QoSRP: A Cross-Layer QoS Channel-Aware Routing Protocol for the Internet of Underwater Acoustic Sensor Networks

Abstract

Quality of service (QoS)-aware data gathering in static-channel based underwater wireless sensor networks (UWSNs) is severely limited due to location and time-dependent acoustic channel communication characteristics. This paper proposes a novel cross-layer QoS-aware multichannel routing protocol called QoSRP for the internet of UWSNs-based time-critical marine monitoring applications. The proposed QoSRP scheme considers the unique characteristics of the acoustic communication in highly dynamic network topology during gathering and relaying events data towards the sink. The proposed QoSRP scheme during the time-critical events data-gathering process employs three basic mechanisms, namely underwater channel detection (UWCD), underwater channel assignment (UWCA) and underwater packets forwarding (UWPF). The UWCD mechanism finds the vacant channels with a high probability of detection and low probability of missed detection and false alarms. The UWCA scheme assigns high data rates channels to acoustic sensor nodes (ASNs) with longer idle probability in a robust manner. Lastly, the UWPF mechanism during conveying information avoids congestion, data path loops and balances the data traffic load in UWSNs. The QoSRP scheme is validated through extensive simulations conducted by NS2 and AquaSim 2.0 in underwater environments (UWEs). The simulation results reveal that the QoSRP protocol performs better compared to existing routing schemes in UWSNs.

1. Introduction

The ocean covers more than 70% of the Earth’s surface and it is vital for human life. It helps in driving weather and regulating Earth’s temperature, provides primary resources for humans and serves as a medium for commerce and transport. However, more than 95% of the volume of the ocean remains unexplored and, even more alarming, is unseen by human eyes due to lack of appropriate acoustic communication technologies for data collection from the ocean. However, the internet of things (IoT) technology is expected to make possible a new era in the vast ocean exploration with the aims to connect ubiquitous devices and facilities with different types of underwater networks to provide efficient and reliable services for all kinds of applications anytime and anywhere [1]. Thus, the use of Internet of Underwater Technology (IoUT) with underwater wireless sensor networks (UWSNs) technology can facilitate the discovery of unexplored marine resources. In fact, communication under water is more challenging due to the harsh nature of underwater environments (UWEs) [2]. Currently, most current underwater networks rely upon acoustic telemetry to communicate, but current acoustic modem technology is very limited by the low signal propagation speed in water (around $1.5\times {10}^3$m/s as opposed to radio propagation of around $3\times {10}^8$ m/s) and no commercial system can exceed a range $\times$ rate product of more than 40 km × 1 kbps [3]. However, the low bandwidth issue is not the only problem that is faced here but also dealing with many other problems such as high noise, path loss, multi-path signal propagation, Doppler spreading and high power consumption [4]. Putting this all together, we can see that acoustic networks have a large propagation latency, low available data transmission capacity, high bit error rate (BER) and dynamic topology structure in UWEs.

There are a few alternatives to acoustic telemetry, such as radio waves and optical communication that can be used [5]. The radio waves have been shown to be unviable for use in underwater networks as radio waves suffer from large signal attenuation and absorption unless extra low frequency (3–300 Hz) signals are used with long antenna size [6]. However, this comes at the expense of extremely high transmission power and low data rates depending upon the physical properties of the medium characteristics on the other hand, the optical modems can be used in conjunction with acoustic telemetry to achieve higher data rates associated and a lower error rate than acoustic telemetry. However, the optical signal can only propagate short distances (e.g., 2−8 m with special lenses) [7]. This solution becomes useful when the acoustic sensor nodes (ASNs) are close to each other and there are no obstructions in the water, such as oil, fish, rocks, etc. The short range means that optical signals are not useful due to heavy scattering and are restricted to short-range-line-of-sight issues in acoustic networks for time-critical events monitoring purposes. Thus, these approaches are expensive, not scalable and intolerant to faults.

To this end, underwater acoustic communication seems to be the best option and, therefore, underwater acoustic sensor networks (UASNs) have been attracting increasing attention from scientific and industrial communities [8]. The ASNs in UASNs have sensing, processing and communication capabilities, thus they are emerging as a fundamental technology for IoUT. The use of underwater sensor nodes enabled with wireless communication capabilities have the potentials to realize real-time underwater monitoring and actuation applications, with an on-line system reconfiguration and failure detection capabilities [9]. Therefore, this technology is making possible a new era in variety of ocean monitoring and exploration applications, such as monitoring natural disasters, mine recognition, navigation assistance, underwater pollution, study of marine life and tactic surveillance applications [10]. These applications will help in filling the gap of our knowledge regarding the ocean and aquatic environments in general. Nevertheless, deployment of UASNs due to their acoustic modem transceiver and the appropriate hardware for protecting the circuitry is expensive and limited to experimental settings. The ASNs deployment missions can take several days since sensors might be attached to docks, anchored buoys, sea floors, underwater autonomous vehicles (UAVs), low-power gliders, or underpowered drifters, depending on the desired network architecture [5].

On the other hand, efficient data collection using ASNs is challenging due to the various aforementioned factors in the UWSNs. Often, underwater ASNs are equipped with acoustic modems to wirelessly communicate with each other. Underwater communication links are affected due to location and time-dependent acoustic channel communication characteristics. Moreover, temporary connectivity loss can occur due to shadow zones. Therefore, the wireless link between neighboring ASNs may perform poorly or even be down at any given moment, which will increase the packet retransmissions as an attempt to deliver data packets result in more packet collisions, delay and energy consumption in UWSNs. Thus, excessive rerouting due to poor link quality consumes nodes residual energy, which can disrupt underwater monitoring and exploration missions prematurely [11]. To address these issues, a number of data collection protocols have been proposed in the literature (see Section 2). The key aim of these routing schemes is to mitigate the impacts of the underwater environments and enhance the poor quality of underwater acoustic physical links by taking advantage of the multicast and broadcast nature of the wireless transmission medium. However, each approach has disadvantages that, when combined, could diminish significantly the performance of UWSNs. Some of their common drawbacks, e.g., poor data reliability and channel capacity, communication void region problem, high latency, and redundant data packet transmission are accented in acoustic communication, which severely diminishes UWSNs performance. Therefore, a highly reliable routing architecture is essential for UWSNs.

In the rest of the paper, Section 2 discusses the existing routing protocols in UWSNs. In Section 3, we briefly describe the proposed quality of service routing protocol (QoSRP). In Section 4, we discuss the acoustic channel and energy consumption models used in the simulation studies. In Section 4, we also provide simulation results and analyze the performance of QoSRP against existing routing protocols in UWEs. Finally, in Section 5, we conclude the paper with future work.

2. Existing Studies and Challenges in Underwater Wireless Sensor Networks (UWSNs)

In the past few years, there has been intensive research in designing routing schemes for UWSNs. In this respect, this section summarizes up to date existing studies on the developments and status of routing schemes in UWSNs. The authors in [12] propose a depth based routing protocol to minimize the end-to-end latency and energy consumption in UWSNs. The designed scheme considers an optimal weight function to compute the speed of receiving signals and transmission loss during relaying data over shortest paths in low-depth regions. The proposed scheme minimizes the latency and energy consumption at the expense of low network throughput and packet delivery ratio in UWSNs. Similarly, a depth-based anycast geographic and opportunistic routing protocol is proposed in [13] for UWSNs. The designed scheme employs a recovery mode procedure based on the ASNs depth adjustment during relaying data packets from ASNs in void regions to the sea surface sink. The research in [14] balances the energy consumption load and avoids the void regions because of considering the normalized depth variance of the relay ASNs in the network. Also, in [15] an idea of pressure routing is discussed to provide reliable packets transmission of the events to any surface sonobuoy. The proposed scheme selects the subset of relay ASNs based on the pressure levels to route data at low interference towards the surface sonobuoy. The work in [2] proposes a location-free pressure routing protocol that considers three basic parameters such as depth, residual energy and link quality of the relay ASNs. Both above routing schemes achieve low latency, energy consumption and high packets delivery with the expense of poor synchronization and corrupted data packets due to detouring forwarding in void regions in UWSNs.

