Intelligent Reflecting Surfaces for Sum-Rate Maximization in Cognitive Radio Enabled Wireless Powered Communication Network

In this paper, we study a cognitive radio (CR)-enabled wireless powered communication network (WPCN) with an intelligent reflecting surface (IRS). A cognitive wireless powered communication network (CWPCN) consisting of one secondary wireless powered communication network, in which a hybrid access point broadcasts wireless energy to multiple users on downlink and receives information signals on uplink, shares the same spectrum with the existing primary wireless communication network. In this paper, we consider an underlay CWPCN in which the secondary network is regulated by a given interference temperature constraint (ITC) such that interference with the primary network is kept no greater than a predefined threshold. To enhance the performance of the CWPCN, we consider an IRS-assisted network in which multiple passive reflection elements are configured to reflect the signals in any desired phase/direction. Our main goal is to maximize the sum throughput of secondary users while managing the interference constraint. For this, we jointly optimize the uplink and downlink phase shift matrices of the IRS elements with optimal time slots for wireless energy transfer (WET) on downlink and wireless information transfer on uplink. In finding the optimal solution, the formulated optimization problem is non-convex, complex, and intractable. In this paper, we propose an alternating optimization (AO)-based solution with a successive convex approximation (SCA) technique. We show through simulation results that the proposed IRS-assisted network provides a significant enhancement in performance over the conventional CWPCN.


I. INTRODUCTION
Recently, intelligent reflecting surface (IRS)-assisted wireless networks are emerging as a promising solution to meet the massive growth of mobile and communication devices by utilizing a smart radio environment [1], [2], [3]. The IRS, which is also known as a re-configurable intelligent surface (RIS), consists of multiple passive elements capa-The associate editor coordinating the review of this manuscript and approving it for publication was Ding Xu . ble of reconfiguring reflecting elements independently to enhance the energy and spectral efficiency of a network [4], [5], [6]. Specifically, the reflective surfaces of the IRS elements can be independently configured to induce a controllable amplitude or phase change in the incident or reflected signal. With the recent advances in metasurfaces and micro-electromechanical systems (MEMS), it is possible to reconfigure IRS elements using controllable phase shifters in real time [7]. Thus, to increase the capacity and efficiency of wireless networks, IRSs can be deployed to control fading VOLUME 11, 2023 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ environment impairments and interference challenges. Considering the significant IRS contributions to energy/spectralefficient networks, we jointly studied the significance of IRSs in cognitive radio (CR)-enabled wireless powered communication networks.
In recent decades, wireless powered communication networks (WPCNs) have emerged as a promising solution for limited-battery-power mobile and communication devices [8], [9], [10]. WPCNs have become a potential candidate for the several low-power applications such as wireless powered IoT networks (WPINs) and RF identification (RFID) networks [11]. WPCNs typically consist of a hybrid access point (H-AP) that coordinates multiple wireless devices for wireless energy transfer (WET) on downlink and wireless information transfer (WIT) on uplink. Moreover, a harvest-then-transmit protocol is considered in WPCNs in which wireless devices have no battery sources, and instead harvest energy from the power transmitter, i.e. the H-AP, and the devices transmit information signals to the H-AP by utilizing this harvested energy [12]. In WPCNs, there is a requirement to jointly design the energy and information transmission for the efficient performance of a network. To achieve the required quality of service (QoS) in the WPCNs, it is mandatory to find the optimal allocation of time for the WET and WIT, because energy and throughput efficiencies are a function of time allocation parameters.
With the massive increase in mobile devices and wireless systems, the available spectrum is greatly limited. Over the last two decades, cognitive radio (CR) technology has been discussed as an efficient approach for opportunistic use of the available spectrum to connect massive devices [13], [14], [15]. By utilizing CR technology in the WPCN, we can enhance the spectral and energy efficiency of wireless communication networks in which a primary network shares spectrum with a secondary network without the need for new spectrum allocation [16], [17], [18], [19]. In particular, we consider a cognitive radio-enabled WPCN (CWPCN) in which a secondary WPCN conducts WET on downlink and WIT on uplink with the existing primary wireless communication. There are two main categories of CWPCN, namely underlay CWPCN and overlay CWPCN. In underlay CWPCN, there is no cooperation between the primary and secondary network thus the secondary network manages to not increase interference with the primary network by more than a predefined threshold. However, in overlay CWPCN, there is a cooperation between the primary and secondary networks thus primary receivers can avoid interference from secondary network by utilizing the data sharing which expense the data sharing overhead [20]. In this paper, we consider an underlay CWPCN, in which there is no cooperation between primary and secondary CWPCN. Hence, it is a great challenge to optimize the configuration of the CWPCN such that the interference from secondary network is avoided and required quality of service is achieved.
In [13], Zhang et al. provided a discussion on the interference challenges for CR technology while scheduling resources in a CR-enabled wireless network. They proposed convex-optimization-based solutions in solving these problems efficiently. In [16], Lee and Zhang, studied a CR-enabled wireless powered communication network under spectrum sharing with primary wireless communication. They maximized the sum throughput of cognitive users while managing the interference threshold for the primary transmissions. They proposed optimal solutions for throughput maximization in underlay-and overlay-based CWPCN models. The authors in [17] discussed cluster-based cooperation with secondary users in a CWPCN to ensure throughput fairness among distant and nearby users. For this, they maximized minimum throughput for the secondary users by optimizing the available resources and managing interference with primary transmissions. In [18], Rezai et al. proposed a full duplex overlay based CWPCN in which they jointly optimize the time and power allocations for secondary WPCN to maximize sum throughput of secondary users. They use the gradient projection technique to convert the non-convex problem into convex problem. In [19], Li et al. investigated a multi-carrier CWPCN in which both offline and online algorithms are proposed to jointly optimize the network resources for secondary user rate maximization under the outage probability constraint. In [20], Kim et al, studied sum-throughput maximization problem for a multi-user MIMO CWPC. They jointly optimize the energy and information covariance matrices along with time allocation for uplink and downlink in secondary WPCN.
In [28], Zheng et al. considered IRS-assisted user cooperation in a WPCN and jointly optimized power and time resources of the network to maximize common throughput. They showed performance from an IRS that improved the energy and spectral efficiency in WET and WIT. The authors in [30] and [31], discussed resource allocations in IRS-assisted WPCNs while considering NOMA technology on uplink and downlink. They proposed an efficient algorithm to maximize system performance while optimizing downlink and uplink phase shift matrices. In [35], Yuan et al. combined cognitive technology with multiple IRSs and maximized the achievable rate for the secondary user by controlling the transmit power of the secondary transmitter and the primary interference temperature constraint. They jointly optimized IRS coefficient matrices and beamforming at the secondary transmitter to maximize throughput. In [36], Wu et al. discussed the spectrum efficiency and security of secondary users in IRS-assisted cognitive radio network (CRN). They investigate the application of IRS in an underlay CRN and jointly design the transmit beamforming and the reflect beamforming at the IRS. They proposed an iterative alternating optimization algorithm to solve the secrecy energy maximization problem. In [38], Jiang et al. studied an application of IRS in MIMO cognitive radio systems. They proposed the weighted minimum mean-square error (WMMSE) method and an alternating optimization based algorithm to investigate the sum-rate maximization problem.

