Review Article Cognitive Radio Transceivers: RF, Spectrum ...downloads.hindawi.com/journals/ijap/2014/548473.pdf · ects of such impairments are reduced through a postprocessing of

Review ArticleCognitive Radio Transceivers RF Spectrum Sensing andLearning Algorithms Review

Lise Safatly1 Mario Bkassiny2 Mohammed Al-Husseini3 and Ali El-Hajj1

1 ECE Department American University of Beirut Beirut 1107 2020 Lebanon2 ECE Department State University of New York Oswego NY 13126 USA3 Beirut Research amp Innovation Center Lebanese Center for Studies amp Research Beirut 2030 8303 Lebanon

Correspondence should be addressed to Lise Safatly les12aubedulb

Received 13 November 2013 Revised 14 January 2014 Accepted 17 January 2014 Published 24 March 2014

Academic Editor Said E El-Khamy

Copyright copy 2014 Lise Safatly et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

A cognitive transceiver is required to opportunistically use vacant spectrum resources licensed to primary usersThus it relies on acomplete adaptive behavior composed of reconfigurable radio frequency (RF) parts enhanced spectrum sensing algorithms andsophisticated machine learning techniques In this paper we present a review of the recent advances in CR transceivers hardwaredesign and algorithms For the RF part three types of antennas are presented UWB antennas frequency-reconfigurabletunableantennas and UWB antennas with reconfigurable band notches The main challenges faced by the design of the other RF blocksare also discussed Sophisticated spectrum sensing algorithms that overcome main sensing challenges such as model uncertaintyhardware impairments and wideband sensing are highlighted The cognitive engine features are discussed Moreover we studyunsupervised classification algorithms and a reinforcement learning (RL) algorithm that has been proposed to perform decision-making in CR networks

1 Introduction

Cognitive radio (CR) provides a solution to the inefficient useof the frequency spectrum [1ndash3] This inefficiency is due tothe current radio spectrum regulations which assign specificbands to particular services and grant licensed bands accessto only licensed users CR implements dynamic spectrumallocation policies by allowing unlicensed users (secondaryusers) to access spectrum bands licensed to primary userswhile avoiding interference with them [2 3]This necessitatesat the RF front end more constraints on the antenna designthe development of algorithms for sensing the surroundingenvironment and autonomously adapting to particular situ-ations through a cognitive engine [1 4ndash9]These three designelements are introduced in this section and presented inmoredetail in the rest of this paper

Spectrum underlay and spectrum overlay represent thetwo main approaches of sharing spectrum between pri-mary users (PUs) and secondary users (SUs) The underlayapproach imposes constraints on the transmission powerwhich can be satisfied using ultrawideband antennas (UWB)

UWB antennas are also used for channel sensing in overlayCR but must be frequency reconfigurable or tunable Inthis case a single-port antenna can have UWB response forsensing and can be reconfigured for tunable narrowbandoperation when needed to communicate over a white spaceIt is also possible to use dual-port antennas for overlay CRin which one port has UWB frequency response and isused for channel sensing and the second port is frequency-reconfigurabletunable and used for communicating In athird possible spectrum sharing approach the antennas couldbe UWB antennas but should have the ability to selectivelyinduce frequency notches in the bands used by PUs thusavoiding any interference to them and giving the UWBtransmitters of the SUs the chance to achieve long-distancecommunication

Aside from the antenna the design of the other RF blocksfaces main challenges related to the ADCDAC the dynamicrange or range of signal strengths to deal with to the linearityof low-noise amplifiers (LNAs) and to the frequency agilityof the duplexer and filters

Hindawi Publishing CorporationInternational Journal of Antennas and PropagationVolume 2014 Article ID 548473 21 pageshttpdxdoiorg1011552014548473

2 International Journal of Antennas and Propagation

In CR smart transceivers scan the spectrum in orderto find white spaces and transmit adaptive signals Thisfunctionality requires sophisticated algorithms to overcomepractical imperfections such as model uncertainty and hard-ware nonideality Some of the well-known spectrum sensingalgorithms are energy detection [10ndash12] matched filter-based detection [11 13] cyclostationarity-based detection[14ndash18] covariance-based sensing [19 20] and eigenvalue-based sensing [21ndash23] Blind detectors were also introducedto elude the model uncertainty problem Spectrum sensingalgorithms could be affected by RF impairments by inducingunwanted frequency components in the collected signal spec-trum The effects of such impairments are reduced througha postprocessing of the signal [24ndash26] In addition a robustdetector based on smart digital signal processing lowers theeffects of RF impairments and guarantees a high sensingaccuracy

The concept of cognitive radios (CRs) goes beyonddynamic spectrum access (DSA) applications and aims toimprove the quality of information (QoI) of users [5] Thisrequires an intelligent radio that uses spectrum sensingtechniques to observe the RF activities and is able toautonomously adapt to particular situations [9] This isachieved through a reasoning engine which executes actionsbased on certain hard-coded rules and strategies [27] Hard-codedpolicies are completely specified by the systemdesignerand may result in the desired performance as long asthe operating conditions do not deviate from the originalassumed model However due to unexpected changes inthe RF environment the hard-coded rules may not lead tooptimal performance making them inefficient in this caseCognitive radios however can overcome this problem byupdating or augmenting their own sets of policies and rulesbased on past experience [27] which may lead to a morereliable communication performance [9 27] This makes thelearning ability a fundamental building block of any CR toachieve autonomous intelligent behavior [9 27ndash29]

In this paper Section 2 presents the antenna designsfor CR ultrawideband (UWB) antennas antennas withreconfigurable band rejection and frequency-reconfigurableor tunable antennas It also briefly discusses the challengesfaced by the design of RF blocks for CR devices Section 3discusses spectrum sensing by presenting challenges novelsolutions and spectrum sensing algorithms RF imperfec-tions and wideband sensing are also studied and recent blinddetectors robust algorithms and wideband techniques arepresented Section 4 presents a cognitive engine and severalunsupervised classification algorithms for autonomous signalclassification in CRs and a reinforcement learning (RL)algorithm that performs decision-making in CR networksConclusive remarks are given in Section 6

2 RF Frontends for Cognitive Radio

CR transceivers are required to look for and operate inwhite spaces which could exist anywhere inside a widefrequency range This comes different from conventionalwireless transceivers which are bound to certain preallocated

frequency bands As a result significant challenges have tobe dealt with when designing the RF components of a CRtransceiver such as the antennas the power amplifiers (PAs)the filtersduplexers and the analog-to-digital and digital-to-analog converters (ADCsDACs)

21 Antennas for Cognitive Radio Two main approaches ofsharing spectrum between primary users (PUs) and sec-ondary users (SUs) exist spectrum underlay and spectrumoverlay In the underlay approach SUs should operate belowthe noise floor of PUs and thus contingent constraints areimposed on their transmission powerUltrawideband (UWB)technology is very suitable as the enabling technology for thisapproach In spectrum overlay CR SUs search for unusedfrequency bands called white spaces and use them forcommunication

UWB antennas are used for underlay CR and alsofor channel sensing in overlay CR For communication inoverlay CR the antenna must be frequency reconfigurableor tunable Single- and dual-port antennas for overlay CRcan be designed In the dual-port case one port has UWBfrequency response and is used for channel sensing and thesecond port which is frequency reconfigurabletunable isused for communicating In themore challenging single-portdesign the sameport can haveUWBresponse for sensing andcan be reconfigured for tunable narrowband operation whenrequired to communicate over a white space

A third possible spectrum sharing approach results fromthe use of the UWB technology in an overlay scheme Herethe antennas could basically be UWB antennas but shouldhave the ability to selectively induce frequency notches inthe bands used by PUs thus avoiding any interference tothem and giving the UWB transmitters of the SUs the chanceto increase their output power and hence to achieve long-distance communication

A concise review of antenna designs for cognitive radiocovering the above three antenna types is given in [30]

211 UWB Antennas UWB antennas were originally meantto radiate very short pulses over short distances They havebeen used in medical applications GPRs and other short-range communications requiring high throughputs The lit-erature is rich with articles pertaining to the design of UWBantennas [31ndash41] For example the authors in [31] present aunidirectional UWB antenna based on a full-ground planeTo keep this full-ground plane they sequentially employ alist of broadbanding techniques (1) resonance overlapping(2) slot (3) parasitic patch (4) Vivaldi blending (5) steppednotch and (6) rectangular and T-shape slits The resultingantenna has an impedance bandwidth from 36GHz to103 GHz while keeping the unidirectionality of the radiationpattern

In general the guidelines to design UWB antennasinclude the following

(i) Theproper selection of the patch shape Round shapesand round edges lead to smoother current flow andasa result to better wideband characteristics

International Journal of Antennas and Propagation 3

50

39

7

10

40

203

403

399209

78

3

39

3

15

4

44

4

44

162

17854

183

Side viewTop viewUnits mm

XZ

Y

Gro

und

plan

e

Isola gigaver 210(120576r = 375)

45∘

Figure 1 Configuration and photo of the UWB antenna in [42] The antenna combines several bandwidth enhancement techniques

2 4 6 8 10 12

SimulatedMeasured

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

minus45

Refle

ctio

n co

effici

ent (

dB)

Frequency (GHz)

Figure 2 Reflection coefficient of the UWB antenna in Figure 1

(ii) The good design of the ground plane Partial groundplanes and ground planes with specially designedslots play a major role in obtaining UWB responseKeeping a full-ground plane is possible but in thatcase an elaborate work has to be done on the patchdesign

(iii) The matching between the feed line and the patchThis is achieved using either tapered connectionsinset feed or slits under the feed in the ground plane

(iv) The use of fractal shapes which are known for theirself-repetitive characteristic used to obtain multi-andwideband operation and their space-filling prop-erty which leads to increasing the electrical lengthof the antenna without tampering with its overallphysical size

Combinations of these guidelines were used in the fol-lowing two examples The UWB design presented in [42]features a microstrip feed line with two 45∘ bends and atapered section for size reduction andmatching respectivelyThe ground plane is partial and comprises a rectangular partand a trapezoidal part The patch is a half ellipse with the cutmade along theminor axis Four slots whose location and sizerelate to a modified Sierpinski carpet with the ellipse as thebasic shape are incorporated into the patchThe geometry ofthis antenna is shown in Figure 1 Four techniques are appliedfor good impedance matching over the UWB range (1) thespecially selected patch shape (2) the tapered connectionbetween the patch and the feed line (3) the optimized partialground plane and (4) the slots whose design is based onthe knowledge of fractal shapes As a result this antennahas an impedance bandwidth over the 2ndash11 GHz range asshown in Figure 2 and thus can operate in the bands usedfor UMTS WLAN WiMAX and UWB applications It hasomnidirectional radiation patterns due to the partial groundplane

The effect of the ground plane on the performance ofUWB antennas is studied in [43] Herein it is proven thatit is possible to obtain an ultrawide impedance bandwidthusing either a partial ground plane or a ground plane withan optimized large slot where in both cases the same exactpatch is usedThis design has a coplanar-waveguide feed thatconnects to an egg-shaped radiator A photo of both versionsis given in Figure 3

212 Antennas with Reconfigurable Band Rejection As pre-viously stated UWB technology is usually associated withthe CR underlay mode It can however be implemented inthe overlay mode The difference between the two modesis the amount of transmitted power In the underlay modeUWB has a considerably restricted power which is spreadover a wide frequency band In the overlay mode however


Figure 3 Two UWB antennas with optimized ground planes [43]One has a partial ground plane and the other has a ground planewith a large slot

the transmitted power can be much higher It actually canbe increased to a level that is comparable to the powerof licensed systems which allows for communication overmedium to long distances But this mode is only applicable iftwo conditions are met (1) if the UWB transmitter ensuresthat the targeted spectrum is completely free of signals ofother systems or if it shapes its pulse to have nulls inthe bands used by these systems and (2) if the regulationsare revised to allow for this mode of operation [44] Pulseadaptation for overlay UWB CR has been discussed in [45]UWB can also operate in both underlay and overlay modessimultaneously This can happen by shaping the transmittedsignal so as to make part of the spectrum occupied in anunderlay mode and some other parts occupied in an overlaymode In the overlay UWB scenario the antenna at the frontend of the CR device should be capable of operating over thewhole UWB range for sensing and determining the bandsthat are being used by primary users and should also be ableto induce band notches in its frequency response to preventinterference to these users Even if the UWB power is notincreased having these band notches prevent raising the noisefloor of primary users

Antennas that allow the use of UWB in overlay CR shouldhave reconfigurable band notches Several band-notchingtechniques are used in such antennas the most famous ofwhich is the use of split-ring resonators (SRRs) [46] andthe complementary split-ring resnators (CSRRs) [47] SRRshave attracted great interest among electromagneticians andmicrowave engineers due to their applications to the synthesisof artificial materials (metamaterials) with negative effectivepermeability From duality arguments CSRRs which are thenegative image of SRRs and roughly behave as their dualcounterparts can generate a negative permittivity media

Some recent UWB antenna designs with fixed bandnotches are reported in [48ndash55]The works in [56ndash59] are forUWB antennas with reconfigurable band notches

