Extraction of The Rice Leaf Disease Image Based On BP Neural Network

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2841888, IEEE Access
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2018.DOI
A Polling-based Traffic-aware MAC

Protocol for Centralized Full-duplex
Wireless Networks
SONG LIU, BIAO HAN (Member, IEEE), and WEI PENG (Member, IEEE).
National Laboratory for Parallel and Distributed Processing, National University of Defense Technology, Changsha, Hunan, China, 410000.
Corresponding author: Wei Peng, Biao Han (e-mail: {wpeng, nudtbill}@nudt.edu.cn).
The work is supported by the National Natural Science Foundation of China under Grant No.61601483.
ABSTRACT Recent advances in self-interference cancellation techniques have enabled in-band full-
duplex (IBFD) transmission, which can double ergodic capacity and reduce end-to-end delay. However,
how to solve the channel contention problem in IBFD radios with inter-node interference and asymmetric
traffic is still a challenge. This paper presents pFD-MAC, a novel polling-based traffic-aware medium access
control (MAC) protocol for centralized full-duplex wireless networks. To solve the channel contention
problem and fully utilize the channel resources under full-duplex mode, we first design a novel polling-based
transmission mechanism and make comprehensive investigations on the effect of polling profile in full-
duplex communication. By characterizing the inter-node interference into a directed non-conflict graph, we
study the polling profile generation problem in which our objective is to minimize the packet transmission
time. The problem is then theoretically formulated and proved to be NP-hard, which means it cannot be
solved in polynomial time. Thus, we develop a heuristic traffic-aware algorithm and apply it to work with
the packet transmission procedure in parallel. Full-duplex communication opportunities are highly exploited
by parallelly organizing per-node upstream/downstream traffic according to the generated polling profile.
Moreover, to achieve fairness without sacrificing throughput, deficit round robin algorithm has been applied
with respect to the access time considering concurrent transmission time. Simulation results reveal that our
proposed protocol can achieve improved performance in terms of throughput and transmission delay while
maintaining fairness, compared with two state-of-the-art centralized MAC protocols.
INDEX TERMS Full duplex, MAC protocol, inter-node interference, asymmetric traffic, fairness
I. INTRODUCTION gation and fairness need to be well investigated at the MAC

Traditional radios operate in half-duplex mode, in which a layer. Specifically, how to resolve contention problem with
wireless device can not receive and transmit on the same inter-node interference and asymmetric traffic in order to
frequency at the same time because of self-interference (SI). improve the full-duplex system performance is still a chal-
Most current wireless systems rely on orthogonal signaling lenging issue. To this end, many MAC protocols have been
division techniques, such as time division duplex (TDD) and proposed. These protocols are either for centralized wireless
frequency division duplex (FDD) [1]–[3]. Recent advances in networks or ad-hoc wireless networks [5]–[9]. In centralized
antenna design and self-interference cancellation (SIC) tech- full-duplex wireless networks, mobile terminals connect to
niques have enabled radios to operate in full-duplex mode the network through access points (APs) or base stations
on a single channel with very low residual self-interference which act as coordinators or central controllers, in which
in the physical layer, called in-band full-duplex (IBFD) trans- full-duplex communications can be managed and scheduled
mission [4], [5]. An IBFD system can simultaneously receive efficiently with the aid of centralized APs.
and transmit packets on the same frequency band. Ideally, With the introduce of IBFD radios, the channel contention
it can improve the spectral efficiency by duplexing ergodic problem become more complicated than that in half-duplex
capacity and reduce end-to-end delay accordingly. mode, especially when inter-node interference and asymmet-
Although IBFD transmission is practical at the physical ric traffic are jointly considered. Here, the asymmetric traffic
layer [4], [5], issues like channel contention, collision miti- refers to traffic that can be used for asymmetric dual links.
VOLUME 4, 2018 1
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
Song Liu et al.: A Polling-based Traffic-aware MAC Protocol for Centralized Full-duplex Wireless Networks
(AP → S2,3,4 and S1,4 → AP ) in half mode. And it is

Interference 1
S3 unfair to S1 because it suffers more interference. However,
Interference 2
i i th transmission S1 will get less access opportunity and access time since
Si: mobile station 4
AP: Access Point
S1 there are less concurrent transmission opportunities for S1
3
in full duplex communications [15]. For example, except
1
S4 2
AP for transmitting alone like in half-duplex mode, S4 can
5 concurrently transmit packets with S1,2,3 while S1 can only
S2
start a concurrent transmission with S4 . As a result, S4 can
FIGURE 1. Illustrative example for exploiting full-duplex communications in a obtain more transmission opportunities and S1 get fewer
wireless network with centralized AP and multiple stations. transmission opportunities, leading to a deterioration of the
Oi i th packet from AP Ii i th packet to AP Ai i th Acknowledgement packet
Collecting information ACK period Ti Transmission time saved
fairness. More details will be discussed in Section IV.
