A new reservation-based Medium-Access Control for full-duplex WLANs

Abstract

In-band simultaneous transmission between an access point (AP) and a node was enabled by recent full-duplex radio communication research. While this event was expected to increase throughput within the feasibility of full-duplex radio communication, the conventional Medium Access Control (MAC) Distributed Coordination Function (DCF) protocol in wireless local area networks (WLAN) limits this performance enhancement. Each node competes for transmission opportunities because DCF is based on half-duplex communication principles. The competition among nodes and an AP creates excessive overhead and many collisions. In this paper, we propose a new MAC protocol called Reservation-Based Medium-Access Control (RMAC) based on node reservation in WLAN with full-duplex radio. RMAC is compatible with DCF. RMAC decreases collisions by reducing the number of competing nodes and increases throughput by reducing competition overhead. Our RMAC assures nodes a transmission opportunity without collision if a node has several packets to send. We show that RMAC achieves at least 86.3% more throughput than full-duplex radio based on DCF. The RMAC protocol also maintains high throughput even in cases of dense node deployment.

Keywords

Wireless LAN computer networks full-duplex radio Medium Access Control

1. Introduction

In conventional WLAN operation, nodes are equipped with half-duplex radio. Because nodes cannot send and receive simultaneously using half-duplex radio, WLAN nodes need a channel access mechanism. DCF is based on the Carrier Sense Multiple Access/Collision Avoidance (CSMA/CA) method and is a widely used MAC protocol in several wireless communication standards. In DCF, nodes choose the back-off number randomly and wait until the back-off number is zero, thus avoiding collisions in a distributed manner. Despite the randomly chosen back-off number, DCF cannot assure that nodes communicate without collisions because some nodes attempt to send simultaneously. As the number of WLAN nodes increases, the number of competing nodes increases, and the probability of simultaneous transmission among nodes also increases. This means that when nodes are densely deployed in a WLAN, the efficiency of DCF is degraded because the competition and collision probability among nodes have increased.

In general, one AP is associated with many nodes; therefore, relatively, APs compete more frequently to get more transmission opportunities. However, some applications, such as peer-to-peer services (e.g., file sharing), VoIP (e.g., skype), and cloud storage (e.g., google drive) require more uplink transmission opportunities and have more uplink traffic volume [7]. VoIP or applications requiring short-interval and continuous transmission incur continuous node competition; therefore, the weight of uplink node transmissions will increase to satisfy various application requirements. When the number of uplink transmission attempts increases, throughput seriously degrades because of the collision increment among nodes. Therefore, we need a new approach to solve these problems.

Recently, in-band full-duplex radio technology has been developed using the Self-Interference Cancellation (SIC) model [1,5]. With in-band full-duplex radio, a node’s ability to send and receive simultaneously not only improves throughput but also enables continued operation even when the nodes are densely deployed. In addition, node transmission competition is halved during competition. Since competition among nodes must be settled within a fixed time, the amount of competition becomes more important as the transmission rate increases. However, since DCF was developed only for the half-duplex radio environment, it limits the performance enhancement of full-duplex radio; thus giving rise to the need for developing a new MAC protocol for full-duplex radio. Several MAC protocols have been proposed, but prior research [4,8,15,16,18] focused on full-duplex communications, especially asymmetric transmission as shown in Fig. 1(b). We focused on full-duplex radio applications to enhance transmission performance even if the nodes were densely deployed because nodes equipped for full-duplex radio can send and receive simultaneously on the same channel.

Fig. 1.

Full-duplex communication cases (a) bi-directional transmission (b) asymmetric transmission.

In this paper, we propose a simple and efficient MAC protocol based on a reservation scheme for full-duplex WLANs. In our research, we chose an environment containing one AP and nodes equipped with either full-duplex or half-duplex radio transmission capability. Our research satisfies the three system problems of design, wasted computational time, and determinism to improve data throughput as described below.

Reduce waste time: Using full-duplex radio protocol, our RMAC reduces competition among nodes to decrease collision probability even when nodes are densely deployed. It has been proven that collision probability is affected by the number of nodes [6,17]. In our research, the number of nodes vying for the right to transmit is decreased, so the competition time is reduced.

Deterministic full-duplex transmission: For full-duplex transmission to maximize its full-duplex radio performance, an AP needs to recognize with low overhead whether a node has a packet to send. In our research, we used a single bit contained in the MAC header for full-duplex transmission. With this, we support deterministic full-duplex transmission between an AP and a node.