B. MOTIVATION AND CONTRIBUTIONS
Inspired by the potential benefits of IRSs, we study the effectiveness of passive elements in IRSs on a CWPCN. In Section I-A, we mentioned several applications of IRS in various networks but the IRS application in CWPCNs is not investigated in literature. This motivates us to investigate the potential benefits of IRS particularly in cognitive radio-enabled WPCNs. In CWPCNs, it is challenging to jointly design the secondary network resources while managing the interference towards primary network. However, in IRS-assisted CWPCN we can configure IRS passive elements to enhance or attenuate the reflected signal in the desired phase/direction. The challenge is to design the phase shift matrices for both downlink and uplink so they increase the performance of the secondary network while managing the minimum interference threshold. In particular, we consider an underlay-based CWPCN in which a secondary WPCN co-exists with the primary network, and we optimize the phase shift matrices and time allocations of the secondary WPCN while managing the interference threshold for the primary network. However, the formulated problem is non-convex and very complex because of highly coupled optimal variables. To solve it, we convert a non-convex problem into equivalent convex problems by using semi-definite relaxation (SDR) method and an alternating optimization (AO)-based algorithm in which we find alternative solutions for optimal variables in the iterations until convergence achieved. We also exploit feasible point pursuit [40] and sequential convex approximation (SCA) methods to get faster convergence in the algorithm.
The main contributions of this paper are summarized as follows.
• We propose an IRS-assisted CWPCN and formulate the sum throughput maximization problem for a secondary WPCN. Our main objective is to maximize the sum throughput of cognitive users in the secondary CWPCN by optimizing the phase shift matrices of IRS elements and time allocations on downlink and uplink, subject to the interference threshold of the primary network and minimum rate requirement of the secondary users. We consider two interference temperature constraints (ITCs) for uplink and downlink to keep the interference towards primary network lower to defined threshold. The proposed problem is non-convex and numerically intractable because of highly coupled variables.
• We propose an AO-based approach to solve the nonconvex sum-throughput optimization problem. For this, we divide the main problem into sub-problems and solve them in separate iterations. We find near-optimal solution for IRS elements for given time allocations.
We use semi-definite relaxation (SDR) and successive convex approximation methods to convert a non-convex problem into convex sub-problems and solve them in iterations. Moreover, we exploit a feasible point pursuit method [40] for finding the initial feasible point to get faster convergence in the algorithm.
• Numerical simulations are presented to evaluate the proposed schemes. Through performance evaluation, we show that the proposed IRS-assisted CWPCN has a significant effect and better performance in comparison to a conventional CWPCN. We summarize the important notations and their definitions in Table 1.

