Integration of Cognitive Radio Technology in NOMA-Based B5G Networks: State of the Art, Challenges, and Enabling Technologies

The integration of cognitive radio (CR) technology and non-orthogonal multiple access (NOMA) techniques, referred to as CR-based NOMA systems, has been recently configured as a promising solution to meet the requirements of beyond fifth-generation (B5G) networks, especially those related to Internet-of-Things (IoT) applications. With such integration, power domain NOMA allows multiple users to share the same orthogonal resource blocks. At the same time, CR technology enables opportunistic bandwidth utilization by permitting secondary users (SUs) to access the licensed spectrum frequency without interrupting the primary users’ (PUs) activities. To support the massive connectivity requirements of IoT-based networks, several multiple-access techniques have been recently combined with CR-based NOMA systems, including orthogonal multiple access (OMA) and multiple-antenna techniques. For example, in CR-based OMA-NOMA systems, the licensed frequency band is split into several channels, and a set of SUs is served on each channel using the NOMA technique. This paper provides an overview and analysis of the state-of-the-art CR-based NOMA network architecture. It summarizes the main design challenges related to the practical implementation of such systems. Furthermore, this paper presents the advances of combining CR-based NOMA with recent multiple-access techniques. The potential capabilities and the design challenges of such integrated systems are also investigated and discussed. On the other hand, this paper demonstrates the potential capabilities of deploying recent technologies in CR-based NOMA networks. The technologies include intelligent reflecting surfaces, terahertz communications, machine learning, unmanned aerial vehicles, and hybrid NOMA systems. Finally, future research directions and open issues are provided and discussed.

are as follows: propose effective multiple access (MA) approaches, investigate various enabling technologies, and propose cost-effective resource allocation strategies [3].
While previous communication generations were primarily developed using multiple-antenna and orthogonal MA techniques (i.e., time-domain MA and frequency-domain MA), power-domain non-orthogonal multiple access (NOMA) has attracted significant attention in the recent advances of 5G, and B5G communication systems [4], [5]. In particular, contrary to traditional orthogonal multiple access (OMA) systems, NOMA enables several users to utilize the same orthogonal resource blocks (RBs), resulting in significant interconnections. In fact, this can be achieved by employing superposition coding (SC) at the transmitter side. With SC, the weak users are assigned larger power compared to those with better channel conditions. This power allocation strategy maintains fairness between users and guarantees the successful implementation of the successive interference cancellation (SIC) technique [6]. SIC decodes and subtracts information meant for poorer users from messages intended for stronger users [7]. To further improve its potential capabilities, NOMA has been integrated with other existing MA techniques, such as hybrid NOMA-OMA systems and multiantenna NOMA systems [8], [9]. Such combined systems offer additional freedom, as several domains can be jointly utilized to serve the users. Furthermore, these hybrid systems facilitate implementing NOMA in dense networks and have a desirable impact on meeting the latency requirements [10].
On the other hand, Cognitive radio (CR) technology has been highlighted as a possible spectrum-management technique that successfully tackles the bandwidth dilemma in legacy licensed wireless communication networks by offering opportunistic, on-demand connection [11]. To be specific, in CR-based networks, there are two types of users, namely the primary (licensed) users (PUs) and the unlicensed users, namely the secondary users (SUs). While PUs have the priority to access the spectrum, the SU users aim to enable opportunistic use of the licensed spectrum [12]. As a result, CR technology is projected to take an active role in constructing massive IoT networks while also meeting the super fast demands of future wireless networks [13]. It is worth mentioning that CR is largely based on Software Defined Radios (SDRs), which include more generic hardware that can be configured through software. CR is not only configurable but also aware of and adaptable to its operating radio environment to maximize spectrum utilization while protecting PU's performance [14].
In fact, combining CR and NOMA, which is referred to as ''CR-based NOMA,'' has recently attracted a lot of attention in the academic community due to its potential benefits over other techniques. This is due to the fact that such CR-based NOMA systems align with the requirements of IoT-based networks. To be specific, CR enables the opportunistic utilization of bandwidth, whereas NOMA permits a larger number of IoT devices to share the same RB [15], [16]. As a result, numerous design challenges and allocating resources approaches for CR-based NOMA systems have been investigated, such as in [17] and [18].

