University of Louisiana at Lafayette, LA 70503, USA. Email: {ogo8842, ama8030, kmk8148, bxd9836, dxp0566, mab0778}@louisiana.edu. AbstractâSpectrum ...
Spectrum-Awareness-based Performance and Scalability of Cognitive Radio Networks Oladiran G. Olaleye, Alaa Ali, Kasem Khalil, Bappaditya Dey, Dmitri Perkins, Magdy Bayoumi The Center for Advanced Computer Studies University of Louisiana at Lafayette, LA 70503, USA Email: {ogo8842, ama8030, kmk8148, bxd9836, dxp0566, mab0778}@louisiana.edu Abstract—Spectrum awareness (SA) stretches the performance bounds of spectrum-sensing-based dynamic spectrum access by intelligently exploiting the big spectrum data (BSD) generated by a network. Hence, in order to analyze the performance and scalability of large scale cognitive radio networks (CRNs), spectrum awareness capacity would take preeminence over spectrum sensing capacity. Although, conventional methods use techniques such as the receiver operating characteristic (ROC) curve and the root mean square error (RMSE) technique to quantify the performance of CRN SA, they do not consider the impact of BSD velocity, variety and volume. Therefore, this research work proposes a novel knowledge-centric method for quantifying, analyzing and comparing the performance and scalability of CRNs based on spectrum awareness. The proposed method considers key performance indices including reliability, computational complexity and latency of the network parameters that are generated by spectrum data acquisition, conversion and dissemination. The steps and applicability of the proposed method in user and network-level performance measurement are also analyzed. Index Terms—Big Spectrum Data, Cognitive Radio, Performance Evaluation, Scalability, Spectrum Awareness
I. I NTRODUCTION Network performance metrics in terms of throughput and bit error rate (BER), which inherently capture the effects of reliability, complexity and latency may not sufficient for comparing Spectrum awareness (SA) techniques and algorithms. Network throughput measures an aggregate of successfully transmitted data, which may have been limited factors from any layer of the entire network architecture, whereas error rates (bit, frame and packet error rates) are layer-specific and only related to utilization rather than availability. Other performance metrics are also specific to particular network layers. Hence, novel metrics and methods are required to measure the efficacy of the techniques for generating SA in cognitive radio network (CRN). The performance of an SA technique and algorithm, in a network, is a function of its reliability [1], complexity [2] and latency [3]. Hence, a suitable performance metric would consider those three indices and would also be useful in characterizing the individual techniques for scalability measurement. Reliability index measures the validity of the input parameters, complexity factor measures the weight of the required computation, while latency measures the delay. When factored together and normalized, the three indices (reliability, computational complexity and latency) can be used to
quantify the improvement in overall network performance due to any referenced technique. SA-based network performance and scalability metrics, therefore, measure the improvement in overall network performance due to SA. They can be used to characterize the algorithms for generating spectrum awareness. For network design optimization and reconfigurability, characterized SA techniques and algorithms would be useful in implementing the appropriate SA modules. Also, since different situations require different combinations of network parameters for SA, multiple SA modules can be implemented at the design stage to boost the reconfigurability of network algorithms during operation and also enhance other desirable network properties such as coexistence and interoperability. We derive the normalized constant of proportionality (SA constant) for the variation of SA-based network performance as a function of specific performance indicators. In this work, we introduce a knowledge-centric approach for modeling the performance of cognitive radio networks based on spectrum situational awareness. Using the parametric analysis of SA techniques and algorithms, we derive the expressions for network performance indicators — reliability, computational complexity and latency. The derived expressions are then normalized and defined as SA constants — a measure of the performance of CRN, based on SA. We apply our proposed method in comparing two SA algorithms (Weighted and Unweighted Least Square Method — WLS and UnWLS respectively), by simulating unlicensed or secondary user (SU) location awareness in MATLAB. The MATLAB simulation is based on a hybrid SA technique, which uses received signal strength and direction of arrival of licensed or primary user (PU) signal. We model the scalability of SA techniques and algorithms by parametric analysis at the network-user level. The rest of the paper is organized as follows: section II describes the problem formulation and system model; the theoretical background is explained in section III; section IV details the case study on hybrid PU localization; while the conclusion and future works are stated in section V. II. P ROBLEM F ORMULATION AND S YSTEM M ODEL A. Network Model We adopt the system model in [4]. The SUs cooperate to detect the activities of PUs. Each SU periodically measures all possible PU parameters (including the total number of
978-1-5386-4328-0/18/$31.00 ©2018 IEEE
Inti , j ^UiI, j , J iI, j ,W iI, j `
Consti , j ^U , J ,W C i, j
C i, j
C i, j
`
Single User (PU/SU)
TABLE I NOMENCLATURE
Meti , j ^UiM,j , J iM,j ,W iM,j `
Symbols AoA AU B Ccap CC CU Consti,j DoA DoT EP F edBi,j GA HA i IC IR IT Inti,j j L, LX , LY , LZ LOLP LsuccT
FedBi , j ^UiF, j , J iF, j ,W iF, j `
Fig. 1. Spectrum data flow diagram for a single opportunistic user, with i = 1, 2, ... and j = 1, 2, ...