To resolve the above issues, we propose a novel polling-
O1 O3 + A1 O4 incur more collisions O2+A4
I1 First, inter-node interference
I3+A3 would I4 + A4 based
I2+A2 traffic-aware MAC protocol for centralized full-duplex
[10]–[12]. As illustrated in Fig.1, considering the case that:
(a) PCF wireless networks, named pFD-MAC. In order to solve the
when S1 is uploading
Scheduling O4 packetsO3to AP (S1O→ 2
AP ) and AP channel contention problem, we design a polling-based trans-
packeta concurrentI4downstream
starts I1 transmission to S2 (AP → mission scheduling mechanism incorporated with pipelined
S2 ), interference exists(b)between Janus mobile stations S1 and S2 . packet transmission for centralized APs. To solve the inter-
O1 If the full-duplex
O3 A1 transmission
O4 Aopportunity
4 O2 is exploited, there node interference problem, after characterizing the inter-node
I1 be a collision
will I3 A2 at S I4since it
2
I2
A4can overhear Athe
3 T1 from T2
signal interference into a directed non-conflict graph, we study the
S1 . Nevertheless,(c) if Our proposed protocol
the interference is weak, one can choose polling profile generation problem in which our objective is
a low data rate to transmit packet, as addressed in [8]. For to minimize the packet transmission time. Specifically, the
example, if interference from S1 to S3 is weak, transmission formulated problem can be transformed to the conventional
AP → S3 is allowed as S3 can tolerate the interference from traveling salesman problem (TSP) with dynamic cost, which
S1 at a low data rate. Therefore, the inter-node interference is proved to be NP-hard. Thus, we develop a heuristic traffic-
affects whether the concurrent transmission can be exploited aware algorithm to solve the problem and apply it to work
and what data rate should be selected. with the packet transmission procedure in parallel. pFD-
Secondly, as asymmetric traffic is quite common in wire- MAC highly exploits full-duplex opportunities by effectively
less communications, it is potential to exploit more full- organizing per-node upstream/downstream traffic in parallel
duplex communication opportunities while it has not been according to the generated polling profile. Moreover, to solve
fully utilized in most current MAC protocols. In Fig.1, it the fairness issue, pFD-MAC refines the fairness metric in
is possible to schedule either S1 → AP or S4 → AP in full-duplex communication and adopts deficit round-robin
parallel with AP → S2 , AP → S3 , or AP → S4 . However, (DRR) algorithm, which can achieve high level fairness and
the channel contention issue under asymmetric traffic cannot maintain good network performance. Through extensive sim-
be resolved effectively in traditional CSMA (Carrier Sense ulation, we validate that our proposed pFD-MAC protocol
Multiple Access) based MAC protocols. The reasons are can improve throughput by 39% and 69% and reduce the
as follows: 1) If AP → S2 starts first, S1 and S4 will average packet delay by 48% and 33%, comparing with Janus
not upload packets for they sense the channel is busy. 2) [8] and IEEE 802.11 PCF (Point Coordination Function)
Assume that there is a symmetric dual link between AP and [16], respectively.
S4 (AP ↔ S4 ). If S4 → AP ends first, S1 → AP can The main contributions can be summarized as follows:
be scheduled. But S1 will not upload until the symmetric • We explore the channel contention problem with inter-
dual link is over since it senses the channel is busy. 3) If node interference and asymmetric traffic for centralized
AP → S4 ends first, S1 senses the channel is idle and full-duplex wireless networks. To the best of our knowl-
will start uploading, resulting in a collision at AP. Jian et edge, pFD-MAC is the first full-duplex MAC protocol
al. in [13] use a busy tone to avoid collisions, but may that highly utilizes the parallelism between transmission
suffer from lower channel utilization ratio as the busy tone scheduling and packet transmission bridged by our pro-
contains no data. Thus, how to schedule transmissions based posed polling-based mechanism.
on asymmetric traffic to fully exploit the potential of full- • We first design a novel polling-based transmission
duplex communication remains a challenge. scheduling mechanism to solve the channel contention
In addition, the fairness among nodes in a network should problem, and make comprehensive investigations on the
be addressed [14]. Fairness issue is a classical problem in effect of polling profile in full-duplex communication.
traditional wireless network, mainly caused by the CSMA • We study the polling profile generation problem by
mechanism. Again, this issue will be aggravated in full- characterizing inter-node interference into a directed
duplex communication because the inter-node interference non-conflict graph, which is then theoretically formu-
is more serious and scheduling patterns are more complex lated and proved to be NP-hard. We develop a heuristic
than in half-duplex mode. As shown in Fig.1, if S1 intends to algorithm to solve the problem and apply it to work with
upload packet, it only needs to contend in five transmissions the packet transmission procedure efficiently.