To solve the issues of above-mentioned depth-based routing schemes, the work in [16] presents a novel concept of vector-based opportunistic routing in which the smallest hop counts are selected towards the sink during conveying data packets. The developed scheme due to its highly stable routing architecture performs well in achieving low latency, energy consumption, and high packet delivery rates in UWSNs. The work in [17] discusses a packets forwarding mechanism for marine monitoring applications. The proposed scheme employs various transmitting power levels and the residual energy of down-stream relay ASNs in the information relaying process. The developed protocol prolonged the network lifetime due to its low energy consumption, however, it faces the issues of high latency and low data delivery rates in UWSNs. Similarly, the study in [18] presents a tree-based routing protocol for reliable data gathering in UWSNs. The entire working mechanism is divided into two phases, namely, dynamic tree constructions and information gathering using an autonomous underwater vehicle. Initially, a set of gateway nodes along the shortest path trees is defined to limit the association count of neighboring ASNs. Then, an autonomous underwater vehicle is used to collect data from gateways to prevent data loss in the network. The proposed protocol achieves low latency performance and increases the packets delivery rates and throughput, but it faces the issues of communication overheads in UWSNs.

The protocol proposed in [19] exploits the link quality of relay ASNs by considering the cross-layer design paradigm during forwarding packets around connectivity voids and shadow zones in UWSNs. Similarly, the research in [20] discusses a cross-layer location-free single-copy protocol for reliable data gathering in UWSNs. The developed scheme due to its dynamic controls transmit power and channel frequency mechanisms avoids overhead introduced by redundant packets, balance energy consumption and increases the packet delivery rate in UWSNs. Also, the study in [21] presents a cross-layer cooperative routing protocol in which the ASNs selects the forwarding relay ASN based on their link quality indicators towards the destination. These schemes perform well in terms of low energy consumption, latency and high packet delivery ratio in UWSNs. However, the first scheme faces the issue of high congestion near to the sink than others while the main problem of the second scheme is invalid data packets due to path loops in the network. On the other hand, the third routing protocol faces excessive overhead issues due to periodically updating the neighboring ASNs in UWSNs. To solve some of the issues faces by above schemes, the authors in [22] discuss a novel deep Q-network based idea to make globally optimal routing decisions. The designed scheme to reduce network overhead employs a hybrid unicast and broadcast communication mechanisms to maintain network topology changes and making new routing decisions when the current data paths become unavailable.

The protocol in [23] divides entire grid routing architecture into many small cubes indicated as clusters to maintain the reliability of data transmission. In each cluster, a cluster leader is appointed based on the location and remaining energy information to maintain the reliability of data transmission. Similarly, the study in [24] provides a bio-inspired distributed spectral clustering routing protocol that exploits link quality by considering signal to noise ratio (SNR) values between relay ASNs to provide highly stable clustering and routing architecture in the network. The research in [10,25,26] propose bio-inspired dynamic cluster-based routing protocols to provide reliable data transmission by considering the number of hop counts and the confidence level of the relay ASNs in UWSNs. Also, in [27], a hybrid data-collection routing protocol is proposed in which the entire working mechanism is divided into the upper layer and lower layer to perform reliable data gathering in UWSNs. The simulation facts show that the proposed schemes achieve low latency, energy consumption and higher packet delivery rates in the network. However, the first scheme faces high routing table management cost, and the second scheme compared to other schemes faces the corrupted data packets due to poor link quality among cluster heads and cluster heads rotating issues in highly dense UWSNs. The study in [28] presents a disjoint multipath disruption-tolerant packets forwarding mechanism in which multiple routing paths by employing the hue, saturation and value colour space that are constructed to greedily convey packets to the sea surface sink. The developed protocol improves the packets delivery rates and reduces the latency, but it faces the issue of low throughput and poor loading balancing in UWSNs.

Han et al. [29] propose an asymmetric link-based reverse routing in which a directional beam width mechanism is used to analyze the link quality between relay ASNs. The proposed protocol eliminates void issues and improves network performances in terms of latency and packets delivery rates. However, this faces the issue of high communication overheads and packets collision due to periodically updating the status of the links in UWSNs. Similarly, the protocol proposed in [30] uses multi-layered architecture to discover a set of feasible relay ASNs during conveying data towards the sink. The proposed protocol achieves better performance in low energy consumption and packet delivery rates with low latency, but it faces the issues of poor synchronization and invalid data packets in UWSNs. To address some of the issues faced by the above schemes, the protocol in [31] overcomes packets loss by considering the forwarder ASNs connectivity issues along a route towards the sink. The performance of the proposed scheme is observed better in low latency, energy consumption and packets error rates in UWSNs. Table 1 shows the comparison between different routing schemes in UWSNs.

The main purpose of the aforementioned routing protocols is to provide reliable data delivery at a low cost in UWSNs. These static-channel based schemes significantly help in designing and development of advanced routing solutions for UWSNs. However, they suffer from major disadvantages. First, most of the schemes employ probabilistic values to estimate the link quality between ASNs in highly dynamic UWEs. Thus, the poor link quality due to the false estimation of the channel is facing severe reliability issues lead to excessive rerouting during monitoring time-critical events in UWSNs. The new routes finding or repairing broken links require a significant amount of control message overheads, which consumes the sensor’s energy and increases the chance of packets collision and brings latency issues in the network. On the other hand, the interference issues further increase the corrupted data packets in UWSNs. Second, during conveying packets these schemes consider shortest paths routing with excessive hop counts, which may balance the network residual energy; however, it consumes more ASNs energy and increases the probability of invalid data packets because of data path loops in UWSNs. In addition, the shortest path routing result in high routing table management cost and congestion problems because of quickly draining the residual energy frequently used by ASNs nearer to the sink. Third, most of these routing protocols are stuck in void regions and instead of finding an alternative route just discards the packets and thus subjected to packet loss in UWSNs. Fourth, the existing static-channel based routing solutions due to lack of dynamic channel adaptation cannot mitigate the inference effects in order to provide high packet delivery rates and network throughput with low latency and corrupted packets in UWSNs.

These facts motivate researchers to propose a novel cross-layer QoS-aware multichannel routing protocol called QoSRP for the internet of UWSNs to mitigate the effects of the underwater environments and improve the overall data collection in UWSNs. The entire routing problem has been modelled using mixed integer linear programming (MILP) in UWSNs. The major contributions of our proposed protocols are as follows:

• We propose an underwater acoustic channel detection mechanism to find the vacant channels with a high probability of detection and low probability of missed detection and false alarms.

• An underwater acoustic channel assignment mechanism is proposed to assign high data rates channels to acoustic sensor nodes with longer idle probability in a robust manner.

• A hybrid underwater routing mechanism is proposed to convey collected data to the sink. The proposed mechanism while conveying information avoids congestion, data path loops and balances the energy consumption load of UWSNs.

• The performance of the QoSRP protocol against existing routing schemes is validated through extensive simulations conducted by NS2 and AquaSim 2.0 in the UWSNs.

The later sections explain the working mechanism of our proposed routing protocol in UWSNs.

3. Proposed Quality of Service Routing Protocol (QoSRP) in UWSNs

The protocol design details are given in the following sections.