II. RELATED SURVEYS AND OUR CONTRIBUTION A. RELATED SURVEYS
Recently, several surveys have been published to address several design aspects of stand-alone NOMA and its potential capabilities on B5G networks. For instance, the authors of [19] outlined the 3GPP views on NOMA, and a set of strategies to lessen the difficulty of using NOMA in dense networks was also offered. In addition, the interplay between NOMA and recent emerging technologies has been surveyed in [20]. Particularly, the combination of NOMA with MIMO, visible light communication (VLC), and energy harvesting (EH) techniques are investigated. In addition, a systematic review of NOMA has been demonstrated in [21]. In specific, the authors have presented the basic concepts, advantages, challenges, and open research issues of NOMA. On the other hand, the application of NOMA in the sixth generation (6G) has been discussed in [22], where the application of machine learning (ML) in NOMA-based systems has also been demonstrated. Furthermore, the applications of NOMA-based machine-type communication (NOMA-MTC) in ultra-dense networks (UDN) have been surveyed in [23].
On the other side, numerous latest surveys have extensively reviewed the applicability and limitations of CR-based systems. For example, the deployment of CR in IoT systems has been demonstrated and surveyed in [24]. In particular, the authors have discussed the spectrum sensing approaches and compared them. In addition, this survey has provided the potential capabilities of combining CR with recent emerging technologies. Furthermore, the authors in [25] have provided a comprehensive survey about the recent MA techniques that can support CR-based networks. Specifically, the authors have compared the different available MA techniques. A detailed overview of the implementation of CR in 5G networks can be found in [26]. Additionally, a brief overview of the CR-based NOMA systems has been presented as a subsection in [5]. Table 1 summarizes the main surveys that discuss NOMA, CR, and CR-NOMA.

B. MOTIVATION AND CONTRIBUTION
While several surveys have been conducted for the stand-alone NOMA or CR, few surveys have demonstrated in depth the potential capabilities of the CR-based NOMA systems. However, it has been widely agreed that such integration will play a dominant role when considering the unprecedented requirements of B5G networks. As a result, we feel inclined to give a detailed survey on the integration of CR-based NOMA technology. Unlike the previous surveys, we provide a detailed discussion about the potential capabilities of such CR NOMA-based systems. We also provide an overview of combining emerging technologies with CR NOMA-based systems. We summarize the main contribution of this survey as follows: • This paper presents an overview of the architecture of NOMA, CR, and CR-based NOMA networks.
• In addition, the challenges of employing CR-based NOMA networks are also provided and discussed, i.e., imperfect channel estimation, imperfect SIC, and resource allocation techniques.
• The integration of CR-based NOMA networks with emerging technologies is also demonstrated. The technologies include OMA, intelligent reflecting surfaces (IRS), simultaneous wireless power and information transfer (SWIPT), and unmanned aerial vehicles (UAVs).
• Finally, the paper discusses some of the open research issues associated with CR-based NOMA systems. The rest of this survey is structured as follows. Section III outlines the architecture of NOMA, CR, and CR-based NOMA networks and their operating environments. In Section IV, we point out the main design challenges related to the large-scale deployment of these networks. In Section V, a set of key enabling technologies that can enhance CR-based NOMA networks are presented and discussed. Finally, Section VI summarizes the paper and provides future research directions and open issues on designing efficient CR-based NOMA networks. For the readers' convenience, the structure of this paper is illustrated in Fig. 1. Furthermore, Table 2 shows the common abbreviations and notions that appear throughout this paper.

III. NETWORK ARCHITECTURE
This section presents the system architectures of the three different networks: NOMA, CR, and CR-based NOMA networks.

A. NOMA NETWORKS
In downlink NOMA transmission, the base station (BS) encodes the messages intended for different users with diverse power levels [6]. This process is referred to as power domain SC [7]. Specifically, the users with weaker channel conditions are allocated higher power levels compared to the stronger users. Users with superior link quality can perform SIC at the receiver by decoding and deleting messages intended for poorer users before receiving their own message [27]. Indeed, effective SIC execution necessitates assigning greater power values to users with poor channel quality. Such power allocation offers a realistic strategy to assist cell-edge users, increasing fairness. Fig. 2 depicts the scenario of NOMA-based transmission, in which a set of two users, namely U 1 and U 2 , communicates with the BS in downlink (DL) and uplink (UL) directions. In UL NOMA, the BS decodes each user's signal by treating interference from other users as noise. Specifically, SIC is performed to detect the message intended for the weaker users, i.e., the ones with a higher signal-to-noise and interference ratio (SINR), by treating interference from other users as noise. The recovered signal is then re-modulated, and the interference produced by the users with greater channel gains is subtracted in order to detect the messages of the users with lower channel gains [28]. Unlike the OMA techniques, NOMA has the potential capability to support a larger number of users within the same orthogonal RB and, thus, can meet the massive connectivity requirement of IoT-based networks [29]. On the opposite side, because users with poorer channel quality are assigned greater power values, the NOMA approach may assist in preserving user fairness. In addition, NOMA has been classified as an ''add-on'' technique, and as such, it can be implemented over the existing communications infrastructure [30]. Given the foregoing, NOMA has been lately designed as a potential MA for IoT-based B5G networks [8], [31]. Nevertheless, various concerns and restrictions must be solved before NOMA may be included in future networks. NOMA, in particular, anticipates that stronger users should be able to do flawless SIC, which is unaffordable in dense networks, where this can only be achieved at the cost of higher latency and higher decoding complexity. Therefore, several research directions have been conducted to address this major limitation. Integration of NOMA and other existing MA approaches, in particular, has indeed been offered as a possible solution for addressing the SIC's practical constraints. NOMA, for instance, is used in conjunction with OMA and multiple antenna approaches. In hybrid OMA-NOMA systems, users are organized into groups; as a result, an orthogonal RB is allocated to every group, and power-domain multiplexing is used to service the users within every group. Several domains can be implemented to service users in such hybrid systems, introducing extra levels of flexibility. However, NOMA may be implemented to service a limited number of users within every group, overcoming the restrictions of installing SIC in dense networks. Though such integration has huge benefits to standalone NOMA, it can't be accomplished without resolving certain design concerns, such as presenting effective resource allocation approaches and optimal clustering algorithms.