2 VOLUME 4, 2018
i i th transmission
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content mayS :change prior to final4 publication. Citation information: DOI
mobile station
i
S1
10.1109/ACCESS.2018.2841888, IEEE Access AP: Access Point 3
S4
1
AP Step
2 O
Song Liu et al.: A Polling-based Traffic-aware MAC Protocol for Centralized Full-duplex Wireless Networks 5
S2
Step
O
Step
O
•We first consider the concurrency transmission in fair- Oi i th packet from AP Ii i th packet to AP Ai i th Acknowledgement packet
ness metrics to encourage full-duplex opportunities, O1 O 3 + A1 O4 O2+A4
I1 I3+A3 I4 + A4 I2+A2
thereby maintaining a high-level fairness while achiev- (a) PCF
ing good network performance. Scheduling O4 O3 O2 Step
packet I4 I1
The rest of the paper is organized as follows. Section (b) Janus
Step
II introduces the system model and problem description. O1
I1
O3
I3
A1
A3
O4
I4 A4
O2
I2
A4
A2 T1 T2
Section III describes our proposed pFD-MAC protocol. The (c) Our proposed protocol
polling profile generation problem is studied and a heuristic FIGURE 2. Transmission procedure of IEEE 802.11 PCF, Janus, and our
algorithm is proposed in Section IV. Experiment results are proposed pFD-MAC
presented in Section V and the paper is concluded finally in LHeader to station
Section VI. Separate Transmission：AP O A O A
I I time
B. PROBLEM DESCRIPTIONO DT Acknowledgement
Concurrent Transmission：AP O A A
II. SYSTEM MODEL AND PROBLEM DESCRIPTION Traditional CSMA-based MAC IprotocolsI [16]–[19] from station time
have
A. SYSTEM MODEL been widely deployed in wireless LANs (WLAN), and many
1) Network and transmission model efforts have been done to extend them for full-duplex wire-
In this paper, we consider an AP-based wireless network less networks [5], [8], [11]–[13], [20]–[23]. However, CSMA
Beacon: Contention-free period starting packet; CF-P: Contention-free Polling packet; ACK: Acknowledgeme
in which the AP is deployed at the center and stations are will reduce the chance
CF-End: Contention-free ofpacket;
period ending full-duplex transmission
CF-A: Contention-free and the
Acknowledge packet; SIFS: Short InterFram
randomly distributed in a fixed area. Both AP and mobile capacity gain [24], [25]. PCF is anotherContention-free half-duplex repetitionMAC
interval
Contention-free period
Contention
stations in the network are equipped with omni-directional technique in IEEESIFS802.11 WLAN
period
SIFS standards,
SIFS which exploits
SIFS SIFS SIFS
CF-P + Data CF-P ACK CF-P + Data A
antennas and are capable of IBFD transmissions. Considering polling
AP
mechanism to
Beacon provide real-time communication ser-
Information (to Station 1) (to Station 2) (to Station 1) (to Station 3) (to S
collection CF-A + Data ACK CF-A + Data CF-A
that a full-duplex link can achieve around 1.84 capacity gain vice. However, it can not be applied to full-duplex com- (from Station 1) (from Station 1) (from Station 2) (from Station 3)
SIFS SIFS SIFS SI

compared with a half-duplex link [10], we assume perfect munications directly as issues like inter-node Set by CF-P
interference Set by CF-P
NAV
self-interference cancellation technique is achieved at PHY- and asymmetric traffic are not considered. As illustrated in Set by Beacon
layer. And this paper focuses on the network-level capacity. Fig.2(a), consider the transmission procedure for Fig.1, S1
To exploit full-duplex communication opportunities, there and S3 cannot initiate concurrent transmission as there is
are two transmission modes in a full-duplex wireless net- inter-node interference between them. Actually, if a low data
work: symmetric transmission and asymmetric transmission. rate is selected, concurrent transmission can be scheduled for
For instance, as shown in Fig.1, when AP and S4 transmit S1 and S3 . But PCF did not utilize this full-duplex transmis-
packets to each other simultaneously (S4 ↔ AP ), we call sion opportunity. Kim et al. [8] provides a centralized MAC
this transmission mode symmetric transmission. When AP protocol called Janus, in which information is collected at
transmits a packet to S3 and receives from another station first and a communication plan is then calculated and packets
S1 (S1 → AP → S3 ), we call this transmission mode are transmitted according to the plan, as shown in Fig.2(b).
asymmetric transmission. In this paper, we consider both By this way, network throughput can be improved. However,
symmetric and asymmetric transmissions. this approach introduced heavy delays as an explicit stage
is required to collect information and schedule the trans-
2) Propagation and interference model missions. Therefore, it needs to schedule more packets to
reduce per-packet overhead, but this will bring more delay.
The propagation model is assumed with deterministic power
Here, an important but challenging question is raised: Can
attenuation and Rayleigh fading is used in this paper. Assume
high network throughput and low delay be achieved in IBFD
that Pt is the transmit power and the distance from a trans-
systems at the same time?
mitter to a receiver is d, then the received signal strength Pr
In order to achieve this goal, three problems need to be
is:
addressed: 1) A novel packet transmission mechanism should
Pr = κPt hd−n (1)
be developed which is independent of scheduling process
where κ is the propagation constant and h is the channel and can maximize the channel utilization, because traditional
coefficient, and both of them are assumed with the fixed transmission mechanisms like CSMA/CA or PCF are not
value. n is the path loss exponent, and we set n = 4 for suitable for IBFD systems. 2) The centralized AP should be
Rayleigh fading in this paper. aware of the inter-node interference and traffic information,
Multi-rate transmission is supported and a mobile station which need to be collected before the packet transmission
can adjust its transmission rate to adapt the communication procedure. 3) The packet scheduling process should work
environment. That is, if the interference is strong, it will with the packet transmission procedure in a parallel manner,
select a lower data rate. Otherwise, a higher data rate will be so that the overhead of scheduling can be eliminated.