A UWB design with a single reconfigurable band notch isproposed in [60]The configuration of this design and a photoof its fabricated prototype are shown in Figure 4 Originallythe antenna is a UWB monopole based on a microstrip linefeed and a partial ground planeThe patch is rectangular withrounded corners A slit is etched in the ground below the

feed for better matching As a result this antenna has animpedance bandwidth that covers the whole UWB frequencyrange Four nested CSRRs are incorporated in the patchThree electronic switches are mounted across the slots Thesequential activation (deactivation) of the switches leads tothe functioning of a larger (smaller) CSRR and thus results ina notch at a lower (higher) frequencyThe following switchingcases are considered Case 1 when all three switches are ONCase 2 when only S3 is deactivated Case 3 when only S1is ON and finally Case 4 when all switches are OFF Theresulting reflection coefficient plots corresponding to thedifferent switching states are shown in Figure 5 The plotsshow one notch which can occur in one of 3 bands or cancompletely disappear In the latter case the antenna retrievesits UWB response which enables it to sense the whole UWBrange

The antenna reported in [61] is capable of concurrentlyinducing three band notches which are independently con-trollable using only three RF switches This is done usingthree CSRRs etched on the patch There are eight switchingcases for this design with one of them being the originalUWB no-notch case A UWB antenna with reconfigurableband notches can also be designed by incorporating abandstop filter in the feed line of a UWB antenna With thisstructure the switching elements will be mounted on thefeed line away from the radiating patch which makes thebias circuit of the switches simpler to design Such a filterantenna (filtenna) with two reconfigurable rejection bands ispresented in [62] Its structure is shown in Figure 6TheUWBantenna is based on a rounded patch and a partial rectangularground plane A reconfigurable filter with two stop bandsis incorporated along its microstrip feed line The filter isbased on one rectangular single-ring CSRR etched on theline and two identical rectangular single-ring SRRs placed inclose proximity to itThe resonance of the CSRR is controlledvia a switch and that of the two SRRs via two switchesthat are operated in parallel As a result there are fourswitching scenarios The resulting reflection coefficient plotsare shown in Figure 7 Case 1 where no band notches existallows the antenna to sense the UWB range to determinethe narrowband primary services that are transmitting insidethe range In the other three cases the notches block theUWB pulse components in the 35 GHz band the 55 GHzband or both It should be noted that notches due to theSRRs and the CSRR around the feed are stronger than thosedue to CSRRs or any notching structures implemented in thepatch This is because energy is concentrated in a smallerarea in the feed and coupling with the SRRsCSRR is higherDue to the location of the switches connecting the DC biaslines especially to the SRRs which are DC-separated fromanything else is an easy task A wire can be used to drive theswitch on the CSRR A note is that extra band notches can beobtained by placing more SRRs around the feed line Tunableversions of these notches can be obtained by replacing the RFswitches with varactors

213 Frequency-ReconfigurableTunable Antennas Anten-nas designed for overlay CR should have the capability tosense the channel and communicate over a small portion


14

4505

12

13

1

7155

31

1

1

24

105 10

S1S2S3

Figure 4 Geometry and photo of an antenna with one reconfigurable rejection band [60]

10

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

Refle

ctio

n co

effici

ent (

dB)

Frequency (GHz)4 5 6 873 9

Case 1 ON-ON-ONCase 2 ON-ON-OFF

Case 3 ON-OFF-OFFCase 4 OFF-OFF-OFF

Figure 5 Reflection coefficient for the different switching cases ofthe antenna in Figure 4

of it These antennas can be implemented as dual portwhere one port is UWB and the other is narrowband andfrequency reconfigurableThey can also be designed as singleport where the same port is used for both sensing andcommunicating and thus should switch between widebandand narrowband operations For the dual-port designs theisolation between the two ports is an issue More elaboratework is required for single-port designs

A dual-port antenna for overlay was proposed in [63] Itsstructure consists of two printed antennas namely wide- andnarrowband antennasThe design in [64] also combines wideand narrow band antennas where the wideband one coversthe 26ndash11 GHz range and the narrowband one is tunableusing a varactor between 685GHz and 720GHz A simpledual-port design is presented in [65] The configurationof this design which comprises two microstrip-line-fedmonopoles sharing a common partial ground is shown inFigure 8 The sensing UWB antenna is based on an egg-shaped patch The UWB response of the sensing antenna isguaranteed by the design of the patch the partial ground

plane and a feed matching section The communicatingantenna is a simple microstrip line matched to the 50 minus Ω

feed line Two electronic switches are incorporated along thisline Controlling these switches leads to various resonancefrequencies within the UWB range as shown in Figure 9 forthree considered switching cases

Dual-port antennas enable simultaneous sensing andcommunicating over the channel but have limitations interms of their relatively large size the coupling betweenthe two ports and the degraded radiation patterns Theselimitations are solved by the use of single-port antennas butthese are only suitable when the channel does not changevery fast and thus sensing and communication are possiblesequentially

Single-port reconfigurable widebandnarrowband anten-nas are reported in [66ndash68] The design in [66] has a widebandwidth mode covering the 10ndash32 GHz range and threenarrowband modes within this range In [67] fifteen PINdiode switches are used on a single-port Vivaldi antennaleading to a wideband operation over the 1ndash3GHz band andsix narrowband states inside this range GaAs field-effecttransistor (FET) switches are used in [68] to connect multiplestubs of different lengths to the main feed line of a UWBcircular-disc monopole The result is an antenna that can beoperated in a UWBmode or in a reconfigurable narrowbandmode over one of three frequency subbands the first covers21ndash26GHz the second covers 36ndash46GHz and finally thethird covers a dual band of 28ndash34GHz and 49ndash58GHzThedesign in [69] is a filtenna based on a reconfigurable bandpassfilter embedded in the feed line of a UWB monopole AUWBand five narrowband operationmodes characterize thisdesign

A printed Yagi-Uda antenna tunable over the 478ndash741MHz UHF TV band is presented in [70] where thenarrowband frequency tunability is obtained by loading thedriver dipole arms and four directors with varactor diodesA miniaturized tunable antenna for TV white spaces isreported in [71] In [72] two PIN diodes and two varactorsare employed for narrow band tuning between 139 and236GHz A Vivaldi-based filtenna with frequency tunabilityover the 61ndash65GHz range is presented in [73] The single-port designs in [70 72 73] are only capable of narrowbandoperation which means that wideband spectrum sensing has


40

20

11

1195

15

60

245

975 79

4152

Units mmTop view Side view

Figure 6 Geometry and photo of a filter antenna with two reconfigurable band notches [62]

Case 1Case 2

Case 3Case 4

0

minus5

minus10

minus15

minus20

minus25

minus30

Refle

ctio

n co

effici

ent (

dB)

2 4 53 6 8 97

Frequency (GHz)


to be done progressively a narrow chunk of frequencies at atime This is also the case with the filtenna design shown inFigure 10 Here a tunable bandpass filter is embedded alongthe microstrip feed line of a UWBmonopole antenna wherethe filter is based on a T-shaped slot incorporated in themicrostrip line between a pair of gaps For the purpose ofachieving frequency tunability a varactor is included in thedesign as indicated Changing the capacitance of the varactorchanges the notch band caused by the T-shaped slot and as aresult the narrow passband of the overall filter The DC linesof the varactors are connected with ease Due to the presenceof the two gaps DC is separated from both the antenna portand the patch Two surface-mount inductors are used over theDC lines as RF chokes The computed reflection coefficientplots are given in Figure 11They show narrowband tunability

over the 45ndash7GHz frequency range for capacitance valuesbetween 03 and 7 pF

22 RF Design Challenges The design challenges for the RFsection of a CR transceiver are well reviewed in [74ndash76]Diagrams of a CR transceiver architecture appear in [74]For wideband operation such as that required for sensingthe main challenges relate to the ADCDAC the dynamicrange or range of signal strengths to deal with which couldbe as wide as 100 dB to the linearity of low-noise amplifiers(LNAs) which has to be high and to achieve impedancematching over a wide frequency range For tunable narrow-band operation the key issues are the frequency agility of theduplexer and filters

221 ADCs and DACs In many situations the desired signalreceived bywireless device could sometimes be 100 dBweakerthan other in-band signals generated by nearby transmittersof the same communication standard or some out-of-bandblockers caused by any transmitter This would demand adynamic range of about 100 dB on the ADC Ideally the RFsignals received by a CR system should be digitized as closeto the antenna as possible so that all the processing is doneat the level of the digital signal processor (DSP) In this casethe ADC and DAC should have the 100 dB dynamic rangeexplained earlier be operable over a UWB frequency rangeand handle significant levels of power These requirementsare still far beyond the limits of available technology that iswhy the more realistic CR receivers reduce both the requireddynamic range and the conversion bandwidth by havingdownconversion and filtering functions implemented beforeADC

On the path towards ideal CR receivers several ADCs canbe used in parallel to widen the conversion bandwidth In[77] a parallel continuous time ΔΣ ADC is presented which


50

50

15 201

2

3

33

16

55

35 57

36

9

12

9

Units mm

S1

S2

Figure 8 Geometry and photo of the dual-port ultrawideband-narrowband antennas in [65]

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

Frequency (GHz)4 5 63

SimulatedMeasured

S1OFFS2OFF

(a)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2OFF

(b)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2ON

(c)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


Case 1 OFF-OFFCase 2 ON-OFF

Case 3 ON-ON

Measurement for the 3 cases

(d)

Figure 9 Reflection coefficient for 3 selected switching cases of the antenna in Figure 8


40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4

Figure 10 Geometry and photo of a tunable narrowband filtenna

0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)

C = 030 pFC = 035 pFC = 040 pFC = 050 pFC = 060 pFC = 070 pF


Figure 11 Reflection coefficient of the filtenna in Figure 10 for theindicated varactor capacitance values

requires very complex digital synthesis filters The hybridfilter bank-basedADCpresented in [78] could be a promisingsolution for CR applications but its use necessitates highresource calibration since it is sensitive to analog filter errorsand imperfections A solution that provides good perfor-mance in terms of speed employs time interleaving [79]but it lacks the resolution and dynamic range To suppresslow order nonlinearities of parallel ADCs and nonlinearitiescaused by pre-ADCanalog components gigital postlineariza-tion is desired The technique in [80 81] combines timemultiplexing and frequency multiplexing by using bandpasscharge sampling filters as analysis filters in hybrid filter banks

architecture This leads to reduced complexity of analoganalysis filters and simultaneously of the sensitivity to analogerrors and imperfections Yet the practical implementationwith the aim of widening the bandwidth and sensitivity stillrequires a deeper investigation

RF DACs are utilized in the fully digital RF transmitterspresented in [82 83] The design of these DACs is lesschallenging than that of ADCs althoughmore improvementsare still necessary

222 Low-Noise Amplifier A receiverrsquos performance is pri-marily determined by the linearity and adequate matchingover a wide frequency range of the broadband or tunableLNA A broadband LNA topology achieving a frequencyrange of 50MHz to 10GHz was proposed in [76] Theinput capacitance is canceled by the LNA using inductivebehavior provided by negative feedback Advanced CMOStechnologies make it easier to design such high-frequencyLNAs without any large inductors On the other hand thelinearity issue becomes crucial because the supply voltagefor core transistors decreases to around 1V in 90 nm ormore advanced processes which results in a limited voltageswing LNAs using thick-oxide transistors may prevent thisproblem [84] The mixer-first RF front ends [85 86] areanother approach for improving linearity performance Animpedance translation technique which uses passive mixersfollowed by capacitive loads can provide low impedance forout-of-band blockers [87 88] which makes it attractive

The even-order nonlinearity of the LNA also should beconsidered in wideband CR systems This is because second-order intermodulation (IM2) products generated by an LNAcan fall within theCRband thus corrupting the desired signaleven before downconversion A differential LNA topologyseems attractive here but a balun is necessary if the antennais single ended Design of low-loss baluns operating acrosstwo or three decades of bandwidth is challenging A usefulbalun-LNA topology is proposed in [89]


223 DownconversionMixer Another issue in thewidebandoperation of a CR device is harmonicmixingTheRF signal isoften downconverted using square-wave local-oscillator (LO)signals that contain large odd-order harmonic componentsThis is done to obtain flat performances of noise and gainacross a wide frequency range The use of a harmonic-rejection mixer (HRM) is one way to remove odd-orderLO harmonics The authors in [90 91] propose techniquesto improve harmonic rejection ratios (HRRs) However ifseventh- and higher-order LO harmonics are to be rejectedthe HRMs become much more complex

224 Wideband Frequency Synthesizer Multiple oscillatorscould be used to cover a wide frequency range This solvesthe issue of the trade-off between tuning range and phasenoises For acceptable phase noise figures the tuning range ofinductor-capacitor (LC-) type oscillators is typically limitedto about plusmn15 at frequencies of tens of gigahertz Frequencysynthesizers for CR having two oscillators to support a widerfrequency range are reported in [84 92]

3 Spectrum Sensing in Cognitive Radios

In a CR scenario secondary users are allowed to detectexploit and use underutilized spectral resources licensedto primary users Such opportunistic behavior urges CRtransceivers to scan the spectrum in order to findwhite spacesand transmit adaptive signals Thus spectrum sensing is thekey function of smart receivers since it enables the cognitivecycle proposed byMitola in [4 93]This functionality createsunique signal processing challenges and requires sophisti-cated algorithms to overcome practical imperfections such asmodel uncertainty and hardware imperfections