B. COGNITIVE RADIO NETWORKS
CR arises as a technique to increase the overall spectrum utilization by the opportunistic utilization of the available spectrum [11]. A typical CR network environment consists of M various types of primary radio (PR) networks and one or more CR networks that coexist geographically within the same area. Furthermore, each PR network is unaware of CR behavior and requires no special capabilities to coexist with CRs [12]. There are two types of users in a CR network, namely PUs, and SUs, where PUs are those who have been granted a license to operate in a certain spectrum band. On the other hand, SUs are often not licensed and can take advantage of the entire PU licensed spectrum but must avoid interfering with the PU transmissions [32]. The cognitive capability is accomplished through perceiving the radio frequency (RF) medium and utilizing the gathered information to understand the local and temporal spectrum utilization [33]. Opportunistic users can dynamically choose the best available channels and adjust their transmission characteristics to minimize mutual interference from competing users. There are three operating paradigms of CR networks based on the employed transmission strategy, the maximum transmitted power of the SUs, and the interaction between the PUs and SUs [34]. As a result, these three major paradigms are summarized as follows: • Underlay Access: Both PU and SU can simultaneously transmit as long as the interference level at the PU remains within acceptable limits. The SUs do not have to wait for transmission until the PUs channel is sensed as idle. Instead, they can jointly communicate as long as the interference temperature experienced by a PU is below a predefined threshold [35].
• Overlay Access: The SU can access a portion of the PU spectrum at the cost of acting as a relay for the PU. The fundamental disadvantage of this paradigm is that it requires complete channel state information (CSI) at both the SUs and PUs, which results in a huge increase in complexity, making it significantly less appealing [36].
• Interweave Access: In this access method, SUs analyze the radio spectrum on a regular basis, identify PUs' occupancy in time, space, and frequency, and opportunistically transmit over the identified spectrum holes or white spaces with almost no interference to the existing PUs [37].

C. CR-BASED NOMA NETWORKS
With NOMA's ability to simultaneously serve multiple users and CR's ability to enhance spectrum utilization, it is anticipated that combining NOMA and CR networks can significantly enhance SE and the number of served users, as well as meet their quality-of-service (QoS) requirements. Specifically, a set of SUs can be concurrently handled over each idle PU channel in a CR-based NOMA network using power domain NOMA. Because of the distinct characteristics of the CR operating conditions, CR-based NOMA networking necessitates adequate communication techniques and protocols that effectively exploit NOMA benefits to improve network operation while considering the distinct characteristics of CR operating conditions. Due to dynamic PU channel availability conditions, channel longevity, and CR connection situations, the CR network environment is inconsistent and time-varying. To serve a larger set of SUs in a CR-based NOMA network, the SUs can be divided into several clusters, each operating over an idle PU channel [38]. These SUs in each cluster can be served using the NOMA technique. The CR-based NOMA networks are expected to support B5G networks with unprecedented capabilities, such as enhanced Mobile Broadband (eMBB), Ultra Reliable Low Latency Communications (URLLC), and massive Machine Type Communications (mMTC) [39]. Providing medium access to a huge number of devices looks to be a big issue for mMTC applications such as autonomy distribution systems in a gigantic power grid, administration of large industrial processes, and monitoring of critical systems. Unlike eMBB, the URLLC seeks ultra-high dependability and low latency for future scenarios such as sophisticated robotics, remote control, monitoring, and augmented and virtual reality. Many applications demand dependability to be near 100%, with a latency of less than 1 ms. In contrast, establishing strict URLLC in 5G is exceedingly difficult, especially when ultra-reliability and low latency are two opposing requirements [40]. Fig. 4 presents a set of potential applications that CR-based NOMA networks can efficiently support. On the other hand, although these CR-based NOMA networks offer an additional degree of flexibility and may serve a wide range of applications, some challenges must be addressed at the design stage, including providing appropriate resource allocation techniques, maintaining the CR requirements, developing efficient user-pairing strategies, and channel assignment protocols [41]. An example of a typical CR-based NOMA network is shown in Fig. 3.