Compatible design: Our proposed MAC protocol operates even when nodes with full-duplex radio coexist with nodes with half-duplex radio using DCF in the same network. Nodes with half-duplex radio capability operate well without any modification.

To achieve the above goals, we had to make reasonable assumptions that can be applied in any practical environment.

Bi-directional transmission: As shown in Fig. 1, there are two types of full-duplex radio communication. Bi-directional transmission is an ideal transmission case because an AP and a node communicate without collisions during full-duplex transmission. In contrast, asymmetric transmission can be applied more frequently in practical environments. However, to support asymmetric transmission, an AP must be able to identify the location of hidden nodes. Despite this requirement, it is difficult to completely prevent collisions during full-duplex transmission. Since collisions degrade throughput, we considered bi-directional transmission only in our research.

Padding: If successive transmissions in either the downlink or uplink do not terminate simultaneously, then transmission padding is added to the shorter packet so that both packets terminate simultaneously as shown in Fig. 2.

Fig. 2.

Padding transmission example for full-duplex radio.

Hidden node problem. We assumed that the problem of hidden nodes was negligible because a full-duplex radio AP is able to detect hidden nodes without any additional method (e.g., RTS/CTS) as depicted in Fig. 3.

Fig. 3.

Principle of hidden node detection in full-duplex WLAN (a) uplink is the preceding transmission (b) downlink is the preceding transmission.

Given the above assumptions, our RMAC achieves high throughput compared with the DCF technique based on both half-duplex radio and full-duplex radio.

The remainder of the paper is organized as follows. We briefly discuss related research in Section 2, our RMAC protocol design is described in Section 3, evaluation results are presented in Section 4, and conclusions in Section 5.

2. Related work

Most MAC protocols for full-duplex radio support asymmetric transmission. Because of the collision probability during asymmetric transmission, nodes should be hidden during asymmetric transmission. In most schemes, to realize asymmetric transmission, information exchange between an AP and a node is required. In [8], nodes used a busy-tone signal during the time between reception of the MAC header and the data for information exchange. In [18], nodes exchanged signatures before sending a packet. In [8,18], it was important to synchronize nodes to implement both time-sensitive schemes. In [4], nodes used RTS and modified CTS (FCTS) to set link-establishments for both bi-directional and asymmetric transmission. Before sending a packet, nodes (bi-directional and asymmetric) send an RTS or an FCTS signal. RTS/FCTS signals in [4] suppressed collisions during asymmetric transmission despite the RTS and FCTS exchange overhead.

In [15,16], the authors claimed that nodes could decode by using a capture effect method. Using a capture effect method, when a node receives two colliding packets, it is able to decode the stronger packet of the colliding packets. In these schemes, nodes use a modified RTS frame to report the signal-to-interference ratio (SIR) and make an SIR map for the establishment of asymmetrical dual links. With this information, an AP can determine whether a node receives a packet from the AP or is sending a packet to the AP.

In [10,12], an orthogonal frequency division multiplexing (OFDM) subcarrier was used to solve the transmission inefficiency resulting from uneven packet sizes between the uplink and downlink. This scheme supports asymmetric transmission without information exchange for finding hidden nodes. By using an OFDM subcarrier, nodes receiving a packet from an AP can recognize collisions. Although this scheme offers additional transmission opportunities, when the nodes are densely deployed, throughput improvement is limited.

In [11], Janus, an AP-based MAC protocol with reservation like our RMAC, was employed. Janus consists of three phases. In Phase 1, the scheduling preparation period, an AP queries all nodes to determine whether they have packets to send. After gathering information from nodes, the AP schedules the transmission sequence and transmission time for nodes and then broadcasts a scheduling packet containing the schedule information. In Phase 2, the AP and nodes exchange packets by following the AP schedule. In this period, asymmetric transmission is allowed because the AP has scheduled and notified all nodes as to when they are to start sending a packet. In Phase 3, the AP and nodes transmit a Request Acknowledgement (RA) and ACK during packet-exchange period. Since the AP is centralized, the Janus protocol supports asymmetric transmission with efficient performance, however, it has limitations. First, the Janus protocol requires excessive overhead because an AP must gather information packets from all candidate nodes. In addition, the AP schedules nodes efficiently, even if the transmissions of nodes are overlapped. This requires a highly complex AP. Also, as the number of nodes increases, the information gathering and scheduling time increases. Second, since Janus is a full-duplex MAC protocol, it is not compatible with the half-duplex DCF technique. Our RMAC, however, not only minimizes the information gathering overhead but also enhances compatibility. We summarize the characteristics of the full-duplex MAC protocols surveyed in this section in Table 1.