selected. In this paper, we define different interference levels In this paper, we consider a polling-based transmission
according to the SIR (Signal-to-Interference Ratio) values. scheme, which is similar to the process of PCF, but optimized
And different transmission rates can be selected according for IBFD transmission. Specifically, according to a generated
to the constructed SIR map. The details of SIR map will be polling profile, an AP can receive packets from a mobile
introduced in Section IV. station and transmit packets to another mobile station simul-
VOLUME 4, 2018 3
for publication in a future issue of this journal, …
t
This article has been accepted
Packets transmission: but has not been fully edited. Content may change prior to final publication. Citation information: DOI bytes 2
)
Packet format: Frame D
Beacon:packets
Handle broadcastaccording
the duration to
time of Polling
the list in a period 10.1109/ACCESS.2018.2841888, IEEE Access
the contention-free
1. 尽可能的使用不干扰
Control
CF-P: Contention-free polling packet
pipeline-like way
CF-A: Contention-free Acknowledge packet
bytes 6
Song Liu et al.: A Polling-based 6
Traffic-aware MAC2 Protocol
1 2. 对于对称双连接可以
for Centralized Full-duplex Wireless Networks
ACK: Data Acknowledge packet Dst Src 3. 首先轮询有上传流量
bytes
type Additional information
address address MAC
Contention period: Network management, station join in and exit, etc.
Contention Period Contention-free Period is specifiedbytessimilar1 with [8]. 1 All
1 stations
1 1keep listening and
Contention-free peirod:
Information
Packet
Station 1 SIR1 Station n SIRn
Greedy Alg
Network Polling-based scheduling sense the signal strength
length from the AP. When stations receive
management: collection: Polling profile
Stations enter or exit
Collect interference Traffic information this Probe packet, they will return the CF-A packet one by
Probe Probe informationProbe
and Probe AP CF-P
AP
the network
(to Station 1) SIFS (to Station 2)traffic
SIFSinformation Packet
(to Station 3) SIFS (to Station 4) transmission
SIFS
… one. At this time, when a station is uploading the CF-A CF-A
Can use the maximal transmission rate
CF-A
(from Station 1)
CF-A
(from Station 2)
CF-A
(from Station 3)
CF-A
(from Station 4)
time packet, its neighbors can sense the interference level from Packet f
bytes 1 1 2 1
FIGURE 3. The workflow of pFD-MAC
Contention period Contention-free period
Initial stage this station, so they can calculate the SIR value for this
Dst ID
station. Src ID
Here, wetype
use theAdditional
Received Signal Strength IndicatorAP bytesP3 6
information
Network management, such as Information Collector: Polling Scheduler: Dst
Stations enter or exit the (RSSI) in the MAC layer to represent the signal/interference FA3
Probe Collect information of the Generate a polling list … addres
(S 1->S2->S3
AP SIFS SIFS SIFS SIFS bytes 1 1 1 1 1
Polling list
network. network, such as interference with minimal
CF-A
level, packets toCF-A
be uploaded. CF-Atransmission time
CF-A time strength. Then the SIR value
Packet cij 1atStation
Station 1 SIR
station j with interference
n SIRn
Packets (from Station 1) (from Station 2) (from Station 3) (from Station 4)
transmission: from station i canlength be calculated by:
… AP P3
Exchange packets using
)
FIGURE 4. Information collection
SIFS procedure in the contention-free
Packets transmission:
SIFS SIFS period
competition mechanism, FA3
such as CSMA/CA. AP RRI1 Handle packets according
RRI2 to the Polling
RRI3 list in a pipeline-like way
RRI4 time cij = Prap (i)/Pri (j) (2)
taneously. Thus, it works in a pipeline-like way to accelerate where Prap (i) is the signal strength at station i from the APAP P3
(2) the Transmission
transmission process.procedure Secondly, unlike Janus in which the and Pri (j) is the interference signal strength at station i from FA3
traffic information Beacon

is collected DIFS
by sending
Numbering probe packets, we station j. Note that, if the interference strength of station i
…
attach the AP traffic DIFS
(probe request)
informationPR packet;
with datastations
PR CF-P:DIFS
packets to be uploaded is weak enough, station j may fail to detect the signal so
Beacon: Contention-free period starting Contention-free Polling packet; ACK: Acknowledgement packet for Data;
Heuristi
… time AP P3
to APs.
CF-End: Considering
Contention-free that packet;
period ending the uplink
Probe
Response 1 CF-A:channel
Probe
Contention-free
Response 2
and downlink
Acknowledge packet; thatSIFS:there
Short is no SIR
InterFrame Spacerecord about i at station j. The AP will
FA3+
channel are entangled in full-duplex transmissions, stations regard that there is no interference from uploading station i
Contention-free repetition interval
in the network cannot distinguish Contention theperiodinterference signal from
while downloading packets to station j. These SIR records
Contention
their neighbors.SIFSBesides, SIFS since stations SIFSdo not move SIFS very SIFS will be included SIFS in theSIFSCF-A packet and uploaded to the AP,
period O
frequentlyBeacon in most Station ID:
cases, theCF-P interference
+ Data 1 information CF-P
2
ACK will
3 as shown
CF-P + Data in Fig.5(c). ACK CF-P
Numbering MAC Header (to Station 1) Address(to1 Station Address 2 (toAddress
Station 1)3 … CF-End S
Informationperiod of time, such as 100ms. Thus, 2) (to Station 3) (to Station 2) (to Station 4)
Here, we use a frame to upload the interference informa- Step 0:
AP not changestations for a certain collection CF-A + Data ACK CF-A + Data CF-A ACK CF-A time O1
… ID is broadcasted AP
interference information can be collected (from Station 1) periodically.(from Station 1) Third- (from Stationtion
2) to(from
AP,Station
as 3)shown inStation
(from Fig.5(c).