II. SYSTEM MODEL AND PROBLEM FORMULATION
In this section, we describe the IRS-assisted cognitive WPCN shown in Fig.1, in which a single pair of primary network co-exist with the secondary network and share the same bandwidth for wireless energy transfer and wireless information transfer. The primary network consists of a primary transmitter (P-TX) and a primary receiver (P-RX), each with a single antenna. The secondary network consists of a single-antenna hybrid access point, N cognitive users denoted as {CU 1 , CU 2 , . . . , CU N } with a single antenna each, and an  IRS surface with L reflecting elements. The communication channels from the transmitters to receivers are shown in Table 2. In addition, we assume that channel state information (CSI) is perfectly known at the transmitter which was commonly used in several papers [4], [30], [31], [32]. All the CUs have embedded energy-storage equipment in the form of rechargeable batteries to store the energy harvested from signals. In our proposed CWPCN system, we assume CUs follow a harvest-then-transmit protocol. The CUs harvest energy from the H-AP and primary transmitter. In addition, CUs have no embedded energy sources, and send an information signal only by utilizing the harvested energy.
In the proposed CWPCN, the P-TX transmits a wireless information signal to the primary receivers for the whole coherence interval. However, the secondary network operates in a time division duplexing (TDD) manner, where the total coherence time interval, denoted by T is divided into two phases, as shown in Fig.2. In the secondary WET phase, all the CUs harvest energy from the energy-carrying signals FIGURE 2. The TDD-frame-based cognitive wireless powered communication network scheme. The primary network transmits for the entire coherence interval. However, for the secondary WPCN, the whole coherence interval is divided into two phases, one for secondary wireless energy transfer and one for secondary wireless information transfer.
broadcast by the H-AP and the P-TX in τ T seconds during downlink transmissions. In the secondary WIT phase, the CUs transmit an information-bearing signal on uplink in the remaining time slot by utilizing the harvested energy. Without loss of generality, it is assumed that the total coherence time is normalized to 1 for convenience. Since the proposed system operates in a TDD manner, channel reciprocity holds for forward and reverse links.

