We integrate the two models of Cognitive Radio (CR), namely, the conventional Sense-and-Scavenge (SS) Model and Symbiotic Cooperative Relaying (SCR). The resultant scheme, called SS-SCR, improves the efficiency of spectrum usage and reliability of the transmission links. SS-SCR is enabled by a suitable cross-layer optimization problem in a multihop multichannel CR network. Its performance is compared for different PU activity patterns with those schemes which consider SS and SCR separately and perform disjoint resource allocation. Simulation results depict the effectiveness of the proposed SS-SCR scheme. We also indicate the usefulness of cloud computing for a practical deployment of the scheme.

1. Introduction

1.1. Cognitive Radio/Dynamic Spectrum Access

The emerging Cognitive Radio (CR) technology is an attempt to alleviate the inefficient utilization of the spectrum, created by the current Command-and-Control spectrum access policy. It temporarily allows unused portions of the spectrum (spectrum holes or white-spaces), owned by the licensed users, known as primary users (PUs), to be accessed by unlicensed users, known as secondary users (SUs), without causing intrusive interference to the former’s communication [1]. This is the Sense-and-Scavenge (SS) Model of conventional CR. A CR node is characterized by an adaptive, multi-dimensionally aware, autonomous radio system empowered by advanced intelligent functionality, which interacts with its operating environment and learns from its experiences to reason, plan, and decide future actions to meet various needs [2].

In the SS model of CR, the temporal PU activity patterns have a significant influence on the opportunities for the SUs. The source traffic for the PU alternates between ON (busy) and OFF (idle) periods. The ON/OFF activity is characterized by suitable statistical models, for predictive estimation of the patterns. Exponential [36] and log-normal [35] distributions are popularly used in the literature to model the ON (and OFF) times of the PU activity. Measurements have also revealed that successive ON and OFF periods are independent, though in some cases long-term correlations exist [4].

1.2. Symbiotic Cooperative Relaying

An interesting paradigm that has surfaced in the research surrounding CR is a symbiotic architecture, which improves the efficiency of spectrum usage and reliability of the transmission links [712]. According to this model, which we refer to as Symbiotic Cooperative Relaying (SCR), the PU seeks to enhance its own communication by leveraging other users in its vicinity, having better channel conditions, as cooperative relays for its transmission and in return provides suitable remuneration to them. The SU nodes, being scavengers of the licensed PU spectrum, are potential candidates as relays, since they are idling when the PU transmission is in progress. Besides, they have cognitive capabilities, which give a large amount of flexibility of reconfiguration and resource allocation during the cooperative relaying process. The cooperation from the SU network results in enhanced transmission rate of the PU, which translates into reduced transmission time for the same amount of information bits of the PU as that transmitted on its direct link. Then, the time saved can be offered to the SUs for their own communication as a reward for cooperating with the PU (with a fixed rate demand). The SUs can achieve their communication in the time incentive without the need for spectrum sensing. In our previous work, we have formulated a cross-layer design to enable the SCR scheme, called Cognitive Relaying with Time Incentive (CRTI), for an Orthogonal Frequency Division Multiplexing-(OFDM-) based multi-hop CR network, with special emphasis on the MAC layer coordination protocol [13]. We have also proposed that it is possible to reward the SUs with incentive frequency bands, that is, Cognitive Relaying with Frequency Incentive (CRFI) [12, 14]. Some unique challenges are faced when the SCR scheme is enabled on the spectra of multiple PUs; we have addressed these in prior work as well [15, 16].

In case of SCR, the PU is assumed to have a constant occupancy state throughout the frame duration (in a frame-based communication); that is, it does not exhibit intermittent ON/OFF periods. During those frames when SCR is enabled, the PU should definitely be ON.