3.1. Network Model

The network model used in the design, development, simulating and testing the QoSRP scheme is illustrated in Figure 1. The proposed architecture consists of a sea surface sink, base station (BS) and ASNs. The randomly deployed ASNs in a geographic area of interest over the ocean bottom are equipped with the main functions of sensing, sampling, and acoustic transmitters. These location-aware ASNs (computed by using self-localization scheme in [32]) deployed for continuous oceanographic data collection are equipped with omnidirectional acoustic transceivers have identical communication range and asymmetric communication links. In UWEs, the ASNs due to constraints of limited residual energy and short communication range transmit packets to the sink in a multiple hops manner. Thus, the ASNs die once the energy runs out. In addition, the ASNs employing different channels on multiple paths with different lengths have different propagation period, the angles of arrival (AoA) and the angle of departure (AoD) in UWEs. In addition, the ASNs move passively with water currents in vertical and horizontal directions with a velocity of 0 to 0.8 m/min and 0 to 1 m/min, respectively. This extremely slow movement in both upward and parallel directions is assumed negligible in the UWEs. On the other hand, the sink floats on the sea surface and is equipped with both acoustic modem and radio modem aiming to gather data from ASNs through acoustic signals and forward the gathered data using radio signals to the offshore BS and then the remote user/s for monitoring purposes. The sea surface sink is embedded with a global positioning system (GPS) and periodically updates the BS about its location information by disseminating periodic beaconing in the network. The energy of both sea surface sink and BS is assumed to be unlimited since they can be charged by suitable green energy resources, such as solar energy, etc. The technology such as carrier sense multiple access (CSMA) is assumed to avoid data packets collision during transmission and reception in UWSNs. Finally, we assume that the remote user/s can configure, control and monitor ASNs by connecting to the BS through one of the highly stable communication technology, such as satellite or cellular. The following section explains the working procedure of QoSRP protocol in detail.

3.2. Underwater Acoustic Channel Detection (UWCD) Algorithm

Recently, the spectrum shortage problem aggravates in an apparent manner because of increasing demands for higher data rates at low frequencies for various UWSN-based applications. The principle of dynamically exploiting the local vacant spectrum bands of the primary users during their silence periods by secondary users is among the best-proposed solutions for this problem. Therefore, the multichannel is proposed as a prominent technology to implement spectrum access for its autonomous, agility, and ability to detect the primary user’s (PU) signal. In multichannel communications, the spectrum sensing is a key element in the sense that it demonstrates how signals of the PUs are identified, sampled, and processed for detecting spectrum holes [33]. The main tasks accomplished by spectrum sensing, such as spectrum monitoring, spectrum analysis, and spectrum decision while the multichannel cycle consists of four core operations, including the above three and data transmission. Thus, the key aim of the spectrum sensing is to search spectrum holes for secondary users by detecting the primary signals in the network. To sense the accessibility of certain portions of the frequency band, the most effective way is to identify the primary users that are active within the range of a secondary user. However, the direct measurement between a transmitter and receiver for a channel is difficult and brings several new challenges for the secondary users in the UWEs. Generally, there exist few signal detection techniques like matched filter detection (MaD), wavelet detection (WeD), cyclostationary detection (CyD) and energy detection (EnD) [34], etc., that can be employed during sensing the spectrum to enhance the detection probability in the UWEs.

In the MaD technique, the $\mathcal{S}\mathcal{U}$ have the prior knowledge of the $\mathcal{P}\mathcal{U}$ signals and a matching filter is applied, which amplifies the SNR of the received acoustic frequency samples. This mechanism, due to coherent detection, achieves a high processing gain and thus reduces the channel detection latency. However, each primary user class requires a dedicated sensing receiver, which is expensive and consumes significant ASNs battery power in UWSNs. Moreover, the channel detection performance of the matched filter can be severely limited because the primary user $\left(\mathcal{P}\mathcal{U}\right)$ signals information is hardly available at the $\mathcal{P}\mathcal{U}{}^{\prime }𝓈$ in UWEs. By contrast, in the WeD approach, the given signal is first split into various frequency components and later every component by matching the resolutions to its scales is studied. The wavelet detection basically employs cosines and sines as key functions [35]. The use of irregularly shaped wavelets in the wavelet transforms as basic functions to identify the local features and sharp changes. This approach offers flexibility in dynamic channel adaptation and low implementation cost. However, the high sampling rates for characterizing the large bandwidth is one of the most critical challenges while implementing the wavelet approach. In the CyD technique, the presence of primary users can be identified even at very low signal-to-noise values by observing the periodicity of the collected $\mathcal{P}\mathcal{U}{}^{\prime }𝓈$ signals in UWEs. In the cyclostationary detection to obtain the periodicity of the primary signals, the modulated signals are usually combined with hopping sequences, spreading code, pulse trains, and cyclic prefixes or sinusoidal carriers. These modulated signals due to exhibiting the characteristics of periodic statistics and spectral correlation are characterized as cyclostationary [36].

Typically, these features in the signal format are introduced intentionally, which enables a receiver to exploit parameter estimation like the direction of arrival, pulse timing, or carrier phase. This detection technique can improve the sensing detection, however, it requires prior knowledge of the primary signal characteristics. In addition, the high computational complexity and significantly extensive sensing delays are the other shortcomings of this approach. The EnD is the widely used channel detection method, which does not require any previous knowledge of the $\mathcal{P}\mathcal{U}$ signals. This method identifies the primary user’s signals based on the sensed energy, where the received signal strength indicator or acoustic-frequency energy is computed to find whether the channel is free or not. Initially, an input signal to obtain the required bandwidth is filtered through a band pass filter and then the obtained signal is squared and combined over the observation interval. Lastly, to find the presence of a primary user signal a predetermined threshold is compared with the output of the integrator. Generally, fast Fourier transform-based techniques are used to analyze the spectral in the digital domain. In particular, in a specific time window, the received signal is sampled and passes through fast Fourier transform equipment to obtain the power spectrum. Then, the power spectrum peak after windowing is detected in the frequency domain. This technique provides better channel detection in underwater since it does not need any previous information about the $\mathcal{P}\mathcal{U}{}^{\prime }𝓈$ signals. However, the high noise, interference, fading, and multipath effects increase the probability of sensing errors in terms of false alarms $\left(\mathcal{F}\mathcal{A}𝓈\right)$ or miss-detection $\left(\mathcal{M}𝒾\mathcal{D}\right)$, which leads to poor channel detection performance in UWSNs. Hence, it is highly desired to reduce the sensing errors during channel detection to enhance the usage level of the vacant channel and to minimize the collision probability with $\mathcal{P}\mathcal{U}$ transmissions.

