In the existing literature, the radio spectrum is divided into licensed and unlicensed bands. The conventional approaches of radio spectrum management give exclusive and permanent resource primarily to licensed users. According to the reports from Federal Communication Committee (FCC) [1] and Shared Spectrum Company [2] , there is a considerable amount of the spectrum detected out of use both in space and time “white space”. Thus, to make full use of the white space, the Cognitive Radio (CR) paradigm has been introduced. The CR technology can be aware of the spectrum usage state and intelligently employs idle states using Software-Defined Radio (SDR) which supports opportunistic multiple radio interfaces for spectrum handoff.

A white space is defined as a frequency band licensed by a Primary User (PU) that is not being utilized at a particular time and a geographical location. In a CR network, a Secondary User (SU) can access the spectrum of the PU in an opportunistic manner (i.e., a SU can access the spectrum of the PU only if the spectrum is idle). In the case of a PU appearing in the spectrum occupied by a SU, the SU moves to a new available spectrum immediately if exists.

In a cognitive radio network, when two SUs need to establish a link and bootstrap communications, they need to find each other first. This process is known as rendezvous or neighbor discovery [3] . A rendezvous is a meeting at an appointed time and place. Many rendezvous protocols exist in the literature. These protocols can be classified into two categories. The protocols in the first category use a predefined Common Control Channel (CCC). In these protocols one of the available channels is assigned as the CCC, and it serves as a rendezvous channel. The protocols in the second category use the Channel Hopping approach (CH). In channel hopping approaches, SUs generate their own channel hopping sequences [4] . When a SU needs to communicate with one of its neighbors, it switches from one channel to another, by following a predefined hopping sequence, until it finds its neighbor. When the SU switches to an available channel, it senses the channel for the presence of PUs and other SUs transmissions. If it senses the channel free, it will broadcast a beacon to inform its neighbors about its request to communicate, so that if the intended receiver is on the current channel and receives the beacon, it can reply with an Acknowledgment (ACK) message. Otherwise, the SU will switch to another channel by following the hopping sequence, and broadcast the beacon message again. This process is repeated, until the SU (transmitter) meets with its intended neighbor (receiver). Once two SUs have rendezvoused on an available channel and exchanged control messages, they can perform negotiation for data communication on the current rendezvous channel.

Several MAC protocols for cognitive radio networks exist in the literature. These protocols can be classified according to how the SU deals with the appearance of the PU in two strategies [5] . The first strategy allows the SU to buffer their connections until the PU disappears. The second approach allows the SU to switch to another idle channel. The disadvantage of the first strategy is that the SU buffers its connection even if there is another free channel. Furthermore, it may happen that the SU will not be able to reestablish its connection after buffering because of continuous PU transmissions which lead to a high delay. The disadvantage of the second strategy is the control message overhead between the transmitter and the receiver to access the new channel.

In [6] , the authors have categorized the MAC protocols for CR into three categories: contention-based protocols, time-slotted protocols and hybrid protocols; that is a combination of the contention-based and time-slotted protocols. Contention-based MAC protocols do not need time synchronization between SUs to access the available channels, and are generally based on the CSMA/CA principle.

Opportunistic Spectrum Access with Backup Channel (SWITCH) [7] and DCA-MAC protocol [8] are examples of this category. Time-slotted MAC protocols need a global synchronization between SUs where the time is divided into slots for both the control channel and the data transmission. Examples of this category are the Cognitive MAC (C-MAC) protocol [9] and the Opportunistic Spectrum Access (OSA-MAC) protocol [10] . Hybrid protocols use a partially slotted transmission, in which the control signaling generally occurs over synchronized time-slots. However, the following data transmission may have random channel access schemes, without time synchronization. An example of this category is the SYNchronizes MAC (SYN-MAC) [11] . SWITCH [7] is a decentralized, asynchronous, contention-based MAC protocol. SWITCH uses a CCC channel so that all the necessary control information is exchanged among SUs via it. SWITCH uses the concept of Backup Channel (BC) to cope with the appearance of the PUs.