1.3. The SS-SCR Scheme

In this paper, we integrate the two aforementioned models of CR, namely, the Sense-and-Scavenge (SS) model of conventional CR and Symbiotic Cooperative Relaying (SCR). We refer to this composite scheme as SS-SCR. SS-SCR entails a multiple PU scenario, with each PU having its own distinct bandwidth of operation. On the PUs’ spectra having a weak direct link, SCR is enabled, while, on the rest of the PUs’ bands, SS is enabled. Since most present day wireless technologies such as IEEE 802.16 [17] and 802.22 [18] are based on OFDM, the multichannel multi-hop networks, thus created, pose a more challenging environment for deployment of the SS-SCR scheme, as opposed to simplistic two-hop or single-channel scenarios addressed in the literature (discussed in Related Literature). Optimum resource (time, bandwidth, power) allocation, which can be achieved by leveraging the channel diversity abundantly available in a multichannel network, will improve spectral efficiency and in turn maximize the transmission opportunities for both the PUs and SUs. With this objective, we present our original contributions in this paper, which are summarized as follows.(1)We propose a scheme for enabling SS-SCR by means of a suitable cross-layer optimization problem which addresses power control, scheduling, and routing. Though the work can readily be extended to any number of PUs, currently a simple scenario with two PUs is assumed—on the spectrum of one we enable SS, while on the other we enable SCR. The SS-SCR scheme jointly considers the resource allocation on both the PUs’ bands to maximize the overall spectral efficiency and mutual benefits of both entities under concern, namely, the PU and SU.(2)For comparison, we also describe two schemes which consider the SS and SCR separately, and the resource allocation on each of the PUs’ bands is disjoint. All the schemes are investigated under various PU ON/OFF traffic models.(3)We propose the use of cloud computing to enhance the performance of SS-SCR in practical CR networks.

To detail our work, the paper has been organized as follows. Section 2 presents related background literature. Section 3 describes the system model and communication scenario. Section 4 methodically explains the generalized cross-layer optimization problem. In Section 5 we propose the SS-SCR scheme, while in Section 6 we describe the problems for the SS and SCR schemes separately considered. Section 7 provides a note on the practical implementation. Section 8 illustrates the use of cloud computing for SS-SCR. In Section 9, we present simulation results and their detailed analysis. Section 10 concludes the paper.

Conventionally, there are two approaches to spectrum sharing in CR [19]: underlay approach, in which the SUs and the PU access the same frequency band by the use of sophisticated spread spectrum techniques, and overlay approach, in which the SUs access the licensed spectra when the PU is not using it. The SS model pertains to the overlay approach—the SUs sense the spectrum to detect a white space and utilize it for their own communication.

Surrounding the concept of SCR for CR, many schools of thought have evolved to accommodate substantially different technologies and solutions. Simeone et al. [7, 8] have used game theoretic tools to analyze the performance of cooperation in a CR network, wherein the PU leases the owned spectrum to an ad hoc network of SUs in exchange for cooperation in the form of transmission power from the SUs. The model proposed by J. Zhang and Q. Zhang [9] is more rational; when the PU’s demand is satisfied, it is willing to enhance its benefit in any other format, for instance, by collecting a higher revenue from the SU. Xue et al. [10] have considered a single full-duplex amplify-and-forward (AF) SU relay to assist the PU transmission. Gong et al. [11] have analyzed the power and diversity gains obtained by AF relaying of the PU’s data by multiple cooperating SUs. All of the aforementioned works in the literature have considered either a single-relay node or single-channel CR networks. The authors have also contributed significantly towards SCR schemes for multichannel multi-hop networks [1216]. The cross-layer formulations in this work are inspired by those of Shi and Hou [20], Zhang et al. [21], and some references therein. While Shi et al. aim at maximizing the sum throughput of the SUs in a multi-hop multichannel CR network, in the proposed SS-SCR scheme, the objective is to perform a joint resource allocation on both the PUs band (SS and SCR) for maximizing the net spectral efficiency. As far as the previous works of the authors are concerned, the concept of CRTI [13] involves a cross-layer optimization problem for a single source, that is, PU Tx, for throughput maximization. The approaches to CRFI [13, 14] are totally different in their objective—that of achieving a specified throughput for the PU while using the least number of frequency bands. Techniques for CRTI for multiple PUs [15, 16] describe the maximization of the time incentive for the SUs, while utilizing multiple PUs spectra optimally. Two methods have been proposed for the same, the formulations for which are distinct, and different from those in the literature [20, 21].

This work differs from the above in the fact that it is a hybrid architecture: it integrates the conventional SS model with SCR, for a multiple PU scenario.