A. UNDERLAY-BASED CWPCN
In the proposed network, during the secondary WET phase, the H-AP broadcasts arbitrary energy signal with transmit power P a , in time τ . The total energy harvested by CU n from the H-AP and P-TX during WET interval τ , denoted by E n is expressed as: where ζ n ∈ (0, 1) for n = 1, 2, 3, . . . , N denotes the energy-harvesting efficiency of CU n ; P t is the total transmit power at P-TX; d = diag(α 1 e jθ 1 , α 2 e jθ 2 , . . . , α L e jθ L ), is the diagonal matrix for the WET phase, with diagonal elements denoting the phase shift matrix of the IRS, where θ l ∈ [0, 2π), ∀ l and α l ∈ [0, 1], respectively, are the phase shift and amplitude reflection coefficient of each element. To maximize signal reflection by the IRS, we set amplitude reflection coefficient α l = 1, ∀ l . The Eq.1 can be transformed into Eq. 2 as follows: The average power available for WIT at CU n is given by In an underlay CWPCN, the H-AP's WET and the CU's WIT are regulated by the given interference temperature constraint such that the interference power at the P-RX is kept no greater than a predefined threshold, denoted by ≥ 0 [13], [16]. Accordingly, during WET, we set the ITC as follows: where P a is the maximum power available at the H-AP, and G During WIT on uplink, the H-AP decodes the information signals by applying successive interference cancellation (SIC) to eliminate multi-user interference. In particular, to decode the information signal of the n th cognitive user, the H-AP first decodes the information signals of all k th CUs, where k < n, and then cancels these signals from the received signal to remove interference in the order of k = 1, 2, . . . , n − 1 [16], [31]. The signal received from the k th user, ∀k > n is treated as noise. Hence, the achievable throughput of n th user in bps/Hz can be expressed as in Eq. 5, as shown at the bottom of the next page. u = diag(β 1 e jω 1 , β 2 e jω 2 , . . . , β L e jω L ) is the IRS phase shift matrix for the WIT phase, with diagonal elements denoting the phase shift matrix of the IRS, where ω l ∈ [0, 2π ), ∀ l and β l ∈ [0, 1], respectively, are the phase shift and amplitude reflection coefficient of each element. To maximize signal reflection by the IRS, we set amplitude reflection coefficient  ). We can rewrite Eq. (5a) to obtain Eq. (5b) by using the mathematical formulation discussed above. In Eq. (5b), Q u is the equivalent phase shift matrix at the IRS for secondary WIT; p n is the harvested power of CU n and σ 2 is the noise variance at the H-AP. However, P t Tr(Q u G pa ) shows the interference from the P-TX with the H-AP. Hence, the achievable sum throughput in the 1 − τ interval for the CUs is given in Eq. 6, as shown at the bottom of the next page [41], [42].
During the secondary WIT phase, ITC is applied so that N n=0 p n Tr(Q u G rc n ) ≤ Here G rc n =ḡ rc nḡ H rc n , whereḡ H rc n = [g H rc n h H rc n ], and g rc n H = h H ic n diag(h ri ).

B. PROBLEM FORMULATION
In this paper, our aim is to maximize the sum throughput of cognitive users by designing and utilizing the available resources in an IRS-assisted CWPCN. We observe from Eq. 5 that sum throughput is a direct function of time allocation and phase shift matrices for uplink and downlink transmissions. If we increase the downlink time allocation, it will increase the harvested energy, but at the same time, cognitive users will have less time to transmit information, which degrades throughput of the network, and vice versa. Hence, there is a need to find the optimal time allocation for downlink and uplink phases to maximize throughput. Similarly, IRS reflecting elements can be adjusted so the beams are constructively added in the desired direction to enhance the performance of the network. Here, to maximize the sum throughput, we jointly optimize the time allocations and phase shift matrices for efficient use of the available resources.
We formulate the optimization problem as The objective in Problem (7) is to maximize sum throughput while optimizing the variables, time allocation τ , and phase shift matrices Q u and Q d for downlink and uplink phases, respectively. Constraint (8b) satisfies the ITC for uplink where the total interference by cognitive users with the P-RX is kept no greater than the predefined threshold. Constraint (8c) satisfies the ITC on downlink to keep to the limit on interference from the H-AP with the P-RX. Constraint (8d) confirms the minimum throughput requirement. Due to the coupling of optimal variables, Problem (8) is non-convex and highly complex, and it is analytically difficult to find an optimal solution for maximum sum throughput.

