Embed Size (px)
Citation preview
2004IEICE TRANS. COMMUN., VOL.E98–B, NO.10 OCTOBER 2015
PAPER Special Section on 5G Radio Access Networks—Part II: Multi-RAT Heterogeneous Networks and Smart Radio Technologies
Measurement-Based Spectrum Database for Flexible SpectrumManagement
Koya SATO†a), Student Member, Masayuki KITAMURA†b), Nonmember, Kei INAGE††c),and Takeo FUJII†d), Members
SUMMARY In this paper, we propose the novel concept of a spectrumdatabase for improving the efficiency of spectrum utilization. In the cur-rent design of TV white space spectrum databases, a propagation modelis utilized to determine the spectrum availability. However, this propaga-tion model has poor accuracy for radio environment estimation becauseit requires a large interference margin for the PU coverage area to ensureprotection of primary users (PUs); thus, it decreases the spectrum sharingefficiency. The proposed spectrum database consists of radio environmentmeasurement results from sensors on mobile terminals such as vehicles andsmart phones. In the proposed database, actual measurements of radio sig-nals are used to estimate radio information regarding PUs. Because thesensors on mobile terminals can gather a large amount of data, accuratepropagation information can be obtained, including information regardingpropagation loss and shadowing. In this paper, we first introduce the archi-tecture of the proposed spectrum database. Then, we present experimentalresults for the database construction using actual TV broadcast signals. Ad-ditionally, from the evaluation results, we discuss the extent to which theproposed database can mitigate the excess interference margin.key words: dynamic spectrum access, cognitive radio, spectrum database,TVWS
1. Introduction
Because of the growth in demand for mobile communica-tion systems, data traffic has significantly increased duringthe last decade. Reference [1] predicts an exponential in-crease in data traffic that corresponds to a 10-fold increasein traffic between 2013 and 2019. Similarly, many industriesforecast a 1,000-fold increase between 2010 and 2020. Be-cause this exponential growth consumes finite spectrum re-sources, traditional spectrum utilization policies with exclu-sive resource allocation must be changed in future wirelesscommunication systems. For this reason, many researchershave shown an interest in spectrum sharing with dynamicspectrum access (DSA), which is an innovative spectrumutilization method [2], [3].
Cognitive radio [4] (CR) is a DSA tool that can solvethe shortage of spectrum resources. Using CR, secondary
Manuscript received January 19, 2015.Manuscript revised May 1, 2015.†The authors are with the Advanced Wireless and Commu-
nication Research Center (AWCC), The University of Electro-Communications, Chofu-shi, 182-8585 Japan.††The author is with Tokyo Metropolitan College of Industrial
Technology, Tokyo, 140-0011 Japan.a) E-mail: k [email protected]) E-mail: [email protected]) E-mail: [email protected]) E-mail: [email protected]
DOI: 10.1587/transcom.E98.B.2004
users (SUs) can utilize spectrum bands that are licensed toprimary users (PUs) under interference constraints for PUs.In the spectrum sharing environment, SUs must not interferewith primary communications. Because SUs are required toset parameters by taking into consideration the worst valueof the radio environment estimation error, low-accuracy es-timations reduce the spectrum sharing efficiency. Therefore,it is important to have accurate knowledge of the radio envi-ronment [5], [6].
Today, spectrum cognition techniques can be catego-rized into two methods: spectrum sensing and spectrumdatabase use. Spectrum sensing uses the detection of PUsignals to characterize radio environments [7]. To providegood protection, signal detection must be performed underthe (strict) condition that the PU signal strength be belowthe noise floor, even under low signal-to-noise ratios (SNRs)and fading conditions. Fluctuations make it difficult for theSUs to achieve stable detection; thus, although this tech-nique can model the radio environment in real time, it is verychallenging to implement. The second method is based onstoring information about spectrum availabilities of each lo-cation in spectrum databases [8]. In this method, SUs shouldquery the database before they utilize the spectrum. Then,the database provides spectrum information to the SUs. Thespectrum database is especially useful for fixed broadcast-ing PUs because the spatial distribution of the average signalpower does not change. The television (TV) band is a suit-able spectrum for database-aided spectrum sharing; its stan-dardization has been discussed in many countries includingthe US [9], the UK [10], and Singapore [11].
In the US, rules for spectrum database utilization forTV white space (TVWS) are defined by the Federal Com-munications Commission (FCC). On the basis of that regu-lation, some service providers have already launched spec-trum databases [12], [13]. The Office of Communications(Ofcom) in the UK started a TVWS pilot program fordatabase-aided TVWS utilization in July, 2014 [10], [14].For the TVWS pilot, some industries, e.g. Spectrum Bridge,Google, Microsoft, and NICT, have been permitted by theOfcom to launch databases.