Table 1
Full-duplex MAC protocol characteristics

MAC Control overhead Information gathering MAC architecture Scheduling Optimization

RCTC [18] fixed no distributed no no

RTS/FCTS [4] fixed no distributed no no

A-duplex [16] fixed yes distributed no no

Janus [11] variable yes centralized yes yes

MAC	Control overhead	Information gathering	MAC architecture	Scheduling	Optimization
RCTC [18]	fixed	no	distributed	no	no
RTS/FCTS [4]	fixed	no	distributed	no	no
A-duplex [16]	fixed	yes	distributed	no	no
Janus [11]	variable	yes	centralized	yes	yes

3. MAC protocol design

Our RMAC protocol based on the node reservation approach was designed for full-duplex WLANs. Because our RMAC uses a centralized AP scheme, APs need to recognize when a node has a packet to send for deterministic full-duplex transmission. In addition, an AP negotiates the sequence of transmission to nodes and the time to transmit. In this section, we describe our experimental environment and the details of our RMAC.

3.1. Overview

In our RMAC scheme, we assume that the APs are equipped with in-band full-duplex radios and the nodes are equipped with either full-duplex radio or half-duplex radio. Our RMAC scheme has two types of periods, one being a contention period for nodes equipped with both a full-duplex and a half-duplex radio, and the other a reservation period for nodes equipped with a full-duplex radio only.

During the contention period, the AP operates in half-duplex communication mode when it communicates with a node equipped with a half-duplex radio. On the other hand, when the AP communicates with a node equipped with full-duplex radio, it performs full-duplex transmission. During communication with full-duplex nodes, the AP gathers information to determine whether a node has more than one frame in its buffer. To minimize the information gathering overhead, we used the More Data bit in the frame control field of the 802.11 MAC header (Fig. 4) [9]. In the previous 802.11 standard, the More Data bit was used to notify a node using Power Save (PS) mode that an AP had more data to send. In our RMAC protocol, we used the More Data bit to check whether a node buffer contained several frames. If the node sends a frame with its More Data bit set, then it considers itself reserved and does not compete during the contention period because our RMAC protocol promises reserved nodes a transmission opportunity without competition or collisions. As a result, the number of competing nodes is reduced.

Fig. 4.

Structure of Frame Control field in 802.11 MAC header.

During the reservation period, the AP sends a frame (data or poll) sequentially to a reserved node. When the node receives a frame from the AP, it has an immediate full-duplex transmission opportunity. If the AP sends and receives frames with all reserved nodes, the AP sets the time for the next reservation period and automatically begins the contention period. In our RMAC protocol, since all nodes do not require information for each period, the AP does not send a notification control frame for each period.

3.2. RMAC protocol design

Since a reservation period is required only for full-duplex transmission, it is important to determine the duration of each period. Although nodes communicate with APs without collision in the reservation period, it is also important to reduce collisions in the contention period because half-duplex nodes, unreserved full-duplex nodes, and new incoming nodes can use a channel only during the contention period. In our RMAC protocol, we classified nodes as reserved and unreserved. Reserved nodes include nodes that succeed in sending a frame with the More Data bit set as 1.

3.2.1. Duration

During the reservation period, the AP sends a frame to all reserved nodes consecutively. Predicting the time of the reservation period is difficult because the AP cannot estimate the uplink duration of a reserved node. Alternatively, our RMAC protocol defines the start time of the reservation period as $T_{r}$ .

We define T as the current time and $T^{'}$ as the reservation period finish time. $T^{'}$ is updated and $T_{r}$ is calculated when a reservation period finishes. When the AP starts, it sets $T^{'}$ to T to initiate reservation period. Then the AP calculates $T_{r}$ . $| N_{all} |$ is the total number of nodes, and $| N_{r} |$ is the number of reserved nodes. $T_{DIFS}$ is the duration of the DIFS (DCF interframe space) interval, and $T_{Slottime}$ is the duration of the slot time.

