Reliability-Security Trade-Off Analysis of Cognitive Radio Networks with Jamming and Licensed Interference

. Cognitive radio networks (CRNs) allow coexistence of unlicensed users (UUs) and licensed users (LUs) and hence, mutual interference between UUs and LUs is neither ignored nor considered as Gaussian-distributed quantity. Additionally, exploiting jamming signals to purposely interfere with signal reception of eavesdroppers is a feasible solution to improve security performance of CRNs. This paper analyzes reliability-security trade-off, which accounts for maximum transmit power constraint, interference power constraint, jamming signal, and Rayleigh fading, and considers interference from LUs as non-Gaussian-distributed quantity. Toward this end, exact closed-form expressions of successful detection probability and successful eavesdropping probability, from which reliability-security trade-off is straightforwardly visible, are first suggested and then validated by Monte-Carlo simulations. Various results demonstrate that interference from LUs considerably decreases both probabilities while jamming signal enlarges the difference between them, emphasizing its effectiveness in improving security performance.


Introduction
High data rate and strong information security are two among obligatory requirements in designing next generation wireless communication networks (e.g., 5G) [1].The former can be achieved by cognitive radio technique whose advantage is high spectrum utilization efficiency by permitting both UUs and LUs to operate in the same frequency band primarily allocated to LUs [2].It is coexistence of both UUs and LUs that causes mutual interference between them [3] and hence, interference from LUs (shortly, licensed interference) is neither ignored nor considered as Gaussiandistributed quantity.On the other hand, physical layer security, which exploits space-time characteristics of wireless channels to secure information transmission, is a feasible solution to satisfy the latter [4][5][6][7][8].Therefore, physical layer security for CRNs, which can meet simultaneously both above requirements, has recently received a great deal of attention from both academia and industry.Among physical layer security solutions, generating jamming signals, which purposely destroys signal reception of eavesdroppers without degrading quality of service (QoS) of legal receivers, is an effective solution to improve security performance [9,10].For a certain physical layer security solution, it is interesting to assess successful detection probability at legal receiver (i.e., probability for legal receiver to successfully restore legitimate signals) and successful eavesdropping probability at eavesdropper (i.e., probability for eavesdropper to successfully decode/steal legitimate signals).Obviously, successful detection probability represents communication reliability while successful eavesdropping probability represents security level.Therefore, reliability-security trade-off is represented by relation between successful detection probability and successful eavesdropping probability.Because the difference (or ratio) between successful detection probability and successful eavesdropping probability has an equivalent physical meanings as the difference (or ratio) between signal power and noise power in communication theory, this difference (or ratio) can represent security performance.This paper analyzes reliability-security trade-off to assess information securing capability of CRNs under both interferences from licensed transmitter and jammer.

Wireless Communications and Mobile Computing
Table 1: Related works where jamming signal and legitimate signal are combined and simultaneously transmitted by a UU.The characteristic marked with "x" implies that it is considered in the corresponding reference.