Current databases usually evaluate white space basedon empirical propagation models. F-Curve, a propagationmodel that is utilized in the FCC-defined database, esti-mates the propagation loss based on the percentage of lo-cations, the percentage of time, frequency, and transmit-ting antenna heights [15]. The Ofcom-defined database
Copyright c© 2015 The Institute of Electronics, Information and Communication Engineers
SATO et al.: MEASUREMENT-BASED SPECTRUM DATABASE FOR FLEXIBLE SPECTRUM MANAGEMENT2005
Fig. 1 Examples of path-loss models.
uses a SEAMCAT extended Hata propagation model [16].This model considers antenna heights of transmitter andreceiver, frequency, and terrain information. However, itis well known that empirical propagation models cannottake into account all of the indeterminacies of radio envi-ronments, such as shadowing effects and differences of lo-cation. Figure 1 shows several simplified path-loss mod-els [17]. These curves follow a path-loss equation L(d) =−20log λ
4πd0+10γlog d
d0[dB], where λ is the wavelength [m],
d0 is the reference distance [m], d is the distance from atransmitter [m], and γ is the path-loss index. For example,at d = 1, 000 [m], there is a difference of 20 [dB] betweenγ = 2 and γ = 3 and also between γ = 3 and γ = 4. BecauseSUs must not interfere with PUs, their communication pa-rameters are determined based on the worst case scenario.For this reason, the conventional approach requires the sys-tem to set large margins to ensure no interference with PUs[15]. We have proposed a pathloss model-tuning method[18] to improve the estimation accuracy. The method esti-mates suitable pathloss parameters, transmission power, andpathloss-index based on measurement datasets using a max-imum likelihood method. However, the method has reducedaccuracy under shadowing effect conditions.
The error characteristics of propagation models havebeen presented in [19], [20]. In [19], the authors measuredthe field strength of radio waves over TV bands using ahighly accurate spectrum analyzer. The statistical resultsshowed that many propagation models perform biased es-timation with wide error variances. Phillips et al. [20] ana-lyzed the efficacy of basic path-loss models when predictingmedian path losses in urban environments. By comparingresults to those of many other path-loss models, the authorsidentified the danger of using basic a priori models to pre-dict the vagaries of the radio environment. It has also beenclaimed that complex models that consider a larger numberof variables (i.e., terrain models) do not necessarily makebetter predictions. To fit propagation models to the regionalradio environment, complex models often require a great
Fig. 2 Concept of the measurement-based spectrum database.
deal of information about the radio environment, e.g., ter-rain, antenna height, antenna pattern, and some measure-ment data. The fitted curve can achieve near-unbiased es-timation over a wide area. However, the information in-creases the calculation complexities, but cannot improve lo-cal accuracy due to shadowing effects. For these reasons,we examined early studies of measurement-based databaseconstruction [21], [22].
In this paper, we propose the novel concept of usinga spectrum database that consists of measurement informa-tion reported by mobile SUs. Figure 2 shows the concept ofthe proposed spectrum database. The proposed database is ahybrid system, combining spectrum sensing and a spectrumdatabase. The spectrum database consists of radio environ-ment information that is measured by mobile SUs (e.g., ve-hicles and smartphones). The SUs measure the received sig-nals from the PUs while the SUs move (without transmittingsignals). The collected dataset is related to the measurementlocation, and is reported to the database. After enough dataare gathered, the database estimates the radio environmentcharacteristics of the PUs by statistical processing with thelarge created dataset. Because the data include actual prop-agation losses and shadowing effects, accurate channel sta-tuses can be determined. SUs can connect the database toa wireless access point such as a cellular network to esti-mate the radio environment around the SUs, and adjust theirown communication parameters, which causes no harmfulinterference to PUs. Today, there are very many vehiclesand smart phones around the world. Because these mobileterminals have high mobility, we consider them particularlysuitable for use as probing sensors. These features of mobileterminals enable the spectrum database to gather measure-ment data rapidly and over a wide range. Thus, the radioenvironment can be estimated accurately. In addition, usingthe accurate knowledge of the received PU signal power,spectrum sharing based on PU signal quality metrics suchas the signal-to-interference-plus-noise ratio (SINR) can beimplemented.
The remainder of this paper is organized as follows.Section 2 describes our proposed hierarchical spectrum
2006IEICE TRANS. COMMUN., VOL.E98–B, NO.10 OCTOBER 2015
database architecture, which enables flexible configurationof spectrum utilization policies. In Sect. 3, we explain howto process the large measurement dataset. Section 4 de-scribes measurement configurations for a field test of thespectrum database construction method; the field test resultsare presented in Sect. 5. We describe future tasks in Sect. 6.Section 7 concludes this paper.