PUs, PUs’ transmit power, channel bandwidth, and PUs’ signal direction-of-arrival) and transmits the measurements to the data fusion center (DFC), which computes, stores, and updates the SA metrics as a REM database. Hence, the SUs cooperatively depends on the DFC for CRN SA. We quantify the amount of knowledge available to user about its spectrum environment by its dynamic parameters. The parametric analysis of an individual user for user-level performance measurement makes it possible to isolate spectrum data streams with direct impact on awareness. Fig. 1 shows the schematics of spectrum data flow diagram for a single user. All the notations are as defined in Table I. In Fig. 1, the arrows represent parametric data streams while the directions of the arrows indicate the sources of data acquisition, dissemination and storage. The box represents the cognitive models for the network data computation and algorithms. Inti,j are the parameters from which a user can derive SA e.g. the RSS at SUs, while M eti,j are those that define the level of awareness e.g. PU transmit power. F edBi,j represents the feedback parameters. Variable i, which indicates the number of parameter types, defines the level of SA, while j, which indicates the number of a particular type of variable, defines the scale of SA. i and j both models the volume of spectrum data. Each SA parameter type is specified using three properties: reliability ρ, computational complexity γ and latency τ . ρ is the degree of accuracy of a parameter based on its source (measurement, estimation, or prediction.) γ is a measure of the numeric computational requirement of a parameter based on the required number and type of mathematical operations. τ is the time required to determine the value of a parameter based on its source and computational complexity. We define a measured parameter as that whose value was obtained by direct physical measurement (e.g. sampling and analysis), an estimated parameter as that whose value was obtained indirectly by measuring a related parameter, and a predicted parameter as that whose value was derived from an existing mathematical model. The SA of the specified N -user network is modeled by the parametric analysis of a single user since all the architectural components of the network have equal potential for generating and transmitting the same variety and volume of data. On the other hand, for a network architecture with components on different data generation and transmission hierarchy (e.g. a centralized or multi-structured clustered network), one representative component from each hierarchy would be required in order to model the entire network’s SA. Hence, the proposed approach is also applicable for heterogeneous network analysis
M eti,j OR OT PA PT PT,max RSS S SC SR SNR TH TDoA ToA U I , γM , γF γi,j i,j i,j γiP ξ F ρIi,j , ρM i,j , ρi,j ρP i I , τM , τF τi,j i,j i,j
τiP ψ
Definition Angle-of-arrival User activity Channel bandwidth Channel capacity Channel count Network user (PU and PR) count jth i-constant SA unit integral Direction-of-arrival Direction-of-transmission Path loss exponent Feedback jth SA unit metric of type i Antenna Gain Antenna height Counter for SA parameter types Cluster index Interference ratio Terrain index jth SA unit integral of type i Counter for i-th type SA parameter User Location Local Oscillator Leakage Power Total load successfully transmitted within a single PU coverage jth SA unit metric of type i Receiver orientation Transmitter orientation Antenna polarization Transmit power Maximum transmit power Received signal strength User mobility Channel Status Receiver sensitivity Signal-to-noise ratio Spectrum hand-off delay Time-difference-of-arrival Time-of-arrival Spectrum utilization Computational complexity of Inti,j , M eti,j , F edBi,j respectively Computational complexity of the data processing technique and algorithm of user i Performance index Reliability index of Inti,j , M eti,j , F edBi,j respectively Reliability index of the data processing technique and algorithm of user i Latency of Inti,j , M eti,j , F edBi,j respectively Latency of the data processing technique and algorithm of user i Scalability factor