III. ALTERNATING OPTIMIZATION (AO)-BASED SOLUTION
In this section, we propose an AO-based solution to solve Problem (8) efficiently, since it is difficult to solve Problem (8) analytically because of the large coupling of optimizing variables. Hence, we use an iterative AO method to find a near-optimal solution for optimizing variables. In particular, we decouple Problem (8) into sub-problems by alternatively fixing the variables, and we maximize the objective function until convergence is achieved. We use SDP and SCA to solve sub-problems for uplink and downlink phase shift matrices, and use an exhaustive search method for optimal time allocation.
A. FIND Q * u BY FIXING τ, AND Q d Here, we fix τ, and Q d , and introduce an auxiliary variable, . VOLUME 11, 2023 We rewrite Problem (8) as We observe Tr[AY ] is a linear function w.r.t. Y, and hence, it is convex. However, Constraint (9b) is non-convex because of the coupling of γ and Q u . For this, we introduce a new auxiliary variable such that t 2 ≥ P t γ Tr(Q u G pa ) . Hence, Problem (9) can be rewritten as In Problem (10), Constraint (10c) is non-convex. As we know, t 2 γ is a convex function w.r.t t and γ . So, we perform first-order Taylor approximation as follows: Hence, we can convert Constraint (10c) into convex by using Eq.11. We rewrite Problem (10) as Note that Problem (12) is still non-convex because of the rank-1 constraint. However, we can convert the rank-1 constraint to a tractable convex constraint by using an exact penalty function method [32], [43]. From (9g), we can write where λ 1 is the largest eigenvalue of Q u . This constraint is added to the objective function and thus Problem (12) can be rewritten as Here, δ is the penalty factor. Increasing the penalty factor will dominate the constraint. Hence, we can start from a smaller value of δ and gradually increase the value until we obtain a rank-1 solution. Considering that λ 1 (Q u ) a convex function, it is still a non-convex problem. We can use SCA to convert it to iterative convex sub-problems. Hence, using Taylor underestimation we derive (1 − τ )log 2 1 + p n Tr(Q u G ac n ) K k=n+1 p k Tr(Q u G ac k ) + σ 2 + P t Tr(Q u G pa ) (6a) Tr(Q d G ac n ) P a + Tr(Q d G pc n ) P t Tr(Q u G ac n ) where x 1 (i) is the corresponding eigenvector of λ 1 in the i th iteration. We can rewrite the Problem (16) as follows: Therefore, the problem is converted into a convex problem and can be solved using the MATLAB CVX solver [44].

2) INITIALIZATION
For fast convergence in the SCA solution, we generate an initial feasible point by exploiting a feasible point pursuit method [40]. For this, we further transform Problem (16) by introducing slack variables s 1 , s 2 , and s 3 , and rewrite a new objective function as follows: By solving Problem (17), we extract the initial values of eigenvalue λ 1 (Q (0) u ) and corresponding eigenvector x 1 (0) by using eigenvalue decomposition (EVD). Then, we solve Problem (16) in iterations until convergence is achieved.
Next, we fix τ, and Q u , given that Q u = Q * u , by solving Problem (16). We rewrite Problem(8) as . Similarly, as discussed in Section III-A1, we convert the rank-1 constraint into a tractable function and rewrite Problem (18) as d )x 1 (i) )) (19b) subject to (18b), (18c), (18d), (18f ), (18g), (18i) (19c) Next, we apply the same initialization method discussed in Section III-A2 to find the initial feasible point of Problem (19) for fast convergence. Then, we iteratively optimize Q u and Q d until convergence is achieved. We use exhaustive search to find the optimal value of τ for the given Q * u and Q * d . We summarize the proposed SDR-and SCA-based solution of Problem (8) with Algorithm 1.

IV. PERFORMANCE EVALUATION A. SIMULATION PARAMETERS FOR UNDERLAY CWPCN
In this section, we discuss the numerical results showing the performance of our proposed solution in an IRS-assisted CWPCN system. In the simulation setup, we consider a 3D coordinate system where the H-AP and the IRS are located at (0m, 0m, 0m) and (2m, 0m, 2m), respectively. The P-TX and P-RX are located 2m apart at (0m, 5m, 0m) and (2m, 5m, 0m), respectively. However cognitive users are randomly and uniformly distributed within a radius of 1.5m centered at (2m, 0m, 0m). We set the maximum power for the P-TX and H-AP at 40dBm with the receiver energy-harvesting efficiency at ζ n = ζ = 0.8, R min n = R min = 0.5 bps/Hz, and noise power at σ 2 = −70dBm. For underlay CWPCN, we set maximum interference threshold, = −60dBm. The distance-dependent path loss is modeled as ρ n = A 0 VOLUME 11, 2023 Algorithm 1 The Proposed SDR-and SCA-Based AO Algorithm to Find Solution to Problem (8) 1: Set the initial values for τ, Q d , δ, tolerance value ξ 2: // main loop: 3: repeat 4: Find the initial feasible solution for γ (0) , t (0) , λ 1 (Q (0) u ) and x Assign

11:
until 12: where A 0 = −30dBm is the free space path loss at the reference distance of d 0 = 1m; d denotes the distance between transmitter and receiver, and α denotes the path loss exponent. For our simulations, we modeled all the direct channels from the H-AP to the CUs, from the P-TX to the P-RX, from the P-TX to the CUs, from the P-RX to the CUs, and from the P-TX to the P-RX as Rayleigh fading channels with a zero mean, where unit variance and path loss component α D = 3.5. However, other channels from the H-AP to the IRS, from the IRS to the CUs, from the IRS to the P-RX, and from the IRS to the P-RX are modeled as Rician fading with path loss component α I = 2.2: where κ is the Rician factor set to 3, h n NLOS denotes the non-line-of-sight (NLOS) standard Rayleigh fading components with zero mean and unit variance, and h LOS n is a line-of-sight (LOS) component that can be modeled using a far-field uniform linear array (ULA) [45],  and θ is the angle-of-departure (AOD) or angle-of-arrival (AOA) of the IRS elements.