To address these challenges CR researchers proposed inthe last decade various detectors that have different complex-ity levels performance results and requirements for imple-mentationWell-known spectrum sensing algorithms sortedin an ascending order of complexity are energy detection(ED) [10ndash12] matched filter (MF-) based detection [11 13]cyclostationarity-based detection (CSD) [14ndash18] covariance-based sensing [19 20] and eigenvalue-based sensing [21ndash23] Some of these methods require a priori knowledge ofnoise andor signal power information these include MFCSD (relying on the full or partial knowledge of signal andnoise levels) and ED (having a threshold dependent on theestimated noise power level) Blind detectors were recentlyproposed to elude the model uncertainty problem relying onadvanced digital signal processing techniques

In a cognitive receiver RF impairments could harm theperformance of the spectrum sensing algorithm by inducingunwanted frequency components in the collected signalspectrum Tomitigate the effects of such impairments ldquoDirtyRFrdquo is applied on the SU receiver inducing a postprocessing ofthe signal thus compensating analog imperfections [24ndash26]A robust detector based on smart digital signal processingshould be able to digitally lower the effects of RF impairmentsand guarantee a high sensing accuracy

The early-designed spectrum sensing algorithms aimedto detect a white space from narrow frequency bands wheremany emerging wireless applications require an opportunis-tic usage of a wideband spectrum [94ndash97] Consequently SUsare forced to scan a wide range of potential spectra and detectavailable holes to be able to transmit One of the main con-cerns of theCR community is to conceivewideband spectrumsensing methods to replace the complicated implementationof high sampling rate ADCs capable of downconvertingwideband signals

The selection of signal processing algorithms and theirparameters reflects the speed and sensing time of thecognitive receiver A complex signal processing algorithmshould respect an optimum sensing value depending on thecapabilities of the radio and its temporal characteristics in theenvironment On the other hand the ADC is considered asthe primary bottleneck of the DSP architecture since it forcesthe clock speed of the system Moreover the selection of thedigital signal processing platformaffects the speed of the frontendAll these parameters influence the sensing frequency andspeed of cognitive radio receivers For that researchers focuson implementing sensing algorithms with low complexityhigh speed and flexibility in order to conceive an adaptiveCR terminal

In the following sections we will describe three mainpractical spectrum sensing challenges and novel solutionsModel uncertainty RF imperfections and wideband sensingare studied and recent blind detectors robust algorithms andwideband techniques are presented Accordingly in the restof this paper we provide an overview of the state of the art ofspectrum sensing algorithms that were proposed to answerthese three major and hot research challenges

31 Blind Detectors As per regulation specifications sec-ondary users are required to detect very weak licensedusers in order to protect primary transmissions [98 99]Any missed detection will enable an unlicensed transmis-sion on a busy channel harming the incumbent primarysignal Unfortunately many detectors reveal performancedegradation at low SNR due to inappropriate estimationof the signal or noise models This phenomenon is knownas SNR wall [10 100] For the ED an estimation of thenoise variance is required to select a suitable thresholdImperfect knowledge of the noise model especially in lowSNR scenarios will consequently deteriorate the efficiency ofthis algorithm The SNR wall phenomenon also harms anydetector based on the received signalrsquos moments Using coop-erative spectrum sensing techniques or relying on calibrationand compensation algorithms are possible solutions to themodel uncertainty problem [100 101] However using totallyblind detectors which detect the presence of a signal withoutany knowledge of signal or noise parameters is consideredthe ideal alternative Two recently proposed blind detectorsare described below

311 Blind Eigenvalue-Based Detector Zeng et al devised ablind detector based on the computation of the minimumand maximum eigenvalues 120582min and 120582max of the samplecovariance matrix R(119873

119878) defined in [22] The test statistics


Initialize Acquire 119871 consecutive data samples and assume that there are (119872 ge 1) receivers (antennas)and (119873

119878

) is the total number of collected samples(a) Define the received vector 119909

119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879

(b) The collection of 119871 consecutive outputs x119899

is defined asx119899

= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879

(c) Compute the sample covariance matrix R(119873119878

)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899

(d) Compute 120582max and 120582min the maximum and minimum eigenvalues of the matrix R(119873119878

)(e) Compute the threshold ] for the test statistics

] =(radic119873119878

+ radic119872119871)2

(radic119873119878

minus radic119872119871)2

(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651

is the Tracy-Widom distribution of order 1 [102]The decision test(f) Decide on 119867

0

or 1198671

by computing the ratio between 120582max and 120582min120582max120582min

≷1198671

1198670]

Algorithm 1 Steps of the MME detector

of this maximum-minimum eigenvalue (MME) detection issimply given by

120582max120582min

≷1198671

1198670

] (1)

where ] is the threshold calculated by using the numberof acquired samples the smoothing factor used for thecalculation ofR(119873

119878) and a selected probability of false alarm

It is expected that noise produces small eigenvalues whereasthe correlation inherited in modulated signals increases theeigenvalues The proposed test statistic does not depend onany knowledge of noise signal or channel models thusit is not sensitive to the model uncertainty problem Thedetailed computational steps of this scheme are described inAlgorithm 1

312 The CAF Symmetry-Based Detector This blind spec-trum sensing detector is based on the symmetry property ofthe cyclic autocorrelation function (CAF) Benefiting fromthe sparsity property of CAF the compressed sensing toolis adopted in this algorithm A test statistic is definedwithout the computation of any threshold by checking if theestimated CAF exhibits symmetry or not As demonstratedin [103] a positive symmetry check affirms the presenceof a primary signal The estimation of the cyclic autocor-relation vector is computed using an iterative optimizationtechnique called the orthogonal matching pursuit (OMP)[104] The computational complexity of this algorithm isreduced by limiting the number of acquired samples and thenumber of needed iterations to ensure its practical feasibilityAlgorithm 2 summarizes the main steps of this detector

32 Robust Sensing Algorithms In practice CR receiversare composed of several sources of hardware imperfections

such as low noise amplifiers (LNA) mixers local oscillators(LO) and analog-to-digital converters (ADC) The mostcritical result of such impairments is the appearance ofnew frequency components in the received signal classi-fied in intermodulation distortions (IM) cross modulation(XM) and phase distortion (AMPM) Consequently strongprimary users could harm the performance of traditionalspectrum sensing techniques by adding unwanted spectrumcomponents via front-endrsquos nonlinearityWhen these compo-nents overlapwithweak secondary users a degradation in thereliability of SU transmissions occurs They can also virtuallyoccupy the whole spectrum thus decreasing the opportunityto find a vacant transmission band In both scenarios theaccuracy of any proposed algorithm will be deteriorated Tomitigate these effects a robust detector algorithm shouldbe equipped with a compensation functionality to digitallyreduce the effects of nonlinearities Possible compensationalgorithms could be based on feed-forward techniques withreference nonlinearity feed-back equalization and train-ing symbol-based equalization A well-known feed-forwardtechnique to alleviate phase noise carrier frequency off-set nonlinearities 119868119876 imbalance or ADC impairments isdescribed below

321 The AIC Algorithm In [25] Valkama et al devised theadaptive interference cancellation (AIC) algorithm whichis a feed-forward algorithm for the mitigation of secondthird and fifth order intermodulation distortion The ideais to model the distortion caused by the interferer and thensubtract them from the received signal A mathematicalformulation of the distortion model and order is studiedbefore implementing the algorithm Then an imitation ofthe distortion products and an adaptive adjustment of theirlevels are performed to compensate the distorted signal


Initialize Acquire 119899 data samples from the spectrum sensing interval formed by 119873

samples and set (119897 + 1) the number of OMP iterations and (119872) the number of delays 120591sFor 119872 different values of 120591(a) Calculate the autocorrelation vector f

1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591

(119905) = 119910(119905)119910(119905 + 120591)(b) Calculate the elements of the matrix 119860 performing the IDFT transform

119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873

TheOMP algorithm(c) Estimate the cyclic autocorrelation vector by solving the following system ofequations 119860r

120591

= f120591

by using an iterative optimization technique calledOrthogonal Matching Pursuit (OMP) that delivers an approximated solution r

120591

Symmetry check(d) Calculate the symmetry index for this value of 120591 by ignoring the first amplitude thatcorresponds to the first iteration of OMP and measuring the mean value of the abscissaof the remaining (119897 minus 1) non zero elements in r

120591

The symmetry index is given by

IND(120591)sym =1

119897

119897

sum119894=1

r120591119894

End ForEquivalent symmetry check

IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0

Algorithm 2 Steps of the CAF symmetry-based detector

The algorithm illustrated in Figure 12 starts by splitting theband of the received signal in order to differentiate betweenthe strong PU and other frequency components (SUs +distortions) The band splitting is accompanied by a coarseenergy detector used to locate the strong interferer Then aparallel block of reference nonlinearities is used to extractpotential distortion products from the strong interferer Anadaptive filter the least mean square (LMS) is used to adjustdigitally created distortions levels The adaptive filter utilizesthe distorted signal resulted from the band splitter as aninput parameter and minimizes the common error signal119890(119905) The adjusted nonlinearities are finally subtracted fromthe received signal to cleanse the band from nonlinearitydistortions It is shown in [26] that the application of theAIC algorithm before the detector increases the detectionreliability in CR devices

33 Wideband Spectrum Sensing Several emerging wirelessapplications and regulation encourage cognitive receiversto scan a wideband spectrum to find potential spectrumholes In contrast to the narrowband techniques mentionedabove wideband spectrum sensing methods aim to sense afrequency bandwidth exceeding the coherence bandwidth ofthe channel A frequent example deals with the design of analgorithm capable of sensing the whole ultra-high-frequency(UHF) TV band (between 300MHz and 3GHz) Practicallywideband scanning could be performed via the following twodifferent methods

(1) By using a filter bank formed by preset multiple nar-rowband pass filters BPFs [105] This hardware-basedsolution requires more hardware components thusincreasing the cost and the RF impairments harmful

SDR RF frontend

band splitter

Referencenonlinearity

Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++

Figure 12 The AIC algorithm

effect and limiting the flexibility of the radio by fixingthe number of filters After each filter a narrowbandstate-of-the-art technique is implemented

(2) By using sophisticated signal processing techniquesIn fact narrowband sensing techniques cannot bedirectly applied to scan a wideband since theyare based on single binary decision for the wholespectrum Thus they cannot simultaneously identifyvacant channels that lie within the wideband spec-trum Recently proposed wideband spectrum sensingcan be broadly categorized into two types

(i) Nyquist wideband sensing processes digital sig-nals taken at or above theNyquist rate for exam-ple the wavelet transform-based technique

(ii) sub-Nyquist wideband sensing acquires signalsusing a sampling rate lower than the Nyquistrate for example the compressive sensing tech-nique


Wideband of interest

Partitioned narrow band

f0 f1 f2 fnminus1 fn fN f

Figure 13 A wideband spectrum seen as a train of narrowbandsignals and presenting frequency irregularities

In the following sections two approaches to performwideband spectrum sensing are discussed

331 Wavelet Transform-Based Technique In this methodthe SU transceiver scans a wideband without using a bankof narrow BPFs Alternatively a wideband receiver will bebased on high-speed digital signal processing to searchover multiple frequency bands in an adaptive manner Theobtained digital signal will be modeled as a train of consec-utive narrow frequency bands as illustrated in Figure 13 Toidentify these bands and search for potential spectrum holesthe wavelet transform will be used to locate edges betweendifferent narrow subbands [106] The corresponding blockdiagram is depicted in Figure 14 Wavelet transform is usedin mathematics to locate irregularities [95] Consequentlyit will be a good candidate to differentiate between thenarrow subbands of wideband signal [97 107] Awavelet edgedetector is able to identify the average power level within eachidentified subband which will lead to the localization of thespectrum holes

The analysis using wavelet transform is based on afunction known as the principal wavelet 120595 which has afinite energy Wavelets are used to transform a given signalinto another representation that models the informationrelated to the signal in a more utile way Wavelets couldbe manipulated in two different ways moved along thefrequency axis or stretched with a variable energy AWavelettransform obtained by summing the product of the signalmultiplied by the wavelet is calculated at different spots ofthe signal and for different combinations of the wavelet Thiscalculation could be monitored to detect the irregularities ofthe signal by observing the different values of the wavelettransform

332 Compressive Sensing Technique Amajor implementa-tion challenge lies in the very high sampling rates requiredby conventional spectral estimation methods which have tooperate at or above the Nyquist rate However to solve thisissue the compressive sampling CS technique is used for theacquisition of sparse signals at rates significantly lower thanthe Nyquist rate Signal reconstruction is no more based onold reconstruction techniques but will be a solution to anoptimization problem Several schemes were suggested in theliterature for the reconstruction of the signal by usingwavelettransforms [97] the autocorrelation of the signal [108] oradvanced algorithms for sparse approximationmethods [104109] It is shown that such methods could preserve the

adaptive response of the algorithm by offering a relativelysmall processing time

Sparse approximation consists of finding a signal or avector with sparseness property that is it has a small numberof nonzero elements that satisfies (approximately) a system ofequations For example consider a linear system of equations119910 = A119909 where A is an 119899 minus by minus 119872 matrix with 119899 lt 119872 SinceA is overcomplete (119899 lt 119872) this problem does not have aunique solution Among all the possible solutions if the trueone is known a priori to be sparse then it happens that thesparsest that is the solution119909 containing asmany as possiblezero components and satisfying 119910 cong 119860119909 is close to the truesolution Reducing the problem complexity to 119899 instead of 119872

increases the adaptation time of the reconstruction algorithmand thus provides better processing time