[5]. III. T HEORETICAL A NALYSIS It is not uncommon to alter the architecture of a network or the rate of consumption of specific network resources in order to improve network performance. There is always a tradeoff between performance and system resources. Hence, since the availability of network resources has a direct impact on reliability, computational complexity and latency, our proposed method — an awareness-based network performance and scal-
ability analysis method — considers the set of all performance and scalability determining factors including computational complexity, knowledge requirement, memory requirement, energy consumption, decision delay, reliability, reconfigurability, and learning algorithm. The decisions about when and how to communicate are made based on the level and module of awareness. The awareness module is defined by the number and types of SA unit metrics while the level of awareness is quantified by the resulting characteristics of derivatives (reliability index, computational complexity and computational latency). Hence, the spectrum access decision making process, based on network awareness or cognition, defines the adaptability and reconfigurability of the network. Provided the specified set of SA unit metrics completely defines a user’s awareness level and there are no redundant parameters, from Fig. 1, P ρM i,j = ρi , ∀j ∈ {1, 2, ...}
(1)
M = γiP , ∀j ∈ {1, 2, ...} γi,j
(2)
M I C F = τiP + max{τi,j , τi,j , τi,j : i, j = 1, 2, ...}, ∀j τi,j
ρP i
=
F f {ρIi,j , ρC i,j , ρi,j
: i, j = 1, 2, ...}
For an accurate scalability analysis, a comprehensive understanding of the relationship among determining factors is required. Where SA level is specified by the number of SA unit metrics required to completely describe the level, we have two types of scalability: SA-based user-level scalability and SA-based network-level scalability. 1) Network-Level Scalability: We define SA-based network scalability as the limit of the derivable level of SA, as the number of network users increases. That is, ψ = lim (iM j M ) j M →∞
ψρ = ψγ =
(4) (5)
M ρF i,j = ρi , ∀j
(6)
F γi,j
γiM , ∀j
(7)
F M = τi,j , ∀j τi,j
(8)
where γiP depends on the SA technique and algorithm and I I C C , τi,j , ρC ρIi,j , γi,j i,j , γi,j and τi,j , all depend on the respective sources of data (measurement, estimation or prediction). The overall analyses for a single user in the specified network are detailed below.
ψτ =
lim (iM j M )
(12)
lim (iM j M )
(13)
lim (iM j M )
(14)
ρM ¯M i,j →ρ i,j
M →¯ M γi,j γi,j
M →¯ M τi,j τi,j
where ψρ , ψγ and ψτ are the scalability factors with respect to reliability, computational complexity and latency respectively; M M ¯i,j and τ¯i,j are the predefined threshold reliability, and ρ¯M i,j , γ computational complexity and latency of M eti,j respectively. Hence, by applying numerical analysis to (12), (13) or (14), the scalability factor is computed. 2) User-Level Scalability: We define SA-based user-level scalability as the limit of the derivable level of SA, for a fixed number of SA unit integrals. That is, ψI = lim (iM j M ) jIi →¯jIi
A. Performance We define the SA-based user-level network performance, ξ, as a measure of the level of spectrum awareness of a user. Thus, ξ=k
ρM i,j M τM γi,j i,j
(9)
where k is the constant of proportionality. On normalizing ξ, ρ, γ, and τ we obtain, ρ¯M i,j ξ¯ = M M γ¯i,j τ¯i,j
(10)
where ξ¯ is referred to as the SA constant. With full awareness of spectrum availability and consumption rate, ξ¯ = 1. However, in practical cases, the SA constant of a SU is limited by the reliability of the measured parameters, the accuracy of the predicted variables and the complexity of the computations involved.