IV. CR-BASED NOMA NETWORKS CHALLENGES
The combination of CR and NOMA networks offers great benefits for both networks. However, these benefits cannot be achieved without introducing new design challenges. This section presents the main practical challenges in designing efficient communication mechanisms for CR-based NOMA networks, which can be summarized as follows: • Imperfect Successive Interference Cancellation (SIC): In CR-based NOMA networks, the perfect implementation of the SIC at the receiving nodes requires the perfect detection of the messages intended for the weaker users at the stronger ones [7]. However, this is hard to guarantee in the realistic deployment of CR-based NOMA networks, leading to imperfect SIC implementation and residual inter-user interference at the near user. An imperfect SIC can happen for several reasons, including channel estimation error and hardware constraints [42]. Such imperfect SIC has several undesirable impacts on CR-based NOMA performance. For instance, an imperfect SIC can cause a higher outage probability and lower achievable rates [43].
• Efficient Resource-Allocation Techniques: Allocating the available resources is one of the major issues that affect the performance of CR-based NOMA networks.
In fact, such an aspect can be addressed by developing and solving appropriate optimization frameworks that satisfy the QoS requirements [44]. To be specific, handling the optimization problems of the CR-based NOMA system is a challenging task compared with stand-alone NOMA and CR systems. This is because the optimization of CR-based NOMA systems requires handling several design parameters simultaneously, such as the allocated bandwidth, the allocated power levels, grouping the users, handling the SU's QoS requirements, channel assignments, and maintaining the PU's activities. Many allocations of resource mechanisms for CR-based NOMA systems have indeed been investigated. However, it is worth mentioning that EE has been considered one of the major requirements for CR-based NOMA systems [9]. This is because the EE achieves a good balance between overall throughput and total transmit power. As a result, various EE models have already been proposed in the literature. For instance, the authors in [45] have solved an EE maximization problem of a hybrid NOMA system for heterogeneous networks (HetNets), HetNets contains a macro base station (MBS) along with small base stations (SBSs). Therefore, more investigations on the EE designs for CR NOMA-based systems should be carried out.
• Channel State Information (CSI) Issue: Another major impediment to the performance of CR-based NOMA networks in practice is the availability of CSI at the transmitting and receiving nodes. To be specific, the availability of CSI is an essential condition for perfect SIC implementation, user pairing, and power allocation.
However, the knowledge of CSI requires continuous channel monitoring, which increases the overall overhead and processing complexity, leading to intolerable delays. The majority of the previous research on NOMA networks has presumed that the transmitting nodes have perfect CSI. A few works were developed assuming that the transmitting nodes have imperfect CSI or statistical CSI [46], [47].
• Dense Networks: When CR-based NOMA is deployed in dense IoT networks, this would increase the inter-cell interference (ICI) as several users are allowed to transmit over the same time-frequency RBs. This affects the system's performance negatively. Accordingly, to enable efficient dense deployment of CR-based NOMA networks, severe ICI must be mitigated, which can be achieved through developing effective transmission power control, and appropriate interference mitigation techniques [48].
• Accurate Sensing Requirements: In a CR network, which is a subset of CR-based NOMA networks, the sensing process attempts to determine the idle channel lists. This can be done through three user identification algorithms: (i) transmitter detection, (ii) primary receiver detection; and (iii) interference temperature control. The majority of current work is centered on the first algorithm (transmitter detection), which can be done based on one of three fundamental schemes: (i) matched filter detection, (ii) energy detection, or (iii) feature detection. Energy detection is the most straightforward way to implement spectrum sensing [49].
• Interference Management and Coexistence Issue: CRbased NOMA networks continue to be extremely interference-limited. In these networks, SUs are subjected to intra-network and inter-network interference due to the employed power domain multiplexing. Furthermore, overall interference experienced at a PU should be limited to a manageable level. As such, interference management is crucial in CR-based NOMA network architecture. Inter-network interference can be reduced by utilizing traditional techniques like interference alignment and transceiver beamforming. Additionally, power distribution on CR-based NOMA networks must be meticulously managed to prevent the detrimental effect of intra-network interference [50].