4 Machine Learning in Cognitive Radios

Cognitive radios (CRs) are considered as intelligent radiodevices that use the methodology of understanding-by-building to learn and adapt to their radio frequency (RF)environment [1] Several CR architectures have been pro-posed over the past years in order to achieve dynamicspectrum access (DSA) [1ndash3] However as initially proposedby [4] the concept of CRs goes beyond DSA applications andaims to improve the quality of information (QoI) of wirelessusers [5] This functionality requires an intelligent radio thatis aware of its RF environment and is able to autonomouslyadapt to the variations in the wireless medium [9]

Cognitive radios are assumed to use spectrum sensingtechniques to identify the RF activities in their surroundingenvironment [6] Based on their observations CRs applytheir reasoning abilities to modify their behavior and adaptto particular situations This is achieved through a reasoningengine which executes actions based on certain rules andstrategies [27] Similar reasoning engines could be identifiedin conventional radios that behave according to a set ofhard-coded rules [27] For example according to the IEEE80211 specifications such hard-coded rules determine theswitching of a radio device among different modulationschemes depending on the signal-to-noise ratio (SNR) [27]Hard-coded policies are completely specified by the systemdesigner and may result in the desired performance aslong as the operating conditions do not deviate from theoriginal assumed model However in situations where theRF environment changes due to unexpected agents or factors(eg jammers interferers extreme fading conditions etc)the hard-coded rules may not lead to optimal performancemaking them inefficient in this case Cognitive radios how-ever can overcome this limitation by updating their ownsets of policies and rules based on past experience [27]For example if a CR is subject to jamming or significantinterference on a certain channel it could come up withnew actions to switch to a new frequency band instead ofsimply modifying its modulation scheme in contrast withthe IEEE 80211 case Hence based on its learning ability aCR can update or augment its set of rules and policies basedon its own experience which may lead to a more reliable


Nyquistsampler FFT Power spectral

densityWavelet

transformLocal maximum

detection

Figure 14 Block diagram of Wavelet transform-based technique

communication performance [9 27]Thismakes the learningability a fundamental building block in any CR to achieveautonomous intelligent behavior [9 27ndash29]

Machine learning techniques are gaining more impor-tance in the current and future wireless networks due to theincrease in system complexity and the heterogeneous natureof the wireless medium [29 110ndash117] In particular significantresearch efforts are being focused on developing link aggrega-tion techniques usingmultiple radio access networks (RANs)[111] For example the authors in [111] have demonstrateda CR system that is able to access a heterogeneous RANaggregation system including HSDPA Wi-Max and W-CDMAsystems In order to ensure efficient systemoperationthe system parameters need to be optimized to maximize theoverall performance However due to the dynamic natureof the wireless network and the large number of systemparameters the operating parameters cannot be optimizedmanually but require intelligent algorithms that are able toautonomously adjust the system parameters leading to opti-mal performance In [111] the support vector regressor (SVR-)-based learning algorithm has been proposed for parameteroptimization This algorithm requires a set of training datato estimate the system model Other learning approachesmay be considered in the future to optimize various systemparameters under different operating environments [111]In this paper however we focus on unsupervised learningmethods to ensure autonomous CR operation as we willdiscuss next

5 The Cognitive Engine

Cognitive radios extend software-defined radios (SDRs) byadding a cognitive engine (CE) to the radio platform [27]According to [27] a CE can be composed of three maincomponents (1) a knowledge base (2) a reasoning engineand (3) a learning engine as illustrated in Figure 15 [27]The reasoning engine executes the actions and policies thatare stored in the knowledge base while the learning engineupdates these policies based on past experience [9 27] Byapplying learning algorithms the learning engine can trans-form the observed data into knowledge thus allowing theCR to be aware of certain characteristics of its environment[4 29]

The machine learning literature is rich with learningalgorithms that can be used in various contexts [29 110ndash117]These learning algorithms can be categorized under eithersupervised or unsupervisedmethods In supervised learninga set of labeled training data is available for the learningagent to specify whether a certain action is correct or wrong[29 118] In unsupervised learning however the learningagent is supposed to identify the correct and wrong actionsbased on its own experience and interactions with the envi-ronment [118] This makes unsupervised learning algorithms

Knowledge base

Reasoningengine

Learningengine

Cognitive engine

Figure 15 The cognitive engine (CE)

more appealing for CR applications compared to supervisedlearning since they lead to autonomous cognitive behavior inthe absence of instructors [9] Hence unsupervised learninghas been the focus of recent autonomous CRs formulations[7 9 119ndash123]

Several unsupervised learning algorithms have been pro-posed for CRs to perform either feature classification ordecision-making [29] Classification algorithms can be usedto infer hidden information about a set of noisy data Theyallow for example inferring both number and types ofwireless systems that are active in a certain environment [120124] In addition classification algorithmswere also proposedfor modulation classification based on Bayesian networks asin [125] On the other hand decision-making algorithms canbe used to update or modify the policies and rules that arestored in the knowledge base of a CR Thus the learningprocess may result in a new set of actions allowing the CRto adapt to completely new RF environments [27]

In the followings we present several unsupervised classi-fication algorithms that have been proposed for autonomoussignal classification in CRs We also present a reinforcementlearning (RL) algorithm that has been proposed to performdecision-making in CR networks

51 Unsupervised Classification Algorithms Classificationalgorithms have been proposed for CRs to extract knowledgefrom noisy data [120 121 124] As we have mentioned abovethe classification algorithms based onmachine learning tech-niques can be divided into two main categories upervisedand unsupervised classifiers [29 120] Supervised classifiersrequire an ldquoinstructorrdquo that specifies whether a certainclassification decision is ldquocorrectrdquo or not These supervisedclassifiers require a set of labeled training data (or expert-annotated data [120]) that specify the correct classes (orclusters) Supervised classifiers can be applied in certain con-ditions when prior knowledge (eg labeled data) is availableto the learning agent However in situations where no such


prior information is available unsupervised classifiers can beused instead Unsupervised classifiers do not require labeleddata and process the detected data to determine the class orcluster of each data element Hence unsupervised classifierscan be considered more suitable for autonomous CRs whenoperating in unknown RF environments [9 124]

In the followings we present three examples of unsu-pervised classifiers that have been proposed for CRs (1) the119870-means (2)119883-means and (3) Dirichlet process mixturemodel (DPMM) classifiersWe give a brief description of eachalgorithm and show their different applications in CRs

511 The 119870-Means Classifier The 119870-means algorithm hasbeen proposed for robust signal classification in CRs [120]It is considered as an unsupervised classifier since it doesnot require labeled training data [29 120] However the 119870-means algorithm requires prior knowledge about the numberof clusters 119870 making it a unsupervised parametric classifier

Given a set of 119873 feature points y119894119873

119894=1

the 119870-meansalgorithm classifies these data points into 119870 clusters where119870 is determined a priori The algorithm starts with a setof 119870 arbitrary centroids c

1 c

119870 defining the centers

of 119870 initial clusters Each feature point y119894is then selected

sequentially and assigned to the closest centroid (in termsof Euclidean distance) Once a feature point is assigned to aparticular cluster the corresponding centroid is updated andcomputed as the mean of the feature points belonging to thatcluster Eventually the means converge to the clusters centers[120]

In CR applications the 119870-means algorithm was appliedto classify different types of RF signals based on their spectralor modulation characteristics For example in [120] the119870-means were allowed to classify primary and secondarysignals transmitting within the TV band The authors in[120] demonstrated the robustness of this classifier againstfluctuations in the signal parameters such as the signal-to-noise ratio (SNR) In general such fluctuations lead tolarger variance in the extracted features which may cause

poor classification performance Nevertheless the 119870-meansalgorithm was shown to be robust against such variationsand achieved good performance in classifying both 8-levelvestigial sideband (8VSB) modulated signals (primary TVtransmission) and BPSK modulated signals (cognitive sec-ondary transmission) [120]

The 119870-means algorithm is characterized by its low com-plexity and fast convergence However it requires accurateknowledge about the number of signal classes which maynot be practical in many cases [9 120 124] For example inmany CR applications it may be required to classify signalsbelonging to an unknown number of systems which requiresnonparametric approaches as we will describe next [9 119120]

512 The 119883-Means Classifier The 119883-means algorithm hasbeen proposed as an extension of the 119870-means algorithmallowing the classifier to estimate the number of clusters fromthe data itself [126] The 119883-means algorithm is formulated asan iterative 119870-means algorithm which computes the optimalnumber of clusters that maximizes either the Bayesian infor-mation criterion (BIC) orAkaike information criterion (AIC)[126] In contrast with the 119870-means the 119883-means algorithmassumes an unknown number of clusters making it suitablefor nonparametric classification

This algorithmic approach was successful in detectingprimary user emulation (PUE) attacks in CR applicationsas discussed in [120] In this case the signal detector canfirst estimate the number of clusters 119883 and then obtainthe different classification regions for each cluster similarto the 119870-means algorithm Both 119870-means and 119883-meansalgorithms can be implemented at low complexity Howeverthey are suitable only for spherical Gaussian mixture models[124] In CRs however feature vectors can be extractedfrom complex observation models which are not necessarilyGaussian [124] This situation can thus be addressed usingthe DPMM classifier which assumes an arbitrary distributionof the observation model but requires higher computationalcomplexity compared to both 119883-means and 119870-means algo-rithms as we will discuss next [124]

120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)

513 The DPMM Classifier The DPMM classifier has beenproposed for unsupervised signal classification in CRs [29121 124] It is considered a Bayesian nonparametric unsuper-vised classifier in the sense of allowing the number of clusters(or classes) to increase with the data size [29 121 124] Thismodel allows classifying a set of feature vectors y

119894119873

119894=1

into 119870

clusters where119870 is to be estimated from the data (in contrastwith the 119870-means which require prior knowledge about 119870)

The DPMM is based on the Chinese restaurant process(CRP) which models the feature vectors as customers joiningspecific tables [127] The CRP has been previously proposedfor both feature classification and decision-making [124 128]In particular [128] proposed a strategic game model basedon the CRP which is referred to as the Chinese restaurantgame This framework has been applied for channel accessin CR networks [128] In this paper on the other hand


we will present the CRP as an underlying framework fornonparametric signal classification in CRs as in [124] Thismodel is formulated based on the DPMM as discussed next

The DPMM assumes that the feature vectors are drawnfrom a mixture model such that [29 124 127]

119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)

where 119866 is a realization of the Dirichlet process 119863119875(1205720 1198660)

with parameters 1205720

gt 0 and a prior distribution 1198660[29 124

127]The nonparametric nature of the DPMM stems from thesupport distribution119866 which is drawn from a nonparametricset of distributions according to the Dirichlet process Therealization 119866 sim 119863119875(120572

0 1198660) is discrete and defined over an

infinite set thus allowing for infinitely many clusters Basedon this model a feature vector y

119894is assumed to be drawn

from a distribution119891120579119894

(y119894) where 120579

119894is drawn from119866 [29 124]

According to this formulation a cluster is defined as a set offeature vectors y

119894rsquos having identical parameters 120579

119894rsquos Thus we

define the clusters parameters120601119896rsquos to denote the unique values

of 120579119894rsquos [29 124 127]By assuming the above DPMM framework the prob-

lem of feature classification can be formulated following aBayesian approach which estimates the parameters 120579

119894rsquos for a

set of feature vectors y119894119873

119894=1

assuming a nonparametric prior119866 sim 119863119875(120572

0 1198660) for 120579

119894rsquos [129 130]The optimal parameters 120579

119894rsquos

can thus be estimated based on the maximum a posterioriprobability (MAP) criterion which finds the parameters120579119894rsquos maximizing the posterior distribution of 119891(120579

1 120579

119873|

y1 y

119873) [124 129] However this posterior distribution

cannot be obtained in closed form under the above DPMMconstruction Thus stochastic simulation approaches havebeen proposed to estimate 120579

119894rsquos by using the Gibbs sampling

method [124 129] By following theGibbs sampling approachthe DPMM-based classifier can be obtained by sampling 120579

119894rsquos

from the posterior distribution 120579119894

| 120579119895119895 = 119894

y1 y

119873in (2)

[29 124 127 129 130]This classification algorithm has been used for signal

classification in CRs to determine the number of wirelesssystems in a certain RF environment [121 124] It was shownto accurately estimate the number of existingwireless systemswithout any prior information about the environment How-ever this algorithm requires extensive computational effortssince it relies on an iterative Gibbs sampling process