3. System Description

We consider a CR system with a network of cognitive SUs and two PU transceivers (Figure 1). Each PU has its own distinct bandwidth of operation. The available bandwidth is divided into frequency flat subchannels by deploying OFDM. The band-sets of the two PUs are denoted by 𝕄1 and 𝕄2, respectively. On the band-set of PU 1, conventional CR mode of operation, that is, SS, is enabled. The SUs are continuously sensing the spectrum for a transmission opportunity; when PU 1 is OFF, the SUs use its spectrum for their own communication. The activity of PU  Tx  1 is detected by all the SU nodes by cooperative spectrum sensing [22]. Band-set 𝕄1 is also referred to as the SS band.

On the other hand, on the band-set of PU 2, SCR is enabled. Rather than using the direct link, the PU  Tx  2 relays its data through the SU network and in return rewards them with a time incentive 𝜆𝑡 for their own communication. If 𝐶dir is the throughput (bits/sec/Hz) obtained on the direct link, 𝐶rel is the maximized throughput (bits/sec/Hz) obtained through the SU relay network, then the incentive in a time frame normalized to unity is 𝜆𝑡=1𝐶dir/𝐶rel, 0𝜆𝑡1. On band-set 𝕄2 (also referred to as the SCR band), PU  Tx  2 acts as the source, PU  Rx  2 as the destination, and the SU nodes act as the relays in the multi-hop relay network (Figure 1). Decode-and-forward multihopping is assumed at each node.

The fading gains for various links are mutually independent and are modeled as zero mean complex circular Gaussian random variables. The protocol interference model is assumed [20]. The channel gains are invariant within a frame but vary over frames (i.e., block-fading channels). We assume that the channel gains from the PU  Tx  2 to SUs, the SUs to the PU  Rx  2, and those among the SUs are good enough to provide a significantly higher end-to-end throughput as compared to the direct link of PU 2, resulting in performance gains for both the PU and the SUs on band-set 𝕄2.

4. Problem Formulation: Cross-Layer Optimization

In the subsequent sections we will be describing the proposed SS-SCR scheme which considers joint resource allocation on both PUs’ bands, as well as the schemes which are disjoint in their resource allocation on the two bands. Each scheme will involve solving a sequence of optimization problems, their objective being maximization of the sum throughput of the users under consideration (PU or SUs or both) within the given resources (time slot, frequency bands, power). To efficiently exploit the channel diversities available in the multi-hop multichannel SU network, we allow flow splitting and spatial reuse of frequencies outside the interference range of nodes. Each optimization problem involves a cross-layer view for power allocation, frequency band scheduling, and routing. A relay with poor channel conditions on all its links will be eliminated from the routes which strive to achieve maximum throughput; thus relay selection is automatically achieved by the problem. We describe the basic structure of such a cross-layer optimization problem which will be suitably adapted for the various schemes to be described subsequently.

Optimization Problem (P1):
max𝑥(𝑚)𝑖𝑗,𝑃(𝑚)𝑖𝑗,𝑓𝑖𝑗(𝑙)𝑙𝕃𝑗𝑇𝑖𝑓𝑖𝑗(𝑙)𝑖=𝑠(𝑙).(1) It is subject to the constraints which are described as follows.

Flow Constraints:

Frequency Domain Scheduling Constraints:

Power Constraints:

Interference Constraints:

Since our objective (1) is to maximize the throughput, it is sufficient to maximize the sum of outgoing flows from the source node [23]. We denote the communication between each unique transmitter-receiver pair as a session. 𝑠(𝑙) and 𝑑(𝑙) represent the source and destination of the session 𝑙,𝑙𝕃, where 𝕃 denotes the set of the sessions.