Reference
Interference from LUs Maximum transmit power constraint Interference power constraint Thermal noise Reliability-security trade-off analysis [14] x x [17] x x x [18][19][20][21] x x x [22] x x [23] x [24] x x x 1.1.Related Works.This subsection reviews only publications relevant to jamming signal generation to guarantee information transmission securely in underlay CRNs (physical layer security solutions are various.For example, [11] deploys a friendly jammer to interfere with the eavesdropper; [12] proposes a relay selection scheme; [13] implements the joint power control in wiretap interference channels.However, most solutions are proposed for noncognitive radio networks (e.g., [11][12][13]).Because this paper focuses on jamming signal generation to guarantee information transmission securely in underlay CRNs, literature review on physical layer security solutions for noncognitive radio networks (e.g., [11][12][13]) should not be further investigated).Such a review is based on typical characteristics such as interference from licensed transmitter, maximum transmit power constraint, interference power constraint, thermal noise, and reliability-security trade-off analysis, and through it, contributions of this paper can be summarized in next subsection.
In [14], an unlicensed source transmits a jamming signal to secure information transmission of LUs in exchange for utilizing licensed spectrum in presence of interference from licensed transmitters.However, [14] does not account for power constraints of unlicensed transmitters such as interference power constraint and maximum transmit power constraint [15,16].In [17], cooperative relaying combines jamming signal generation to guarantee information transmission of LUs securely in cognitive two-way networks where one of two UUs sends both jamming signal and legitimate signal while the other UU merely amplifies and forwards the received signal together with its own signal.Optimum power allocation to jamming signal and legitimate signal is proposed there without being subjected to interference power constraint even though both jamming signal and interference from LUs and maximum transmit power constraint are accounted.In [18][19][20][21], beamforming vectors for jamming signal and legitimate signal are designed to optimize power allocation to jamming signal and legitimate signal at an unlicensed transmitter subject to maximum transmit power constraint and interference power constraint.Nevertheless, interference from licensed transmitter is not considered there.In [22], a UU transmits both jamming signal and legitimate signal under a simple context, which ignores maximum transmit power constraint and interference from licensed transmitter, to optimize the secrecy throughput of the wiretap channel for CRNs.The same optimization problem as [22] is revisited in [23].However, [23] considers a different context where interference from licensed transmitter, the connection outage constraint (i.e., the probability for signalto-interference-plus-noise ratio (SINR) at licensed receiver below a preset level is less than a certain value), and no thermal noise at corresponding receivers, and all randomly located users are investigated.It is noted that the connection outage constraint makes transmit power of UUs unchanged.Therefore, the connection outage constraint together with ignored thermal noise considerably simplifies information securing capability analysis in [23].The authors in [24] design a beamforming vector at the multiantenna unlicensed source to maximize the minimum secrecy rate of the unlicensed network while guaranteeing the minimum secrecy rate of the licensed network and satisfying maximum transmit power constraint for cognitive multicast communications.Besides beamforming the legitimate signals, the source transmits the jamming signal together with them to further secure information transmission.However, [24] ignores interference power constraint.Table 1 summarizes typical characteristics considered in [14,[17][18][19][20][21][22][23][24] with an emphasis that reliabilitysecurity trade-off analysis is not implemented there.
In contrast to [14,[17][18][19][20][21][22][23][24] where jamming signal and legitimate signal are merged and concurrently broadcasted by an unlicensed transmitter, [25][26][27][28][29][30][31][32] suggest new techniques to prevent eavesdroppers from stealing confidential information in CRNs where jamming signal and legitimate signal are transmitted by two different users.To be more specific, jamming signal and legitimate signal are separately transmitted by two different UUs to guarantee LUs' communication securely while maintaining QoS of UUs [25].Nevertheless, [25] neglects interference from licensed transmitter and interference power constraint.The authors in [26] complement the idea in [25] by further considering interference power constraint.Similar to [25,26] where jamming signal and legitimate signal are transmitted by different UUs, [27] proposes design of a weight vector for jamming signals forwarded by different relays to achieve maximum signal-tonoise ratio (SNR) at legal receiver while restricting SNR at the eavesdropper below a certain level in a context where both interference power constraint and maximum transmit power constraint are accounted.However, [27] ignores interference from licensed transmitter.The authors in [28] complement the work in [27] by accounting for this interference to maximize the secrecy rate.Instead of designing a weight x x x [30] x [31] x x [32] x x x vector for jamming signals forwarded by different relays, each equipped with single antenna as in [27,28], the authors in [29] implement design of a beamforming vector at a single multiantenna relay.Therefore, [27][28][29] solve quite similar problems.Furthermore, interference from licensed transmitter is ignored in [29].In [30], a UU is selected to transmit merely jamming signals not only to deteriorate QoS of the eavesdropper but also to assist secrete communication of another UU.However, power constraints and interference from licensed transmitter are neglected there, which are not reasonable for CRNs where UUs and LUs concurrently operate.For secure transmission of an unlicensed source, [31] adopts two unlicensed relays while [32] chooses an unlicensed relay and an unlicensed destination.However, [31,32] neglect interference from LUs.Furthermore, [31] does not investigate maximum transmit power constraint.Table 2 summarizes typical characteristics considered in [25][26][27][28][29][30][31][32] with an emphasis that reliability-security trade-off analysis is not implemented there.