52 Reinforcement Learning Algorithms In addition to itsability of classifying wireless signals a CR is assumed touse machine learning techniques for decision-making [4 927 29 131ndash133] This includes the ability to develop andadapt new strategies allowing us to maximize certain perfor-mance measures In particular the RL algorithms have beenproposed to achieve such unsupervised decision-making inCRs [7 29 122 123 133ndash135] The concept of RL is basedon learning from experience by trial and error [29 118]After executing a certain action the learning agent receivesa certain reward showing how good it is to take a particular

action in a certain environment state [29 118 122] As aresult the learning agent (the CR in this case) will selectcertain actions that lead to the highest rewards in a particularstateThe corresponding action selection method is based onan exploration-exploitation strategy that selects the highestreward action with a higher probability compared to theother available actions [118]This can be usually implementedusing the 120598-greedy approach which selects a greedy actionwith a probability 1 minus 120598 and a random action with smallprobability 120598 thus allowing us to avoid local optima [7 118]

The119876-learning algorithm is one of the RL algorithms thathas been proposed for CR applications [7 118 122] Under aMarkov decision process (MDP) framework the 119876-learningcan lead to optimal policy yet without knowledge of the statetransition probabilities [136 137] An MDP is characterizedby the following elements [7 29 122 133 137]

(i) a finite set S of states for the agent (ie secondaryuser)

(ii) a finite setA of actions that are available to the agent(iii) a nonnegative function 119901

119905(1199041015840 | 119904 119886) denoting the

probability that the system is in state 1199041015840 at time epoch119905 + 1 when the decision-maker chooses action 119886 isin Ain state 119904 isin S at time 119905

(iv) a real-valued function 119903MDP119905

(119904 119886) defined for state 119904 isin

S and action 119886 isin A to denote the value at time 119905 ofthe reward received in period 119905 [137]

At each time epoch 119905 the agent observes the current state119904 and chooses an action 119886 The objective is to find the optimalpolicy120587 thatmaximizes the expected discounted return [118]

119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)

where 119904119905and 119886119905are respectively the state and action at time

119905 isin 119885The optimal policy of the MDP can be based on the 119876-

function (or action-value function) which determines howgood it is to take a particular action 119886 in a given state 119904Formally the 119876-function is defined as the value of takingaction 119886 in state 119904 under a policy 120587 [118]

119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)

This function can be computed using an iterative proce-dure as follows [7 29 122 133 136]

119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)

TheRL algorithmcan be represented in the block diagramof Figure 16 in which the learning agent receives the stateobservation 119900

119905and the reward function 119903

119905at each instant

119905 [29 118] It then updates its 119876-function at the learningstage and selects an appropriate action 119886

119905 Under the MDP

assumption the119876-learning can guarantee convergence of the119876-function to its optimal value [136]Thus the optimal policy


RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at

Figure 16 The RL cycle

120587 can be obtained in function of the 119876-function such that[29 118]

119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)

where 119880(A) is the discrete uniform probability distributionover the set of actions A and 119886lowast(119904) is the optimal actionselected in state 119904 and corresponding to the optimal policy120587

The 119876-learning algorithm has been proposed for twomain CR applications

(1) aggregate interference control [8 122](2) spectrum sensing policy [7 123]

In aggregate interference control the 119876-learning wasproposed to optimize the power transmission of secondaryCRs in a WRAN IEEE 80222 CR network (CRN) [122] Theobjective is tomaintain the aggregated interference caused bythe secondary networks to the DTV network below a certainthreshold In this scenario the CRs constitute a distributednetwork and each radio tries to determine how much powerit can transmit so that the aggregated interference on theprimary receivers does not exceed a certain threshold level[122] Simulation results have shown that the 119876-learningalgorithm can successfully control the aggregate interferencein a WRAN scenario [122]

On the other hand the 119876-learning algorithm has beenproposed for opportunistic spectrum access (OSA) appli-cations to coordinate the actions of CRs in a distributedCRN [7] Given a set of CRs the 119876-learning can determinethe channel that should be sensed by each CR at eachtime instant in order to maximize the average utilizationof idle primary channels while limiting collision amongsecondary cognitive users [7] The 119876-learning algorithm wasshown to achieve near-optimal performance in such OSAapplications [7] However it cannot guarantee optimal policy

for the decentralized partially observable decision-makingproblem which is considered one of the most challengingproblems for MDPs in general [137 138] However given thelimited amount of information the119876-learning algorithm canbe considered as one of the most effective low-complexityapproaches for distributed partially observable decision-making scenarios [7 122 133]

6 Conclusion

This paper presented a review of three major front-end CRelements the RF part spectrum sensing and machine learn-ing For the RF part three types of antennas were presentedUWB antennas used for spectrum sensing frequency-reconfigurabletunable antennas for communicating overwhite spaces (also for sequential channel sensing) and UWBantennas with reconfigurable band notches for overlay UWBCR Also for the RF part it was shown that the main designchallenges are those pertaining to the ADCsDACs dynamicrange LNAs filters mixers and synthesizers Sophisticatedspectrum sensing algorithms that overcome the challengesresulting from the adaptive behavior of CR transceivers needto be developed to relax RF designs and provide accuratedecisions A CE executes actions based on certain rulesand policies that are learnt from past experience Withthe growing complexity of the current wireless networksmore sophisticated learning algorithms should be developedtaking into account the heterogeneous structure of existingand future communication networks

Conflict of Interests

The authors declare that there is no conflict of interests regar-ding the publication of this paper

References

[1] S Haykin ldquoCognitive radio brain-empowered wireless com-municationsrdquo IEEE Journal on Selected Areas in Communica-tions vol 23 no 2 pp 201ndash220 2005

[2] A Goldsmith S A Jafar I Maric and S Srinivasa ldquoBreakingspectrum gridlock with cognitive radios an information the-oretic perspectiverdquo Proceedings of the IEEE vol 97 no 5 pp894ndash914 2009

[3] N Devroye M Vu and V Tarokh ldquoCognitive radio networkshighlights of information theoretic limits models and designrdquoIEEE Signal Processing Magazine vol 25 no 6 pp 12ndash23 2008

[4] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[5] J Mitola III ldquoCognitive radio architecture evolutionrdquo Proceed-ings of the IEEE vol 97 no 4 pp 626ndash641 2009

[6] T Yucek and H Arslan ldquoA survey of spectrum sensing algo-rithms for cognitive radio applicationsrdquo IEEE CommunicationsSurveys and Tutorials vol 11 no 1 pp 116ndash130 2009

[7] M Bkassiny S K Jayaweera and K A Avery ldquoDistributedReinforcement Learning basedMAC protocols for autonomouscognitive secondary usersrdquo in Proceedings of the 20th AnnualWireless and Optical Communications Conference (WOCC rsquo11)pp 1ndash6 Newark NJ USA April 2011


[8] L Giupponi A Galindo-Serrano P Blasco and M DohlerldquoDocitive networks an emerging paradigm for dynamic spec-trum managementrdquo IEEE Wireless Communications vol 17 no4 pp 47ndash54 2010

[9] S K Jayaweera and C G Christodoulou ldquoRadiobots architec-ture algorithms and realtime reconfigurable antenna designsfor autonomous self-learning future cognitive radiosrdquo TechRep EECE-TR-11-0001 University of New Mexico 2011httprepositoryunmeduhandle192812306

[10] A Sonnenschein and P M Fishman ldquoRadiometric detectionof spread-spectrum signals in noise of uncertain powerrdquo IEEETransactions on Aerospace and Electronic Systems vol 28 no 3pp 654ndash660 1992

[11] S M Kay Fundamentals of Statistical Signal Processing Detec-tion Theory Prentice-Hall Upper Saddle River NJ USA 1998

[12] Z Ye G Memik and J Grosspietsch ldquoEnergy detection usingestimated noise variance for spectrum sensing in cognitiveradio networksrdquo in Proceedings of the IEEE Wireless Commu-nications and Networking Conference (WCNC rsquo08) pp 711ndash716Las Vegas Nev USA March-April 2008

[13] H S ChenW Gao and D G Daut ldquoSignature based spectrumsensing algorithms for IEEE 80222 WRANrdquo in Proceedingsof the IEEE International Conference on Communications (ICCrsquo07) pp 6487ndash6492 Glasgow UK June 2007

[14] S Xu Z Zhao and J Shang ldquoSpectrum sensing based oncyclostationarityrdquo inProceedings of the IEEEWorkshop on PowerElectronics and Intelligent Transportation System (PEITS rsquo08)pp 171ndash174 Guangzhou China August 2008

[15] W A Gardner ldquoExploitation of spectral redundancy in cyclo-stationary signalsrdquo IEEE Signal Processing Magazine vol 8 no2 pp 14ndash36 1991

[16] W A Gardner A Napolitano and L Paura ldquoCyclostationarityhalf a century of researchrdquo Signal Processing vol 86 no 4 pp639ndash697 2006

[17] WAGardner ldquoSpectral correlation ofmodulated signalsmdashpartI analog modulationrdquo IEEE Transactions on Communicationsvol 35 no 6 pp 584ndash595 1987

[18] W A Gardner W A Brown III and C Chen ldquoSpectralcorrelation of modulated signalsmdashpart II digital modulationrdquoIEEE Transactions on Communications vol 35 no 6 pp 595ndash601 1987

[19] Y H Zeng and Y C Liang ldquoCovariance based signal detectionsfor cognitive radiordquo in Proceedings of the 2nd IEEE InternationalSymposium on New Frontiers in Dynamic Spectrum AccessNetworks (DySPAN rsquo07) pp 202ndash207 Dublin Ireland April2007

[20] Y H Zeng and Y C Liang ldquoSpectrum-sensing algorithms forcognitive radio based on statistical covariancesrdquo IEEE Transac-tions onVehicular Technology vol 58 no 4 pp 1804ndash1815 2009

[21] Y H Zeng and Y C Liang ldquoEigenvalue based sensing algo-rithmsrdquo IEEE 802 22-060118r0 2006

[22] Y H Zeng and Y C Liang ldquoMaximum-minimum eigenvaluedetection for cognitive radiordquo in Proceedings of the 18th AnnualIEEE International Symposium on Personal Indoor and MobileRadio Communications (PIMRC rsquo07) Athens Greece Septem-ber 2007

[23] Y H Zeng and Y C Liang ldquoEigenvalue-based spectrumsensing algorithms for cognitive radiordquo IEEE Transactions onCommunications vol 57 no 6 pp 1784ndash1793 2009

[24] G FettweisM Lohning D PetrovicMWindisch P Zillmannand W Rave ldquoDirty RF a new paradigmrdquo in Proceedings of

the IEEE 16th International Symposium on Personal Indoor andMobile Radio Communications (PIMRC rsquo05) vol 4 pp 2347ndash2355 Berlin Germany September 2005

[25] M Valkama A S H Ghadam L Anttila and M RenforsldquoAdvanced digital signal processing techniques for compen-sation of nonlinear distortion in wideband multicarrier radioreceiversrdquo IEEE Transactions on Microwave Theory and Tech-niques vol 54 no 6 pp 2356ndash2366 2006

[26] M Grimm R K Sharma M A Hein and R S Thoma ldquoDSP-based mitigation of RF front-end non-linearity in cognitivewideband receiversrdquo Frequenz vol 6 no 9-10 pp 303ndash3102012

[27] C Clancy J Hecker E Stuntebeck and T OrsquoShea ldquoApplicationsof machine learning to cognitive radio networksrdquo IEEEWirelessCommunications vol 14 no 4 pp 47ndash52 2007

[28] S K Jayaweera Y Li M Bkassiny C Christodoulou andK A Avery ldquoRadiobots the autonomous self-learning futurecognitive radiosrdquo in Proceedings of the International Symposiumon Intelligent Signal Processing and Communication Systems(ISPACS rsquo11) pp 1ndash5 Chiang Mai Thailand December 2011

[29] M Bkassiny Y Li and S K Jayaweera ldquoA survey on machine-learning techniques in cognitive radiosrdquo IEEE CommunicationsSurveys Tutorials vol 15 no 3 pp 1136ndash1159 2013

[30] M Al-Husseini K Y Kabalan A El-Hajj and C GChristodoulou ldquoReconfigurable microstrip antennas for cogni-tive radiordquo in Advancement in Microstrip Antennas with RecentApplications A Kishk Ed chapter 14 pp 337ndash362 InTechRijeka Croatia 2013

[31] P Samal P Soh and G Vandenbosch ldquoA systematic designprocedure formicrostrip-based unidirectional UWB antennasrdquoProgress In Electromagnetics Research vol 143 pp 105ndash130 2013

[32] B Gong J Li Q R Zheng Y Z Yin and X S Ren ldquoAcompact inductively loadedmonopole antenna for future UWBapplicationsrdquo Progress In Electromagnetics Research vol 139 pp265ndash275 2013

[33] W Liu Y Yin W Xu and S Zuo ldquoCompact open-slot antennawith bandwidth enhancementrdquo IEEE Antennas and WirelessPropagation Letters vol 10 pp 850ndash853 2011

[34] M-RGhaderi and FMohajeri ldquoA compact hexagonalwide-slotantenna with microstrip-fed monopole for UWB applicationrdquoIEEE Antennas and Wireless Propagation Letters vol 10 pp682ndash685 2011

[35] HOraizi and SHedayati ldquoMiniaturizedUWBmonopolemicr-ostrip antenna design by the combination of Giusepe Peanoand Sierpinski carpet fractalsrdquo IEEE Antennas and WirelessPropagation Letters vol 10 pp 67ndash70 2011

[36] P C Ooi and K T Selvan ldquoThe effect of ground plane on theperformance of a square loop CPW-fed printed antennardquo Pro-gress In Electromagnetics Research Letters vol 19 pp 103ndash1112010