Bidirectional links are assumed; that is, in the network graph each node 𝑖 has an transmit/receive set of nodes 𝑇𝑖. 𝑓𝑖𝑗(𝑙) is the data flow (bits/sec) from node 𝑖 to node 𝑗 for session 𝑙. Equation (2) indicates that, except for the source and destination nodes, the inflow into a node is equal to the outflow. Equation (3) ensures that all the flows are nonnegative. Equation (4) refers to the fact that the sum of the flows on a link cannot exceed the capacity of a link according to Shannon’s channel capacity theorem [24]. Each link has |𝕄| orthogonal frequency bands, and the net achievable throughput is the sum throughput of the individual bands. (𝑚)𝑖𝑗 denotes the channel power gain on band 𝑚, and 𝑃(𝑚)𝑖𝑗 denotes the corresponding power allocation. We have assumed unit bandwidth of each band. In (4), the log function contains only 𝜎2 in the denominator due to the use of an interference model, which ensures that when node 𝑖 is transmitting to node 𝑗 on band 𝑚, the interference from all other nodes in this band must remain negligible due to the frequency domain scheduling and interference constraints. denotes the node set of the network and 𝔼 denotes the edge set.

Equation (5) suggests that if a node 𝑖 has used a band 𝑚 for transmission or reception, it cannot be used by node 𝑖 again for any other transmission or reception. Note that 𝑥(𝑚)𝑖𝑗 is a binary variable which takes the value 1 if and only if band 𝑚 is active on link (𝑖,𝑗).

Equation (7) ensure that 𝑃(𝑚)𝑖𝑗[𝑃𝑇(𝑚)𝑖𝑗,𝑃peak] if the band 𝑚 is selected and 𝑃(𝑚)𝑖𝑗=0 if the band is not selected. The data transmission from node 𝑖 to node 𝑗 is successful only if the received transmission power exceeds a power threshold 𝑃𝑇, from which we can calculate the minimum required transmission power on a band 𝑚 at node 𝑖 as 𝑃𝑇(𝑚)𝑖𝑗=𝑃𝑇/(𝑚)𝑖𝑗. 𝑃peak denotes the maximum power that can be allocated to any band 𝑚, under which we compute the interference set 𝐼𝑚𝑗 of a receiving node 𝑗. Equation (8) is to ensure that the total power transmitted on all the active bands at node 𝑖 does not exceed the power available at the node 𝑃avl𝑖.

Equation (9) ensures that for a successful transmission on link 𝑖 to 𝑗, on an interfering link 𝑘 to , the transmit power on any band 𝑚 cannot exceed a threshold 𝑃peak if 𝑥(𝑚)𝑖𝑗=0, and if 𝑥(𝑚)𝑖𝑗=1, then 𝑘𝐼𝑚𝑗𝑃(𝑚)𝑘(𝑚)𝑘𝑗𝑃𝐼. The complete list of symbols with their description is given in Table 1.

In the above optimization problem (𝑚)𝑖𝑗, 𝜎2, 𝑃𝑇, 𝑃𝐼, 𝑃peak, and 𝑃avl𝑖 are all constants, while 𝑥(𝑚)𝑖𝑗, 𝑃(𝑚)𝑖𝑗, and 𝑓𝑖𝑗(𝑙) are the optimization variables. The formulation is a mixed integer nonlinear programming problem (MINLP). Based on the discussion on similar problems in [20, 21] and the references therein, we conjecture that the given problem is NP-hard. We are thus motivated to investigate a linear formulation, which will greatly simplify the problem (which is observed in terms of reduced computation time during simulation). This entails employing three tangential supports to the log term in (4), as its approximation [20]. The tangential supports are drawn at points 1, 2, and 3 on the log curve (Figure 2), namely, (0,0), (𝛽,𝑓(𝛽)), and (𝑃peak,𝑓(𝑃peak)). 𝛽 denotes the 𝑥-coordinate of the point of intersection of the tangents drawn at points 1 and 3. The solution to the log relaxed problem provides an upper bound to the original maximization problem P1.

A Feasible Centralized Solution
We suggest an approach to obtain a feasible suboptimum solution to the original problem by decoupling the operations of power allocation and band scheduling and that of flow computation. The solution consists of two steps.(1)The power allocation and band scheduling (𝑃𝑚𝑖𝑗,𝑥𝑚𝑖𝑗) are obtained from the log relaxed problem with tangential supports. This solution, however, may violate the flow constraints.(2)The above (𝑃𝑚𝑖𝑗,𝑥𝑚𝑖𝑗) are substituted in the original problem, which is then solved only with respect to 𝑓𝑖𝑗 as the optimization variable. The overall result represents a feasible solution to the original problem P1.