The coexistence issue is seen as one of the main limiting factors in developing effective CR communications.
To be specific, in a CR network context, three types of harmful interference have a negative influence on the performance of communication systems, namely interference between PUs and SUs, i.e., PU-to-SU, SU-to-PU, and SU-to-SU. While several cooperative solutions have been established to effectively address the PU-to-SU interference problem, the SU-to-SU and SU-to-PU interference issues continue to be challenging tasks [51].
• Security Challenge: Because CR-based NOMA networks are susceptible to interference, security is an essential and demanding problem. Hence, there is a vulnerability that hackers can use to tamper with and disrupt SIC execution at NOMA-based SUs. Furthermore, for cooperative CR-based NOMA networks, an untrusted relay can be another main security threat that can eavesdrop on confidential information [52], [53]. In fact, mobile applications are used in everyday life to share sensitive communications such as bank account information, e-health statistics, key military communications, and control messages in industrial processes. As a result, secure communication is a critical feature. [54], for example, studied an effective end-to-end encrypted communication of an N-pair cellular network. More particularly, the NOMA and physical layer network coding (PLNC) (NOMA-PLNC) schemes were used to increase spectral and temporal efficiency. Furthermore, physical-layer security was considered in their suggested approach to enabling safe wireless communication. Increase the source transmit power or decrease the data rate to increase the dependability of the communications systems. This minimizes the risk of a destination outage while raising the possibility of capture via a wiretap connection by the external observer. As a result, various security considerations must be addressed throughout the implementation of a CR-based NOMA network, such as distinct transmit powers, SIC implementation affecting user privacy, and diversified user security [55]. For instance, legal proactive eavesdropping is one of the effective ways of monitoring the flow of information amongst suspected users, who may use phones for illegal operations. As a result, eavesdroppers are accepted as legitimate monitors for wireless surveillance [56].
• QoS and Design Constraints: In CR networks, the SU transmissions over PU channels should be carefully controlled to protect the performance of the PUs. To address this issue, two unique power control schemes have been adopted; binary and multi-level transmission power schemes. In the binary-level scheme, which is the most widely used power control scheme in CR networks, the SU does not transmit power when the PU channel is busy and utilizes the highest possible power when the PU channel is idle. This scheme ensures collision-free spectrum sharing between SUs and PUs, but it requires an accurate spectrum sensing protocol [57]. On the other hand, a multi-level transmission power strategy allows SUs to share the available spectrum with PUs simultaneously, which can significantly improve spectrum utilization. However, this scheme requires real-time RF sensing to account for the dynamic PU activities. Furthermore, controlling the SU-to-PU interference with this approach is a challenging problem. Therefore, this adaptive scheme is dynamically adjusted based on PU traffic activity, and interference margins [58]. As a result, when CR is integrated into NOMA networks, the designed communications, and protocol should ensure predefined QoS requirements for SUs sharing the same spectrum. Thus, ensuring a set of predefined QoS requirements in CR-based NOMA networks for all users should be further investigated [59]. Furthermore, to design efficient CR-based NOMA networks, several constraints (either from CR or NOMA networks) must be taken into account, such as: 1) SIC constraint: Because multiple SUs in a CR-based NOMA network can be fulfilled using the same NOMA channel, the stronger users should be able to utilize perfect SIC [60], in which SUs with best channel characteristics must be capable of decoding messages meant for users with poorer link quality before decoding their own messages. As a result, the power allocation levels are subject to an extra limitation. In particular, weaker users should be assigned larger power levels than users with better channel conditions. This constraint is known as the SIC constraint. 2) Minimum-rate demand constraint: To fulfill the individual application requirements of all users in the CR-based NOMA network and to preserve user fairness, a minimum rate demand for all SUs must always be ensured.