2. Hierarchical Database Architecture
Figure 3 summarizes the two spectrum database utiliza-tion methods: propagation model-based spectrum databasesand proposed databases. For the databases populated usingpropagation modeling, spectrum availability is determinedin two steps. First, the service area of a PU is estimatedusing the propagation model. Second, the expected ser-vice area is extended by an additional margin to guaranteeinterference-free operation for the PUs located at the bound-ary of the primary coverage area. Because the margin isadded without any regard for the actual propagation condi-tions, the SU transmission power and the SU coverage areaare unnecessarily reduced. As a result, the total spectrumsharing performance is degraded. On the other hand, theproposed database can provide accurate propagation infor-mation regarding the PU signals to the SUs. Therefore, ifthe location of the PU receiver is known, aggressive spec-trum sharing while controlling the SU transmission powercan be realized. This flexible and sophisticated transmis-sion power control method can obtain highly efficient spec-trum sharing performance between PUs and SUs becausethe spectrum database can help to accurately determine thereceived power of PU signals at any location and any fre-quency if a perfect spectrum database can be generated.
The proposed spectrum database can realize highly ef-ficient spectrum utilization with strict interference manage-ment by employing measurement datasets. However, if allof the measurement data are uploaded to the database cov-ering a large area, a large amount of data must be processedto estimate the primary coverage, which makes it difficultto provide spectrum usage information to SUs. Therefore,we consider a hierarchical database structure that supportsstoring data particles of flexible size in each layer, as shownin Fig. 4. In the hierarchical architecture, the size of an areasupported by the databases at each layer is different. Thedatabases at the lowest layers operate with high-resolutionraw data, but they can only support small areas. Higher lay-ers, with information of lower resolution, are statisticallycalculated from the lower-layer data. Because of the re-duced size of the stored information, the databases on higherlayers can support wider areas.
As shown in Fig. 4, the lowest-layer database is locatedat each SU. SUs probe the radio environment during dis-placement and store the most recent measurement results inthese databases at high resolution. However, because an in-dividual local database cannot cover a sufficiently large area,any comprehensive view of the primary signal propagationmust be derived from the locally collected data. Hence, a
Fig. 3 Operation of the spectrum database.
Fig. 4 Hierarchical database architecture.
second-layer database is used to store the stochastic infor-mation gathered from surrounding SUs. This layer consistsof many databases supporting small areas of a few squarekilometers in size. SUs upload measurement datasets whenthe second-layer database requires updates to its own statis-tical information. After enough data are gathered, SUs canutilize the statistical information provided by the second-layer database. We consider the second-layer database tobe fixed and to be managed by industry, which is permit-ted to manage the database by governments. If there is animmediate demand for white-space utilization, the second-layer database with mobility can be added for the requiredarea. Because it is difficult to store all of the measurementdata, stochastic information with lower resolutions is storedaccording to the location coordinates based on a grid struc-ture. A database of this size can gather information on sig-nal power from surrounding SUs and can store the averagepower for each frequency and location. This database cansupport SU interference management so as to achieve stricttransmit power control without degradation of PU receptionquality. The database at the highest layer implements a spec-trum utilization policy provided by the regulatory organiza-
SATO et al.: MEASUREMENT-BASED SPECTRUM DATABASE FOR FLEXIBLE SPECTRUM MANAGEMENT2007
tion in charge of a region, such as the FCC in the US, Ofcomin the UK, or the Ministry of Internal Affairs and Communi-cations (MIC) in Japan. As a result, metrological spectrummanagement can be realized with highly efficient spectrumsharing.
The proposed architecture can be used for various pri-mary systems. Here, various parameters of the database areconstructed according to the spectrum characteristics of theprimary system. We assume that the frequency of data up-load and the spectrum allocation period are the main param-eters used for database construction, and that the databasemust adjust these parameters based on the spectrum char-acteristics. In this paper, we specifically consider the com-patibility with the current TVWS spectrum database design.According to the current rules of TVWS utilization, SUsmust access the database at intervals of several hours: twohours in the UK and 24 hours in the US. We follow thesetime frames for the spectrum allocation periods. Becausebroadcast TV transmitters are fixed, their spectrum occu-pancies are static in the time domain. Therefore, in termsof TV protection, the importance SUs need to know is thespatial distribution of the average received signal power. Be-cause the variation of the distribution is gradual in the timedomain, the frequency of the dataset upload by each mea-surement terminals is expected to be low. From the practicalstandpoint of database access cost, frequent dataset upload-ing is undesirable. Dataset upload should be conducted onlywhen the node can access the wireless networks with suffi-cient capacity, such as public wireless LAN (WLAN), homeWLAN, and cellular networks.
Of course the measurement-based database design canbe applied to other networks with frequent changes of spec-trum usage: unlicensed bands, cellular systems, radar bands,and so on. However, spectrum occupancies of these systemsfluctuate drastically in the time domain; thus, other char-acteristics of white-spaces, e.g., variance, duty cycle, andtransition ratio, are required. We speculate that the require-ments increase the complexity of measurement systems andincrease the upload cost. Generalization of such complexi-ties and the database design should be studied further.