3) The
(from station
Station 4) I1
by the AP in the contention period. We use one byte to Step 1:
ly, since theWhen AP there
transmits
are stations packets
join in or according SIFS toAP
exit the network, the polling
numbersSIFS each station. SIFS SIFS SIFS
O1
profile, we can schedule packet transmissions by acting on
Set by CF-P
representSetthe SIR value, as Tang et
by CF-P
al. have done in [20]. The
Set by CF-P
AP
I1
NAV the polling profile. In this way, the scheduling process and first bit is used to indicate the sign of SIR and thetime last 7 bits Step 2:
Set by Beacon O1
the packet transmission procedure can be decoupled and give an absolute SIR value. Thus, the SIR value ranges from AP
I1
operated in parallel. The key idea of our proposed protocol −64dB to 64dB, which is enough for a practical wireless AP
O1
is shown in Fig.2(c). Probe SIFS SIFS SIFS SIFS system. Besides, someone may say that why don’t you use1.the 尽可能的使用
Step 3:
(Registered node)
AP
AID (association ID) in IEEE 802.11 standard to represent2.
AP
a
对于对称双连 O1
I1
III. THE PROPOSED PFD-MAC CF-A CF-APROTOCOL CF-A CF-A CF-A station. I think it’s also a solution to label a station. 3. 首先轮询有上
However, Step 4:
(from (from (from (from (from time
In our proposed pFD-MAC Station 1)
protocol,
Station 2) Station 3)
contention period
Station 4)
(CP)
Station 5) AID shares with the Duration field in the frame as it’s useless AP
O1
I1
and contention-free bytes period (CFP) 2alternate, 1as shown 1in Packet in PCF format:while this field is useful in pFD-MAC.
Beacon: broadcast the duration time6 of the contention-free
6 1
period
CF-P: Fig.3. In CP,polling
Contention-free network packet management
Dst address Src address type is Station
carried Station
out. …In CFP, Station Greedy
ID1 ID2 IDn bytes 6 6 2 1
CF-A:packets
Contention-free Acknowledge packet
will be transmitted in a pipelined way. The AP polls 2)DstTraffic information
ACK: Data Acknowledge packet Src
type Additional information
stations in the network according to a polling profile. The address
Generally, packets to be transferred will be stored in the AP
address
traffic
Contention information
period: Network management,will be station uploaded join in to and AP during packet
exit, etc. sending buffer. Therefore, we can read the traffic information
Contention-free peirod: and the polling profile will be updated by bytes 1 1 1 1 1
transmission to be transmittedPacket from the sending buffer. Information to be
polling-based scheduling process. The scheduling process downloaded canlength be read from the AP directly. Packets to
and packet transmission process work in parallel. be uploaded in the next polling round need to be collected AP P
Probe Probe Probe Probe
… in the packet transmission procedure. And this information
AP (to Station 1) SIFS (to Station 2) SIFS (to Station 3) SIFS (to Station 4) SIFS Can use the maximal transmission rate
A. INFORMATION CF-A COLLECTION
CF-A CF-A CF-A time will be attached with the uploading packet. Since the traffic
(from Station 1) (from Station 2) (from Station 3) (from Station 4)
In order to solve the inter-node interference problem and bytesinformation 1 1 2 1
is sent from mobile stations to AP, we make a
maximize the network throughput, the centralized AP needs Dst ID change
slight Src ID to the type Data Additional
frame in the IEEE 802.11 standard. AP
information P
to collect some necessary information Initial stage from the network. The And this kind of frame is only used in the packet from
information includes inter-node interference information and stations to bytes AP. As 1
shown
Packet
in1Fig.5(b), 1 1 byte
one 1 is added in the
traffic information. Station 1 SIR1 Station n SIRn

MAC header, representing length the packet length to be uploaded AP P
in the next polling round. Since a byte can only represent a
1) Inter-node interference information range of 0-255, this value is not exactly the packet length. For
Beacon DIFS Numbering
…
pFD-MAC
AP
uses
(probe request) SIR to quantify the
stationslevel of inter-node inter- example, if the value of the byte is 200, it indicates the packet
DIFS PR [20]. PR DIFS
ference like in [8], To calculate
Probe Probe the SIR … time value, we need length is 200∗4 = 800 bytes. For the AP, it needs to parse this AP P
Response 1 Response 2
signal strength from the AP and the interference strength traffic information in its protocol stack and it needs one byte
from its neighbors. The AP collects inter-node interference to reserve the traffic information in its memory. However, the
Contention period
information as shown in Fig.4. It first starts broadcasting a additional cost for the AP to reserve the traffic information is
Probe packet, in which the reply sequence for each station negligible relative to the entire protocol stack.
Station ID: 1 2 3
4 Numbering MAC Header Address 1 Address 2 Address 3 VOLUME 4, 2018
stations
AP
When there are stations join in or http://www.ieee.org/publications_standards/publications/rights/index.html
exit the network, AP numbers each station. for more information.