3) Probability of success constraint: This ensures
that each SU can successfully transmit its data with a probability greater than a predefined threshold, γ . 4) Power mask constraint: Any communicating SU pair in any cluster should not transmit more power than a power threshold, p mask , over the associated channel. This upper bound is set by the Federal Communications Commission (FCC) [61].
Any protocol or mechanism that needs to be designed for CR-based NOMA networks should deal with these challenges. Several solutions have been proposed, but most of them have considered only a subset of these challenges. Table 3 provides a detailed comparison of recent work on CR-based NOMA networks in terms of transmission direction, main performance metrics, such as throughput and sum rate, EE, SE, outage probability (OP), and whether the CR constraints are addressed or not. These constraints include the minimum rate requirement, success probability, and power mask.

V. ENABLING TECHNOLOGIES
In this section, we present some of the recent enabling technologies and the potential capabilities of integrating them with CR-based NOMA networks.

A. MILLIMETER WAVE (mmWave)
Compared to the sub-6 GHz frequency ranges, there are much wider bandwidths available in the TeraHertz (THz) frequency bands, 1 which range from 0.3 to 10 THz. Unfortunately, the utilization of such bands was limited in the past as semiconductor devices failed to efficiently convert electrical energy into electromagnetic energy at THz frequency bands. For instance, at these frequencies, electrons cannot travel the required distance to enable a semiconductor device to function before the polarity of the device is reversed. As a result, these transceiver-level problems have delayed THz frequency access in the real world [80]. However, the aforementioned problems at the transceiver level are rapidly alleviating because of the recent advancements in plasmonic devices and graphene-based solutions. With this, such bands are seen as a critical facilitator of producing ultra-high data rates and wide connectivity of different devices with various service requirements in B5G networks. [81]. Because of the large offered bandwidth in the mmWave spectrum, several THz (mmWave) communications protocols were designed for NOMA-based mmWave communication systems (e.g., [82]). Furthermore, the combination of device-to-device (D2D) communication, mmWave technology, and NOMA has always been one of the potential methods for supporting up to 50 billion device connections by 2025. Numerous and closely situated devices' signals, which operate in the mmWave spectrum region, are stacked in the power domain in NOMAbased D2D networks to improve overall spectral efficiency (SE) [83].

B. INTELLIGENT REFLECTING SURFACES (IRS)
IRS, also known as software-controlled metasurfaces (SCM) or reconfigurable intelligent surfaces (RIS), is a promising technology to enhance the coverage and performance of wireless networks by controlling the propagation of the RF signals. The IRS is made up of several passive elements that can independently reflect signals using a customizable phase shift to create three-dimensional (3D) passive beamforming collaboration without the need for any costly RF links or complex signal processing [84]. It is typically used as a multi-antenna relay. However, it is fundamentally different from a traditional relay. The IRS, in particular, acts as a reconfigurable scatterer, requiring no specific energy supply for decoding, channel estimation, and transmission [85]. As a result, it is considered an efficient approach in terms of both spectrum and energy with low hardware costs, which is critical for realizing sustainable B5G development with scalable costs. Like the IRS, backscatters are also deployed to enhance the performance of wireless communication systems by controlling the reflection of the incident signals [86]. However, unlike the IRS, a backscatter modulates its information before reflecting it. In fact, backscattering NOMA-based systems have been widely considered, such as the work in [87]. Accordingly, further elaboration on the deployment of backscatters-assisted and IRS-assisted CR-based NOMA systems should be carried out.

C. UNMANNED AERIAL VEHICLES (UAVs)
UAVs can help traditional communication networks by functioning as flying base stations (UAV-BSs) and managing traffic demand on an on-demand basis, such as during sporting events, concerts, disaster situations, search-andrescue situations, and traffic jams, among others, [88]. There are currently over 500, 000 commercial UAVs and over 1.5 million recreational UAVs registered with the Federal Aviation Administration (FAA) in the United States. This number is predicted to increase substantially by 2024 [89]. These factors help the widespread integration of UAVs into CR-based NOMA networks due to the UAV's unique advantages in enhancing spectral efficiency and supporting massive connectivity [90].

D. SIMULTANEOUS WIRELESS INFORMATION AND POWER TRANSFER (SWIPT)
Because most IoT devices are energy-limited devices, offering sustainable energy resources has become one of the key aspects of future networks, including CR-based NOMA networks. Accordingly, simultaneous power and information transfer (SWIPT) has been considered a potential solution to provide IoT devices with their required energy [91].
In SWIPT, the RF signal can be simultaneously used for information decoding (ID) and energy harvesting (EH) [92]. Specifically, the EH models can be broadly divided into linear and nonlinear models. While the linear model is ideal and easier to be designed, the non-linear model, on the other side, is more practical but challenging in terms of design [93].
In fact, proposing energy-efficient communication systems is one of the major issues in B5G and IoT networks. Thus, energy harvesting through SWIPT is expected to play a crucial role in handling the exponential growth in power consumption. This becomes of vital importance when considering the NOMA CR-based IoT networks.