3. Low-Layer Configuration in TVWS
This section explains the low-layer configuration in TVWS.As already mentioned, the most important spectrum char-acteristic in TVWS is the average received signal power.To estimate the spatial distribution effectively, we assumethat measurement nodes probe signals with short-term av-eraging. In addition, the second-layer databases collect thedatasets and average the data per short-size grid.
3.1 Lowest Layer: Local Database at Node
A layer is implemented at each measurement node. Eachnode probes signals during displacement and stores the sig-nal data in its own database. Because each node moves withthe user, the explicit spatial border of this layer is not de-
termined. To remove signal variations due to small-scalefading, each node periodically measure signals (with short-term averaging), as shown in Eq. (1).
Pi =1M
M−1∑t=0
|hi[t]si[t] + wi[t]| , (1)
where hi[t] represents the channel coefficient at the i-th nodelocation, which includes propagation loss, shadowing, andsmall-scale fading. s[t] is the signal from the primary user.wi[t] is additive white Gaussian noise (AWGN). M is thenumber of averaged samples of each sensor. The mobileterminal equipped with a GPS device stores the value Pi andthe current location, time, and observed TV channel in thelocal database of the node. These values are uploaded tosecond-layer databases via wireless access networks. Afterupload, the node can erase the dataset from its own database.
3.2 Second Layer: Local Database
The second layer typically consists of fixed multi-databasesmanaged by industries which are allowed to utilize it bygovernments. Each database supports a small area of a fewkilometers, which matches the typical coverage of TV trans-mitters. Higher-layer databases divide datasets from mobilenodes into corresponding second-layer databases, based onmeasurement location and the supported area of the second-layer databases. After the data has been gathered, the spatialdistribution of the average received signals can be estimated.In addition to the difficulty of storing a large number of mea-surements in a single database, another important problemto overcome is the limited accuracy of localization systemsin the mobile terminals. A typical GPS device experienceserrors on the order of several meters (up to more than 10meters). For these reasons, we use a grid structure for lo-cation indexing, as shown Fig. 5. We divide each area of afew square kilometers (which is assigned to a database onthe layer immediately above the local databases on the mo-bile terminals) into a square grid with a side length of l [m].Each square cell in the grid in the second-layer databasesis represented by the average received signal power of allmeasurement data that was collected from the cell describedwith coordinates x and y that satisfy the condition
Fig. 5 Structure of the local database.
2008IEICE TRANS. COMMUN., VOL.E98–B, NO.10 OCTOBER 2015
Fig. 6 Measurement area.
√(x − xc)2 + (y − yc)2 ≤ l√
2, (2)
where xc and yc represent the coordinates of the grid field’scenter. In other words, the PU power in a given grid field isestimated as the mean μ,
μ =1N
N∑i=1
Pi, (3)
where N is the number of observations that satisfy Eq. (2).Note that there are borders of the estimation areas be-
tween the databases. Thus, white-space determination meth-ods taking into account the information of databases, whichare located on the both ends of the boundary, are required.For example, average of two values which are stored in theboth ends of databases is a simple connection method.
4. Measurement Campaign over TV Bands
We conducted a large-scale measurement campaign to eval-uate the radio environment estimation performance of theproposed spectrum database [23], [24]. We used five vehi-cles with spectrum sensors. A TV transmitter was treated asthe primary system, and measured the signal power from thetransmitter. Two one-week measurement campaigns wereconducted, in October 2013 and in February 2014. The priormeasurement datasets were stored in the spectrum database.On the other hand, the posterior measurement datasets weretreated as instantaneous measurement data and were usedfor strict evaluation of the statistical estimation error char-acteristics.
4.1 Measurement Object and Measurement Area
Figure 6 shows the measurement area and the object of
Fig. 7 Measurement equipment.
Fig. 8 Experimental configuration in a vehicle.
measurement. This figure describes a 40 [km] × 45 [km]square area. In this experiment, Kumagaya relay stationwas treated as the PU transmitter. This station is locatedin Kumagaya city, Saitama, Japan, which is a suburban areanear metropolitan Tokyo. We mainly measured the signalof 13CH (center frequency of 473.14 MHz and bandwidthof 6 MHz). The signal is a vertical polarized signal with anEIRP at 31 W.
4.2 Experimental Structure
Figure 7 shows the sensing equipment used. The spectrumsensing function was implemented on a software-defined ra-dio platform, Universal Software Radio Peripheral (USRP)N210, using GNU Radio software, and run on a laptop com-puter. The TV signal was sampled using a fast Fouriertransform (FFT) with a 200-kHz sampling rate; the num-ber of samples M was set to 2,048. Five vehicles measuredthe signal while driving on roads. The observation resultswere stored on the laptop, together with location informa-tion collected using a Garmin GPS18xUSB GPS unit. Toallow data collection in a short time, we had two sensingdevices in each vehicle, as shown in Fig. 8. We used fivevehicles in total; thus, ten measurement units were utilized.Each USRP has individual (linear) differences in measure-ment values. Therefore, we had to offset the difference byemploying tone signals in the TV band, provided by a signalgenerator (Rohde-Schwarz, SMU200A).