(11)
where iM is the counter for a specific type of SA unit metric; j M and j I are the respective counters for SA unit metric and integral; and all the counters can be expressed as functions of ρM , γ M and τ M . Therefore, on applying design constraints to (14) — fixing the reliability index ρM , the computational complexity γ M and the latency τ M at a threshold, we have
(3)
τiP = f {γiP : i, j = 1, 2, ...}
=
B. Scalability
(15)
where ψI is the scalability factor with respect to SA integral; jIi is the counter for the i − type SA unit integral; ¯jIi is the fixed number of i − type SA unit integral; and (15) can also be expressed in the same forms as (15), (16) and (17). IV. C ASE S TUDY: RSS/D OA- BASED H YBRID P RIMARY U SER L OCALIZATION A. Basis The goal of this illustration is to demonstrate the need for novel SA-based network performance metric and validate our proposed approach. We adopt the hybrid SA technique (RSS/DoA-based SU location awareness [6]), which was analyzed in [7]. The authors of [7] compare three different SA techniques: RSS-based, DoA-based and hybrid RSS/DoAbased location awareness by deriving the respective CramerRao bounds and justifies the results using the RMSE method. They adopted two SA algorithms: MUltiple SIgnal Classification (MUSIC) and Stansfield algorithms, which are used for
DoA estimation and data fusion respectively. Hence, based on the adopted performance measuring techniques (Cramer-Rao bounds and RMSE), the authors of [7] were only able to justify the accuracy of the three SA techniques. For design purposes, however, the analysis on the accuracy of the techniques and algorithms would have to be combined with other design factors such as latency and computation complexity in order to make sound judgments on expected network performance. The significance of composite design factors become more apparent as networks become bigger and heterogeneous. Inter-connectivity is achieved in large-scale networks by protocol coexistence at different layer of network architecture. Since the physical layer is the first point of consideration in establishing network coexistence, here, we focus on physical layer parameters. The SA unit integrals, for this illustrative analysis, are RSS and DoA while the SA unit metrics are LX and LY (the 2-dimensional location coordinates of the PUs.) The feedback metrics are also LX and LY , as shown in Fig. 2. RSSi , j ^UiR,j , J iR,j ,W iR,j `
DoAi , j ^UiD,j , J iD,j ,W iD,j `
LX i , j ^UiL, j , J iL, j ,W iL, j ` SU
LY i , j ^UiL, j , J iL, j ,W iL, j `
LX i , j ^UiL, j , J iL, j ,W iL, j `
Fig. 2. Flowchart for the hybrid RSS/DoA-based PU localization of primary users based on the proposed methodology. The reliability, computational complexity and latency are as defined in the section III.
SU Coordinates Actual PU Coodinates 600 400 200 0 −200 −400 −600 −600 −400 −200
0
200
400
600
Fig. 3. Ad hoc cognitive radio network geometry consisting of 25 static and radially located secondary users and 25 uniformly distributed primary user grid for the MATLAB simulation of hybrid RSS/DoA-based primary user localization.
C. Assumptions We assume an ideal system and do not consider redundant integrals. We simulate the latency and computational complexity of the SA unit metrics as a uniform distributed variable with bounds of 0.75 and 1.0 for latency and 0.8 and 1.0 for computational complexity. The variance of latency is larger than that of computational complexity because the former is less predictable. The reliabilities of the unit integrals are assumed independent of the angular distance from the actual PU coordinates and vary only in the radial direction on a 2dimensional plane. Hence, the reliabilities of measured RSS values are equated to the derived probability of detection of PUs while the reliability of measured DoA is assumed to vary uniformly between 0.5 and 1.0. The probability of false alarm, on the hand was fixed at 0.01. D. Cognitive Model
B. Objectives The two integrals (specified as basis) can be used in estimating and predicting different unit metrics including PU mobility, transmission range, maximum transmit-power and PR density. For this illustration, however, we compare the performance of WLS and UnWLS methods for PU localization. For empirical justification, we simulated the technique in MATLAB. The simulation was grouped into three cases. In the first case, we simulated the localization of PUs by a single SU. In the second case, we simulated the cooperative localization of PUs by UnWLS method in a network with equal number of PUs and SUs. The third case is similar to the second case but with UnWLS replaced with WLS method. The topology of the network is shown in Fig. 3. We simulated a maximum of 25 static SUs and PUs within a 500 × 500 square meter network coverage. The PUs are distributed in a square grid pattern while the SUs are radially located about the origin ((0, 0) coordinate). Each PU transmits at 20 dBm (100 Watts) with an antenna gain of 1 dB and a signal bandwidth 6MHz. The path loss exponent and sensing time are fixed at 5 and 0.2 milliseconds respectively. At the SUs, the antenna gain and the estimated power of the assumed additive white Gaussian noise (AWGN) are 1 and -65 dBm respectively.