Acknowledgement
This article has been accepted forTransmission：
Concurrent O issue O
publication inAPa future A but has notAbeen fully edited. Content may change prior to final publication. Citation information: DOI
of this journal,
I I from station time
Greedy algorithm 10.1109/ACCESS.2018.2841888, IEEE Access
Beacon: Contention-free period starting packet; CF-P: Contention-free Polling packet; ACK: Acknowledgement packet for Data;
CF-End: Contention-free period ending packet; CF-A: Contention-free Acknowledge packet; SIFS: Short InterFrame Space
Contention-free repetition interval
Contention SIFS SIFS SIFS SIFS SIFS SIFS SIFS
period
CF-P + Data CF-P ACK CF-P + Data ACK CF-P
Beacon Information … CF-End
(to Station 1) (to Station 2) (to Station 1) (to Station 3) (to Station 2) (to Station 4)
AP collection CF-A + Data ACK CF-A + Data CF-A ACK CF-A … time
(from Station 1) (from Station 1) (from Station 2) (from Station 3) (from Station 3) (from Station 4)
Probe
…
Client (1, 2, 3, 4) SIFS SIFS SIFS SIFS SIFS SIFS SIFS SIFS SIFS
CF-A CF-A CF-A Set by CF-P CF-A Set by CF-P Set by CF-P
AP
NAV (from Client 1)
time
(from Client 2) (from Client 3) (from Client 4)
Set by Beacon time
FIGURE 6. The schematic diagram of packet transmission procedure in our proposed pFD-MAC
Beacon: CFP starting packet; CF-P: Contention-free Polling packet; SIFS: S

CF-End: CFP ending packet; CF-A: Contention-free Acknowledge packet; Pi+Dj:P
Station ID: 1 2 3 out in CP. Especially, when there is only a few data to be
MAC Header Address 1 Address 2 Address 3
(a) Frame structure for numbering Stations
transmitted or only a few stations in the network, CSMA
bytes 2 2 6 6 6 2 1 0-2312 4 mechanism may be a better way since the idle stations
SIFS SIFS will SIF
Frame Duration Seq- Packet Frame
Control ID
DA SA BSSID
ctl length Body
FCS not be polled. AndContention
we can also develop a hybrid scheme in
CF-P + Data
(b) Frame structure for uploading traffic information which data transmissionperiod Beacon
switching in CP Information
and CFP according(P4+D2)
bytes 25 1 1 1 1 4 to the trafficAPin future work. In this way,collection
the CFP can be CF-A + Da
MAC Header Station 1 SIR1 … Station n SIRn FCS
(c) Frame structure for uploading interference information
thought as a period to provide real-time service. In this paper, (Station 4)
bytes 2 2 3 3 6 6 2 0-2312 4 we mainly focus on the design of CFP.
Frame Duration Seq- Frame
DA PA SA BSSID FCS Set by CF-P
Control ID ctl Body
(d) Frame structure for polling header CF-P NAVin contention-free period
2) Operations
FIGURE 5. Key frame structures used in pFD-MAC
The CFP starts with a frame called Beacon and ends with
a frame called CF-End. The duration time of CFP, denoted
Beacon: CFP starting packet; CF-P: Contention-free Polling packet; SIFS: S
Station ID: 1 2 3 by TCF P , is specified in ending
CF-End: CFP frame Beacon
packet; and stations
CF-A: Contention-free cannot
Acknowledge packet; Pi+Dj:P
MAC Header
B. POLLING-BASED Address 1
SCHEDULING Address 2 Address 3 upload packets until they are polled by AP during this time.
(a) Frame structure for numbering Stations
The polling-based 2
scheduling process is responsible for gen- The AP polls each station according to the polling profile.
bytes 2 6 6 6 2 1 0-2312 4
eratingFrame
a polling
Durationprofile used in packet Seq- Packettransmission
Frame proce- When all nodes in the polling profileSIFS have been visited,SIFS we SIF
DA SA BSSID FCS
Control ID ctl length Body Contention
dure. Unlike some contention resolution schemes
(b) Frame structure for uploading traffic information
like CS- say an access round is over and the polling profile will
CF-P be + Data
period Beacon
updated. Once (P4+D2)
MA/CA, the media accessing order is determined in advance AP the network enters into CFP, the AP will col-
Information
bytes 25 1 1 1 1 4
collection
by the polling profile inClient
MAC Header pFD-MAC.
1 SIR1 … Thus, Client the
n SIR scheduling
n FCS lect the interference information and the traffic information CF-A + Da
process is the key (c) Frame structure for uploading interference information
to solve problems introduced by inter- for the first access round. And with these information, we (Client 4)
node interference and asymmetric traffic. In addition, other can determine the polling profile and the packet transmission
optimizing objectives can be considered when generating the procedure can be started. Set by CF-P
NAV
polling profile, such as fairness and quality-of-service (QoS),
etc. The details of the polling-based scheduling procedure 3) Polling procedure
will be described in Section IV. First, the AP starts a transmission using a polling packet,
denoted by CF-P. If it has data to the station, the packet
C. PACKET TRANSMISSION will be CF-P + Data. The frame structure for CF-P is shown
In an AP-based network, traffic can be upstream traffic and in Fig.5(d), in which PA denotes the polling address and S3
downstream traffic. Upstream traffic is the traffic from sta- DA is the destination address (the address for Data part).
tions to the AP and the downstream traffic is the traffic in Here, we use the 3bytes (PA/DA address) to represent the
the reverse direction. Fig.6 shows the transmission procedure 1bytes station ID to reuse the frame structure in IEEE 802.11 Weak int
of pFD-MAC, in which packets above the time axis belong standard, and we set the subtype in FrameControl to 0100.
to downstream traffic and packets below the time axis are The address is a station ID numbered by Fig.5(a). Stations in
upstream traffic. the network keep listening and check whether current polling
Packet
node is itself. If it is, the station will return a contention-free
1) Operations in contention period S 4
acknowledge packet, denoted by CF-A, after a duration time S
Packet
The CSMA mechanism is mainly used to fulfill network indicated in the CF-P packet. And if this station has data to
management in our protocol. For example, if there are mobile be uploaded, the packet will be CF-A + Data. Otherwise, Accessif Point
stations
Step0 that enter or exit the network, the AP will broadcast a station is not the current polling node, it keeps listening.