Here, the measurement parameters dictate that the mea-
SATO et al.: MEASUREMENT-BASED SPECTRUM DATABASE FOR FLEXIBLE SPECTRUM MANAGEMENT2009
Fig. 9 Calibration equipment for investing the difference between theband-limited measurement power and the full band signal power.
Fig. 10 Measurement power of USRP N210: Measurement of 6-MHzISDB-T signal using FFT with 200-kHz sampling rate.
surement system probes only a 200-kHz wide band withinthe (total) 6-MHz bandwidth. Therefore, we need to under-stand the relationship between the measured signal powerand the full-band signal power to allow calibration. To in-vestigate this relationship, we measured the signal based onIntegrated Services Digital Broadcasting-Terrestrial (ISDB-T). Figure 9 presents the experimental system. The ISDB-T signals in 13CH were generated by the signal generator.After the signals were attenuated, the USRP measured thesignal with the parameters presented above. The measure-ment result is shown in Fig. 10. In this figure, input poweris defined as PTx − Γ [dBm], where PTx is the transmissionpower of the signal generator in full band [dBm], and Γ isthe attenuated value at the output of the attenuator [dB]. Themeasurement value shows the received signal power aver-aged over a sufficiently large set of FFT signals. In the linearregion, the difference between the full band signal and themeasured signal is roughly −16.7 dB. Because of the 2.0-dBattenuation due to the two SMA cables, the measurementsignal power is 14.7 dB lower than that of the full-band sig-nal power. Because the ratio of the bandwidth between fullband and limited band is 10log{(6.0 × 106)/(2.0 × 103)} �14.77 [dB], the experimental result is considered to have theproper attenuation ratio.
4.3 Spectrum Database Construction
For statistical processing of the measurement dataset, weprepared a MySQL database server at our university. Thedatabase server has two tables: a raw data table and a statis-tical data table. The raw data table stores the large measure-ment dataset with no changes. Each measurement datumis associated with measurement parameters: location, fre-quency, power, and so on. The statistical data table has the
Fig. 11 Generated radio environment map at 473.14 MHz.
average received signal power for each spatial grid cell con-taining raw data. In addition, the database uses statisticalprocessing functions that generate the statistical data tablefrom the raw data table. These functions produce spatially-distributed average received signal information when the ap-propriate queries are made.
After the field test, all measurement data were recordedon the MySQL database server. To estimate the spatially-distributed average received signal information, the priormeasurement datasets were averaged over spatial grids witha 10-m side length.
5. Measurement Results
Figure 11 shows an average received signal map at473.14 MHz. This map consists of datasets measured ina prior measurement campaign and stored in the MySQLdatabase server. The strong trend of path loss increasingwith increased distance from the transmitter is clearly vis-ible. The map also indicates that some regions have a re-ceived power value significantly different from the receivedpower value measured in adjacent cells. Because of timevariation factors, which include AWGN, are averaged outusing long-term measurement, these variations are attributedto shadowing effects.
In this section, we evaluate the estimation error perfor-mance of the constructed spectrum database. The generatedmap is an instance of the proposed spectrum database. Thedataset that is measured a posteriori is used to obtain themost probable value and is compared with both the map anda propagation model.
5.1 Comparison with a Propagation Model
To clarify the probabilistic error performances, we evaluatedthe cumulative distribution function (CDF) of the radio envi-ronment estimation error. As a comparison method, we used
2010IEICE TRANS. COMMUN., VOL.E98–B, NO.10 OCTOBER 2015
Fig. 12 CDF of estimation error: Comparison with a propagation model,Notice 640.
Notice 640, which is a propagation model for Japanese TVbroadcast systems. This propagation model is defined bythe Japanese spectrum regulator, MIC; it is virtually identi-cal to the propagation model defined by the FCC. JapaneseTV broadcast operators often use this model to estimate thecommunication areas of TV transmitters because MIC li-censes the use of TV spectrums based on the estimation re-sults for Notice 640. According to MIC, the propagationcurve considers multipathing and diffraction due to terrain;however, the theoretical rationale is unclear. To estimatethe radio environment based on Notice 640, we used a radiopropagation simulator called area kakube [25]. This soft-ware was made for Japanese TV broadcast operators. Notice640 based electrical field strength can be calculated usingstored Japanese terrain information. Note that the measure-ment signal power is attenuated because of the band limit onthe measurement, as discussed in Chapter 4. Therefore, wesubtracted 14.7 dB from the calculation results taken fromarea kakube.