In this respect, the detection threshold that precisely detects the presence of primary signals plays an important role in the multichannel acoustic sensor networks (MCASNs). The less error in computing the SNR value decreases the probability of $\mathcal{F}\mathcal{A}𝓈$ in UWEs. However, there exists a limit for the noise level. The increase in noise level above this limit increases the probability of $\mathcal{F}\mathcal{A}𝓈$ while it reduces with the decrease in noise below this threshold level. In the proposed scheme, the channel detection threshold level is determined by considering the various underwater environment parameters and set to a minimum level of the transmitted signal power value in the underwater. This detection sensitivity identifies the received SNR ratio at the secondary user $\left(\mathcal{S}\mathcal{U}\right)$ with a satisfied probability of detection conditions in MCASNs. Thus, the optimal channel detection threshold significantly reduces the sensing errors, enhance spectrum utilization and provides enough protection to $\mathcal{P}\mathcal{U}$ transmissions in the MCASNs. In the designed scheme, a simple and effective energy detection mechanism appropriately detects the existence of the $\mathcal{P}\mathcal{U}$ signal based on its received energy by comparing it to a detection threshold. Herein, we employ an energy-based signal detection $\left(E\right)$ method revealed in [37] to identify idle spectrums in UWSNs. The key objective function of the EnD mechanism $\left({\varphi }_{UWCD\left(1\right)}\right)$ of the proposed algorithm is numerically indicated as:

$\forall 𝒾= \left\{1,2,\dots ,𝓃\right\};\forall 𝒿= \left\{1,2,\dots ,𝓂\right\};\forall 𝓀= \left\{1,2,\dots ,\ell \right\};$

(1) ${\varphi }_{UWCD\left(1\right)}={\displaystyle \sum }_{i=1}^{\left|n\right|}\underset{i}{\max }{\rho }_{\mathcal{D}ℯ\mathcal{T}}+\underset{i}{\min }{\rho }_{\mathcal{F}\mathcal{A}𝓈}+ \underset{i}{\min }{\rho }_{\mathcal{M}𝒾\mathcal{D}}$

The key aims of the objective function is to maximize the probability of detection $\left({\rho }_{\mathcal{D}ℯ\mathcal{T}}\right)$ and minimize the probability of $\mathcal{F}\mathcal{A}𝓈$ $\left({\rho }_{\mathcal{F}\mathcal{A}𝓈}\right)$ and missed-detection $\left({\rho }_{\mathcal{M}𝒾\mathcal{D}}\right)$ in UWSNs. Consequently, the test statics of energy detection $\left({\mathcal{T}}_{\mathsf{{\rm E}}\mathcal{D}}\right)$ of the received signal at the $\mathcal{S}{\mathcal{U}}_{𝒾}$ can be numerically written as:

(2)${\mathcal{T}}_{\mathsf{{\rm E}}\mathrm{n}\mathcal{D}}={\displaystyle \sum }_{𝓃=1}^{{\mathcal{L}}_{𝓈}}{\left|\mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}\left[𝓃\right]\right|}^2$

subject to:

(2a)${\mathcal{S}}_{\mathcal{L}}\left[𝓃\right]\le {\mathsf{\tau }}_{\varsigma } ;$

(2b) $0<{\mathcal{M}}_{sg\left(i\right)}\le Ms{g}_{sg\left(j\right)}\le 1;$

(2c) $𝓃>1\le k;$

(2d) $\mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}\left[𝓃\right]\ge {\mathsf{{\rm E}}}_{\varsigma };$

(2e) ${\rho }_{ℯ}\le {\rho }_{ℯ}\left(ϵ\right)<1;$

(2f) ${\mathsf{\tau }}_{\varsigma }>0 1\le \varsigma \le \mathrm{T}\forall 𝓃\in \mathcal{P}{\mathcal{U}}_{𝒿};$

(2g) ${\mathsf{{\rm E}}}_{\varsigma }\ge 0 1\le \varsigma \le E \forall E\left[𝓃\right]\in \mathcal{P}{\mathcal{U}}_{𝒿};$

(2h) ${\rho }_{ℯ}\ge 0 1\le e\le E;$

In Equation (2), $\mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}$ and ${\mathcal{S}}_{\mathcal{L}}\left[𝓃\right]$ are, the received signal energy and the $𝓃$-sample of the ${\mathcal{S}}_{\mathcal{L}}$ while ${\mathcal{L}}_{𝓈}$ is the sum of the length of $𝓃$ samples over an interval which helps to obtain a level of performance under certain $\mathcal{S}\mathcal{N}\mathcal{R}$ conditions in the network. The number of samples collected by individual $\mathcal{S}\mathcal{U}$ depends upon the sensing time and the longer sensing delay within in predefined time intervals leads to better detection performance of the primary signals. However, the longer sensing time results in less available time for packets transmission, which minimizes the throughput of the $\mathcal{S}\mathcal{U}𝓈$ is constrained by Equation (2a). Moreover, due to additional sensing time the cooperation overhead with the increasing number of cooperating users’ increases, which lead to a huge volume of information that need to process by the requested $\mathcal{S}\mathcal{U}$ to make a local decision is constrained by Equation (2b). Therefore, the number of samples of the received signal energy of ${\mathcal{L}}_{𝓈}\left(n\right)$ such that $𝓃$ $>1$ using the central limit theorem over an interval are considered to obtain some level of performances under certain $\mathcal{S}\mathcal{N}\mathcal{R}$ conditions is constrained by Equations (2c) and (2b) in the network. Therefore, it is indicated as one of the basic functions of the $\mathcal{S}\mathcal{N}\mathcal{R}$ in the UWEs. The transmission of $\mathcal{P}\mathcal{U}$ signals is a random process, which follows an independent identically distributed pattern with mean zero and variance ${\sigma }_{𝓈}^2$. In the underwater, the noise $\left({𝓃}_{oise}\right)$ is a real-valued Gaussian variable that considers various underwater environment parameters with zero mean and variance ${\sigma }_{𝓋}^2$ is constrained by Equation (2e) within a certain error probability. Constraints in Equations (2f, h) satisfy the limits in the network. To this end, the $\mathcal{S}\mathcal{N}\mathcal{R}$ in the underwater at the $𝒾\mathrm{th}$ secondary user $\mathcal{S}{\mathcal{U}}_{𝒾}$ is computed as $\mathcal{S}\mathcal{N}\mathcal{R}\left(\mathcal{S}{\mathcal{U}}_{𝒾}\right)={\left|{\mathcal{C}}_{ℊ}\right|}^2{\sigma }_{𝓈}^2/{\sigma }_{𝓋}^2$. The primary user signal and noise are assumed to be independent and therefore a binary hypothesis testing problem of received energy signals at the $\mathcal{S}\mathcal{U}$ can be formulated as:

(3)${\mathcal{T}}_{\mathsf{{\rm E}}\mathrm{n}\mathcal{D}}{\left(\mathcal{S}{\mathcal{U}}_{𝒾}\right)}_{{\mathcal{S}}_{\mathcal{L}}}={\displaystyle \sum }_{𝒾=1}^{𝓂}{\left|{𝓃}_{o\left(𝒾\right)}\right|}^2\in {\mathcal{H}}_0$

(4)${\mathcal{T}}_{\mathsf{{\rm E}}\mathrm{n}\mathcal{D}}{\left(\mathcal{S}{\mathcal{U}}_{𝒾}\right)}_{{\mathcal{S}}_{\mathcal{L}}}={\displaystyle \sum }_{𝒾=1}^{𝓂}{\left|{\mathcal{C}}_{ℊ\left(𝒾\right)}{\mathcal{S}}_{\mathcal{L}\left(𝒾\right)}+{𝓃}_{o\left(𝒾\right)}\right|}^2\in {\mathcal{H}}_1$

in which $𝓂$, ${\mathcal{C}}_{ℊ\left(𝒾\right)}$, ${𝓃}_{oise\left(𝒾\right)}$ are the time-bandwidth product, the channel gain between $\mathcal{P}\mathcal{U}$ and $\mathcal{S}\mathcal{U}$, and noise in the ${𝒾}^{𝓉𝒽}$ time slot, respectively. The received energy at each $\mathcal{S}{\mathcal{U}}_{𝒾}$ is compared to the predefined detection threshold $\left(\sigma \right)$ to reach a decision about the presence or absence of $\mathcal{P}{\mathcal{U}}_{𝒿}$. In locally sensing mechanism each $\mathcal{S}\mathcal{U}$ decides whether the $𝒾\mathrm{th}$ channel is available (${\tau }_{\mathrm{off}})$ or occupied (${\tau }_{\mathrm{o}\mathrm{n}}$) by using a predefined threshold in the network. Thus, the hypothesis at each secondary user is saying that the $\mathcal{P}\mathcal{U}$ is active only if the received signal energy is greater than the defined limit and idle when it is less than the defined threshold value, which can be numerically indicated as:

(5)${\mathcal{T}}_{\mathsf{{\rm E}}\mathrm{n}\mathcal{D}}{\left(\mathcal{S}{\mathcal{U}}_{𝒾}\right)}_{{\mathcal{S}}_{\mathcal{L}}}=\left[{\mathcal{H}}_0\in \mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}>\sigma |{\mathcal{H}}_1\in \mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}<\sigma \right]\in {\tau }_{\mathrm{s}}$

subject to:

(5a)${\mathcal{T}}_{\mathsf{{\rm E}}\mathrm{n}\mathcal{D}}{\left(\mathcal{S}{\mathcal{U}}_{𝒾}\right)}_{{\mathcal{S}}_{\mathcal{L}}}= {\tau }_{\mathrm{on}}|{\tau }_{\mathrm{off}}\in {\mathcal{C}}_{𝒾}\subseteq {\mathcal{C}}_{𝓃}$

(5b)${\mathcal{H}}_1\left({\mathcal{C}}_i\right)\in {\mathcal{C}}_n=1\left({\tau }_{\mathrm{on}}\right)$

(5c)${\mathcal{H}}_0\left({\mathcal{C}}_i\right)\in {\mathcal{C}}_n=0\left({\tau }_{\mathrm{off}}\right)$

(5d)$\mathcal{P}{\mathcal{U}}_{𝒿}\left({\mathcal{C}}_i\right)\in {\mathcal{C}}_n\ne {\tau }_{\mathrm{on}}$

(5e)${\mathcal{C}}_i\in {\mathcal{H}}_1|{\mathcal{H}}_0 \le 1$

(5f)$\mathcal{S}{\mathcal{U}}_{𝒾}\left({\mathcal{C}}_i\right)\in {\mathcal{C}}_n\left({\mathcal{R}}_{𝒾}\right)=\tau -{\tau }_{\mathrm{s}}$

(5g)$\mathcal{S}{\mathcal{U}}_{𝒾}\notin {\mathcal{C}}_i\left({\mathcal{R}}_{𝒾}\right)=\tau$

(5h)$\mathcal{S}{\mathcal{U}}_{𝒾}{\left({\mathcal{C}}_i\right)}_{busy}\in {\mathcal{C}}_n\left({\mathcal{R}}_{𝒾}\right)\ne {\tau }_{\mathrm{s}}$

in which $\sigma$ is the threshold value of energy is used to compare is the local spectrum sensing decision of a $\mathcal{S}{\mathcal{U}}_{𝒾}$,${\mathcal{H}}_1$ denotes the $𝒾\mathrm{th}$ channel is not available because of the $\mathcal{P}{\mathcal{U}}_{𝒾}$ activity with probability ${\rho }_{\left(\mathcal{P}{\mathcal{U}}_{𝒾}\right)}$ in the sensing time ${\tau }_{\mathrm{s}}$ is constraint by Equation (5a). Equations (5b) and (5c) constraints guarantee that the $\mathcal{P}\mathcal{U}$ is active and inactive for a particular channel ${\mathcal{C}}_i$ in a time interval in the network. Equation (5d) constraints state that a channel belongs to the $\mathcal{P}{\mathcal{U}}_{𝒿}$ is available to $\mathcal{S}{\mathcal{U}}_{𝒾}$ for the time ${\tau }_{\mathrm{s}}$, while the constaints in Equation (5e) satisfy the satement specified in Equation (5d). Constraints in Equation (5f) verify that the $\mathcal{S}{\mathcal{U}}_{𝒾}$senses a channel ${\mathcal{C}}_i$ in the region ${\mathcal{R}}_{𝒾}$ and holds the channel for a specific time once it is found free in the network. Constraints in Equation (5f) explain that the $\mathcal{S}{\mathcal{U}}_{𝒾}$ must not hold the channel for the entire time since the $\mathcal{P}{\mathcal{U}}_{𝒿}$ can reclaim the channel any time in the network. Constraints in Equation (5h) state that the $\mathcal{S}{\mathcal{U}}_{𝒾}$senses a channel ${\mathcal{C}}_i$ in a region ${\mathcal{R}}_{𝒾}$ and will not be sensed the same channel immediately once it is found busy in the network. On the other hand, ${\mathcal{H}}_0$ illustrates that the $𝒾\mathrm{th}$ spectrum is available with probability${\rho }_{𝒾𝒹ℯ𝒶\ell }$, i.e., ${\rho }_{\left(\mathcal{P}{\mathcal{U}}_{𝒾}\right)}+{\rho }_{𝒾𝒹ℯ𝒶\ell }=1$ in the sensing time ${\tau }_{\mathrm{s}}$. The performance of locally detecting spectrum by each $\mathcal{S}\mathcal{U}$ is computed by the probability of detection when a $\mathcal{P}{\mathcal{U}}_{𝒾}$ is idle and the probability of $\mathcal{F}\mathcal{A}𝓈$ when a $\mathcal{P}{\mathcal{U}}_{𝒾}$ is active and can be numerically expressed as

(6) ${\rho }_{\mathcal{D}ℯ\mathcal{T}}=\rho \left({\mathcal{H}}_1|{\mathcal{H}}_1\right)\in \rho \left(\mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}>\sigma |{\mathcal{H}}_1\right)$

(7) ${\rho }_{\mathcal{M}𝒾\mathcal{D}}=1-{\rho }_{\mathcal{D}ℯ\mathcal{T}}\left(\mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}<\sigma |{\mathcal{H}}_1\right)$

(8) ${\rho }_{\mathcal{F}\mathcal{A}𝓈}=\rho \left({\mathcal{H}}_1|{\mathcal{H}}_0\right)\in \rho \left(\mathsf{{\rm E}}{\mathcal{S}}_{\mathcal{L}}>\sigma |{\mathcal{H}}_0\right)$

Equations $\left(6\right)$, $\left(7\right)$ and $\left(8\right)$ show the $\mathcal{D}ℯ\mathcal{T}$ probability when $\mathcal{S}\mathcal{U}$ determines ${\mathcal{H}}_1$ when ${\mathcal{H}}_1$ exits, the $\mathcal{M}𝒾\mathcal{D}$ probability when $\mathcal{S}\mathcal{U}$ decides ${\mathcal{H}}_0$ but ${\mathcal{H}}_1$ exists, and the $\mathcal{F}\mathcal{A}𝓈$ probability when $\mathcal{S}\mathcal{U}$ decides ${\mathcal{H}}_1$ but ${\mathcal{H}}_0$ exists, respectively. In addition, Equations $\left(7\right)$ and $\left(8\right)$ clearly indicate that $\mathcal{S}\mathcal{U}$ provides reliable protection to $\mathcal{P}\mathcal{U}$ when the detection probability is high while the $\mathcal{S}\mathcal{U}$ loses spectrum access opportunities when the false alarm probability is high in the network. Some common numeric terms used throughout the paper are given in Table 2.