E. HYBRID NOMA
Combining NOMA with other MA approaches has recently been identified as a viable option to address the huge connection of IoT devices in B5G networks to further boost the potential capabilities of CR-based NOMA networks. Those systems are known as hybrid systems. The power domain provided by NOMA, in conjunction with standard MA approaches, can bring an extra degree of design freedom to such hybrid systems. The hybrid NOMA systems are classified as follows: a-Hybrid OMA-NOMA systems: In such systems, the orthogonal RBs, offered by OMA techniques, are utilized along with NOMA to increase the number of the connected SUs [94], as shown in Fig. 5. Specifically, unlike the conventional orthogonal frequency division multiple access (OFDMA) systems, each idle channel is dedicated to serving a group of SUs in the hybrid OFDMA-NOMA systems. This, as a result, improves the resources' utilization while facilitating the implementation of SIC [95]. b-Hybrid multiple-antenna-NOMA systems: The multipleantenna techniques, including multiple-input multipleoutput (MIMO) and multiple-input single-output (MISO) systems, employ the spatial domain to improve the SE and EE of the communication systems [96], [97], [98]. Accordingly, the combination of NOMA with the existing multiple-antenna techniques can further improve the performance of the communication systems, especially for CR-based networks. Considering cluster-based MISO-NOMA CR-based systems, an individual beamforming vector, offered by the multiple antennas at the CR BS, can be dedicated to serving a group of SUs (i.e., a cluster). To be specific, the SUs in each cluster are served using NOMA . Similarly, VOLUME 11, 2023 in massive MIMO, the CR base station is outfitted with a large number of antennas that can serve several SUs by employing NOMA. This considerably enhances the SUs' performance while also efficiently managing CR network interference [99]. This is because the CR-BS broadcasts at low power, which does not impact the PUs' performance. Furthermore, the EE may be increased further by creating energy-efficient precoding matrices for the various SUs [100]. While the MIMO-NOMA systems have the potential capabilities to provide a considerable diversity gain over the conventional OMA-NOMA systems, combining the sparse code multiple access (SCMA) can introduce an additional degree of freedom [101] and improve diversity and multiplexing gains. Additionally, such systems can overcome the high detection complexity of the stand-alone MIMO-NOMA systems. It is worth mentioning here that, because of the restricted spatial multiplexing capabilities, even multi-antenna approaches cannot interconnect many devices to support B5G networks. As a result, iterative linear receivers can be used to overcome this issue [102].

F. MACHINE LEARNING (ML)
Machine learning, i.e., ML, has been recently configured as an emerging trend that is expected to play a dominant role in the shifting paradigm towards 6G [103].To be specific, ML algorithms, i.e., deep neural network (DNN), and reinforcement learning (RL) algorithms, have the potential to handle several communication difficulties, such as network traffic control, mobile edge computing, and resource allocation problems [104]. Specifically, with ML algorithms, a network's components should be able to learn, predict, and make decisions without pre-defined rules [105]. This, as a result, improves the latency while reducing the processing overhead in wireless communication systems. Several recent NOMA-based systems have employed ML algorithms to handle a set of design challenges. For example, the authors in [106] have deployed the deep Q algorithm to evaluate the beamforming vectors of a massive MIMO-NOMA system. With such a deep RL approach, the collected empirical data is used to discover the best policy. Accordingly, RL and DNN algorithms are anticipated to play a vital role in solving several CR NOMA-based related problems, such as resource allocation problems, clustering algorithms, channel sensing, and channel estimation.

G. SHORT PACKET COMMUNICATIONS IN NOMA VLC SYSTEMS AND UNDERWATER VLC SYSTEMS
Visible light communications (VLC) have recently been regarded as a promising modern technology for augmenting and (or) offloading RF communications systems in a wide range of indoor user-dense scenarios, including households, workplace halls, convention facilities, aircraft, and public transport, along with some exterior and vehicle-to-everything (V2X) applications in 5G and 6G networks [107]. Short packet communication (SPC), like VLC, has recently been offered as a potential alternative in 5G and beyond to improve network performance in terms of latency and jitter. SPC, in particular, employs finite blocklength codes to enable low-latency communications [108].

H. MOBILE EDGE COMPUTING (MEC)
Mobile edge computing (MEC) can also be investigated as a potential solution to overcome the low-computation ability of the size constraints and non-rechargeable IoT devices [109]. Such a MEC approach acts as a computing server for the CR-NOMA-based IoT systems. However, further research should be carried out to handle several design challenges for applying MEC in CR NOMA-based systems, such as the trade-off between communication and computation and the efficient resource allocation techniques.

I. COOPERATIVE CR-BASED NOMA SYSTEMS
The utilization of cooperative transmissions is also another effective approach that can improve CR-based NOMA networks. NOMA SUs can collaborate with PUs in this approach to obtain the spectrum access opportunity. This is performed by stacking PU messages on SU messages [110]. Nonetheless, selecting the right NOMA SUs to communicate with the PUs effectively is crucial to safeguard the QoS of the PUs and increase the SE of CR-based NOMA networks. As a result, understanding how and where to develop optimum cooperative NOMA SU scheduling systems is crucial. In the conventional full-duplex (FD) cooperative NOMA systems, the near user decodes the messages of the far users and thus acts as a relay to improve the QoS of the far users [111]. Accordingly, this FD cooperative communication protocol can potentially improve CR NOMA-based systems' performance, especially considering the EH circuits' nonlinearity.