[37] M Al-Husseini A Ramadan Y Tawk A El-Hajj and K YKabalan ldquoDesign and ground plane consideration of a CPW-Fed UWB antennardquo in Proceedings of the International Confer-ence on Electrical and Electronics Engineering (ELECO rsquo09) ppII-151ndashII-153 Bursa Turkey November 2009

[38] MAl-Husseini A RamadanA El-Hajj andKKabalan ldquoDesi-gn of a compact and low-cost fractal-based UWB PCB Anten-nardquo in Proceedings of the National Radio Science Conference(NRSC rsquo09) pp 1ndash8 Cairo Egypt March 2009

[39] A Mehdipour K Mohammadpour-Aghdam R Faraji-Danaand M-R Kashani-Khatib ldquoA novel coplanar waveguide-fed


slot antenna for ultrawideband applicationsrdquo IEEE Transactionson Antennas and Propagation vol 56 no 12 pp 3857ndash38622008

[40] M Al-Husseini A El-Hajj and K Y Kabalan ldquoA 19-135 GHzlow-cost microstrip antennardquo in Proceedings of the InternationalWireless Communications and Mobile Computing Conference(IWCMC rsquo08) pp 1023ndash1025 Crete IslandGreeceAugust 2008

[41] Z N Low J H Cheong and C L Law ldquoLow-cost PCB antennaforUWBapplicationsrdquo IEEEAntennas andWireless PropagationLetters vol 4 no 1 pp 237ndash239 2005

[42] M Al-Husseini Y Tawk A El-Hajj and K Y Kabalan ldquoAlow-costmicrostrip antenna for 3GWLANWiMAX andUWBapplicationsrdquo in Proceedings of the International Conference onAdvances in Computational Tools for Engineering Applications(ACTEA rsquo09) pp 68ndash70 Zouk Mosbeh Lebanon July 2009

[43] M Al-Husseini A Ramadan Y Tawk A El-Hajj and K YKabalan ldquoDesign and ground plane optimization of a CPW-fedultra-wideband antennardquo Turkish Journal of Electrical Engineer-ing and Computer Sciences vol 19 no 2 pp 243ndash250 2011

[44] H Zhang X Zhou and T Chen ldquoUltra-wideband cognitiveradio for dynamic spectrum accessing networksrdquo in CognitiveRadio Networks Y Xiao and F Hu Eds pp 353ndash382 CRCPress Boca Raton Fla USA 2009

[45] L Safatly M Al-Husseini A El-Hajj and K Kabalan ldquoAdva-nced techniques and antenna design for pulse shaping in UWBcognitive radiordquo International Journal of Antennas and Propa-gation vol 2012 Article ID 390280 8 pages 2012

[46] J B Pendry A J Holden D J Robbins andW J Stewart ldquoMag-netism from conductors and enhanced nonlinear phenomenardquoIEEE Transactions onMicrowaveTheory and Techniques vol 47no 11 pp 2075ndash2084 1999

[47] F Falcone T Lopetegi M A G Laso et al ldquoBabinet principleapplied to the design of metasurfaces and metamaterialsrdquoPhysical Review Letters vol 93 no 19 Article ID 197401 4 pages2004

[48] P Moeikham C Mahatthanajatuphat and P AkkaraekthalinldquoA compact uwb antenna with a quarter-wavelength strip in arectangular slot for 55 GHz band notchrdquo International Journalof Antennas and Propagation vol 2013 Article ID 574128 9pages 2013

[49] B Yan D Jiang R M Xu and Y H Xu ldquoA UWB band-passantenna with triple-notched band using common directionrectangular complementary split-ring resonatorsrdquo InternationalJournal of Antennas and Propagation vol 2013 Article ID934802 6 pages 2013

[50] X Liu Y-Z Yin P Liu J Wang and B Xu ldquoA CPW-fed dualbandnotchedUWB antennawith a pair of bended dual-L-shapeparasitic branchesrdquo Progress In Electromagnetics Research vol136 pp 623ndash634 2013

[51] R Azim and M Islam ldquoCompact planar UWB antenna withband notch characteristics for WLAN and DSRCrdquo Progress InElectromagnetics Research vol 133 pp 391ndash406 2013

[52] J R Kelly P S Hall and P Gardner ldquoBand-notched UWB ant-enna incorporating a microstrip open-loop resonatorrdquo IEEETransactions on Antennas and Propagation vol 59 no 8 pp3045ndash3048 2011

[53] T-D Nguyen D-H Lee and H-C Park ldquoDesign and analysisof compact printed triple band-notched UWB antennardquo IEEEAntennas andWireless Propagation Letters vol 10 pp 403ndash4062011

[54] M Almalkawi and V Devabhaktuni ldquoUltrawideband antennawith triple band-notched characteristics using closed-loop ring

resonatorsrdquo IEEE Antennas and Wireless Propagation Lettersvol 10 pp 959ndash962 2011

[55] D-O Kim N-I Jo H-A Jang and C-Y Kim ldquoDesign of theultrawideband antenna with a quadruple-band rejection char-acteristics using a combination of the complementary split ringresonatorsrdquo Progress In Electromagnetics Research vol 112 pp93ndash107 2011

[56] Y LiW Li andQ Ye ldquoA reconfigurable triple-notch-band ante-nna integrated with defected microstrip structure band-stopfilter for ultrawideband cognitive radio applicationsrdquo Interna-tional Journal of Antennas and Propagation vol 2013 Article ID472645 13 pages 2013

[57] T Aboufoul A Alomainy and C Parini ldquoReconfigured andnotched tapered slot UWB antenna for cognitive radio applica-tionsrdquo International Journal of Antennas and Propagation vol2012 Article ID 160219 8 pages 2012

[58] J Perruisseau-Carrier P Pardo-Carrera and P MiskovskyldquoModeling design and characterization of a very wideband slotantenna with reconfigurable band rejectionrdquo IEEE Transactionson Antennas and Propagation vol 58 no 7 pp 2218ndash2226 2010

[59] C-Y Sim W-T Chung and C-H Lee ldquoPlanar uwb antennawith 5ghz band rejec-tion switching function at ground planerdquoProgress In Electromagnetics Research vol 106 pp 321ndash333 2010

[60] MAl-Husseini J Costantine CGChristodoulou S E BarbinA El-Hajj and K Y Kabalan ldquoA reconfigurable frequency-notched UWB antenna with split-ring resonatorsrdquo in Proceed-ings of the Asia-Pacific Microwave Conference (APMC rsquo10) pp618ndash621 Yokohama Japan December 2010

[61] MAl-Husseini A Ramadan A El-Hajj K Y Kabalan Y Tawkand C G Christodoulou ldquoDesign based on complementarysplit-ring resonators of an antenna with controllable bandnotches forUWBcognitive radio applicationsrdquo inProceedings ofthe IEEE International Symposium onAntennas and Propagation(APSURSI) pp 1120ndash1122 Spokane Wash USA July 2011

[62] M Al-Husseini L Safatly A Ramadan A El-Hajj K Y Kab-alan and C G Christodoulou ldquoReconfigurable filter antennasfor pulse adaptation in UWB cognitive radio systemsrdquo ProgressIn Electromagnetics Research B no 37 pp 327ndash342 2011

[63] E Ebrahimi J R Kelly and P S Hall ldquoIntegrated wide-nar-rowband antenna for multi-standard radiordquo IEEE Transactionson Antennas and Propagation vol 59 no 7 pp 2628ndash2635 2011

[64] G Augustin and A Denidni ldquoAn integrated ultra widebandnarrow band antenna in uniplanar configuration for cognitiveradio systemsrdquo IEEETransactions onAntennas and Propagationvol 60 no 11 pp 5479ndash5484 2012

[65] MAl-Husseini Y Tawk CG Christodoulou A ElHajj andKY Kabalan ldquoA reconfigurable cognitive radio antenna designrdquoin Proceedings of the IEEE Antennas and Propagation SocietyInternational Symposium (APSURSI) pp 1ndash4 Toronto CanadaJuly 2010

[66] M R Hamid P Gardner P S Hall and F Ghanem ldquoSwitched-band Vivaldi antennardquo IEEE Transactions on Antennas and Pro-pagation vol 59 no 5 pp 1472ndash1480 2011

[67] M R Hamid P Gardner P S Hall and F Ghanem ldquoVivaldiantenna with integrated switchable band pass resonatorrdquo IEEETransactions on Antennas and Propagation vol 59 no 11 pp4008ndash4015 2011

[68] T Aboufoul A Alomainy and C Parini ldquoReconfiguring UWBmonopole antenna for cognitive radio applications using GaAsFET switchesrdquo IEEE Antennas andWireless Propagation Lettersvol 11 pp 392ndash394 2012


[69] M Al-Husseini A Ramadan M E Zamudio C G Christodo-ulou A El-Hajj and K Y Kabalan ldquoAUWB antenna combinedwith a reconfigurable bandpass filter for cognitive radio appli-cationsrdquo in Proceedings of the IEEE-APS Topical Conference onAntennas and Propagation inWireless Communications (APWCrsquo11) pp 902ndash904 Torino Italy September 2011

[70] Y Cai Y Guo and T Bird ldquoA frequency reconfigurable printedyagiuda dipole antenna for cognitive radio applicationsrdquo IEEETransactions on Antennas and Propagation vol 60 no 6 pp2905ndash2912 2012

[71] M Y A Shahine M Al-Husseini Y Nasser K Y Kabalan andA El-Hajj ldquoA reconfigurableminiaturized spiral monopole ant-enna for TV white spacesrdquo in Proceedings of the 34th Progress inElectromagnetics Research Symposium (PIERS rsquo13) StockholmSweden August 2013

[72] G Mansour P Hall P Gardner and M-K A Rahim ldquoTunableslotloaded patch antenna for cognitive radiordquo in Proceedings ofthe Loughborough Antennas and Propagation Conference (LAPCrsquo12) Loughborough UK November 2012

[73] Y Tawk M E Zamudio J Costantine and C ChristodoulouldquoA cognitive radio reconfigurable lsquofiltennarsquordquo in Proceedings of the6th European Conference onAntennas and Propagation (EUCAPrsquo12) pp 3565ndash3568 Prague Czech Republic March 2012

[74] V Nguyen F Villain and Y Le Guillou ldquoCognitive radio RFoverview and challengesrdquo VLSI Design 2012

[75] M Kitsunezuka K Kunihiro andM Fukaishi ldquoEfficient use ofthe spectrumrdquo IEEE Microwave Magazine vol 13 no 1 pp 55ndash63 2012

[76] B Razavi ldquoCognitive radio design challenges and techniquesrdquoIEEE Journal of Solid-State Circuits vol 45 no 8 pp 1542ndash15532010

[77] J Arias L Quintanilla J Segundo L Enrıquez J Vicente andJ M Hernandez-Mangas ldquoParallel continuous-time ΔΣ ADCfor OFDM UWB receiversrdquo IEEE Transactions on Circuits andSystems I vol 56 no 7 pp 1478ndash1487 2009

[78] C Lelandais-Perrault T Petrescu D Poulton P Duhamel andJ Oksman ldquoWideband bandpass and versatile hybrid filterbank AD conversion for software radiordquo IEEE Transactions onCircuits and Systems I vol 56 no 8 pp 1772ndash1782 2009

[79] C-C Huang C-YWang and J-TWu ldquoA CMOS 6-Bit 16-GSstime-interleaved ADC using digital background calibrationtechniquesrdquo IEEE Journal of Solid-State Circuits vol 46 no 4pp 848ndash858 2011

[80] A Gruget M Roger V T Nguyen C Lelandais-Perrault PBenabes and P Loumeau ldquoWide-band multipath A to D conv-erter for cognitive radio applicationsrdquo in Proceedings of the IEEEInternational Microwave Workshop Series on RF Front-ends forSoftware Defined and Cognitive Radio Solutions (IMWS rsquo10) pp73ndash76 Aveiro Portugal February 2010

[81] A Gruget M Roger V T Nguyen C Lelandais-Perrault PBenabes and P Loumeau ldquoOptimization of bandpass chargesampling filters in hybrid filter banks converters for cognitiveradio applicationsrdquo in Proceedings of the 20th European Confer-ence on Circuit Theory and Design (ECCTD rsquo11) pp 785ndash788Linkoping Sweden August 2011

[82] Z Boos A Menkhoff F Kuttner et al ldquoA fully digital multi-mode polar transmitter employing 17b RF DAC in 3G moderdquoin Proceedings of the IEEE International Solid-State CircuitsConference (ISSCC rsquo11) pp 376ndash377 San Francisco Calif USAFebruary 2011

[83] A Pozsgay T Zounes R Hossain M Boulemnakher V Kno-pik and S Grange ldquoA fully digital 65 nmCMOS transmitter for

the 24-to-27GHz WiFiWiMAX bands using 54GHz 120575120590 RFDACsrdquo in Proceedings of the IEEE International Solid StateCircuits Conference (ISSCC rsquo08) pp 355ndash619 San FranciscoCalif USA February 2008

[84] J Borremans G Mandal V Giannini et al ldquoA 40 nm CMOS04ndash6GHz receiver resilient to out-of-band blockersrdquo IEEEJournal of Solid-State Circuits vol 46 no 7 pp 1659ndash1671 2011