B. RESULTS AND DISCUSSION
We first show the convergence of the proposed algorithm. Fig.3 shows that the proposed algorithm has very fast convergence and achieves convergence at or before the 5 th iteration for L = [8,12,16] and N = [2,4,6]. In addition, we studied when the number of IRS elements increases, the sum throughput of cognitive users increases, which also shows the effectiveness of IRS elements. Fig. 4 shows the sum throughput for different numbers of users along with increased IRS elements. Here, we compare the results with different schemes: (a) the proposed algorithm based on alternating optimization, (b) CWPCN without considering IRS elements (c) a random phase IRS based on random selection of IRS phase shift matrices. The sum throughput increases with the increasing number of IRS elements and when we increase the number of cognitive users. This shows the constructive beam effects of the IRS elements towards cognitive users. Moreover, we can see that the proposed algorithm performs better than the other schemes because the proposed algorithm is based on joint optimization of time allocation and IRS phase shift matrices. Fig. 5 represents the feasibility analysis of our proposed algorithm for different number of IRS elements. We perform simulations for 100 channels of randomly distributed. Here, we observe the proposed algorithm has maximum feasibility for L = [4,8,12,16,20,24,28]. However, the scheme with random IRS phase shifts achieves less feasible solutions even below 20%. This is because random phase shifts allocation cannot guarantee that IRS beams on uplink and downlink are aligned with the communication channels such that the network performance can be improved. Hence, this shows the importance of optimization of IRS passive beamforming on uplink and downlink.    6 depicts the performance of the proposed algorithm under the given interference temperature constraint for predefined threshold . We can see that the throughput ratio increases when we increase the IRS elements, which means there is a significant effect from constructive beams towards cognitive users. In addition, this is also because of destructive beam effects towards the primary network satisfying Constraint (8b), thus causing less interference with the primary network and enhancing the sum throughput for secondary users. Sum throughput also increased when we increase the interference temperature limit, because a cognitive user can transmit information with more power. Fig. 7 shows the sum throughput versus the transmission power available at the P-TX for different numbers of IRS elements. Here, we observe that the sum throughput decreased when we increased the power available at the P-TX. When the power at the P-TX increases, it increases interference with the H-AP during uplink transmission, which decreases the sum throughput for cognitive users. However, we also  observe the effectiveness of IRS elements as we increase their number, which lowers the effect of power at the P-TX. This is because of the constructive behavior of beams in the desired direction and the destructive behavior from interference by the reflecting elements. In addition, we also plot the sum throughput with respect to power available at the H-AP in Fig. 8, which shows that the sum throughput increases with the increase in available power at the H-AP. The harvested energy increases with the increase in power at the H-AP, which increases the sum throughput. We can also see that harvested energy is the function of power available at the H-AP, but there is a weaker channel between the P-TX and the CUs compared to the channel between the P-TX and the H-AP. Hence, there is a stronger effect from interference with the P-TX on uplink than the effect from harvested energy in Fig. 7. However, there is a tradeoff between power at the P-TX and at the H-AP. We need to choose optimum values for both, so that throughput for cognitive users can increase. VOLUME 11, 2023

V. CONCLUSION
In this paper, we discussed an IRS-assisted CWPCN to maximize throughput for secondary users. We proposed an alternating optimization algorithm for the maximization problem. Specifically, time allocations for downlink and uplink transmission and IRS phase shift matrices are jointly optimized to maximize the sum throughput for secondary users while managing the minimum throughput requirement of the secondary network. To ensure efficiency in primary communications, we impose an interference temperature constraint on downlink and uplink to keep the interference level no greater than a predefined threshold. We exploit SDP and SCA techniques to find an efficient solution with fast convergence. Simulation results found that IRSs can significantly enhance sum throughput for secondary users.