Each SU measured the RSS and DoA of a PU signal and computes the estimated PU coordinates. The estimate is then shared with other SUs where they are fused by UnWLS and WLS method. Hence, the cognitive operation, as dictated by the proposed method, are the mathematical models for converting RSS and DoA to LX and LY and the least square algorithms. For the UnWLS method, the shared locations coordinates were fused by averaging while a novel weighting technique was proposed for the WLS — reliability-based WLS method. For the RSS (dBm) and DoA (radian) measurements, ⎧ λ × 10(0.1×(PT +Gtx +Grx +PN −RSS)) ⎨ L = 4π (16) ⎩ L= (xSU − xP U )2 + (ySU − yP U )2 DoA = arctan
ySU − yP U xSU − xP U
+ nDoA
(17)
where λ is the signal wavelength; Gtx is the PU antenna gain; Grx is the SU antenna gain; PN is the estimated noise power with a variance, σ 2 = 1; (xSU , ySU ) is the SU location coordinate; (xP U , yP U ) is the PU location coordinate; and nDoA is the zero mean DoA measurement error.
For UnWLS method, we apply the iterative NewtonRaphson procedure, thus ⎧ est ˜ est 2 2 LPU (X) = (xP U − xest ⎪ PU ) + (yP U − yPU ) ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ˜ X= argminLest PU (X) (18) ˜ X ⎪ ⎪ ⎪ ⎪
⎪ ⎪ ⎩ X i+1 = i − H −1 Lest (X i ) × ∇ Lest (X i) X PU PU ˜ where Lest PU (X) is the distance between the actual PU loest cation and the estimated location; (xest PU , yPU ) are the esti−1 i mated PU coordinates from each SU; H Lest PU (X ) and
i ∇ Lest are the corresponding Hessian and gradient PU (X ) vector at the ith iteration of PU location estimate. For the reliability-based WLS method, est N SU xPU est = rPU (19) X est yPU i=1 ⎧ est
LPU ⎪ 1.0 − ⎪ ⎨ Lorigin PU est (20) = rPU
⎪ ⎪ ⎩ 0.0 if Lest ≥ Lorigin PU
PU
est where rPU is the reliability of the estimated PU coordinate; origin and LPU is the distance between the actual PU location and the origin. The above weighting technique assumes the reliability of an estimated coordinate is zero when the RMSE equals the distance between the actual PU coordinate and the origin and 100% when RMSE is zero. This novel approach is most suitable for infrastructure-based network, where it is safe to assume that the farther away a user is from the centralized access point, the less significant RMSE becomes. For future projects, we would be analyzing the behavior of RMSE for location estimates, as radial distance from the actual location coordinate increases. The focus would be on the characterization of RMSE for improved convergence of location estimates.