[85] MCM Soer E AMKlumperink Z Ru F E vanVliet andBNauta ldquoA 02-to-20GHz 65 nm CMOS receiver without LNAachieving gt11 dBm IIP3 and lt65 dB NFrdquo in Proceedings of theIEEE International Solid-State Circuits Conference (ISSCC rsquo09)pp 222ndash223 San Francisco Calif USA February 2009

[86] C Andrews and A C Molnar ldquoA passive mixer-first receiverwith digitally controlled and widely tunable RF interfacerdquo IEEEJournal of Solid-State Circuits vol 45 no 12 pp 2696ndash27082010

[87] A Mirzaie A Yazdi Z Zhou E Chang P Suri and H Dar-abi ldquoA 65 nm CMOS quad-band SAW-less receiver for GSMGPRSEDGErdquo in Proceedings of the Symposium onVLSI Circuits(VLSIC rsquo10) pp 179ndash180 Honolulu Hawaii USA June 2010

[88] A Ghaffari E A M Klumperink M C M Soer and BNauta ldquoTunable high-qN-Path Band-Pass filters modeling andverificationrdquo IEEE Journal of Solid-State Circuits vol 46 no 5pp 998ndash1010 2011

[89] S C Blaakmeer E A M Klumperink B Nauta and D M WLeenaerts ldquoAn inductorless wideband balun-LNA in 65 nmCMOS with balanced outputrdquo in Proceedings of the 33rd Euro-pean Solid-State Circuits Conference (ESSCIRC rsquo07) pp 364ndash367 Munich Germany September 2007

[90] Z Ru N A Moseley E A M Klumperink and B Nauta ldquoDig-itally enhanced software-defined radio receiver robust to out-of-band interferencerdquo IEEE Journal of Solid-State Circuits vol44 no 12 pp 3359ndash3375 2009

[91] A A Rafi A Piovaccari P Vancorenland and T Tuttle ldquoA har-monic rejection mixer robust to RF device mismatchesrdquo in Pro-ceedings of the IEEE International Solid-State Circuits Conference(ISSCC rsquo11) pp 66ndash68 San Francisco Calif USA February 2011

[92] J Borremans K Vengattaramane V Giannini B Debaillie Wvan Thillo and J Craninckx ldquoA 86MHzndash12GHz digital-inten-sive PLL for software-defined radios using a 6 fJstep TDC in40 nm digital CMOSrdquo IEEE Journal of Solid-State Circuits vol45 no 10 pp 2116ndash2129 2010

[93] J Mitola Cognitive radio an integrated agent architecture forsoftware defined radio [PhD thesis] Royal Institute of Technol-ogy KTH 2000

[94] Z Quan S Cui A H Sayed and H V Poor ldquoOptimal multib-and joint detection for spectrum sensing in cognitive radio net-worksrdquo IEEE Transactions on Signal Processing vol 57 no 3 pp1128ndash1140 2009

[95] Z Tian and G B Giannakis ldquoA wavelet approach to widebandspectrum sensing for cognitive radiosrdquo in Proceedings of the 1stInternational Conference on Cognitive Radio Oriented WirelessNetworks and Communications (CROWNCOM rsquo06) pp 1ndash5Mykonos Island Greece June 2006

[96] W Liu and S Ding ldquoMultiuser detections based on global opti-mality necessary conditions for binary quadratic programm-ingrdquo in Proceedings of the 2nd International Conference on Inn-ovative Computing Information andControl (ICICIC rsquo07) Kum-amoto Japan September 2007

[97] Z Tian andG BGiannakis ldquoCompressed sensing forwidebandcognitive radiosrdquo in Proceedings of the IEEE International


Conference on Acoustics Speech and Signal Processing (ICASSPrsquo07) vol 4 pp IV-1357ndashIV-1360 Honolulu Hawaii USA April2007

[98] E G Larsson and M Skoglund ldquoCognitive radio in afrequency-planned environment some basic limitsrdquo IEEETransactions on Wireless Communications vol 7 no 12 pp4800ndash4806 2008

[99] A Sahai N Hoven and R Tandra ldquoSome fundamental limitson cognitive radiordquo in Proceedings of the 42nd Annual AllertonConference on Communication Control and Computing pp1662ndash1671 Monticello Ill USA October 2004

[100] R Tandra and A Sahai ldquoSNR walls for signal detectionrdquo IEEEJournal on Selected Topics in Signal Processing vol 2 no 1 pp4ndash17 2008

[101] E G Larsson and G Regnoli ldquoPrimary system detection forcognitive radio does small-scale fading helprdquo IEEE Commu-nications Letters vol 11 no 10 pp 799ndash801 2007

[102] IM Johnstone ldquoOn the distribution of the largest eigenvalue inprincipal components analysisrdquo Annals of Statistics vol 29 no2 pp 295ndash327 2001

[103] Z Khalaf A Nafkha and J Palicot ldquoBlind spectrum detectorfor cognitive radio using compressed sensing and symmetryproperty of the second order cyclic autocorrelationrdquo in Proceed-ings of the 7th International ICST Conference on Cognitive RadioOriented Wireless Networks and Communications (CROWN-COM rsquo12) pp 291ndash296 Stockholm Sweden June 2012

[104] G Davis S Mallat and M Avellaneda ldquoAdaptive greedy appr-oximationsrdquo Constructive Approximation vol 13 no 1 pp 57ndash98 1997

[105] A Sahai and D Cabric ldquoSpectrum sensing fundamental limitsand practical challengesrdquo in Proceedings of the IEEE Interna-tional Symposium onNew Frontiers in Dynamic SpectrumAccessNetworks (DySPAN rsquo05) 2005

[106] H Sun A Nallanathan C X Wang and Y Chen ldquoWidebandspectrum sensing for cognitive radio networks a surveyrdquo IEEEWireless Communications vol 20 no 2 pp 74ndash81 2013

[107] S Chantaraskul and K Moessner ldquoImplementation of waveletanalysis for spectrum opportunity detectionrdquo in Proceedings ofthe IEEE 20th Personal Indoor and Mobile Radio Communi-cations Symposium (PIMRC rsquo09) pp 2310ndash2314 Tokyo JapanSeptember 2009

[108] Y L Polo Y Wang A Pandharipande and G Leus ldquoCompres-sive wide-band spectrum sensingrdquo in Proceedings of the IEEEInternational Conference on Acoustics Speech and Signal Pro-cessing (ICASSP rsquo09) pp 2337ndash2340 Taipei City Taiwan April2009

[109] S G Mallat and Z Zhang ldquoMatching pursuits with time-freq-uency dictionariesrdquo IEEE Transactions on Signal Processing vol41 no 12 pp 3397ndash3415 1993

[110] Y Zhao and L Morales-Tirado ldquoCognitive radio technologyprinciples and practicerdquo in Proceedings of the International Con-ference on Computing Networking and Communications (ICNCrsquo12) pp 650ndash654 Maui Hawaii USA February 2012

[111] Y Kon M Ito N Hassel M Hasegawa K Ishizu and HHarada ldquoAutonomous parameter optimization of a heterogene-ous wireless network aggregation system using machine learn-ing algorithmsrdquo in Proceedings of the IEEE Consumer Commu-nications and Networking Conference (CCNC rsquo12) pp 894ndash898Las Vegas Nev USA January 2012

[112] P Venkatraman B Hamdaoui and M Guizani ldquoOpportunis-tic bandwidth sharing through reinforcement learningrdquo IEEE

Transactions on Vehicular Technology vol 59 no 6 pp 3148ndash3153 2010

[113] J Oksanen J Lunden and V Koivunen ldquoReinforcement lear-ning method for energy efficient cooperative multiband spec-trum sensingrdquo in Proceedings of the IEEE International Work-shop on Machine Learning for Signal Processing (MLSP rsquo10) pp59ndash64 Kittila Finland September 2010

[114] Z Chen and R C Qiu ldquoQ-learning based bidding algorithm forspectrum auction in cognitive radiordquo in Proceedings of the IEEESoutheastCon 2011 pp 409ndash412 Nashville Tenn USA March2011

[115] K-LA Yau P Komisarczuk and P D Teal ldquoPerformance anal-ysis of reinforcement learning for achieving context awarenessand intelligence in mobile cognitive radio networksrdquo in Pro-ceedings of the IEEE International Conference on Advanced Info-rmation Networking and Applications (AINA rsquo11) pp 1ndash8 Singa-pore March 2011

[116] P Venkatraman and B Hamdaoui ldquoCooperative Q-learning formultiple secondary users in dynamic spectrum accessrdquo in Proc-eedings of the 7th International Wireless Communications andMobile Computing Conference (IWCMC rsquo11) pp 238ndash242 Istan-bul Turkey July 2011

[117] D Kumar N Kanagaraj and R Srilakshmi ldquoHarmonized q-learning for radio resource management in lte based networksrdquoin Proceedings of ITU Kaleidoscope Building Sustainable Com-munities 2013

[118] R S Sutton andA G BartoReinforcement Learning An Introd-uction MIT Press Cambridge Mass USA 1998

[119] M Bkassiny S K Jayaweera Y Li and K A Avery ldquoWidebandspectrum sensing and non-parametric signal classification forautonomous self-learning cognitive radiosrdquo IEEE Transactionson Wireless Communications vol 11 no 7 pp 2596ndash2605 2012

[120] T C Clancy A Khawar and T R Newman ldquoRobust signal clas-sification using unsupervised learningrdquo IEEE Transactions onWireless Communications vol 10 no 4 pp 1289ndash1299 2011

[121] N Shetty S Pollin and P Pawelczak ldquoIdentifying spectrumusage by unknown systems using experiments in machine lear-ningrdquo in Proceedings of the IEEE Wireless Communications andNetworking Conference (WCNC rsquo09) pp 1ndash6 Budapest Hun-gary April 2009

[122] A Galindo-Serrano and L Giupponi ldquoDistributed Q-learningfor aggregated interference control in cognitive radio networksrdquoIEEE Transactions on Vehicular Technology vol 59 no 4 pp1823ndash1834 2010

[123] K-L A Yau P Komisarczuk and P D Teal ldquoApplications ofreinforcement learning to cognitive radio networksrdquo inProceed-ings of the IEEE International Conference on CommunicationsWorkshops ICC 2010 pp 1ndash6 Capetown South Africa May2010

[124] M Bkassiny S K Jayaweera and Y Li ldquoMultidimensional Diri-chlet process-based non-parametric signal classification for aut-onomous selflearning cognitive radiosrdquo IEEE Transactions onWireless Communications vol 12 no 11 pp 5413ndash5423 2013

[125] C Phelps and R Buehrer ldquoSignal classification by probabilisticreasoningrdquo in Proceedings of the IEEE Radio and WirelessSymposium (RWS rsquo13) pp 154ndash156 Austin Tex USA January2013

[126] D Pelleg and A Moore ldquoX-means extending k-means withefficient estimation of the number of clustersrdquo in Proceedingsof the 7th International Conference on Machine Learning (ICMLrsquo00) Stanford Calif USA June-July 2000


[127] Y W Teh M I Jordan M J Beal and D M Blei ldquoHierarchicalDirichlet processesrdquo Journal of the American Statistical Associa-tion vol 101 no 476 pp 1566ndash1581 2006

[128] C-YWang Y Chen andK Liu ldquoSequential Chinese restaurantgamerdquo IEEE Transactions on Signal Processing vol 61 no 3 pp571ndash584 2013

[129] M D Escobar and M West ldquoBayesian density estimation andinference using mixturesrdquo Journal of the American StatisticalAssociation vol 90 no 430 pp 577ndash588 1995

[130] M D Escobar ldquoEstimating normal means with a dirichlet pro-cess priorrdquo Journal of the American Statistical Association vol89 no 425 pp 268ndash277 1994

[131] J Mitola Cognitive radio an integrated agent architecture forsoftware defined radio [PhD thesis] KTH 2000

[132] V Stavroulaki A Bantouna Y Kritikou et al ldquoKnowledgeman-agement toolbox machine learning for cognitive radio net-worksrdquo IEEE Vehicular Technology Magazine vol 7 no 2 pp91ndash99 2012

[133] L Gavrilovska V Atanasovski I Macaluso and L DaSilvaldquoLearning and reasoning in cognitive radio networksrdquo IEEECommunications Surveys and Tutorials vol 15 no 4 pp 1761ndash1777 2013

[134] Y Li S K Jayaweera M Bkassiny and C Ghosh ldquoLearning-aided sub-band selection algorithms for spectrum sensing inwide-band cognitive radiosrdquo IEEE Transactions on WirelessCommunications In press

[135] Y Li S K Jayaweera C Ghosh and M Bkassiny ldquoLearning-aided sensing scheduling for wide-band cognitive radiosrdquo inProceedings of the IEEE 78th Vehicular Technology Conference(VTC Fall rsquo13) pp 1ndash5 Las Vegas Nev USA September 2013

[136] C J C H Watkins and P Dayan ldquoQ-learningrdquo Machine Lear-ning vol 8 no 3-4 pp 279ndash292 1992

[137] M L PutermanMarkov Decision Processes Discrete StochasticDynamic Programming John Wiley amp Sons New York NYUSA 1994