5. The SS-SCR Scheme

As described earlier, PU 1’s activity is changing on band-set 𝕄1 (SS band), providing intermittent periods for the SUs to communicate; on band-set 𝕄2 (SCR band), PU 2 is ON and relaying its data through the SU network. It is on this band that a time incentive will be offered to the SUs in return for their cooperation. In SS-SCR, we solve a joint resource allocation problem on both the PUs’ bands; that is, 𝕄1𝕄2, in every such time interval that PU 1’s activity changes. There are totally four possibilities (Figure 3): PU 1 is OFF and PU 2 is relaying on 𝕄2, PU 1 is ON and PU 2 is relaying on 𝕄2, PU 1 is OFF and SUs are using the time incentive on 𝕄2, and PU 1 is ON and SUs are using the time incentive on 𝕄2. Cross-layer optimization problems are formulated for the aforementioned possibilities, as follows.(Ia) PU 1 is OFF and PU 2 is relaying on 𝕄2. In this case, the joint problem entails maximizing the sum throughput of the SUs and PU 2; the SUs want to make the best utilization of the OFF time of PU 1, while PU 2 wants to maximize its throughput through the SU network so that in turn it can maximize the time incentive offered to the cooperating SUs. The complete band-set 𝕄1𝕄2 and the total node power budget 𝑃node𝑖 are available for the problem.(Ib) PU 1 is ON and PU 2 is relaying on 𝕄2. PU 2 can maximize its throughput through the SU network only on 𝕄2 with the total node power budget 𝑃node𝑖.(Ic) PU 1 is OFF and SUs are using the time incentive on 𝕄2. The SUs can now use the complete band-set 𝕄1𝕄2 with the total node power budget 𝑃node𝑖 to maximize their sum throughput.(Id) PU 1 is ON and SUs are using the time incentive on 𝕄2. The SUs can only use 𝕄2 with the total node power budget 𝑃node𝑖 to maximize their sum throughput.

To enable SS-SCR, the following parameters should be set in problem P1 (Table 2).

6. Disjoint Resource Allocation for SS and SCR

In this section, we describe schemes based on disjoint resource allocation on the SS and SCR bands, considering them as separate problems.

6.1. Scheme A

This scheme gives priority to the activity on the SS band and second preference to the SCR band. It is devised for that situation in which the OFF periods of PU 1 are high. The following steps are adopted (Figure 4(a)).(IIa) First, the SUs’ sum throughput maximization problem is solved on band-set 𝕄1 (SS band). The SUs will be sensing for a spectrum opportunity on this band. In the OFF time of PU 1, they will utilize this band for their own communication. The total node power budget 𝑃node𝑖 is available for them at each node 𝑖.(IIb) Secondly, on band-set 𝕄2 (SCR band), PU Tx 2 will relay its data through the SU network with maximized throughput. Since the communication happens concurrently with the SU’s communication on 𝕄1, now the power available at each node 𝑖 is the node power budget minus the power consumed in step (IIa), that is, 𝑃node𝑖𝑃cons𝑖. The channel diversity and consequently the higher throughput obtained from the SU network will diminish the transmission time for the same number of bits as those transmitted on the direct link of PU 2. The time saved is offered as an incentive to the SUs for their own communication.(IIc) In the time incentive obtained from PU 2, the SUs maximize their sum throughput on 𝕄2. The power available at each node 𝑖 is 𝑃node𝑖𝑃cons𝑖.

To enable Scheme A, the following parameters should be set in problem P1 (Table 3).

6.2. Scheme B

This scheme gives priority to the activity on the SCR band and second preference to the SS band. It is devised for that situation in which the ON periods of PU 1 are high. The following steps are adopted (Figure 4(b)).(IIIa) First, on band-set 𝕄2 (SCR band), PU  Tx  2 will relay its data through the SU network with maximized throughput. The total node power budget 𝑃node𝑖 is available for its communication. The higher throughput achieved, as compared to the direct link of PU 2, will generate a time incentive for the SUs on 𝕄2.(IIIb) Next, in the time incentive obtained from PU 2, the SUs maximize their sum throughput on band-set 𝕄2. The power available at each node 𝑖 is 𝑃node𝑖.(IIIc) Lastly, the SUs’ sum throughput maximization problem is solved on band-set 𝕄1 (SS band). The SUs will be sensing for a spectrum opportunity on this band. In the OFF time of the PU, they will utilize this band for their own communication. Since this transmission is concurrent with that on 𝕄2, the power available for them at each node 𝑖 is minimum of that left after consumption in the relaying interval and the incentive period, that is, min(𝑃node𝑖𝑃consIII𝑎𝑖, 𝑃node𝑖𝑃consIII𝑏𝑖).