Since the reservation and contention periods are iterated, to minimize the wasting time when no nodes contend, we define $| T_{w} |$ , the interval time between two reservation periods, as Eq. (1) $\begin{matrix} (1) & | T_{w} | = T_{DIFS} + (| N_{all} | - | N_{r} | + 1) \times T_{Slottime} \end{matrix}$

Then, $T_{r}$ is calculated by Eq. (2). Note that, $| T_{w} |$ is the minimum duration for the contention period. $\begin{matrix} (2) & T_{r} = T^{'} + | T_{w} | \end{matrix}$

Furthermore, if all nodes associated with an AP use the full-duplex radio mode and are reserved, then the number of unreserved nodes ( $| N_{all} | - | N_{r} |$ ) is 0. Even if $| N_{all} | = | N_{r} |$ is satisfied, an AP needs time of at least $T_{DIFS} + T_{Slottime}$ to associate with new incoming nodes because they follow DCF protocol to associate with an AP [3]. $| T_{w} |$ is recalculated whenever the reservation period finishes because the number of reserved nodes can change.

$T_{r}$ is determined by Eq. (1) and Eq. (2). A reservation period can begin only if the following conditions in Eq. (3) are verified: $\begin{matrix} (3) & \{\begin{matrix} T > T_{r} \\ N_{r} > 0 \\ T_{idle} > T_{PIFS} \end{matrix} \end{matrix}$ where $T_{idle}$ is the duration time that a channel is kept idle, and $T_{PIFS}$ is the duration of the PIFS (PCF interframe space) interval.

The AP waits during $T_{PIFS}$ when it begins the reservation period in order to avoid collisions with the ACK frame or data frames during the contention period (see Fig. 5).

Fig. 5.

Example of beginning reservation period for each case: (a) If the channel is busy when the time has arrived to begin the reservation period, the AP waits until the channel is idle and then starts the reservation period after sensing the channel during PIFS; (b) If the channel is idle when the time has arrived to begin the reservation period, the AP waits during PIFS to begin the reservation period; and (c) If the channel is busy when the AP senses the channel state during PIFS, it waits until the channel is idle and operates like case (a).

The reservation period is finished when the AP sends a frame to every reserved node or when the AP fails to receive an ACK frame during the reservation period.

When the reservation period finishes, the AP begins its contention period. If a channel is idle during the DIFS, the AP and unreserved nodes begin competition automatically. Modification of half-duplex nodes is not required because the reservation period is the same as the channel-busy state.

3.2.2. Competition and transmission

There is no competition during the reservation period. The AP sends a data frame first. Then, the node receiving the frame starts sending a data frame with no competition. The AP continues to send one frame to all the reserved nodes. If the AP does not have a data frame for a reserved node, it sends a null data frame (MAC header plus padding bits only) to provide transmission opportunity for other reserved nodes.

If a node has more frames in its buffer, it can retain its reserved state by setting its More Data bit to ‘1’, thereby allowing it to have a transmission opportunity at the next reservation period without the threat of collision. On the other hand, if a reserved node has no frames in its buffer, it sets its More Data bit to ‘0’.

During the contention period, similar to the DCF technique, only unreserved nodes and the AP compete for the right to transmit. Since reserved nodes do not compete during the contention period, the collision probability decreases because the number of competition nodes has reduced. In this period, if a full-duplex node receives a frame, it has an opportunity to transmit. During transmission, nodes check whether or not they have more frames in their buffers. During the contention period, new reserved nodes can be added. The overall operation is described in Fig. 6.

Fig. 6.

Operation example with three full-duplex nodes (S1, S2, S3) and one half-duplex node (S4).

3.3. Overhead

In this section, we compare network overhead during competition in several cases. We compared half-duplex and full-duplex based on the DCF technique and the RMAC protocol. We assumed that half-duplex nodes and full-duplex nodes followed the IEEE 802.11 DCF without the use of RTS/CTS signals. In addition, we used an expected value of contention window (CW) defined as $E [CW]$ (i.e., $E [16] = 8$ ) for simplification. The denotations used in our analysis are provided in Table 2.

