelligence, regulation and economics. WUN CogCom was established in January 2009 and now has members from over 90 organizations. The editors and lead authors of the book are all members, and it was felt that this opportunity to write a book in this area was a timely way to disseminate the latest thinking from a subset of its members. Although officially classed as an edited book, it is hoped that through tight selection and control of its contents, coupled with strict editing, the book is comparable in style to authored books often seen in the technical literature. The editors and authors, many of them leading experts, are all highly active in this area, and regularly participate in related activities be they research projects, practical implementations, or regulatory/standards contributions. When writing the material we made every effort to suitably reference other publicly available information sources such as journal and conference papers, technical reports and recommendations from various international bodies. It is recommended that these be used for an even more detailed treatment of a specific subject.

The book is aimed at serving as a reference book and it is our hope that it will enthuse a new generation of researchers and PhD students to take up this exciting research area, as well as providing informative advice to motivate the existing research, regulatory and business communities to take forward the state-of-the-art in new ways.

The book is structured in five parts and provides a comprehensive overview of the state-of-the-art of cognitive communications and its key enabling technologies. The parts of the book are of differing lengths intentionally, which has allowed us to place greater emphasis on those areas which we feel are most under-represented in the literature today and those that will be of greater importance in the years to come. The first part of the book provides a short introduction to the area of cognitive communications. This is followed by the wireless communications part where we discuss key wireless aspects of the field. In Part Three we discuss in detail the application of Distributed Artificial Intelligence, and how it can be applied in different forms to communications systems. Part Four examines the current regulatory thinking behind the application of cognitive communications, particularly the latest initiatives of applying cognitive radio to TV White Space. The final part of the book addresses implementation aspects. We look at several examples of proposed application of cognitive communications, from its more conventional application to TV white space, electronic device implications, to the novel subject area of cognitive acoustics.

Although overall the book is edited by us, and we also contribute as authors, it would not have been possible to publish a book of this quality and breadth without the other authors contributing to each chapter. We are also very grateful to other long-time collaborators in several projects and WUN CogCom in general, for their contributions, guidance and valuable advice.

Finally we would like to thank the John Wiley & Sons, Ltd editorial team, who showed a lot of patience, enthusiasm and support during the preparation of this book, especially Susan Barclay and Anna Smart.

David Grace and Honggang Zhang

York, UK and Hangzhou, China

Part I

Introduction

Chapter 1

Introduction to Cognitive Communications

David Grace

Department of Electronics, University of York, Heslington, UK

1.1 Introduction

Communication devices today are becoming ever more sophisticated and diverse, delivering a plethora of new services and applications. The last hop to the end user, a person or device, is increasingly being delivered wirelessly. This sophistication brings with it complexity, making conventional approaches to organization, implementation and regulation increasingly inadequate.

This is seen especially in the case of usage of the radio spectrum, which has manifested itself as a perceived shortage of spectrum, but this shortage is mainly due to inadequate command and control regulation, and conventional technical understanding – studies have shown up to 90% of the radio spectrum remains idle in any one geographical location.

The rapid improvements in functionality will come to an end if standard approaches to communications delivery are not radically updated. This presents a global challenge that impacts not only on device manufacture, software and firmware, but also on changes to radio regulation, business models and economics. The key to the next revolution in communications delivery is the application of distributed artificial intelligence (i.e. cognition) to the communications devices. This will enable intelligent local decisions to be made on network routing, and spectrum and resource usage, based on interaction with other devices and the local environment. Such decisions can take into account mixed systems and applications, and even devices that break the rules.

Two fields are already emerging: cognitive radio (CR), which deals with the intelligent assignment and use of the radio spectrum; and cognitive networking, which deals with the intelligent routing of information through a network, taking into account various local constraints. However, this application of distributed artificial intelligence can be extended to other areas in communications that today rely on fixed-rule adaptivity, allowing for the first time, flexible changes to complex varying local circumstances.

Thus the fields of cognitive radio and cognitive networks represent the tip of the iceberg when it comes to how distributed artificial intelligence can be used in communications systems. This is the major reason why we use the more general term Cognitive Communications as the main title of this book, so as not to limit ourselves to this somewhat blinkered vision of the application of cognition.

Cognitive Communications brings with it many fundamental challenges, given its breadth and multi-disciplinary nature, taking elements from the main established areas of:

Wireless communications,

Distributed artificial intelligence,

Regulatory policy and economics,

Implementation.