Figure 12 shows a statistical comparison of residual-error performance. The first curve indicates the CDF ofdifferences between average values in the constructed ra-dio environment map and the dataset of instantaneous re-ceived signals, which are collected in the second measure-ment campaign. The second curve shows the residual errorperformance of Notice 640. This figure confirms that thecurve of the proposed database rises rapidly from −10 dB.On the other hand, the curve of Notice 640 rises slowly from−80 dB to +80 dB.
Here, we show the CDF of the absolute error to con-firm the residual error level, as shown in Fig. 13. In the90th percentile, the proposed database has a residual error ofroughly 7.0 dB, and the propagation model has a residual er-ror of roughly 37.0 dB. This figure shows the poor accuracyof the propagation model. On the other hand, the proposeddatabase is shown to achieve high estimation accuracy.
Fig. 13 CDF of absolute estimation error.
Fig. 14 Comparison of the residual error in cumulative probability.
5.2 Impact of the Proposed Spectrum Database on Reduc-tion of the Excess Interference Margin
When SUs share a spectrum, they must set an interferencemargin after due consideration of the probabilistic error per-formances based on the radio environment estimation. IfSUs estimate the radio environment with a low-precisionmethod, SUs need to set exceed margin. This means thatSUs lose opportunities for spectrum sharing so as to pro-tect primary communications. Thus, the estimation error incumulative probability indicates the extent to which SUs de-termine the interference margin directly. This section evalu-ates the extent to which the proposed spectrum database canimprove the error performance in terms of cumulative prob-ability. Figure 14 presents the upside cumulative probabilityperformance and downside cumulative probability perfor-mance; the former shows underestimation characteristics,and the latter shows overestimation characteristics. Eachvalue is plotted with reference to Fig. 12. In the evalua-tion results, the proposed database achieves more than 20dB performance improvement, for both the upside perfor-mance and the downside performance. The improvement in
SATO et al.: MEASUREMENT-BASED SPECTRUM DATABASE FOR FLEXIBLE SPECTRUM MANAGEMENT2011
Fig. 15 Impact of the proposed spectrum database on the improvementof estimation error.
the estimation error is shown in Fig. 15; this figure showsthe difference between the proposed spectrum database andNotice 640. This figure clarifies that the proposed spec-trum database can mitigate the excess interference marginby more than 20 dB.
6. Future Tasks
We have confirmed that the proposed database can achievehighly-accurate radio environment estimation. It is expectedthat the proposed architecture will improve spectrum utiliza-tion efficiency in the future. However, some unsolved issuesremain. At the end of the paper, we describe the major tasksthat we have identified during this study that remain to becompleted.
If the compatibility with the current TVWS databaseis considered, the most spectrum use policies in the high-layer is roughly equivalent to the current policies. On theother hand, to achieve highly efficient spectrum sharing, thedatabase should modify several parameters for white-spaceestimation: interference margin is a main parameter. Theparameters depend on estimation results and estimation er-ror; the facts should be discussed carefully. This sectionmainly focuses on measurement indeterminacies due to col-lecting data from mobile terminals of citizens. In the mea-surement campaign, an identical experimental system wasstructured. We confirmed highly accurate estimation underthese ideal measurements. However, it is difficult to con-struct the system in realistic situations. The indeterminaciesdegrade the estimation accuracy, thus, we should try to solvethe issues.
6.1 Accuracy of Measurement Results due to DifferentDevices in Various Installation
We mounted antennas on the roof of each vehicle during themeasurement campaign. Although the installation locationaffects the measurement results, it is difficult for all mea-surement nodes to be suitably located. For example, many
users want to avoid mounting the antenna on the roof oftheir vehicles because of design limitations. Smartphoneusers will likely observe that the signal is changed basedon where the phone is located: in a hand, in a pocket, etc.Because these differences increase the individual differencesbetween measurement data, calibration techniques will be akey technology.
6.2 Reliable Estimation under Few Datasets
We have confirmed highly accurate received power estima-tion using the proposed method for a sufficient number ofdatasets. In realistic situations, because database construc-tion depends on the number and mobility of SUs, theremight be some areas through which only a small number ofSUs pass (and therefore, can measure). When the number ofmeasurement samples is small, the accuracy of the averagereceived power estimation decreases, causing harmful inter-ference to PUs in the area because of poor secondary com-munication parameter settings. This issue is specific to theproposed database; it is caused by using mobile terminals asthe measurement sensors. However, SUs must protect pri-mary communications, even if there are only a few samples.We have already begun the initial examination of a possiblesolution [26].
6.3 Upload Timing of Datasets
It is clear that frequent dataset gathering increases the es-timation accuracy. However, because of the practical issueof database access cost, frequent uploading is undesirable,as we explained in Sect. 2. On the other hand, keeping thedatasets over a long period of time in the local databaseat the node requires high-capacity storage. Therefore, weshould develop data exchange algorithms that consider thistrade-off.