Table 2
Denotation for analysis

Parameter Description

$N_{all}$ The number of all non-AP nodes

$N_{f}$ The number of full-duplex nodes (without AP)

$N_{h}$ The number of half-duplex nodes (without AP)

$N_{r}$ The number of reserved nodes (without AP)

$N_{nr}$ The number of unreserved nodes (without AP)

$T_{DL}$ Downlink frame transmission time

$T_{UL}$ Uplink frame transmission time

$T_{ACK}$ ACK frame transmission time

$T_{DIFS}$ Duration of DIFS ( $SIFS + 2 slot times$ )

$T_{PIFS}$ Duration of PIFS ( $SIFS + slot time$ )

$T_{SIFS}$ Duration of SIFS

$T_{slottime}$ Duration of slot time

Parameter	Description
$N_{all}$	The number of all non-AP nodes
$N_{f}$	The number of full-duplex nodes (without AP)
$N_{h}$	The number of half-duplex nodes (without AP)
$N_{r}$	The number of reserved nodes (without AP)
$N_{nr}$	The number of unreserved nodes (without AP)
$T_{DL}$	Downlink frame transmission time
$T_{UL}$	Uplink frame transmission time
$T_{ACK}$	ACK frame transmission time
$T_{DIFS}$	Duration of DIFS ( $SIFS + 2 slot times$ )
$T_{PIFS}$	Duration of PIFS ( $SIFS + slot time$ )
$T_{SIFS}$	Duration of SIFS
$T_{slottime}$	Duration of slot time

In our analysis, one round consisted of $N_{all}$ times of downlink and $N_{all}$ times of uplink, therefore, $2 \times N_{all}$ data frames were exchanged in one round. We analyzed the overhead of each case for K rounds.

When a node sends a frame in the half-duplex mode, the transmission time T was calculated with Eq. (4) as referred to in [2], $\begin{matrix} (4) & T = T_{DIFS} + E [CW] + (T_{DL} ∣ T_{UL}) + T_{SIFS} + T_{ACK} \end{matrix}$

The number of all nodes was set as $N_{all} = N_{h}$ when all nodes were equipped for half-duplex transmission. During one round, at least $2 \times N_{all}$ competitions occurred, and overhead during one round ( $O_{h - 1}$ ) was calculated as $\begin{matrix} (5) & O_{h 1} = (T_{DIFS} + E [CW] \times T_{slottime} + T_{SIFS} + T_{ACK}) \times 2 N_{all} \end{matrix}$

Overhead during K rounds ( $O_{h K}$ ) was simply calculated as $\begin{matrix} (6) & \begin{matrix} O_{h K} & = O_{1} \times K \\ = (T_{DIFS} + E [CW] \times T_{slottime} + T_{SIFS} + T_{ACK}) \times 2 K N_{all} \end{matrix} \end{matrix}$

If half-duplex radio nodes successfully transmitted without collision, the overall overhead was calculated with Eq. (6). If we consider collisions during DCF, then $2 K N_{all}$ in Eq. (6) is replaced by ( $2 K N_{all} + N_{col}$ ), where $N_{col}$ is the number of collisions depending on the number of nodes.

The overhead of full-duplex with DCF ( $O_{f 1}$ ) and that of RMAC ( $O_{r 1}$ ) are the same at the first round because the number of reserved nodes is zero.

Therefore, the competition overhead of both cases is calculated as $\begin{matrix} (7) & O_{r 1} = O_{f 1} = (T_{DIFS} + E [CW] \times T_{slottime} + T_{SIFS} + T_{ACK}) \times N_{all} \end{matrix}$ where the number of nodes is set to $N_{all} = N_{f}$ . In Eq. (7), the full-duplex overhead is halved compared to the half-duplex operation because the number of competitions required for the same number of transmissions is reduced by simultaneous transmission. However, the number of reserved nodes changes as the transmission proceeds, thereby creating a difference between the overhead of full-duplex with DCF and that of RMAC. When K rounds occur, the overhead of full-duplex with DCF is calculated as $\begin{matrix} (8) & \begin{matrix} O_{f K} & = O_{r 1} \times K \\ = (T_{DIFS} + E [CW] \times T_{slottime} + T_{SIFS} + T_{ACK}) \times N_{all} \times K \end{matrix} \end{matrix}$

To calculate the overhead of the RMAC protocol, the number of nodes was set to $N_{all} = N_{f}$ and the number of reserved nodes was set to $N_{r} = N_{f}$ because all nodes will be reserved after the first round finished and traffic is saturated. In this case, the overhead of the RMAC protocol was calculated as $\begin{matrix} (9) & \begin{matrix} O_{r K} & = (T_{DIFS} + E [CW] \times T_{slottime} + T_{SIFS} + T_{ACK}) \times N_{all} \\ + (T_{PIFS} + T_{SIFS} + T_{ACK}) \times (K - 1) \\ + (T_{SIFS} + T_{SIFS} + T_{ACK}) \times (N_{all} - 1) \times (K - 1) \\ + (T_{DIFS} + T_{slottime}) \times (K - 1) \end{matrix} \end{matrix}$