System Model
Figure 1 demonstrates a system model for cognitive radio networks under investigation where an unlicensed source U S transmits confidential information to an unlicensed destination U D (namely, legal receiver) and this transmission is illegally wire-tapped by an eavesdropper E. Unlicensed users operate in the underlay paradigm [33] and hence, communication between U S and U D concurrently takes place with communication between a licensed transmitter L T and a licensed receiver L R , inducing mutual interference to each other.All terminals are assumed to be equipped with single antenna and operated in the half-duplex mode.In order to secure communication between U S and U D , an unlicensed jammer U J transmits a jamming signal √  V  at the same time that U S transmits a legitimate signal √  V  to purposely interfere with signal reception of E where   and   are transmit powers of U J and U S , respectively; V  and V  are correspondingly a jamming signal and a legitimate signal which are assumed to be Gaussian-distributed and have unity power (i.e., V  ∼ CN(0, 1) and V  ∼ CN(0, 1) where ℎ ab ∼ CN(0,  ab ) stands for a circular symmetric complex Gaussian random variable (CSCGRV) with zero mean and variance  ab ).
where   is power of the licensed transmitter L T ; V  is the signal transmitted by L T and is normalized to have unity power and Gaussian-distributed (i.e., V  ∼ CN(0, 1));   and   are additive white Gaussian noise at U D and E, correspondingly, each modelled as CN(0,  0 ); ℎ ab stands for a channel coefficient between a transmitter a and a receiver b, which is assumed to be independent, frequency flat, and Rayleigh distributed and hence, ℎ ab is modelled as a CSCGRV with zero mean and variance  ab , that is, is the interference power caused by U J to L R .Therefore, the total interference power induced by U S and To guarantee reliable signal reception at LUs, this interference power must be controlled below maximum interference power   that licensed receivers can tolerate, that is, In addition, each licensed transmitter is designed with a certain maximum transmit power (i.e., P for U S and P for U J ) and hence,   and   are limited by P and P , that is,   ≤ P and   ≤ P .Briefly, the transmit powers of U S and U J are subjected to the following constraints: where ( 3) and ( 4) are maximum transmit power constraints while ( 5) is interference power constraint.
From (5), one can select where 0 <  ≤ 1 is an interference power distribution factor for U S and U J . = 1 means no jammer. can be optimally selected not only to control interference to L R but also to increase interference to E.
Combining maximum transmit power constraint in (3) and interference power constraint in (6), transmit power of U S must satisfy   ≤ min(  /|ℎ  | 2 , P ).For maximum radio coverage, the equality should hold, that is, Similarly, to meet both maximum transmit power constraint in (4) and interference power constraint in (7) as well as maximize radio coverage, transmit power of U J is established as The jamming signal V  is intentionally generated by U J to only cause interference to E without inducing performance degradation of U D .This can be achieved by allowing U J to share V  with U D (e.g., the seed of the jamming signal generator at U J is shared with U D in a secure manner through a cooperation agreement only between U J and U D before information transmission starts).Such a jamming signal generation is widely accepted in open literature (e.g., [5, 8, 19-22, 25, 30-32] and references therein).As such, the legal receiver U D can accurately predict the jamming signal and completely eliminate it out of U D 's received signal, ultimately resulting in the jamming-free signal as ũ = ℎ  √  V  + ℎ  √  V  +   .Based on the jamming-free signal, one can infer SINR at U D as where E ,,... {⋅} denotes the expectation operation over random variables , , . ... Because the knowledge of the jamming signal V  is only shared between U J and U D for information securing purpose, the eavesdropper does not know it.Therefore, the SINR at E is inferred from (2) as It is worth mentioning that the licensed transmitter L T introduces interference powers at U D and E as |ℎ  | 2   and |ℎ  | 2   , respectively, according to (10) and (11).These interference powers are neither ignored nor considered as Gaussian-distributed quantity but rather treated as exponentially distributed random variables.Additionally, U J deliberately generates jamming power   |ℎ  | 2 to the eavesdropper.As such, increasing this jamming power is expected to secure information transmission of U S .
Channel capacities at U D and E are, respectively, given by [34] which are helpful in analyzing successful detection probability of the legal receiver U D and successful eavesdropping probability of the eavesdropper E in next section.

Performance Analysis
This section analyzes successful detection probability of the legal receiver U D ,  det , and successful eavesdropping probability of the eavesdropper E,  eav .According to information theory,  det and  eav are correspondingly defined as probabilities for channel capacities of U D and E to be larger than a required level.In other words, successful detection probability of U D and successful eavesdropping probability of E are probabilities for U D and E to successfully decode U S 's information, respectively.As such, they are important metrics for evaluating communication reliability and information securing capability.The larger  det , the more reliable the signal reception at U D .Meanwhile, the larger  eav , the less secure the information transmission of U S .Therefore, the relation between  det and  eav represents reliability-security trade-off.Furthermore, the difference (or ratio) between successful detection probability and successful eavesdropping probability has an equivalent physical meanings as the difference (or ratio) between signal power and noise power in communication theory and hence, this difference (or ratio) can represent security performance.
The successful detection probability of U D and the successful eavesdropping probability of E are respectively expressed as where  th is the threshold for successfully decoding U S 's information,  = 2  th − 1, and Pr{Q} is the probability of the event Q.

Theorem 1. The exact closed-form representation of the successful detection probability of the legal receiver U D is given by
where () is the exponential integral function [35].
Proof.Please see Appendix A.
Before deriving the exact closed-form expression of the successful eavesdropping probability of E, we introduce two special results in the following lemmas.