(AP)
a packet to number the stations registered in the AP as shown Then, when the upload packet arrives at the AP, the AP can
in Fig.5(a). When mobile stations in the network receive this obtain the duration time of the packet by parsing the packet
packet, they will number stations according to the order they header, denoted by TCF −A and the AP can start polling the (a)
T32 Step1 in the packet, i.e., Address 1 indicates station number
appear next node even if current polling node’s receiving process is
is 1 and so on. Actually, data transmission can also be carried not over. Moreover, if downloading packet is a CF-P + Data
VOLUME 4, 2018 5
2169-3536
Step2(c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
10.1109/ACCESS.2018.2841888, IEEE Access Traffic
SIFS SIFS DRR Policy A heuristic algorithm
Interference information Get the
to solve the TSP
ACK CF-P information of
Song Liu et al.: A Polling-based Traffic-aware MAC Protocol for Centralized Full-duplex polling
… CF-End problem ofWireless
non- Networks
(to Station 2) (to Station 4) the network SIR map Non-conflict graph list
F-A ACK CF-A time conflict graph
…
Station 3) (from Station 3) (from Station 4)
SIFS packet,SIFSthe polled station should return an acknowledgment Traffic A heuristic

DRR Policy Get the
t by CF-P
information algorithm to solve
(ACK) afterSetpacket by CF-P CF-A is transmitted. If the upload packet
the polling profile
polling
time SIR map Non-conflict graph profile
has a Data frame to be confirmed, the AP will also return an generation problem
ACK when the receiving process of current polling node is TSP with dynamic cost
FIGURE 7. Full-duplex polling-based scheduling procedure
over. In addition, if the next polling node conflicts with the
1. 尽可能的使用不干扰的互补对
current one, we cannot poll the next node until the receiving
2. 对于对称双连接可以使用最大速率进行发送
process of current polling node is3.over. 首先轮询有上传流量的节点
Note that, there is two A. FAIRNESS SCHEDULING
kinds of ACK packets, ACK from the AP to stations and ACK In order to solve the fairness issue and improve throughput,
mat: from stations to the AP. The ACK packet fromAlgorithm
AP to stations
Greedy two basic questions should be considered: 1) what is the
6 2 can1 be easily controlled at AP. But for the ACK packet from
Src
metric of fairness? 2) what mechanisms can be used to
type stations to information
Additional AP, we expect stations return an ACK as soon asCF-P + Dataachieve
ddress AP CF-P + Data fairness based
CF-P P4+Data on theACK metric?
ACK Algorithm: traffic-aware
the transmission of CF-A packet is over. If there CF-A is no ACKACK CF-A ACK FA3+Data
First, the fairness in MAC protocolFA4+Data ACK
can be understood Input: upstream,
and downstre
bytes 1 at expected
1 1 time, 1 AP 1will consider that the data packet to the (a) Before Scheduling Output: a polling list
Packet
defined in many ways, among which the channel access time
lengthstation has failed and retransmit in the next access round.
AP P3
is widely used for as the fairness metric. However, the chan-
1) initialize the values: po
FA3+Data nel access time in full-duplex mode is different with that in
for i in [0, length(pollin
4) Calculation
aximal transmission rate of Tγ
half-duplex mode. In half-duplex mode, the wireless channel calculate T_alone
1 2 The1 duration time in the CF-P packet, denoted by Tγ , means is shared by each station and a station will exclusively obtain calculate T_conc
rc ID type
uplink channel reservation time instead
Additional information AP of the
P3 durationP1+Datatime ACK the channel. In full-duplex mode, the wireless channel is still delta_time.appen
of the packet CF-P itself. The uplink channel FA3+Data shouldFA1be ACK
shared by each station but two stations can share the channel 2) for i in [0,length(pollin
bytes 1 reserved 1 for 1 Tγ time
1 to
1 avoid a collision, and the next polling to transmit packets at the same time. for j in unvisited Sta
Packet if min(delta_time
lengthnode should be notified of the reservation time. For P1+Data
example,
AP P3 ACK However,
P4+Data recently
ACK studies [11], [15] have not considered delta_time =
when the AP polls S2 , it should tell S2 the duration FA3+Data time FA1of FA4+Data
ACK the characteristics ACK of the channel access time in full-duplex
for j in unvisi
uplink channel used by S1 . Therefore, AP cannot poll the communication, but simply thought Timethe
Savedpacket transmission calculate
next node until it obtains the duration time of the current time as the channel access time, result in discouraging full- calculate
AP P3 P1+Data ACK P4+Data ACK P2+Data
uploading packet. delta_tim
FA3+Data FA1 ACK duplex opportunities
FA4+Data ACK and reducing
FA2 ACK the throughput improve-
Let TR be the time consumed from the beginning of the if min(delta_t
ment. For example, there are two transmissions in the net-
(b) After Scheduling
current uploading packet and Vt be the current data rate. And transmit
work, S4 → AP → S3 and S1 → AP → S2 , as shown in else:
the packet length of ACK is denoted by LACK . Packets from Fig.1. If the packets length are the same, S4 → AP → S3 index =
different stations should be separated with the timeO of Short I A only O consumes I half
A transmission time of S1 → AP → S2 . trasmit s
Interframe Space (TSIF S ). If there is no data packetScheduled to be But recent Unscheduled D Tcompletion
studies think that they have consumed the same else:
Step 0:
confirmed at the station, Tγ can be calculated AP as:O1 A1 Ochannel
2 access O3 time. Therefore, O4 A4 half-duplex transmission, index = inde
the
I1 I2 A2 I3 A3 I4 A4
Step 1: such as S1 → AP → S2 , will in fact obtain more channel
Tγ = TCF −A − TR +APTSIF S O1 (3)
A1 O2 O3 O4 A4
I1 I2 accessA2time Ithan 3
full-duplex
A3 I4 transmissions.