In Eq. (9), in the first round, all nodes need competition overhead $(T_{DIFS} + E [CW] \times T_{slottime} + T_{SIFS} + T_{ACK}) \times N_{all}$ to be included in the reservation period. After finishing the first round, because all nodes are reserved, the AP begins transmitting after waiting a PIFS time interval during $(K - 1)$ rounds. After sending the first transmission of each round, the AP uses an SIFS time interval to send the downlink frame. When each reservation period is finished, the AP waits a minimum time equal to the sum of the DIFS time interval and one slot time for the contention period. When we use the RMAC protocol, the overhead decrement compared with full-duplex with DCF is calculated as $\begin{matrix} (10) & \begin{matrix} O_{f K} - O_{r K} & = (T_{DIFS} - T_{SIFS} + E [CW] \times T_{slottime}) \times (N_{all} - 1) \times (K - 1) \\ + ((E [CW] - 1) \times T_{slottime} - T_{PIFS}) \times (K - 1) \\ = ((E [CW] + 2) \times T_{slottime}) \times (N_{all} - 1) \times (K - 1) \\ + ((E [CW] - 1) \times T_{slottime} - T_{PIFS}) \times (K - 1) \end{matrix} \end{matrix}$

As presented in Eq. (10), when a round proceeds in saturated traffic, the RMAC protocol reduces overhead more efficiently than full-duplex using DCF.

4. Performance evaluation

In this section, we describe the evaluation of our RMAC protocol to prove the performance improvement. To evaluate RMAC, we developed a WLAN simulator that supports full-duplex radio communication based on SIC by using $C + +$ and compared the RMAC with IEEE 802.11 DCF with half-duplex radio and full-duplex radio. We considered the metrics of normalized throughput and fairness for evaluation. The simulation environment and experimental results are described in Sections 4.1 and 4.2, respectively.

4.1. Simulation environment

When we simulated IEEE DCF with half-duplex radio and full-duplex radio and RMAC, we considered only a single AP with multiple nodes. We assumed that the size of network is $25 \times 25 (m^{2})$ and the coverage of each node including AP is 37 (m) to cover the whole network. The AP was located at the center of the network, and the nodes were located at random positions in the network. During MAC-protocol simulation, we changed the number of nodes from 5 to 60, and we measured the performance degradation when nodes were densely deployed. In addition, we assumed saturated traffic. When the AP and nodes generated packets, the packet size was decided by the distribution shown in Fig. 7 based on [13,14], so the frame size of the uplink and downlink was different. Theoretically, the throughput of the full-duplex radio was twice as large as that of the half-duplex radio if the frame size of the uplink and downlink were equal. However, in a practical environment as shown in Fig. 7, the downlink frame size was larger than that of the uplink. Therefore, the increased downlink traffic of a practical environment achieved a higher throughput. In addition, we considered only a 20-MHz channel bandwidth with a single stream because channel extension schemes can be used in specific environments (i.e., multiple channels are idle).

Fig. 7.

Packet size distribution.

Since we assumed that all nodes were one-hop, all nodes were able to receive or overhear and decode all frames in the network. Therefore, when we simulated IEEE 802.11 DCF with half-duplex radio, we did not employ RTS/CTS. Other simulation parameters are described in Table 3.

Table 3

Parameters for simulation

Description	Value
Duration of simulation	10 (s)
Physical overhead (header, preamble, signal)	40 (μs)
Data rate	65 (Mbps)
Slot time	9 (μs)
SIFS	16 (μs)
PIFS	25 (μs)
DIFS	34 (μs)
Contention Window Size [ $Minimum / Maximum$ ]	[ $16 / 1024$ ]
ACK	14 (bytes)

For precise simulation results, we created two scenarios. First, when we simulated IEEE DCF with half-duplex radio, the AP and all nodes operated only in the half-duplex mode. On the other hand, when IEEE DCF with full-duplex radio and RMAC were simulated, the AP and all nodes had the capability of supporting full-duplex communication.

Second, we compared two MAC protocols (DCF and RMAC). The two MAC protocols operated with various numbers of full-duplex nodes and half-duplex nodes. This scenario proved the ability of full-duplex nodes to coexist with half-duplex nodes.