7. Conclusion
In this paper, we proposed the novel concept of ameasurement-based spectrum database. The spectrumdatabase consists of a large number of measurement resultsfor the radio environment reported from mobile terminalssuch as vehicles and smart phones. To meet regional spec-trum use policies, the proposed spectrum database has a hi-erarchical architecture. The low layer gathers the large mea-surement dataset. The high layer stores the spectrum usepolicies and dictates how the dataset in the low layer utilizesthe spectrum. This paper focused specifically on low-layerdatabase construction. To evaluate the estimation error char-acteristics of the proposed spectrum database, a field experi-ment for radio environment probing was conducted over TVbands. From the spectrum database construction results, weconfirmed that the proposed database dramatically reducesthe estimation error of the radio environment information.In addition, we showed that the proposed database can re-duce the interference margin.
2012IEICE TRANS. COMMUN., VOL.E98–B, NO.10 OCTOBER 2015
Acknowledgement
A part of this work is supported from the Ministry of InternalAffairs and Communications (MIC) of Japan under R&D forNetwork Establishment Utilizing Vehicle-to-Vehicle Com-munications.
References
[1] G. Wunder, M. Kasparick, S. ten Brink, F. Schaich, T. Wild,I. Gaspar, E. Ohlmer, S. Krone, N. Michailow, A. Navarro, G.Fettweis, D. Ktenas, V. Berg, M. Dryjanski, S. Pietrzyk, and B.Eged, “5GNOW: Challenging the LTE design paradigms of orthog-onality and synchronicity,” Proc. 2013 IEEE 77th Vehicular Tech-nology Conference (VTC Spring), pp.1–5, June 2013.
[2] S.-Y. Lien, K.-C. Chen, Y.-C. Liang, and Y. Lin, “Cognitive radioresource management for future cellular networks,” IEEE WirelessCommun., vol.21, no.1, pp.70–79, Feb. 2014.
[3] C.-X. Wang, F. Haider, X. Gao, X.-H. You, Y. Yang, D. Yuan, H.Aggoune, H. Haas, S. Fletcher, and E. Hepsaydir, “Cellular archi-tecture and key technologies for 5G wireless communication net-works,” IEEE Commun. Mag., vol.52, no.2, pp.122–130, Feb. 2014.
[4] S. Haykin, “Cognitive radio: Brain-empowered wireless communi-cations,” IEEE J. Sel. Areas. Commun., vol.23, no.2, pp.201–220,Feb. 2005.
[5] H.A. Suraweera, P.J. Smith, and M. Shafi, “Capacity limits and per-formance analysis of cognitive radio with imperfect channel knowl-edge,” IEEE Trans. Veh. Technol., vol.59, no.4, pp.1811–1822, May2010.
[6] P.J. Smith, P.A. Dmochowski, H.A. Suraweera, and M. Shafi, “Theeffects of limited channel knowledge on cognitive radio system ca-pacity,” IEEE Trans. Veh. Technol., vol.62, no.2, pp.927–933, Feb.2013.
[7] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithmsfor cognitive radio applications,” IEEE Commun. Surv. Tutorials,vol.11, no.1, pp.116–130, March 2009.
[8] D. Gurney, G. Buchwald, L. Ecklund, S.L. Kuffner, and J.Grosspietsch, “Geo-location database techniques for incumbent pro-tection in the TV white space,” Proc. 2008 3rd IEEE Symposium onNew Frontiers in Dynamic Spectrum Access Networks, pp.1–9, Oct.2008.
[9] Federal Communications Commission, “Unlicensed operation inthe TV broadcast bands, third memorandum opinion and order,”FCC.12-36, April 2012.
[10] Office of Communications, “TV white spaces: Pilot databaseprovider contract,” Feb. 2014.
[11] Infocomm Development Authority of Singapore, “Regulatoryframework for TV white space operations in the VHF/UHF bands,”June 2014.
[12] Spectrum Bridge, “Showmywhitespace.” http://whitespaces. spec-trumbridge.com/whitespaces/home.aspx, accessed Jan. 19. 2015.
[13] Google, “Spectrum database.” http://www.google.org/spectrum/whitespace/, accessed Jan. 19. 2015.
[14] ETSI EN 301 598 V1.1.1, “White space devices (WSD); wirelessaccess systems operating in the 470 MHz to 790 MHz TV broadcastband,” April 2014.
[15] “Second memorandum opinion and order in the matter of unlicensedoperation in the TV broadcast bands, additional spectrum for un-licensed devices below 900 MHz and in the 3 GHz band,” FederalCommunications Commission, 2010, FCC 10–174.
[16] Office of Communications, “TV white spaces: approach to coexis-tence,” Dec. 2013.
[17] A. Goldsmith, Wireless Communications, Cambridge UniversityPress, 2005.
[18] K. Sato, K. Inage, and T. Fujii, “Parameter estimation method of pri-mary user using measurement-based spectrum database,” Proc. 2014IEEE 11th Consumer Communications and Networking Conference(CCNC), pp.535–536, Jan. 2014.