In this paper, we present a modified protocol from the existing SWITCH protocol named Opportunistic Spectrum Access with backup channel and Buffered data with Resume (OSA-BR). In OSA-BR, instead of using the CCC approach, the blind rendezvous [12] under the PUs activities is used. In addition, instead of using the BC only to deal with the suddenly appearance of PUs, the buffer concept is introduced in OSA-BR to increase the overall throughput.

The current paper is organized as follows. In section 2, the proposed protocol is described in details. In section 3, an example is presented to simplify the description of the proposed protocol. In section 4, the simulation model and its results are presented. In section 5, we draw a conclusion.

In this section, the details of the proposed protocol are introduced.

To simplify the description of the spectrum sharing process of the OSA-BR protocol, an example is presented. Suppose that the considered network has five secondary users su₁, su₂, su₃, su₄, and su₅, as shown in Figure 4. Each SU builds its IUL and PLCH based on the sensing information that is collected during the spectrum sensing process. There are two classical users Cu₁, and Cu₂ and two primary users Pu₁ and Pu₂. The number of the available LCs within the transmission ranges of the secondary users is set to four and indexed as Lu₁ trough Lu₄. While the number of the available UCs is assumed to be six and indexed from UC₁ to UC₆.

The unlicensed channel UC₆ and the licensed channel LC₂ are utilized by Cu₂ and Pu₁ respectively. Hence, as shown in Figure 4, they are not listed in both IUL and PLCH data structures. The remaining three LCs (i.e., LC₁, LC₃ and LC₄) and five UCs (i.e., UC₁ to UC₅) are available for starting connections. We assume that the Su₂ (transmitter) wants to establish a connection with Su₃ (receiver) to start its data transmission. The Su₂ starts the rendezvous process to finds its receiver su₃ at time t₀ as shown in Figure 5. Assuming that the su₂ finds its receiver su₃ on the channel Lc₁ which serves as a common control channel. At time t₁ the modified two-way handshake RTS/CTS, 802.11 MAC, is started. In this example, we have three cases depending on the exchanged control information between su₂ and su₃ during the two-way handshake RTS/CTS process.

Case1: The receiver (su₃) agrees with transmitter’s (su₂) proposed control information. In Figure 5, the proposal of Su₂ is BC = 3, IUL = 1, 3, 4 and PLCH = 3, 4. After the two-way handshake RTS/CTS is completed, both Su₂ and Su₃ start data transmission on LC₁. Here LC₁ is utilized as CCC and operational channel. If a PU appears during the data transmission between Su₂ and Su₃, both SUs switch to the BC (BC = 3) during T_S switching time. The switching time T_S can be defined as the time required by the SU to sense and switch to the BC. This process is called spectrum handoff. The T_S time should be less than the distributed Coordination Function Interframe Space (DIFS) of IEEE 802.11. If this condition is not satisfied then there is a probability that another SU in the vicinity of the transmitter wins the contention and thus the switching fails [7] . If the BC is available, both users will not perform any additional handoff since the UCs are free from PUs.

Case 2: This happened when the receiver (su₃) did not agree with transmitter’s (su₂) proposal about the BC number. In this case the proposal of su₂ is BC = 3, IUL = 1, 3, 4 and PLCH = 3, 4. Upon receiving the RTS, the SU₃ responses with CTS = {BC = −1, PLCH = 0}. This means that there is no agreement about the BC and full agreement with the PLCH. Based on this matching, both nodes will follow the channel hopping sequence identified by PLCH in the case of a PU appears as shown in Figure 6. This figure illustrates multiple handoffs due to the subsequent appearance of the PU.

Case 3: When the receiver (Su₃) did not agree with transmitter’s (Su₂) proposal about both the BC number and the PLCH. The Su₃ responses with CTS = {BC = −1, PLCH = −1}. This means that there is no agreement about the BC and the PLCH. Based on this matching, the data transmission of Su₂ will be queued in its buffer for specific time in the case of a PU appears as shown in Figure 7. If the queued time is expired before the SU₂ win a new channel to resume its transmission, the data will be dropped.