3.2.1. Decision Rules for the Fused Information in Distributed Cooperative Sensing (DCoS)

Distributed cooperative spectrum sensing consists of main two types of data fusion, namely hard decision and soft decision mechanisms in the MCASNs. The soft decision mechanism due to heavily sharing test statistics faces excessive cooperation overheads compared to the hard decision mechanism, which consumes the sensor’s energy and increases the probability of channel assignment delay in MCASNs. On the other hand, the hard decision mechanism performs best when the channel state information between the $\mathcal{P}\mathcal{U}𝓈$ and $\mathcal{S}\mathcal{U}𝓈$ varies in time and location in UWEs. The hard decision mechanism after considering the individual $\mathcal{S}\mathcal{U}𝓈$ local decisions conveniently applies the linear fusion rules such as AND, OR, and K-out-of-M to obtain a reliable cooperative decision. In the AND fusion rule, the fusion center declares ${\mathcal{H}}_1$ only if all independent $\mathcal{S}\mathcal{U}𝓈$ decide on ${\mathcal{H}}_1$ and declares ${\mathcal{H}}_1$ for the OR rule only if any of independent $\mathcal{S}\mathcal{U}𝓈$ decides on ${\mathcal{H}}_1$ in the cooperative decision. On the contrary, the fusion center in K-out-of-M rule declares the ultimate decision that there is a $\mathcal{P}\mathcal{U}{}^{\prime }𝓈$ transmission only if at least K secondary users out of M selected local detectors decide about the presence of a $\mathcal{P}\mathcal{U}{}^{\prime }𝓈$ signal. In the proposed scheme, the probability of $\mathcal{F}\mathcal{A}𝓈$, $\mathcal{M}𝒾\mathcal{D}$ and $\mathcal{D}ℯ\mathcal{T}$ using K-out-of-M rule can be written as:

(9)${\rho }_{\mathcal{D}ℯ\mathcal{T}}={\displaystyle \sum }_{\ell =𝓀=1}^{\left|𝓃\right|}{\displaystyle \sum }_{𝒿=1}^{\left|𝓂\right|}{\rho }_{\mathcal{D}ℯ\mathcal{T}}{\left(\mathcal{S}{\mathcal{U}}_{𝒿}\right)}^{\ell } {\left(1-{\rho }_{\mathcal{D}ℯ\mathcal{T}\left(\mathcal{S}{\mathcal{U}}_{𝒿}\right)}\right)}^{\left|𝓂-\ell \right|}$

(10) ${\rho }_{\mathcal{M}𝒾\mathcal{D}}=1-{\displaystyle \sum }_{𝒿=1}^{\left|𝓂\right|}\left[{\rho }_{\mathcal{D}ℯ\mathcal{T}}\left(1-{\rho }_{ℯ}\right)+\left(1-{\rho }_{ℯ\left(\mathcal{S}{\mathcal{U}}_{ 𝒿}\right)}\right)+{\rho }_{ℯ\left(\mathcal{S}{\mathcal{U}}_{ 𝒿}\right)}\right]$

(11) ${\rho }_{\mathcal{F}\mathcal{A}𝓈}=1-{\displaystyle \sum }_{𝒿=1}^{\left|𝓂\right|}\left[\left(1-{\rho }_{\mathcal{F}\mathcal{A}𝓈}\right)\left(1-{\rho }_{ℯ}\right)\mathcal{S}{\mathcal{U}}_{ 𝒿}+{\rho }_{\mathcal{F}\mathcal{A}𝓈\left(\mathcal{S}{\mathcal{U}}_{ 𝒿}\right)},{\rho }_{ℯ\left(\mathcal{S}{\mathcal{U}}_{ 𝒿}\right)}\right]$

Equations $\left(9\right)$, $\left(10\right)$ and $\left(11\right)$ illustrate the probability of $\mathcal{D}ℯ\mathcal{T}$ when all$𝓂$ secondary users sensed the existence of a $\mathcal{P}{\mathcal{U}}_{𝒾}$ such that $\mathcal{S}{\mathcal{U}}_{𝒿}>\mathcal{S}{\mathcal{U}}_i$ determine ${\mathcal{H}}_1$ when ${\mathcal{H}}_1$ exits, the $\mathcal{M}𝒾\mathcal{D}$ probability when the secondary users $\mathcal{S}{\mathcal{U}}_{𝒿}>\mathcal{S}{\mathcal{U}}_i$ decides ${\mathcal{H}}_0$ but ${\mathcal{H}}_1$ exists, and the $\mathcal{F}\mathcal{A}𝓈$ probability when the secondary users $\mathcal{S}{\mathcal{U}}_{𝒿}>\mathcal{S}{\mathcal{U}}_i$ decide ${\mathcal{H}}_0$ but ${\mathcal{H}}_1$ exists, respectively.

3.2.2. Distributed Cooperative Sensing (DCoS)

In the UWCD scheme, the local channel detection made by the individual $\mathcal{S}\mathcal{U}$ for the desired capacity-aware channel is preferred. However, the high underwater noise, fading, and interferences have destructive effects caused by imperfect reporting channels on the wireless link quality leading to poor cooperative spectrum sensing in the network. This problem becomes more severe between the sender and the receiver secondary users due to an increasing probability of errors over the reporting channels of the transmitted signal in the network. Therefore, finding a channel with high data rates become more difficult in local spectrum sensing in UWEs. To this end, the distributed cooperative sensing (DCoS) provides better results in highly dynamic UWEs. The $\mathcal{S}\mathcal{U}𝓈$ in the proposed DCoS architecture performs the channel sensing tasks on demand only if the required capacity-aware channel is not found locally by the $\mathcal{S}\mathcal{U}$. To do so, the $\mathcal{S}\mathcal{U}$ that requires the channel information sends multicast request messages to its neighboring ASNs. The request message includes the required channel information, identity, and location information in the network. Upon receiving the request message, the neighboring $\mathcal{S}\mathcal{U}𝓈$ start to sense the required channels and share their channel detection statistics by exploiting the spatial diversity in the observations. Thus, each $\mathcal{S}\mathcal{U}$ based on its local sensing observation exchanges decision on the existence or non-existence of the primary signals in the UWSNs.

The receiver $\mathcal{S}\mathcal{U}$ combines the received sensing information with its own information and after some necessary calculations makes an ultimate decision whether the $\mathcal{P}\mathcal{U}$ is active or not in the network. This decision information is disseminated to each sender $\mathcal{S}\mathcal{U}𝓈$ in the network. This combined decision mechanism significantly improves the performance of channel detection and relaxed sensitivity requirements for the $\mathcal{S}\mathcal{U}$ in the network. Thus, each receiver $\mathcal{S}\mathcal{U}$ acts as a fusion center as shown in Figure 2. In Figure 2, a $\mathcal{S}{\mathcal{U}}_{𝒾}$ that requires a specific channel/s information sends a request message to its neighboring nodes. The neighboring $\mathcal{S}\mathcal{U}𝓈$ after successfully receiving the request message start to observe the channel activity in the network. Then each $\mathcal{S}\mathcal{U}$ based on local observation decides the presence or absence of the primary signal and forwards its final decision to the request $\mathcal{S}\mathcal{U}$ by considering the CSMA mechanism in the MCASNs. The selected $\mathcal{S}\mathcal{U}𝓈$ perform channel detection in a group manner in a way such that if required then more than one channel are sensed in parallel in every sensing interval as shown in Figure 2. During the information exchange process, an acknowledgment message is forwarded by the receiving node, which guarantees that the information has been received successfully.

This mechanism significantly reduces the cooperation overhead in terms of latency for searching an appropriate spectrum since multiple channels are sensed in a single sensing interval. Moreover, it also significantly minimizes the network implementation cost by eliminating the need for centralized intelligent devices in the UWSNs. In addition, the fusion mechanism improves the robustness of the decision-making process at the fusion node on the presence or absence of $\mathcal{P}\mathcal{U}𝓈$ in UWSNs. The cooperative decentralized parallel channel sensing architecture of our proposed scheme is indicated in Figure 3. It illustrates that different $\mathcal{S}\mathcal{U}𝓈$ observe the designated channel/s and each of them makes an individual decision and forwards its one-bit decision to the fusion center in a parallel manner in UWSNs.