J. SATELLITE COMMUNICATION (SatCom) CR-BASED NOMA SYSTEMS
NOMA CR-based system can also be deployed in Satellite communication (SatCom) systems, where SatCom has been a key component of B5G networks, will serve as an excellent alternative to the existing aerial communications because of their unique characteristics such as high available bandwidth, wide coverage, high reliability, and disregarding geographical barriers. However, the fundamental issue with SatCom is the extremely long distance, which results in poor SIC implementation [112].

A. OPEN RESEARCH ISSUES
Even though several enabling technologies have been discussed in this paper, these technologies need extensive investigation before they can be integrated into CR-based NOMA networks. We summarize open research issues when considering several CR-based NOMA-based systems. THz (or mmWave) circuit design issues must be resolved first in THz communications and the THz device technology barrier. THz signal transmission results in narrow beams, high path losses, low diffraction, high scattering, high sensitivity to obstacles, and significant differences in path gains across line-of-sight and non-line-of-sight conditions. Furthermore, to get the most benefits from utilizing IRS in CR-based NOMA networks, the resource allocation techniques should consider the IRS phase shift matrix and other resources when formulating the resource allocation problems. However, solving joint optimization frameworks is more challenging than conventional CR-based NOMA resource allocation problems. For hybrid OMA-NOMA, if all SUs utilize NOMA to utilize the same spectrum simultaneously, the implementation complexity of the SU receivers is quite high owing to multi-user detection equipment. To fully exploit the benefits of OMA-NOMA schemes, user clustering mechanisms that pair the SUs into numerous clusters must be further investigated. In particular, further investigation of the deployment of SWIPT in heterogeneous NOMA CR-based networks is required. This is because interference management becomes a more challenging aspect [113]. In addition, there are two main concerns with utilizing EH in CR-based NOMA networks: energy harvesting efficiency (EHE) and security. The EH circuit determines the EHE, and the resource allocation strategy developed using the EH model. Security concerns, on the other hand, are exceedingly difficult. This is because malevolent energy-harvesting SUs may act as licensed NOMA SUs, taking advantage of the services offered by the CR base station. This is why it is necessary to investigate how to use physical-layer security measures to improve the security of CR-based NOMA networks. For ML, there is a dilemma of where to apply the ML algorithms so the available resources can be optimally allocated among the users. Two potential solutions are proposed in the literature, i.e., on the edge users or the cloud itself. If we push the processing to the cloud, even though the cloud has the capability to perform such huge computations, this will introduce an unwanted delay. If we perform the computations at the edge users or nodes, this will be beyond their capabilities, as the IoT devices are usually power and resource-limited. Hence, this problem needs much more investigation.
Massive MIMO transmission, on the other hand, can be an effective strategy for improving the functionality of a CR-based NOMA network. However, two critical design difficulties must always be solved. First, because of the nonorthogonal resources, the estimation of the channels that appear among the CR BS and one NOMA SU is substantially contaminated by the channel between that CR BS and another NOMA SU. This is termed as ''pilot contamination''. As a result, developing blind estimation approaches to prevent this issue in CR-based NOMA networks is crucial. Second, a substantial portion of CSI is required to construct suitable precoding methods that maximize the achievable rate of the NOMA CR network or the EE of the secondary networks. However, because of the restricted resources and the presence of quantization errors, obtaining a perfect CSI is a challenging problem in realistic scenarios. Another unresolved issue is building strong precoding techniques to deal with imprecise multi-user detection. In addition, the integration between UAV and RIS, referred to as ''UAV-assisted RIS,'' is expected to play a crucial role in the practical deployment of IoT CR-based NOMA systems. With this integration, UAVs can be used as mobile base stations, where the IRS unit can be deployed to improve the quality of communication systems [114]. Moreover, due to the UAV's ability to move, IRS units can be implemented over the UAV, generating LOS communication with the intended users. However, several design challenges should be addressed when considering the UAV-assisted RIS with a CR-based NOMA system, including channel modeling, phase shift of IRS, resource allocation techniques, and CR constraints.

VII. CONCLUSION
CR-based NOMA networks can significantly improve spectrum utilization and meet the massive demand for IoT and B5G networks. We begin this paper by highlighting the primary architecture of NOMA, CR, and CR-based NOMA networks. Then, we discussed the characteristics of their operational environment that must be considered when designing effective communication protocols for these networks. We then demonstrated some design challenges related to these networks. Finally, we highlighted future research directions and enabling technologies that can be integrated with the CR-based NOMA systems.