[19] A. Achtzehn, J. Riihijarvi, and P. Mahonen, “Improving accuracyfor TVWS geolocation databases: Results from measurement-drivenestimation approaches,” Proc. 2014 IEEE International Symposiumon Dynamic Spectrum Access Networks (DYSPAN), pp.392–403,April 2014.
[20] C. Phillips, D. Sicker, and D. Grunwald, “Bounding the error of pathloss models,” Proc. 2011 IEEE International Symposium on Dy-namic Spectrum Access Networks (DySPAN), pp.71–82, May 2011.
[21] T. Fujii, “Hierarchical radio environment database for efficient spec-trum sharing,” Proc. APWCS 2012, Aug. 2012.
[22] T. Fujii, K. Inage, M. Kitamura, O. Altintas, H. Kremo, and H.Tanaka, “Short paper: Probing the spectrum with vehicles: Towardsan advanced spectrum database,” Proc. 2013 IEEE Vehicular Net-working Conference, pp.226–229, Dec. 2013.
[23] M. Kitamura, K. Inage, Y. Ohue, and T. Fujii, “Development of mea-surement based spectrum database for efficient spectrum sharing,”Proc. SDR–WInnComm 2014, Feb. 2014.
[24] K. Sato, T. Kosugi, and T. Fujii, “Demo: A field test of measure-ment-based spectrum database using probing vehicles,” Proc. 2014IEEE Vehicular Networking Conference (VNC), pp.205–206, Dec.2014.
[25] RCC Chugoku Broadcasting Company, “Area kakube.” http://kakube.rcc.ne.jp/ (in Japanese), accessed Jan. 19. 2015.
[26] K. Sato and T. Fujii, “Average interference power constraint withmeasurement-based spectrum database,” Proc. 2015 IEEE Wire-less Communications and Networking Conference Workshops (WC-NCW), pp.211–216, March 2015.
Koya Sato was born in Miyagi, Japan,in 1991. He received the B.E. degree in elec-trical engineering from Yamagata University in2013 and the M.E. degree from The Universityof Electro-Communications in 2015. He is cur-rently working toward the Ph.D. degree in TheUniversity of Electro-Communications. His re-search topic is dynamic spectrum access. He isa member of IEEE.
Masayuki Kitamura was born in Hok-kaido, Japan, in 1989. He received the B.E. andM.E. degrees in electrical engineering from TheUniversity of Electro-Communications, Japan,in 2012 and 2014. Since April 2014, he has beenwith the Panasonic System Networks Co., Ltd.,Japan. His research topic in the university wascognitive radio.
SATO et al.: MEASUREMENT-BASED SPECTRUM DATABASE FOR FLEXIBLE SPECTRUM MANAGEMENT2013
Kei Inage was born in Chiba, Japan, in1986. He received the B.E., M.E. in ElectronicEngineering and Ph.D. degree in Communica-tion Engineering and Informatics from The Uni-versity of Electro-Communications, Japan, in2009, 2011 and 2014 respectively. From 2012to 2014, he was a Research Fellow DC2 at JapanSociety for the Promotion of Science (JSPS).From 2014 to 2015, he joined National Instituteof Information and Communications Technol-ogy (NICT) as a researcher in the Smart Wire-
less Laboratory, Wireless Network Research Institute. Currently, he is anassistant professor in Electrical and Electronics Engineering Course, To-kyo Metropolitan College of Industrial Technology. His current researchinterests are in cognitive radio, ad-hoc networks and cellular networks. Hereceived Young Researcher’s Award from the IEICE in 20012. He is amember of IEEE.
Takeo Fujii was born in Tokyo, Japan, in1974. He received the B.E., M.E. and Ph.D. de-grees in electrical engineering from Keio Uni-versity, Yokohama, Japan, in 1997, 1999 and2002 respectively. From 2000 to 2002, he wasa research associate in the Department of Infor-mation and Computer Science, Keio University.From 2002 to 2006, he was an assistant profes-sor in the Department of Electrical and Elec-tronic Engineering, Tokyo University of Agri-culture and Technology. From 2006 to 2014, he
has been an associate professor in Advanced Wireless Communication Re-search Center, The University of Electro-Communications. Currently, heis a professor in Advanced Wireless and Communication Research Cen-ter, The University of Electro-Communications. His current research in-terests are in cognitive radio and ad-hoc wireless networks. He receivedBest Paper Award in IEEE VTC 1999-Fall, 2001 Active Research Award inRadio Communication Systems from IEICE technical committee of RCS,2001 Ericsson Young Scientist Award, Young Researcher’s Award from theIEICE in 2004, The Young Researcher Study Encouragement Award fromIEICE technical committee of AN in 2009, and Best Paper Award in IEEECCNC 2013. He is a member of IEEE.