3.2.3. User Selection in Cooperative Mechanism

In UWSNs, the fusion center cooperating with all users does not essentially attain the optimal performance since the channel fluctuates over time and location due to the harsh nature of the UWEs. This might lead to assigning a poor-quality channel to the $\mathcal{S}\mathcal{U}$ which degrades the performance of the network. Therefore, the selection of $\mathcal{S}\mathcal{U}𝓈$ for obtaining the decisions for specific primary signals plays an important role since it can be used to increase the sensing performance at the fusion center. Thus, the most appropriate $\mathcal{S}\mathcal{U}𝓈$ among the others which have better $\mathcal{D}ℯ\mathcal{T}$ probability by considering a given $\mathcal{F}\mathcal{A}𝓈$ probability in a region must be given the opportunity to participate in the DCoS. In the UWCD scheme, the appropriate cooperating $\mathcal{S}\mathcal{U}𝓈$ raise the probability of $\mathcal{D}ℯ\mathcal{T}$ and minimize the total error probability for the $\mathcal{S}\mathcal{U}$, and thus improve the total throughput of the network. In addition, it provides sufficient protection to the $\mathcal{P}\mathcal{U}𝓈$ in terms of interference in the network. A set of secondary users selected ${\mathcal{S}}_{\mathrm{e}}\left(\mathcal{S}{\mathcal{U}}_{𝒾}\right)$ from the rest of the secondary users $\mathcal{S}{\mathcal{U}}_{𝒿}$ in the decision-making procedure can be numerically shown as:

(12)${\mathcal{S}}_{\mathrm{e}}\left(\mathcal{S}{\mathcal{U}}_{𝒾}\right)={\displaystyle \sum }_{𝒾=1}^{𝓃}{\displaystyle \sum }_{𝒿=1}^{𝓀}\mathcal{S}{\mathcal{U}}_{𝒿}-\mathcal{S}{\mathcal{U}}_{𝒾}$

subject to:

(12a)$\mathcal{S}{\mathcal{U}}_{𝒾}\subseteq \mathcal{S}{\mathcal{U}}_{𝒿}\in {\mathcal{R}}_{𝒾}\left({\mathcal{C}}_i\right)<{\mathcal{R}}_{𝓃}$

(12b) $\mathcal{S}{\mathcal{U}}_{𝒾}\in {\mathcal{I}}_{𝓫ℯ𝓈𝓉}\left({\mathcal{H}}_1|{\mathcal{H}}_0\right)\cong 1$

(12c) ${\rho }_{ℯ}\mathcal{S}{\mathcal{U}}_{𝒾}\left({\mathcal{C}}_i\right)<\delta$

(12d) $\mathcal{S}{\mathcal{U}}_{𝓀}\in {\mathcal{I}}_{𝓅oo𝓇}\left({\mathcal{H}}_1|{\mathcal{H}}_0\right)=0$

(12e) $\mathcal{S}{\mathcal{U}}_{𝒾}\left({\mathcal{H}}_1|{\mathcal{H}}_0\right)\le {\tau }_{\mathrm{s}} <\tau$

Constraints in (12a) state that the selected secondary users are less than the total number of secondary users involved in the sensing process for the distinct channels in a region in the network. Constraints in (12b) guarantee that the selected secondary users perform the best $\left({\mathcal{I}}_{𝓫ℯ𝓈𝓉}\right)$ in terms of channel detections with low error probability $\left(\delta \right)$ is satisfied by Equation (12c). Constraints in (12d) state that the secondary users with poor detection probability $\left({\mathcal{I}}_{𝓅oo𝓇}\right)$ are removed from the channel detection group in the network. Constraints in (12e) make it sure that the selected secondary users detect the channel in predefined time intervals $\left({\tau }_{\mathrm{s}}\right)$.

3.2.4. Sensing Overhead at the Fusion Center

The local sensing and data reporting to the fusion center spend a notable amount of $\mathcal{S}\mathcal{U}𝓈$ energy in the MCASNs. In this respect, the limited sensing information by following certain criteria or constraints, which avoids unnecessary or uninformative data reporting can significantly improve energy efficiency in cooperative detection in the MCASNs. Realizing the facts, the proposed scheme employs a few sensing bits from each neighboring so that $\mathcal{S}\mathcal{U}$ is reported to the requested $\mathcal{S}\mathcal{U}$ in the MCASNs. This notably reduces the average number of sensing bits reported by the selected neighboring $\mathcal{S}\mathcal{U}𝓈$ to the fusion center in the MCASNs. In the proposed scheme, each $\mathcal{S}\mathcal{U}$ after capturing the primary signal sample computes its energy and detects the presence or absence by comparing with the defined threshold value. Then, a one-bit decision 0 for idle and 1 for active with the channel information is sent to the fusion center. However, if no decision is made by the neighboring $\mathcal{S}\mathcal{U}$ then it simply drops the request without reporting the fusion center. The excessive number of $\mathcal{S}\mathcal{U}𝓈$ with no reply may degrade the false alarm probability but the reported local decisions are significantly reduced in the proposed scheme, which saves energy, latency, and processing time consumed in the decision-making process. However, after predefined iterations, if there is no reply for these $\mathcal{S}\mathcal{U}𝓈$ in the specific time intervals then they are removed from the channel detection and reporting group list by the fusion center. Consequently, the fusion center combined the neighboring $\mathcal{S}\mathcal{U}𝓈$ information and computes the combined likelihood ratio like in a Neyman-Pearson test $\left(\mathcal{T}\right)$to make the final decision on the presence or absence of the primary signals which can be numerically indicated as

(13)$\mathcal{T}\left(𝓎\right)={\displaystyle \prod }_{𝒾=1}^{\mathcal{S}{\mathcal{U}}_{𝓃}}\frac{\rho \left({𝓍}_{𝒾}|{\mathcal{H}}_1\right)}{\rho \left({𝓍}_{𝒾}|{\mathcal{H}}_0\right)}\begin{array}{c}{\mathcal{H}}_1\\ ⋛\\ {\mathcal{H}}_0\end{array}\aleph$

in which $\mathcal{S}{\mathcal{U}}_{𝓃}$ shows that there are $𝓃$ number of $\mathcal{S}\mathcal{U}𝓈$ and the observations at each $\mathcal{S}\mathcal{U}$ are indicated by ${𝓍}_{𝒾}$ such that $𝒾=\left\{1, 2, 3,\dots \mathcal{S}{\mathcal{U}}_{𝓃}\right\}$ at the fusion center $𝓎$ such that $𝓎=\left\{{𝓎}_1,{𝓎}_2,{𝓎}_3,\dots \mathcal{S}{\mathcal{U}}_{𝓃-𝓀}\right\}$ and $\aleph$ is the defined threshold value. This mechanism guarantees the target probability of false alarm while maximizing the probability of detection when making a global decision for the presence or absence of the $\mathcal{P}\mathcal{U}$ on a channel in the MCASNs. To further enhance the probability of detection, the fusion center adds only those $\mathcal{S}\mathcal{U}𝓈$ in a group whose detection probability are higher than others above the defined threshold. The fusion center periodically computes the detection probability of each $\mathcal{S}\mathcal{U}$ based on its measurements in different rounds and updates its table. Thus, a fewer $\mathcal{S}\mathcal{U}𝓈$ with the highest detection probability sense and provide the most accurate results with reduced sensing overhead in UWSNs.