These areas in themselves derive from the disciplines of electronic engineering, computer science, and economics, creating a complex challenge of how to further this new field. Given the centre of gravity of the research and development today, this has only been partially successful. The vast majority of research is focused on the wireless communications aspects of cognitive radio, including spectrum sensing, dynamic spectrum access, with a tiny minority focused on application of distributed artificial intelligence. Regulatory aspects are largely restricted to application of cognitive radio to the TV White Space (TVWS) bands. The purpose of this book is to try and rebalance and reprioritise this research in the forthcoming chapters, to highlight their contributions to the wider field of Cognitive Communications, and to thereby encourage the existing and future generations of researchers to think further outside of the box; to investigate these new exciting challenges and opportunities that this new way of thinking brings.

In the remaining part of this chapter we delve more deeply into this new way of thinking; we place Cognitive Communications within a wider historical context. We discuss the key components of Cognitive Communications, and finally we provide a short overview of the rest of the book, illustrating how each chapter plays its part in characterizing the different areas of the subject.

1.2 A New Way of Thinking

At the core of this new way of thinking is the use of distributed artificial intelligence, where individual agents, but with reference to other agents, make decisions on their next action(s). These local actions replace centralized control or fixed rules, with the aim of better exploiting resources, or controlling behaviour, based on the local environment. What part of communications system or sub-system constitutes an ‘agent’ is open for definition and study, but in most cases a communication node is considered as a single agent, but there is no reason why an agent cannot be a collection of nodes, or even different parts of a radio system can operate as individual agents, depending on the scale of operation.

Intelligence can be considered as the ability to act appropriately in an uncertain environment, where an appropriate action is one which increases the probability of success, and success is the achievement of behavioural sub-goals that support the system's ultimate goal; that is order is created, rather than anarchy. Intelligence is one of the prerequisites to autonomy. The study of intelligence, learning, and reasoning has been around for a number of years, but it is only now that concepts such as reinforcement-based learning, game theory, evolutionary computation and neural networks are being actively applied to heterogeneous cognitive radio-based systems. With the exception of game theory, historically this work has been focused on centralized homogeneous schemes, which aim to optimize channel usage for a particular configuration [1–3]. Here, in this book we mainly focus on fully distributed techniques. There have been some recent activities using reinforcement based learning [4] and game theoretic approaches [5–8]. However, this has still only been applied to the cognitive radio area of Cognitive Communications.

A significant portion of this book focuses on distributed artificial intelligence strategies using a multi-agent system mapped directly on to the nodes of a heterogeneous wireless system. As we shall see, agents often employ negotiation strategies, such as auctions, to achieve local and global goals. With multi-agent systems the emergent properties of self-organization, robustness, adaptivity and tolerance arise naturally to a wide variety of disturbances (e.g. from interference-, failure-, and change-related issues).

In order for such a distributed set of agents to reach successful decisions they need to interact with each other and the wider environment. Figure 1.1 shows how different agents interact within a cognitive network environment. Two loops are important: the Action and Sensing loop, and the Reasoning and Learning loop. Dealing first with the outer loop, each node observes, or takes in inputs from the environment, for example context of operation, traffic, interference level, primary user operation, which are then processed. Constraints are then applied and this information is then processed by the Reasoning and Learning loop where the information is further processed, taking into account historical context and learned behaviour. A decision is made as to what action to take next, for example a frequency resource to use is selected. The action is then carried out which causes the environment to change. For example a new transmission takes place on a particular frequency band, causing the interference level on the frequency to change. This change in the environment is observed by all the other entities in the system who may in turn choose to react and change their behaviour. This cycle is repeated until there are no further changes to the environment – in practice there is constant change, as it is likely that some outside stimuli, for example node mobility, new message, will cause the environment to change. The inner loop, Reasoning and Learning Loop, is probably the area that requires the most further research and development, and is ultimately responsible for giving the agent (and the loops collectively in each agent, that is the system) its intelligence. This is made up of reasoning and learning. Traditionally dynamic systems and dynamic spectrum assignment make use of reasoning – the application of fixed rules to decide the course of action. An example of this could be ‘to pick the best channel’. In Cognitive Communications, such strategies can be influenced by learning (e.g. machine learning), which may be used to take into account historical information, and/or manipulate competing parameter inputs taking into account behaviour of other nodes in the system. This provides a greater degree of flexibility, allowing the behaviour of the agent/node or system to adapt to changes in the system. A good example of this is shown in a later chapter, where users (transmitter and receiver pairs) can learn over time to avoid one another's transmissions.

Figure 1.1 Example of how distributed artificial intelligence is used in a cognitive network.

Significantly, more discussion is given on how different forms of learning can be used in different situations later on in this book.