Lemma 2. The integrand
can be represented in closed form as where Proof.Please see Appendix B.

Lemma 3. The integrand
can be represented in closed form as where = 1 Proof.Please see Appendix C.
Using the results in Lemmas 2 and 3, one can represent the successful eavesdropping probability of E in closed form as follows.

Theorem 4. The successful eavesdropping probability of the eavesdropper E has an exact closed form as
Proof.Please see Appendix D.
It is worth mentioning that although the exact closedform expressions of the successful detection probability of U D and the successful eavesdropping probability of E in (15) and (30), respectively, are relatively long, they are simply computed since they encompass simple built-in functions such as the Whittaker function and the exponential integral function.Therefore, (15) and (30) are helpful in quickly assessing the trade-off between the communication reliability and the information security in underlay CRNs under interference from licensed transmitter, jamming signal, maximum transmit power constraint, and interference power constraint, without exhaustive Monte-Carlo simulations.

Results and Discussion
This section provides numerous results to illustrate the tradeoff between the communication reliability and the information security in CRNs under interference from licensed transmitter and jamming signal.Theoretical results are obtained from analytical expressions in ( 15) and (30) while simulation results are generated from Monte-Carlo simulations with 10 6 channel realizations.The coordinates of L T , L R , U S , U D , and E are arbitrarily chosen as (0.1, 0.9), (0.8, 0.7), (0.0, 0.0), (1.0, 0.0), and (0.9, 0.2), correspondingly, for illustration purpose while U J is on the straight line connecting U S and U D ; that is, the coordinate of U J is (, 0) where 0 <  < 1.Additionally, the fading power for the a-b channel is modelled as  ab =  − ab according to [36] where  is the path-loss exponent and  ab is the distance between the transmitter a and the Figure 2 illustrates the trade-off between the communication reliability and the information security in CRNs with respect to the maximum interference power-to-noise power ratio   / 0 for the maximum transmit power-to-noise power ratio   / 0 = 20 dB, the required channel capacity  th = 0.1 bits/s/Hz, different interference levels from licensed transmitter (  / 0 = 15, 18, 21 dB), the interference power distribution factor  = 0.3, and  = 0.3.It is observed that the analysis accurately matches the simulation, confirming the validity of ( 15) and (30).Additionally, both the successful detection probability  det and the successful eavesdropping probability  eav are proportional to   .This is due to the fact that   upper-bounds the power of unlicensed transmitters according to   = min(  /|ℎ  | 2 ,   ) and   = min((1 − )  /|ℎ  | 2 ,   ) and hence, the increase in   raises the transmit power, ultimately increasing   and   according to (10) and (11).In other words, the probabilities for   and   to be greater than a threshold increase proportionally to   .The fact that both  det and  eav are proportional to   shows that the communication reliability (i.e.,  det ) trades off the security capability (i.e.,  eav ).Moreover, interference from licensed transmitter decreases SINRs at both U D and E and hence, the decrease in SINRs at U D and E drastically reduces the successful detection probability and the successful eavesdropping probability.However, the successful detection probability is significantly larger than the successful eavesdropping probability.This signifies that the probability for the legal receiver U D to successfully recover the source information is larger than the probability for the illegal receiver E to do so, creating high security performance.In the same context as Figure 2 except   / 0 = 12 dB, Figure 3 demonstrates the trade-off between the communication reliability and the information security in CRNs with respect to the change of   / 0 .It is seen that the simulation and the analysis are in excellent agreement.This again validates the accuracy of ( 15) and (30).Additionally, same remarks as Figure 2 can be withdrawn such as the increase in the successful detection probability and the successful eavesdropping probability with respect to the increase in   , the degradation of the successful detection probability and the successful eavesdropping probability with respect to the increase in interference level from licensed transmitter, and the superiority of the successful detection probability to the successful eavesdropping probability which indicates a high information securing capability.
Figure 4 shows the trade-off between the communication reliability and the information security in CRNs with respect to the required channel capacity  th for  = 0.3,   / 0 = 19 dB,   / 0 = 15 dB,  = 0.5, and different interference levels from licensed transmitter (  / 0 = 15, 18, 21 dB).It is observed that the analysis coincides with the simulation, confirming the validity of ( 15) and (30).Moreover, interference from licensed transmitter (i.e.,   increases) dramatically reduces the successful detection probability and the successful eavesdropping probability.Furthermore, higher required channel capacity lowers these probabilities.This is reasonable because, given operation parameters, channel capacities at U D and E are only achieved at a certain level and hence, more stringent channel capacity requirement reduces the successful detection probability and the successful eavesdropping probability.Nevertheless, the successful detection probability outperforms the successful eavesdropping probability, indicating high security performance.Figure 5 illustrates the trade-off between the communication reliability and the information security in CRNs with respect to the jammer's position (i.e., ) for   / 0 =   / 0 = 15 dB,  th = 0.1 bits/s/Hz,  = 0.5, and different interference levels from licensed transmitter (  / 0 = 15, 18, 21 dB).It is observed that the simulation accurately matches the analysis, validating (15) and (30).Additionally, interference from licensed transmitter (i.e.,   increases) significantly decreases the successful detection probability and the successful eavesdropping probability.Moreover, the successful detection probability is independent of the jammer's position.This is because the jamming signal is to purposely harm the signal reception of E without degrading the performance of U D .However, the successful eavesdropping probability is severely dependent on the jammer's position.More specifically, because the jammer is located at (0.9, 0.2), when the jammer moves toward the eavesdropper (i.e.,  increases from 0 to 0.9), the successful eavesdropping probability drops abruptly due to the increase in the receive power of the jamming signal at E. This enlarges the difference between the successful detection probability and the successful eavesdropping probability, securing information transmission of U S and emphasizing the role of the jammer in improving security performance.Furthermore, when the jammer moves away from the eavesdropper (i.e.,  increases from 0.9 to 1), the successful eavesdropping probability increases significantly owing to the decrease in the receive power of the jamming signal at E, reducing security performance.For all the jammer's position under consideration in Figure 5, the difference between the successful detection probability and the successful eavesdropping probability is very large, indicating high security performance.
Figure 6 demonstrates the trade-off between the communication reliability and the information security in CRNs with respect to the interference power distribution factor (i.e., ) for   / 0 = 15 dB,   / 0 = 12 dB,  th = 0.1 bits/s/Hz,  = 0.7, and different interference levels from licensed transmitter (  / 0 = 15, 18, 21 dB).It is seen that the simulation and the analysis are in a perfect agreement, validating (15) and (30).Moreover, interference from licensed transmitter (i.e.,   increases) considerably reduces the successful detection probability and the successful eavesdropping probability.Furthermore, the successful detection probability (or the successful eavesdropping probability) slowly (or quickly) increases with the increase of , making the difference between them inversely proportional to .The large difference between them represents the dominance of the successful detection probability to the successful eavesdropping probability (equivalently, high security capability).Therefore, Figure 6 shows that small values of  are good for information security.