3.3. Underwater Acoustic Channel Assignment Algorithm (UWCA)

The $\mathcal{S}\mathcal{U}𝓈$ for a channel ${\mathcal{C}}_i$ sense the temporal non-existence of the $\mathcal{P}{\mathcal{U}}_i$ and find this channel busy if occupied. The $\mathcal{S}\mathcal{U}𝓈$ must leave the detected channel ${\mathcal{C}}_i$ and immediately switch to an alternative channel ${\mathcal{C}}_{𝒿}$ by identifying the spectrum holes for transferring packets in UWSNs. The spectrum detecting approaches to find vacant spectrum bands can be categorized into a cooperative (CoS) and non-cooperative (NCoS) in UWSNs. In the CoS approach, the secondary users sense the vacant channels and cooperate closely with the neighboring secondary users to make the decision for the vacant channels in the network. In the NCoS, a secondary user makes the local decision for the vacant channels based on its own spectrum measurement in the UWSNS. The cooperative sensing can be individual (ICoS) or group-based (GCoS) sensing. In the ICoS mechanism, some randomly selected ASNs monitor the data transmission activities of a specific channel independently. By contrast, in the GCoS approach, a set of predefined ASNs is assigned to monitor the activities of a channel in a group. Consequently, the cooperative sensing based on the architecture can be divided into two types, namely centralized cooperative sensing (CCoS) and decentralized cooperative sensing (DCoS) [38]. In the CCoS approach,$\mathcal{S}\mathcal{U}𝓈$ for a channel ${\mathcal{C}}_i$ sense the temporal nonexistence of the $\mathcal{P}{\mathcal{U}}_i$ and send their local observation data to a specialized fusion center, which combines the received channel results and decides the use of the channel ${\mathcal{C}}_i$ for packets transmission. The specialized fusion centers periodically forward the local decision updates to neighboring specialized devices since they are aware of the existing vacant channels at any given time.

Thus, the key aim of the specialized devices is to allocate vacant channels efficiently to $\mathcal{S}\mathcal{U}𝓈$ with least information message-sharing in the UWSNs. On the contrary, in the DCoS approach, secondary users sense the vacant channels and cooperate closely with the neighboring secondary users and make the decision without using the centralized device for the vacant channels in the network. The CCoS can considerably intensify the systems aptitude in identifying and evading the $\mathcal{P}\mathcal{U}$ signals in the network. However, it faces several issues such as high information exchange overheads and particular region cut off issues due to a single expert device failure, and overall network deployment cost due to the high price of these controllers. Thus, it could not be the best choice for the UWSNs. In this respect, the DCoS approach due to its flexibility in deployment for autonomous decision making seems to be the best choice for UWSNs. In DCoS, a set of $\mathcal{S}\mathcal{U}𝓈$ without the coordination of the centralized expert devices locally detects the $\mathcal{P}\mathcal{U}$ signals and maintains the spectrum information to make the decision by itself. In the proposed scheme, the DCoS overcomes the drawbacks of CCoS, but the sensing capabilities of $\mathcal{S}\mathcal{U}𝓈$ in DCoS are usually limited due to software-defined hardware limitations such as low computation, signal processing, memory, etc. The DCoS is further divided into restricted spectrum sensing (DRS) and whole spectrum sensing (DWS) [39].

The DRS allows secondary users to sense for the vacant channels in a specific region. Although, secondary users sequentially sense the specific region for available channels might raise the probability of detecting an available channel more quickly. However, the possibility of finding a poor capacity-aware channel with high interference also arises that affects the neighboring ASNs leading to low link qualities in the network. This results in packets loss, significant delay and degrades secondary users’ throughput in the network. On the contrary, the DWS allows secondary users to allow to sense the whole region for the desired channel instead of sensing the specific region. Compared to DRS, the DWS in the proposed scheme increases the chance of finding high data rates channels at low interference, sensing overheads and latency due to allowing various$\mathcal{S}\mathcal{U}𝓈$ to sense different channels concurrently in UWSNs. Consequently, there are three types of channel assignment mechanism for multichannel schemes, namely static channel assignment (SCA), dynamic channel assignment (DCA) and semi-dynamic channel assignment (SDA) in UWSNs. In SCA, the acoustic environments such as interferences, data traffic pattern, etc., are generally known [40]. The protocols based on SCA mechanism usually assign channels for long durations relative to the channel switching time or permanently relative to the acoustic interfaces in the network. The static channel assignment is easiest to implement and beneficial to delay-sensitive applications since it avoids additional switching delay during data communication in the network. This type of channel assignment offers several advantages such as low complexity and overhead in the network. However, it is not suitable for the applications whose links are varied and data traffic is unknown due to highly dynamic UWEs.

In UWSNs, the data traffic passing over an ASN differs based on its location along a data path. Generally, the ASNs closer to the sink convey more packets compared to the sensors that are far away. Therefore, data reliability is one of the main requirements for various underwater monitoring and control applications. The static channel assignment mechanisms in underwater suffer from link burstiness due to high interferences and channel variation in the network. This result in packet loss between a sender and receiver for long time intervals, resulting instability and severe communication delays in the scheme. This issue becomes more severe when routing topology varies dynamically. Thus, the SCA limits channel utilization and cannot provide data reliability in the underwater. To this end, the DCA is introduced to mitigate the impact of the interference on the link dynamics due to acoustic channel variations and channel traffic changes in the underwater. The DCA allows the channel assignment schemes to allocate channels with more accurate decisions to secondary users. The DCA provides high throughput in harsh nature dynamic underwater environments. However, it is with the expense of freshness of the data delivery and overhead due to very frequent channel assignment, typically before each transmission. The SDA approach takes the advantages of both static and dynamic channel assignments and provides better trade-off between low overheads and traffic changes with high adaptation to channel variety in the proposed scheme. In this mechanism, the channels are assigned periodically or based on events in the network. Generally, high capacity links with low interferences are allocated statically to the sensors closer to the sink that have the heaviest traffic loads in the network.

The rest of the available channels are allocated for other sensors to pursue a dynamic channel assignment. The semi-dynamic mechanism due to its adaptability to links dynamics, data traffic variations, and interferences while managing the long switching latencies provides a high throughput for delay-sensitive underwater events monitoring applications. To this end, the key objective function of UWCA algorithm can be numerically indicated as

(14) ${\varphi }_{UWCA\left(2\right)}={\displaystyle \sum }_{i=1}^{\left|n\right|}\underset{i}{\max }\left({\mathcal{C}}_{𝓊}+ {\mathcal{T}}_{𝓅}\right)+\underset{i}{\min }\left({\mathcal{I}}_{𝓃𝓉𝒻}+{\mathcal{P}}_{ℯ𝓇}+{\mathcal{C}}_{𝓉}\right)$

In the proposed scheme, the $\mathcal{P}\mathcal{U}$ activity information can provide an opportunity to $\mathcal{S}\mathcal{U}$s for their reliable data transmission in the network. Therefore, the statistical model of the $\mathcal{P}\mathcal{U}𝓈$ behavior must be simple enough that describes precisely how the $\mathcal{P}\mathcal{U}𝓈$ are envisioned to operate in the network. The widely considered model that accurately describes the behavior of the $\mathcal{P}\mathcal{U}$ in an ON–OFF model. In this model, there exist two states, namely ON state and OFF state are used to describe the activity of each channel, indicating the channel is occupied by a $\mathcal{P}\mathcal{U}$ and the channel is idle, respectively. The $\mathcal{S}\mathcal{U}$ can use the oFF period of the $\mathcal{P}\mathcal{U}$ channel to transmit its own data in the network. However, if the OFF period of the $\mathcal{P}\mathcal{U}$ channel is too small then the spectrum state may not change for the $\mathcal{S}\mathcal{U}𝓈$ due to not utilizing the channel in case the primary user is absent. Generally, the exponential model provides a good approximation to measure the duration of the $\mathcal{P}\mathcal{U}$ states. In this model, the durations of both ON and OFF periods is considered as exponentially distributed and independent with means $1/{\mathcal{I}}_{𝒹\ell ℯ}$ and $1/{\mathcal{B}}_{𝓊𝓈𝓎}$