E. Analysis L L In this illustration, ρLi,j , γi,j and τi,j are derived from the results our simulation in MATLAB and hence, used to ¯ The resulting PU location estimates from the compute ξ. scenarios with isolated SUs and the ones with cooperating SUs are as plotted in Fig. 4. Fig. 4(a) shows the estimated PU coordinates from four static SUs at locations: (0, 0), (-216.5, 125), (250, 0) and (500, 0). The SUs measured and computed the PU coordinates without cooperation. The results reveal the limited PU detection capability of SUs even when equipped with one detector per PU. The probability of false alarm (Pf ) was fixed at 0.01, while the probability of detection varied between 0.1 and 1. The resulting PU estimates vary immensely from the actual PU coordinates. The SU at (0, 0), for example, was only able to localize the uniformly distributed 500 m × 500 m PU grid to within 200 meters radial distance. Similar pattern is observable
in the other three SUs. The results of Fig. 4(a) emphasis the significance of SU cooperation and need for improved SA techniques and algorithms. In Fig. 4(b), the SU cooperate to estimate the PU coordinates. Fig. 4(b) shows the estimated PU coordinates after applying the UnWLS and WLS algorithm, to cooperatively localize the PUs. The plot directly compares the accuracy of the two SA algorithms. As expected [8], the WLS performs better than the UnWLS method, since its estimates are more spread out and closer to the actual PU coordinates. Although, the results of SU cooperation by UnWLS method are better than those obtained without cooperation (Fig. 4(a)), WLS performance better than UnWLS method and hence validates our proposed reliability-based weighting technique. F. Results and Evaluation Figs. 5 and 6 compare the two SA algorithms (LS and WLS method) in terms of reliability, complexity and latency while Fig. 7 compares the scalability of the two algorithms. Although, the results of the simulation, as shown in Fig. 7, reveal that the UnWLS is more scalable than WLS method, the essence of the plots is not only to evaluate the performance of UnWLS and WLS but also assess the suitability of the normalized variables as performance metrics. Figs. 5 and 6 show the plot of normalized reliability, latency, computational complexity and awareness-based network performance index versus the number of PUs, for a single SU network (Fig. 5), for cooperating SU using UnWLS and WLS method for location coordinate fusion (Fig. 6), and for a direct comparison of UnWLS and WLS (Fig. 6). For the purpose of this illustration, the number of SUs and PUs were equal. As observed in Fig. 5, the performance of the single SU localization of PU coordinates is very low. In addition, Fig. 6 clearly shows that WLS performs better than UnWLS for the simulated network model, since for equal number of PU, the performance curve of WLS is always higher than that of UnWLS . Fig. 6 also shows the differences in the use of reliability, computational complexity and latency as metrics for awareness-based network performance. From the plots, it becomes obvious that the performance index, which is a combination of all other metrics, is the best measure for comparing SA techniques and algorithms. Besides reducing the number of plots to be compared, the performance plot also considered a broader dynamics of the two algorithms being compared. For our case study, the dynamics include variation in accuracy (reliability), rate of network resource consumption (computational complexity) and delay (latency). It is noteworthy that the complexity and latency of the reliability evaluation stage is negligible when compared to the entire SA module. Therefore, a combination of the three metrics (reliability, complexity and latency) is the best indicator for the performance of opportunistic networks based on spectrum awareness. V. C ONCLUSION We propose a novel method for quantifying the awarenessbased performance and scalability of cognitive radio networks.
SU Coordinates Actual PU Coodinates Estimated PU Coord from the SU at (0,0) Estimated PU Coord from the SU at (−216.5,125) Estimated PU Coord from the SU at (250,0) 800 Estimated PU Coord from the SU at (−500,0) 600
SU Coordinates Actual PU Coodinates 800 Estimated PU Coodinates from all SUs (Unweighted) Estimated PU Coodinates from all SUs (Weighted) 600
400
400
200
200
0
0
−200
−200
−400
−400
−600 −600 −400 −200
0
200
400
−600 −600 −400 −200
600
(a)
0
200
400
600
(b)
Fig. 4. Simulation results for the MATLAB simulation of hybrid RSS/DoA-based primary user localization. (a) shows the PU location estimates, computed individually and without cooperation, by four different secondary users, located at (0, 0), (-216.5, 125), (250, 0) and (-500, 0). (b) shows the PU estimates computed by cooperating SUs.
Normalized Metrics
1
performance metric for cognitive radio networks. The need for the simulation of complex networks is also minimized, provided that provable models, analytical or empirical, exist for desired network dynamics. Also following the simulation trend WLS performs comparatively better than that of scalable UnWLS . For future work, we would be evaluating empirical models for the dynamic behavior of opportunistic networks.