To enable Scheme B, the following parameters should be set in problem P1 (Table 4).

7. A Note on the Practical Implementation

To make the SS-SCR scheme a practical reality, a MAC schedule is needed to coordinate all the operations. The MAC frame consists of a control interval in which estimation of the channel states, prediction of PU activity, solving the optimization problems at a centralized controller, and dissemination of the decision throughout the network, are conducted [13]. It is followed by the data interval in which the PUs and SUs communicate using the designated resources. Based on the predicted PU activity, it can be decided when the different solutions of the joint resource allocation are to be applied. The prediction may be corroborated with spectrum sensing to protect the PU 1 from the SU’s interference. The time incentive can be computed in the control interval itself, to determine when the SUs can access the SCR band. An important underlying assumption for the successful execution of the SS-SCR scheme, as well as Schemes A and B (included for comparison), is that the solution time for the optimization problem on the available spectrum is less than the spectrum hole created by the inactivity of PU 1.

Discontiguous OFDM (D-OFDM) is used at the physical layer, which allows the relays to decode only a fraction of the total subcarriers. A control channel is dedicated for all the signalling that enables and coordinates the entire SS-SCR scheme.

8. Cloud Computing for SS-SCR

In SS-SCR, the SU nodes are involved with the following tasks, (i) spectrum sensing, (ii) collaborative spectrum sensing decision algorithms, (iii) machine learning algorithms for PU activity prediction based on recorded history, (iv) solving the cross-layer optimization problems for resource allocation and (v) software defined radio (SDR) technologies for reconfiguration. Most of these operations involve both processing vast volume of data (depending on the network size and parameters) and processing it fast. The cognitive SU nodes may have limited computing and storage capability, which may prevent them from realizing their full potential. In such a situation, shifting some of the operations to the cloud may drastically improve the performance of the system [2527]. Cloud computing is a recent technology revolution that is shaping the world. However, the decision to exploit the vast computational resources of the cloud should be governed by the volume of data and computational complexity, as well as time sensitivity. Primarily for the tasks of PU activity prediction and solving the cross-layer optimization (especially in a large network), the cloud may be of great use in SS-SCR (Figure 5). A low latency, high-bandwidth, reliable link is needed between the SU network and the cloud; else the connectivity may become a performance bottleneck.

9. Simulation Results and Discussion

We have simulated a network with the nodes randomly distributed in an area of 10 square units (Figure 8). Nodes 1 and 9 represent PU  Tx  1-PU  Rx  1, on the band-set of which Sense-and-Scavenge (SS) takes place. Nodes 10 and 11 represent PU  Tx  2-PU  Rx  2, on the band-set of which Symbiotic Cooperative Relaying (SCR) takes place. Nodes 2 to 8 represent the SU relay nodes.

All the links undergo the Rayleigh multipath fading, defined in the time domain by 𝐿1𝑙=0𝑙𝛿(𝑡𝑙𝑇) where 𝑙 is the complex amplitude of path 𝑙 and 𝐿 is the number of channel taps. The 𝑙th channel coefficient between two nodes with a distance 𝑑 between them is distributed as 𝑁(0,1/𝑑𝜂), and the frequency domain channel is given by its Fourier transform. The path loss exponent 𝜂=2.5. The AWGN variance 𝜎2=1𝑒4. A 16 band OFDM system is considered on each link. Bands 1–8 are the SS bands, while 9–16 are the SCR bands. The OFDM subcarrier bandwidth is unit Hz.

The detection threshold is 𝑃𝑇=0.01W, the interference threshold is 𝑃𝐼=0.001W, the peak power constraint on each frequency band is 𝑃peak=0.5W, and the node power constraint is 𝑃node𝑖=3W (it is the same on each node 𝑖).

The environment has been simulated in MATLAB, while the LINGO [28] software has been used to solve the MINLP problem.