4.2. Simulation result

In this section, the normalized throughput of each MAC protocol and the average number of uplink transmissions are compared. Figure 8 shows the normalized throughput as the number of nodes was increased. As shown in Fig. 8, the normalized throughput of DCF with full-duplex nodes was 167.7% and 61.4% greater than that of DCF with half-duplex nodes when the number of nodes was 5 and 60, respectively. Since the number of downlinks of full-duplex radio was more than that of the half-duplex radio as mentioned in Section 4.1, the normalized throughput of DCF with full-duplex nodes was more than double that of the half-duplex nodes. In addition, the RMAC protocol throughput with full-duplex nodes only (presented as RMAC-P in Fig. 8) outperformed that of the DCF with full-duplex nodes by 86.3 %. Moreover, with 60 network nodes, the normalized throughput of DCF with full-duplex nodes and half-duplex nodes decreased 56.1% and 27.1%, respectively. However, the normalized throughput using the RMAC protocol increased 5.2% because the number of competing nodes had decreased and the number of transmission had increased, demonstrating that the RMAC protocol operated well even in a network with a dense node deployment.

Fig. 8.

Normalized throughput with saturated traffic for several cases; RMAC-P: full-duplex nodes only, RMAC-C: mixture of the same number of full-duplex nodes and half-duplex nodes.

In addition, Fig. 8 shows the normalized throughput of a network consisting of an AP, full-duplex nodes, and half-duplex nodes (presented as RMAC-C in Fig. 8). During these simulations, we set the number of full-duplex nodes equal to the number of half-duplex nodes to verify that RMAC operates well and maintains throughput when half-duplex nodes coexist. As the number of nodes increased, the normalized throughput using DCF decreased because the number of competing nodes increased. The greater the number of competing nodes, the greater the number of collisions that occurred. On the other hand, using the RMAC protocol shows that as the number of full-duplex nodes increased, the normalized throughput also increased as shown in Fig. 8 even with the same number of half-duplex nodes. Using the RMAC protocol, since the reserved full-duplex nodes did not compete during the contention period, the number of competing nodes is not proportional to the total number of full-duplex and half-duplex nodes, and likewise the collision probability and competition overhead decreased. With saturated traffic, all full-duplex nodes will eventually be reserved and transmit without competition during the reservation period with the RMAC protocol. Therefore, the increment of full-duplex nodes leads the increment of proportion of the reservation period in one cycle consisting of one reservation period and one contention period.

The results of the two scenarios show that the RMAC protocol achieved high throughput when nodes frequently sent many frames because the nodes remained reserved until their buffers were empty. The efficiency of the RMAC protocol was maintained even when a large number of nodes were deployed with heavy traffic.

Figure 9 shows the average number of uplink transmissions of each MAC protocol when the number of nodes is 60. In general, DCF provides long-term fairness among nodes. The number of uplink transmissions with RMAC protocol was 108.6% and 62.2% greater than that of DCF with full-duplex radio and half-duplex radio, respectively. Furthermore, RMAC protocol achieves the minimum deviation of the number of uplink transmissions because nodes already reserved had equitable transmission opportunities during reservation period. As a result, RMAC also provides fairness well among nodes.

Fig. 9.

The average number of uplink transmission for three MAC protocols (DCF w Full-duplex nodes, DCF w half duplex nodes and RMAC) when the number of nodes is 60.

5. Conclusion

The increasing number of nodes in WLANs, the overhead of collision prevention, and the competition among nodes are crucial issues to be dealt with in the development of WLAN communication equipment. Despite the advantages of full-duplex radio technology, most prior research has focused on the feasibility of symmetric and asymmetric transmission for frame exchange.

In this paper, we proposed an RMAC protocol consisting of a reservation-based MAC protocol for a full-duplex equipped WLAN. Our RMAC protocol reduced competition overhead and collision probability by providing transmission opportunities that avoided competition. We demonstrated that our RMAC protocol performed very well with a large number of nodes and heavy traffic. In addition, our RMAC protocol was designed to be compatible with IEEE 802.11 DCF with half-duplex; therefore, we believe our RMAC protocol is suitable for propagating and generalizing full-duplex WLANs.