In this section, we present our simulation and compare the proposed scheme with the drop tail scheme. For the simulation proposes, we carried out a discrete-event simulation program using JAVA language. All simulation results have been obtained by running the simulation for 1,000,000 milliseconds. The operational simulation parameters are shown in Table 1.

The efficiency of CR-MAC protocols is evaluated by three metrics: saturation throughput, communications overhead and dropping probability. Saturation throughput is defined as the number of data packets transmitted by the SUs. Communications overhead is defined as the number of control packets transmitted during the transmission negotiation process. Dropping probability is defined as the number of SU’s transmissions rejected due to lack of the channels availability. The efficiency of the OSA-BR is compared with the SWITCH protocol. The reason for choosing the SWITCH protocol is that it reacts efficiently with the appearance of the PUs by using a so-called backup channel. On the other hand, the SWITCH protocol increases the throughput compared to

DCA-MAC protocol in which the LCs only are used by 91.7% and 63.5%, respectively [7] .

The scenario used in our simulation can be described as follows: there are 50 SUs, 12 CUs and 12 PUs. Each two SUs establish a session. We assume that each SU has always a packet in its queue to send. The SUs coexist with both the PUs and the CUs. Each SU in this network independently generates traffic of fixed-size packets. The obtained results are averages over 10 different runs of the developed simulation program. The running time of each run is set to 10⁸ μs. Unlike SWTICH which uses a dedicated CCC for control packets exchange, OSA_BR uses the blind rendezvous to select the channel for control packets exchange in a dynamic manner.

The PU traffic load has a great impact on the performance of SUs since once a PU appears in a channel occupied by a SU; the SU should vacate this channel and determines another free one. Figure 8 shows the saturation throughput of the SUs using the OSA-BR protocol compared to SWITCH and DCA-MAC vs. the PU traffic load. For OSA-BR and SWITCH, the numbers of LCs and UCs are set to 12 and 2 channels respectively. For fairness evaluation, the number of LCs assigned for DCA-MAC is set to 14 channels due to the fact that the DCA-MAC uses the LCs only. The performance of these protocols is evaluated under both low and high traffic of the CUs. The figure shows that the OSA-BR outperforms the performance of both SWITCH and DCA-MAC in both high and low traffic of the CUs. The throughput of OSA-BR increases compared to SWITCH by 38%.

Figure 9 shows the impact of the UCs on the throughput of the OSA-BR. To make a fair comparison when using LCs only and a combination of LCs and UCs, we use the same number of channels in each case. From the figure, we can conclude that, the UCs cannot be ignored when evaluating the performance of the CR MAC protocols.

One of the most important metric used to evaluate the CR MAC protocol is the communications overhead.

The communications overhead is defined as the number of control packets transmitted during the communication setup. As we noted earlier, the SWITCH protocol uses CCC which means that when a PU appears in the CCC, all SUs must leave this channel and wait until the PU completes its transmission. If the PU transmission period is long, it may block the channel access for SUs. Thus the overall throughput of the network degrades. However, a new control packet called NTR is added to the SWICH protocol during the communication setup process which resulted in more overhead.

Figure 10 shows the average number of control packets exchanged between two SUs to complete their connection setup versus the PU traffic load of the OSA-BR and SWTICH protocols. This figure also indicates that the OSA-BR protocol reduces the communications overhead traffic if it is compared with SWITCH protocol.

Figure 11 illustrates the dropping probability against the PU traffic load. From the figure, we note that, the

dropping probability increases in both SWITCH and OSA-BR protocols with the increase of the PUs traffic load. On the other hand, the OSA-BR protocol outperforms the SWITCH protocol because of using a buffer for hosting the data of the interrupted transmissions. The buffered data of the interrupted transmissions is resumed later if the transmitter wins a new idle channel within a maximum predefined buffered time.

We have presented a decentralized, asynchronous and contention-based MAC protocol named OSA-BR. OSA-BR is a modified protocol from the existing SWITCH protocol. It uses the concepts of rendezvous and buffering. The simulation results have been compared with the results of the SWITCH protocol. The results of the model show that the OSA-BR protocol outperforms the well-known SWITCH protocol in terms of throughput, dropping probability and communications overhead.