Figures 6(a) and 6(b) depict the sum SU throughput (bits/sec/Hz) for the proposed SS-SCR scheme with respect to 30 independent channel instances. It is compared with Schemes A and B, which consider SS and SCR separately, on their respective bands. Each of the values are averaged over 100 time frames, each of 10 sec duration. Two SU sessions are assumed, with nodes 2–7 forming the first pair and nodes 3–8 forming the second pair. The ON and OFF periods of PU 1 are each assumed to follow a log-normal distribution. In Figure 6(a), the mean ON time of PU 1 (𝜇ON) is 2 and the mean OFF time (𝜇OFF) is 8, while the variance of each distribution (𝜎2ON=𝜎2OFF) is 10. It is observed that Scheme A performs better (on an average) than Scheme B since it gives preference to the SUs to communicate on the SS band, which is free most of the time (mean OFF time of PU 1 is higher). In Figure 6(b), the mean ON time of PU 1 (𝜇ON) is 8 and the mean OFF time (𝜇OFF) is 2, while the variance of each distribution (𝜎2ON=𝜎2OFF) is 10. It is observed that Scheme B performs better (on an average) than Scheme A because it gives preference to PU 2’s relaying and consequently creates a higher time incentive for the SUs to communicate, while PU 1 provides few opportunities for the SUs to communicate on its band (mean OFF time of PU 1 is lower). SS-SCR consistently performs better than the disjoint SS and SCR schemes, since the complete band-set, 𝕄1𝕄2, is available in every time interval for the SU’s communication with the total node power budget. Figures 7(a) and 7(b) depict a similar trend for the exponential distribution of PU 1. In Figure 7(a), 𝜇OFF=𝜎OFF=8 and 𝜇ON=𝜎ON=2, while in Figure 7(b), 𝜇OFF=𝜎OFF=2 and 𝜇ON=𝜎ON=8.

To illustrate the results of the cross-layer optimization problems, the band assignment and power allocation for a particular channel instance for SS-SCR (Case Ia) are shown in Table 5. The corresponding flow (bits/sec/Hz) is shown in Figure 8.

Figure 9 demonstrates the average sum SU throughput with different mean ON and OFF times of the log-normal and exponential distributions (fixed variance 𝜎2ON=𝜎2OFF=10). It is observed that when the mean OFF time is higher and ON time is lower, Scheme A performs better than Scheme B, for reasons described earlier. But as the OFF time reduces and the ON time increases, the trend reverses. For equal mean ON and OFF times, both Schemes A and B perform similarly. SS-SCR is consistently better than the previous two schemes, but its performance degrades and approaches that of Scheme B as the mean ON time increases. This is because the band-set of PU 1 is available for too short a duration for it to exploit the channel diversity. The above discussion holds true for log-normal and exponentially distributed ON/OFF periods of PU 1.

10. Conclusion

We have proposed a novel SS-SCR scheme to be deployed in CR networks with multiple PUs, some of which have weak direct links. On the spectra of such licensed users SCR is enabled, while on the other PUs’ spectra conventional SS is implemented. The hybrid SS-SCR scheme results in a better utilization of the available resources (time, bandwidth, power) by means of the formulated cross-layer optimization problems. Its performance is compared, for different PU activity patterns on the SS bands, with those schemes which consider SS and SCR separately and perform disjoint resource allocation. Simulation results depict that the SS-SCR scheme with joint resource allocation gives a higher net SU throughput as compared to the other schemes. Further, the usefulness of cloud computing is illustrated to realize the full potential of SS-SCR.


If 𝐷off is the random variable which describes the OFF period of the PU activity and if it follows the log-normal distribution, its probability density function (PDF) is given by foff(t;μ,σ)=1tσ2πe-(lnt-μ)2/2σ2,t>0.

𝜇 and 𝜎 denote the mean and standard deviation, respectively.

In case of the exponential distribution, foff(t;λ)=λe-λt,t0.

The mean and standard deviation are both given by 1/𝜆.


This work has been supported in part by the Ministry of Communication and Information Technology, Government of India, New Delhi. The work has also been supported by Microsoft Corporation and Microsoft Research India under the Microsoft Research India PhD Fellowship Award 2009.