1.3 History of Cognitive Communications

To obtain a complete picture of the history of Cognitive Communications, one has to delve back into the little-known research done on distributed dynamic channel assignment in the early 1990s [9–11]. This dealt with the dynamic assignment of radio resources based on fixed rules, where all nodes operated the same algorithm. The relevant research reached maturity in the late 1990s. Early applications were military-based, and largely for ad hoc networks, and used to ensure the radio spectrum could be (re)assigned to combat enemy threats and jamming [12]. At the end of the 1990s, with the use of the 2.4 GHz unlicensed band, protocols such as IEEE 802.11, and Bluetooth adopted dynamic techniques based on listen before talk strategies, along with DECT at 1.9 GHz, for indoor use.

In 1999 Joseph Mitola III, suggested ‘cognitive radio’ be developed, where artificial intelligence was used to control radios [13, 14], allowing them to dynamically access the radio spectrum with the appropriate protocol, taking into account context and usage information. Refinements by Simon Haykin [15], suggested new protocols that the techniques could be applied to dynamic spectrum access, now considered as cognitive radio. These two major innovations have now made the research field mainstream over the last five years.

Within two years it is anticipated that cognitive radio will be used in the TV bands by devices that are used to exploit the ‘white space’ spectrum; a spectrum that is unoccupied in a particular geographical location [16]. To do this, devices need to be aware of their surroundings and who is operating in the spectrum. A mixture of primary user spectrum database and sensing to locate secondary users has been put forward as a solution in both the UK [17] and USA [18].

Mainstream research, as we mentioned earlier, is now focussed on a number of narrow, but relevant, fields within cognitive radio and spectrum sensing [19, 20], both individual and cooperative forms have received much attention recently. Latterly, research into architecture configurations and assignment techniques that can exploit spectrum databases has been put forward. Research is also underway on the spectrum assignment techniques themselves [21, 22] and how cognitive radio can be used to assign the radio spectrum. Work including artificial intelligence is still in its infancy, with early suggestions to use techniques of a centralized nature [23], or in a distributed form for part of the system control [24, 25]. More recently distributed artificial intelligence techniques have been used to help prioritize radio resources (spectrum and other resources) [26–28]. These are largely based on reinforcement learning.

A key focus within cognitive radio has been to replace the conventional command and control approach with something more flexible that improves the utilization of the radio spectrum and efficiency. Focus has been on the development of bandwidth efficient systems, rather than on something that improves the spectrum utilization. This bandwidth efficient design strategy is a result of decades of radio regulatory policy [29], because communication systems were not sufficiently intelligent or adaptive. This has resulted in a perceived spectrum shortage, despite studies that show that up to 90% of the radio spectrum is unoccupied at any one location or time [30]. Today, this is not the case; systems can be made increasingly agile [13, 31], making it now possible to improve the use of the radio spectrum in order to reduce the energy requirement, while also encouraging new applications and users. In principle, users and systems should aim to maximize their use of unoccupied or free spectrum (in order to use more energy efficient, low rate modulation), while also avoiding interference to/from other users. The key issues are how to intelligently select the free spectrum, and how such intelligent choices of spectrum by one user/system will positively change the behaviour of other like-minded users operating in the same geographical area.

Significant resources are now being put into the research topic. A well-known early project in the USA dealing with cognitive aspects was the DARPA XG project [21]. There are a number of European Union projects in this general area looking at different cognitive approaches for exploiting the radio spectrum. These include BuNGee [32], COGEU [33], FARAMIR [34], ARAGORN [35], E3 [36], and QOSMOS [37]. Cognitive techniques are also being directly incorporated into LTE-A [38], showing that in general key cognitive aspects are now reaching mainstream applications.

Most recently, using cognitive principles are being considered as way to improve energy efficiency of radio systems, with a number of projects and papers discussing the relevant options. The FP7 project SACRA [39] applied multi-band cognitive radio (CR) technology for energy and spectrum efficiency in a single broadband communication device, rather than the whole system. This kind of research application is very much in its infancy, but given the rise of ‘green issues’ and ‘green radio’, we can expect significant growth in this area over the next four or five years.

1.4 Key Components of Cognitive Communications

There are several key components of Cognitive Communications and we will briefly outline them here:

Awareness of the environment of operation – This is fundamental to the awareness of all cognitive systems. Without some form of input stimuli, it would be impractical for devices and systems to make decisions on how to act. Obvious examples of this include spectrum sensing, but also higher level context information may be important, regulatory policy, and even economics. Recently the agenda has shifted to also consider the use of spectrum databases, as a more centralized way of making devices and systems aware of their environment.

Wireless architectures – The architecture of cognitive systems and devices still is not a mature subject area, as we will see in later chapters. There is no common agreement of whether a cognitive communications device in the broadest sense should be based on a software radio or software defined radio, or perhaps something with less sophisticated capabilities. Key aspects of any device will include the ability to reason, maybe to learn, but that can be achieved with very simple processing elements. Complex approaches involving high levels of signal processing, in order to perform spectrum sensing, may not be required. One can even