For the future work, we will consider various aspects such as the performance enhancement, the security, and the applicability to other wireless networks. Since the proposed RMAC covers only symmetric transmissions, it is possible to modify RMAC to support asymmetric transmissions as well. Then the modified RMAC may enhance its throughput by mitigating the collision probability further. What if a node does not follow RMAC protocol? Then the reservation period is not ensured so that RMAC may function just like DCF. It is necessary to protect RMAC from any security breach. Finally, we may consider other wireless networks such as Bluetooth or Zigbee carefully to apply RMAC to if the full-duplex radio is evaluated as a good technology for them in future.

Footnotes

Acknowledgements

This study was supported by Brain Korea 21 PLUS project for POSTECH Computer Science & Engineering Institute. This research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Korea, Under the “ICT Consilience Creative Program” (IITP-R0346-16-1007) supervised by the IITP (Institute for Information & Communications Technology Promotion).

References

Bharadia,

McMilin and

Katti, Full duplex radios, ACM SIGCOMM Computer Communication Review 43(4) (2013), 375–386. doi:10.1145/2534169.2486033.

Bianchi, Performance analysis of the IEEE 802.11 distributed coordination function, IEEE Journal on Selected Areas in Communications 18(3) (2000), 535–547. doi:10.1109/49.840210.

Brenner, A technical tutorial on the IEEE 802.11 protocol, BreezeCom Wireless Communications 1 (1997).

Cheng,

Zhang and

Zhang, RTS/FCTS mechanism based full-duplex MAC protocol for wireless networks, in: Global Communications Conference (GLOBECOM), IEEE, 2013, pp. 5017–5022.

J.I.

Choi,

Jain,

Srinivasan,

Levis and

Katti, Achieving single channel, full duplex wireless communication, in: Proceedings of the 16th Annual International Conference on Mobile Computing and Networking, MobiCom’10, ACM, 2010.

Dong and

Dargie, Analysis of collision probability in unsaturated situation, in: Proceedings of the 2010 ACM Symposium on Applied Computing, ACM, 2010, pp. 772–777.

Ericsson , Ericsson mobility report, [online], 2012, available at: https://www.ericsson.com/res/docs/2012/ericsson-mobility-report-november-2012.pdf.

Goyal,

Liu,

Gurbuz,

Erkip and

Panwar, A distributed MAC protocol for full duplex radio, in: Forty Seventh Asilomar Conference on Signals, Systems, and Computers (ASILOMAR), 2013.

IEEE standard for information technology – Telecommunications and information exchange between systems – Local and metropolitan are networks – Specific requirements. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications, IEEE Std., March 2012.

10.

Kim and

Kim, A full duplex MAC protocol for efficient asymmetric transmission in WLAN, in: 2016 International Conference on Computing, Networking and Communications (ICNC), Kauai, HI, 2016, pp. 1–5.

11.

J.Y.

Kim,

Mashayekhi,

Qu,

Kazadiieva and

Levis, Janus: A novel MAC protocol for full duplex radio, CSTR 2(7) (2013), 23.

12.

Lee,

Ahn and

Kim, In-frame querying to utilize full duplex communication in IEEE 802.11ax, in: Standards for Communications and Networking (CSCN), 2015 IEEE Conference on, Tokyo, 2015.

13.

Packet size distribution comparison between Internet links, in 1998 and 2008, Available at: www.caida.org/research/traffic-analysis/pkt_size_distribution/graphs.xml.

14.

Palit,

Naik and

Singh, Anatomy of WiFi access traffic of smartphones and implications for energy saving techniques, International Journal of Energy, Information and Communication (IJEIC) 3(1) (2012), 1–16.

15.

Tang and

Wang, Medium access control for a wireless LAN with a full duplex AP and half duplex stations, in: Global Communications Conference (GLOBECOM), IEEE, 2014, pp. 4732–4737.

16.

Tang and

Wang, A-duplex: Medium access control for efficient coexistence between full duplex and half duplex communications, Wireless Communications, IEEE Transactions on 14(10) (2015), 5871–5885. doi:10.1109/TWC.2015.2443792.

17.

H.L.

Vu and

Sakurai, Collision probability in saturated IEEE 802.11 networks, in: Australian Telecommunication Networks & Applications Conference (ATNAC), Australia, 2006.

18.

Zhou,

Srinivasan and

Sinha, RCTC: Rapid concurrent transmission coordination in full duplex wireless networks, in: 2013 21st IEEE International Conference on Network Protocols (ICNP), IEEE, 2013, pp. 1–10.