[138] L Busoniu R Babuska and B de Schutter ldquoMulti-agent reinfo-rcement learning a surveyrdquo in Proceedings of the 9th Interna-tional Conference on Control Automation Robotics and Vision(ICARCV rsquo06) pp 1ndash6 Singapore December 2006

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



In CR smart transceivers scan the spectrum in orderto find white spaces and transmit adaptive signals Thisfunctionality requires sophisticated algorithms to overcomepractical imperfections such as model uncertainty and hard-ware nonideality Some of the well-known spectrum sensingalgorithms are energy detection [10ndash12] matched filter-based detection [11 13] cyclostationarity-based detection[14ndash18] covariance-based sensing [19 20] and eigenvalue-based sensing [21ndash23] Blind detectors were also introducedto elude the model uncertainty problem Spectrum sensingalgorithms could be affected by RF impairments by inducingunwanted frequency components in the collected signal spec-trum The effects of such impairments are reduced througha postprocessing of the signal [24ndash26] In addition a robustdetector based on smart digital signal processing lowers theeffects of RF impairments and guarantees a high sensingaccuracy

The concept of cognitive radios (CRs) goes beyonddynamic spectrum access (DSA) applications and aims toimprove the quality of information (QoI) of users [5] Thisrequires an intelligent radio that uses spectrum sensingtechniques to observe the RF activities and is able toautonomously adapt to particular situations [9] This isachieved through a reasoning engine which executes actionsbased on certain hard-coded rules and strategies [27] Hard-codedpolicies are completely specified by the systemdesignerand may result in the desired performance as long asthe operating conditions do not deviate from the originalassumed model However due to unexpected changes inthe RF environment the hard-coded rules may not lead tooptimal performance making them inefficient in this caseCognitive radios however can overcome this problem byupdating or augmenting their own sets of policies and rulesbased on past experience [27] which may lead to a morereliable communication performance [9 27] This makes thelearning ability a fundamental building block of any CR toachieve autonomous intelligent behavior [9 27ndash29]

In this paper Section 2 presents the antenna designsfor CR ultrawideband (UWB) antennas antennas withreconfigurable band rejection and frequency-reconfigurableor tunable antennas It also briefly discusses the challengesfaced by the design of RF blocks for CR devices Section 3discusses spectrum sensing by presenting challenges novelsolutions and spectrum sensing algorithms RF imperfec-tions and wideband sensing are also studied and recent blinddetectors robust algorithms and wideband techniques arepresented Section 4 presents a cognitive engine and severalunsupervised classification algorithms for autonomous signalclassification in CRs and a reinforcement learning (RL)algorithm that performs decision-making in CR networksConclusive remarks are given in Section 6

2 RF Frontends for Cognitive Radio

CR transceivers are required to look for and operate inwhite spaces which could exist anywhere inside a widefrequency range This comes different from conventionalwireless transceivers which are bound to certain preallocated

frequency bands As a result significant challenges have tobe dealt with when designing the RF components of a CRtransceiver such as the antennas the power amplifiers (PAs)the filtersduplexers and the analog-to-digital and digital-to-analog converters (ADCsDACs)

21 Antennas for Cognitive Radio Two main approaches ofsharing spectrum between primary users (PUs) and sec-ondary users (SUs) exist spectrum underlay and spectrumoverlay In the underlay approach SUs should operate belowthe noise floor of PUs and thus contingent constraints areimposed on their transmission powerUltrawideband (UWB)technology is very suitable as the enabling technology for thisapproach In spectrum overlay CR SUs search for unusedfrequency bands called white spaces and use them forcommunication

UWB antennas are used for underlay CR and alsofor channel sensing in overlay CR For communication inoverlay CR the antenna must be frequency reconfigurableor tunable Single- and dual-port antennas for overlay CRcan be designed In the dual-port case one port has UWBfrequency response and is used for channel sensing and thesecond port which is frequency reconfigurabletunable isused for communicating In themore challenging single-portdesign the sameport can haveUWBresponse for sensing andcan be reconfigured for tunable narrowband operation whenrequired to communicate over a white space

A third possible spectrum sharing approach results fromthe use of the UWB technology in an overlay scheme Herethe antennas could basically be UWB antennas but shouldhave the ability to selectively induce frequency notches inthe bands used by PUs thus avoiding any interference tothem and giving the UWB transmitters of the SUs the chanceto increase their output power and hence to achieve long-distance communication

A concise review of antenna designs for cognitive radiocovering the above three antenna types is given in [30]

211 UWB Antennas UWB antennas were originally meantto radiate very short pulses over short distances They havebeen used in medical applications GPRs and other short-range communications requiring high throughputs The lit-erature is rich with articles pertaining to the design of UWBantennas [31ndash41] For example the authors in [31] present aunidirectional UWB antenna based on a full-ground planeTo keep this full-ground plane they sequentially employ alist of broadbanding techniques (1) resonance overlapping(2) slot (3) parasitic patch (4) Vivaldi blending (5) steppednotch and (6) rectangular and T-shape slits The resultingantenna has an impedance bandwidth from 36GHz to103 GHz while keeping the unidirectionality of the radiationpattern

In general the guidelines to design UWB antennasinclude the following

(i) Theproper selection of the patch shape Round shapesand round edges lead to smoother current flow andasa result to better wideband characteristics


50

39

7

10

40

203

403

399209

78

3

39

3

15

4

44

4

44

162

17854

183


XZ

Y

Gro

und

plan

e


45∘


2 4 6 8 10 12

SimulatedMeasured

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

minus45

Refle

ctio

n co

effici

ent (

dB)

Frequency (GHz)


















14

4505

12

13

1

7155

31

1

1

24

105 10

S1S2S3


10

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

Refle

ctio

n co

effici

ent (

dB)













40

20

11

1195

15

60

245

975 79

4152



Case 1Case 2

Case 3Case 4

0

minus5

minus10

minus15

minus20

minus25

minus30

Refle

ctio

n co

effici

ent (

dB)

2 4 53 6 8 97

Frequency (GHz)








50

50

15 201

2

3

33

16

55

35 57

36

9

12

9

Units mm

S1

S2


0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)


SimulatedMeasured

S1OFFS2OFF

(a)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2OFF

(b)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2ON

(c)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11



Case 3 ON-ON


(d)



40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4


0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










50

39

7

10

40

203

403

399209

78

3

39

3

15

4

44

4

44

162

17854

183


XZ

Y

Gro

und

plan

e


45∘


2 4 6 8 10 12

SimulatedMeasured

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

minus45

Refle

ctio

n co

effici

ent (

dB)

Frequency (GHz)


















14

4505

12

13

1

7155

31

1

1

24

105 10

S1S2S3


10

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

Refle

ctio

n co

effici

ent (

dB)













40

20

11

1195

15

60

245

975 79

4152



Case 1Case 2

Case 3Case 4

0

minus5

minus10

minus15

minus20

minus25

minus30

Refle

ctio

n co

effici

ent (

dB)

2 4 53 6 8 97

Frequency (GHz)








50

50

15 201

2

3

33

16

55

35 57

36

9

12

9

Units mm

S1

S2


0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)


SimulatedMeasured

S1OFFS2OFF

(a)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2OFF

(b)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2ON

(c)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11



Case 3 ON-ON


(d)



40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4


0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation



















14

4505

12

13

1

7155

31

1

1

24

105 10

S1S2S3


10

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

Refle

ctio

n co

effici

ent (

dB)













40

20

11

1195

15

60

245

975 79

4152



Case 1Case 2

Case 3Case 4

0

minus5

minus10

minus15

minus20

minus25

minus30

Refle

ctio

n co

effici

ent (

dB)

2 4 53 6 8 97

Frequency (GHz)








50

50

15 201

2

3

33

16

55

35 57

36

9

12

9

Units mm

S1

S2


0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)


SimulatedMeasured

S1OFFS2OFF

(a)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2OFF

(b)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2ON

(c)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11



Case 3 ON-ON


(d)



40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4


0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










14

4505

12

13

1

7155

31

1

1

24

105 10

S1S2S3


10

0

minus5

minus10

minus15

minus20

minus25

minus30

minus35

minus40

Refle

ctio

n co

effici

ent (

dB)













40

20

11

1195

15

60

245

975 79

4152



Case 1Case 2

Case 3Case 4

0

minus5

minus10

minus15

minus20

minus25

minus30

Refle

ctio

n co

effici

ent (

dB)

2 4 53 6 8 97

Frequency (GHz)








50

50

15 201

2

3

33

16

55

35 57

36

9

12

9

Units mm

S1

S2


0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)


SimulatedMeasured

S1OFFS2OFF

(a)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2OFF

(b)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2ON

(c)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11



Case 3 ON-ON


(d)



40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4


0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










40

20

11

1195

15

60

245

975 79

4152



Case 1Case 2

Case 3Case 4

0

minus5

minus10

minus15

minus20

minus25

minus30

Refle

ctio

n co

effici

ent (

dB)

2 4 53 6 8 97

Frequency (GHz)








50

50

15 201

2

3

33

16

55

35 57

36

9

12

9

Units mm

S1

S2


0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)


SimulatedMeasured

S1OFFS2OFF

(a)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2OFF

(b)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2ON

(c)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11



Case 3 ON-ON


(d)



40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4


0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










50

50

15 201

2

3

33

16

55

35 57

36

9

12

9

Units mm

S1

S2


0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)


SimulatedMeasured

S1OFFS2OFF

(a)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2OFF

(b)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11


SimulatedMeasured

S1ONS2ON

(c)

0

minus10

minus20

minus30

minus40

Refle

ctio

n co

effici

ent (

dB)

10 11



Case 3 ON-ON


(d)



40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4


0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










40

20

11

1195

15

60

245

025

4

20310

775

152RF choke

DC line

Varactor RF choke

Via


4


0

minus5

minus10

minus15

minus20

minus25

Refle

ctio

n co

effici

ent (

dB)

4 5 6 87

Frequency (GHz)
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation
























119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











119878


119899

given byx(119899) = [119909

1(119899) 119909

2(119899) 119909

119872(119899)]119879



= [x119879 (119899) x119879 (119899 minus 1) x119879 (119899 minus 119871 + 1)]119879


)

R (119873119878

) =1

119873119878

119871minus1+119873119878

sum119899=119871

x119899

x119867119899



] =(radic119873119878

+ radic119872119871)2

(radic119873119878


(1 +(radic119873119878

+ radic119872119871)minus23

(119873119878

119872119871)16

119865minus1

1

(1 minus 119875119865119860

))

where 1198651


0

or 1198671


≷1198671

1198670]



120582max120582min

≷1198671

1198670

] (1)











1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












1205910given by

f120591

= [119891120591

(0) 119891120591

(1) 119891120591

(119873 minus 1)]119879

where 119891120591


119886(119901119902)

1198902119894120587(119901minus1)(119902minus1)119873


120591

= f120591


120591


120591


IND(120591)sym =1

119897

119897

sum119894=1

r120591119894


IND(equ)sym =1

119872

119872

sum119894=1

10038161003816100381610038161003816IND(120591119894)sym

10038161003816100381610038161003816lt 0





SDR RF frontend

band splitter


Adaptivefilter

Baseband processing

PU

SU

Coarse ED ++























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation


























densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











densityWavelet


detection







Knowledge base

Reasoningengine

Learningengine

Cognitive engine











119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation














119894=1


1 c









120579119894| 120579119895119895 = 119894

y1 y

119873

= 120579119895

with prob 119902119895

=119891120579119895

(y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

sim 119891 (120579119894| y119894) with prob 119902

0=

1205720119891 (y119894)

1205720119891 (y119894) + sum119873

119895=1119895 = 119894

119891120579119895

(y119894)

(2)


119894119873

119894=1

into 119870






119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












119866 sim 119863119875 (1205720 1198660)

120579119894| 119866 sim 119866

y119894| 120579119894sim 119891120579119894

(y119894)

(3)









(y119894) where 120579

119894is drawn from119866 [29 124]







119894rsquos for a


119894=1


0 1198660) for 120579


119894rsquos


1 120579

119873|

y1 y





119894rsquos


| 120579119895119895 = 119894

y1 y

119873in (2)








119905(1199041015840 | 119904 119886) denoting the






119877 (119905) =

infin

sum119896=0

120574119896

119903119905+119896+1

(119904119905+119896

119886119905+119896

) (4)




119876120587

(119904 119886) = E120587

119877 (119905) | 119904119905

= 119904 119886119905

= 119886 (5)


119876 (119904119905 119886119905) larr997888 (1 minus 120572) 119876 (119904

119905 119886119905)

+ 120572 [119903119905+1

(119904119905 119886119905) + 120574max

119886

119876 (119904119905+1

119886)] (6)








RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










RFenvironment

Reward rt+1

Reward rt

State st+1

State st

Sensors

Observe

Learn

Decide

Observation ot

Action at



119886lowast

(119904) = argmax119886isinA

119876 (119904 119886) with Pr = 1 minus 120576

sim 119880 (A) with Pr = 120576(7)







6 Conclusion




References
























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation

























































































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation























































































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation























































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation
























































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation
























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Documents

Review Article Cognitive Radio Transceivers: RF, Spectrum ...downloads.hindawi.com/journals/ijap/2014/548473.pdf · ects of such impairments are reduced through a postprocessing of