Reliability per PU Latency per PU Computational Complexity per PU Performance per PU
0.8 0.6 0.4 0.2 0 0
5 10 15 20 Number of Primary Users in the Network
25
Fig. 5. Location-awareness-based performance plot for a single secondary user, with (0, 0) coordinates, in a cognitive radio ad hoc network with 1 secondary user and 25 primary users. The awareness-based performance of the secondary user was measured in terms of reliability, latency, and computational complexity. Normalized Metrics
1
This work was supported in part by the National Science Foundation (NSF) under grants 1454835 and 1429526. R EFERENCES
Reliability per PU (UwLSM)
0.8
Latency per PU (UwLSM) Computational Complexity per PU (UwLSM)
0.6
Performance per PU (UwLSM) Reliability per PU (WLSM)
0.4
Latency per PU (WLSM)
0.2
Computational Complexity per PU (WLSM)
0 0 5 10 15 20 25 Number of Primary Users in the Network
Performance per PU (WLSM)
Fig. 6. Normalized plots for direct comparison of the two simulated SA algorithms (Weighted and Unweighted Least Square Methods), based on reliability, computational complexity and latency. 25 Number of Primary Users
ACKNOWLEDGMENT
20 15 10 5 0 0
Unweigthed Least Square Method (UwLSM) Weigthed Least Square Method (WLSM) 0.1
0.2
0.3
0.4 0.5 0.6 Normalized Performance
0.7
0.8
0.9
1
Fig. 7. The scalability plot of location-awareness-based performance against the number of primary users, comparing two SA algorithms: the Unweighted and Weighted Least Square Methods.
The proposed method, a knowledge-centric approach, analyzes and compares network architectures based on reliability, computational complexity and latency of network dynamic variables. The results of applying the proposed method in the simulation of RSS/DoA primary user localization shows validates the combination of reliability, computational complexity and latency as the suitable spectrum-awareness-based
[1] L. Cao and H. Zheng, “Balancing reliability and utilization in dynamic spectrum access,” IEEE/ACM Transactions on Networking, vol. 20, no. 3, pp. 651–661, June 2012. [2] W. Dong and M. Kam, “Detection performance vs. complexity in parallel decentralized bayesian decision fusion,” in 2017 51st Annual Conference on Information Sciences and Systems (CISS), March 2017, pp. 1–6. [3] L. C. Wang, Y. C. Lu, C. W. Wang, and D. S. L. Wei, “Latency analysis for dynamic spectrum access in cognitive radio: Dedicated or embedded control channel?” in 2007 IEEE 18th International Symposium on Personal, Indoor and Mobile Radio Communications, Sept 2007, pp. 1–5. [4] O. G. Olaleye, A. Ali, A. Aly, M. A. Iqbal, D. Perkins, and M. Bayoumi, “Framework for generating and designing spectrum awareness modules for opportunistic networking,” in 2017 IEEE 8th Annual Ubiquitous Computing, Electronics and Mobile Communication Conference (UEMCON), Oct 2017, pp. 522–527. [5] S. H. Alnabelsi and A. E. Kamal, “Performance modeling of secondary users in crns with heterogeneous channels,” in 2012 IEEE Global Communications Conference (GLOBECOM), Dec 2012, pp. 1326–1331. [6] S. Wang, B. R. Jackson, and R. Inkol, “Hybrid rss/aoa emitter location estimation based on least squares and maximum likelihood criteria,” in 2012 26th Biennial Symposium on Communications (QBSC), May 2012, pp. 24–29. [7] J. Wang, J. Chen, and D. Cabric, “Cramer-rao bounds for joint rss/doabased primary-user localization in cognitive radio networks,” IEEE Transactions on Wireless Communications, vol. 12, no. 3, pp. 1363–1375, March 2013. [8] P. Stoica and A. Nehorai, “Music, maximum likelihood, and cramer-rao bound,” IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 37, no. 5, pp. 720–741, May 1989.