Conclusions
The successful detection probability and the successful eavesdropping probability in underlay cognitive networks are analyzed in this paper under interferences from licensed transmitter and jammer, the maximum transmit power constraint, the interference power constraint, and Rayleigh fading to assess the communication reliability and the information security of CRNs without exhaustive simulations.Exact closed-form expressions of the successful detection probability and the successful eavesdropping probability are derived and verified by Monte-Carlo simulations.Numerous results illustrate that interference from licensed transmitter considerably reduces both probabilities while interference from the jammer makes the successful detection probability significantly larger than the successful eavesdropping probability, demonstrating the jammer's efficacy in enhancing security performance.

B. Proof of Lemma 2
Performing the integral by part on ( 16), one obtains By denoting it is straightforward to express the integrand in (B.2) in terms of the (⋅) function as (18).Given the H(, , ) function in (B.2), one can express (B.1) as (17), completing the proof.
Applying the partial fraction decomposition,  can be simplified as where  and   are defined in (25) and (26), respectively.By denoting and noting that the first integrand of (C.10) can be represented in terms of the H(⋅) function in (B.2), it is obvious that (C.10) becomes (24).The B(, , V) function can be represented as (27) by firstly performing the variable change and then using [35, eq. (3.381.6)].

D. Proof of Theorem 4
Plugging ( 11) into ( 14) and after some simplification, one obtains where Applying the definition of the expectation, one obtains the closed form of  as where By plugging ( 9) into (D.6), one can decompose  as By substituting (D.8) into (D.7) and then plugging the result together with (8) into (D.5), one can partition  eav into two terms as where Inserting (D.12) and (D.14) into (D.9) results in (30), completing the proof.

Figure 4 :Figure 5 :
Figure 4: Successful detection probability and successful eavesdropping probability versus required channel capacity.

Figure 6 :
Figure 6: Successful detection probability and successful eavesdropping probability versus the interference power distribution factor.

Table 2 :
Related works where jamming signal and legitimate signal are transmitted by two different users.The characteristic marked with "x" implies that it is considered in the corresponding reference.