A4
Step 2:
If there is data packet to be confirmed at the station: In this paper, we distinguish the concurrent transmission
O1 O2 A1 O3 O4 A4
AP I1 I2 A2 and Iexclusive
3 A3 transmission
I4 A4 in calculating
+ the channel access
Tγ = TCF −A − TR + 2TSIF S + LACK /VO1t (4) O3 time, which A1 we think O2 is more Oreasonable
4 A4 in full-duplex com-
AP I1 I3 A3 I2 A2 I4 A4
Step 3: munications. Let α be the proportion- of exclusive transmis-
IV. POLLING PROFILE GENERATION O1 O2 A1 sion Otime4 A4 LjO3can be Dj or Uj , the channel access time
and
Since pFD-MAC transmits packets APaccording I2
to aI1 polling A2 I4 A4 I3 A3
Step 4: consumed by station j can be calculated as:
profile, all the scheduling strategies should be reflected
O1 in
O2the A1 O4 A4 O3
polling profile. In turn, we can optimize the polling I1
profile I2
to A2 I4 Lj I3
A4 LAj3 1 Lj
AP tc (j) = ·α+ · · (1 − α) = (1 + α) (5)
achieve our scheduling purposes, such as channel utilization Vt Vt 2 2Vt
maximization, fairness, and so on. Second, DRR is a widely scheduling algorithm with low
Fig.7 shows the scheduling procedure of pFD-MAC. First- complexity and it can be realized in a centralized or distribut-
ly, an SIR map is constructed using the interference infor- ed way. Singh et al [11] try to alleviate fairness problem by
mation, which is collected at the beginning of CFP. And a penalizing the access opportunity based on the access history
non-conflict graph is then generated based on the SIR map in a distributed way, resulting in lower throughput. Kim et
and deficit round robin (DRR) policy [26]. The problem to al [8] implemented DRR in their protocol in a centralized
calculate a polling profile with the minimum transmission manner but introduces unnecessary overhead. In our proto-
time in a given non-conflict graph is transformed into an col, pFD-MAC provides the same access opportunities for
optimization problem, which is proved to be NP-hard. For each station. However, some stations may always have longer
the scheduling process to work with the packet transmission packet length than others, resulting in the unfairness with
in a parallel way, we need the scheduling algorithm to be very respect to access time. Thus, we will use DRR to solve the
fast. Therefore, a heuristic traffic-aware algorithm has been fairness issue in our protocol. More specifically, the fairness
developed in pFD-MAC protocol. Finally, a polling profile is scheduling process handling N stations is configured with
generated for AP to poll and transmit data packet. one quantum Tbase (j) for each station, and this pre-allocated
6 VOLUME 4, 2018
access time is called deficit. At round n, the station j can (1 ≤ i ≤ N ) and Ptap be the transmit power of the AP. Then,
access the wireless channel at most Tj (n): the theoretical value of SIR c0ij can be calculated as:
κPtap hd−n
0j
Tj (n) = Tbase (j) + Tj (n − 1) · Ij (n − 1) (6) c0ij = 10log10 (Prap (j)/Pri (j)) = 10log10 ( )
κPtsta hd−n
ij
(9)
where Tbase (j) is a constant value for each round, Tj (n − 1)
P ap dij n
is the remaining time in the last round. Ij (n − 1) = 1 if the = 10log10 ( tsta ) + 10log10 ( )
queue of station j is not empty and Tj (n − 1) > 0, otherwise Pt d0j
Ij (n − 1) = 0. = δ + 10 · n · log(dij /d0j )
Since the medium access order is determined by the trans- where Prap (j) means the received signal strength from the AP
mission scheduling process, an accurate access time tc (j) to station j and Pri (j) means the received interference signal
for station j can be obtained after a packet is scheduled. strength from station i to station j. d0j is the distance from
After getting this transmission time by feedback from the the AP to station j and dij is the distance from station i to
scheduling process, we can update the deficit of station j for station j. We use δ to represent the transmit power difference
round n as: between the AP and stations.
From Eqs.9 we can see that the SIR value is determined
Tj (n) = Tj (n) − tc (j) (7)
by the transmit power difference δ, path loss exponent n and
the distribution of two stations in a network. Generally, n is
To encourage full-duplex communications, we will sched- a fixed value in the specific wireless environment. And the
ule the packet if Tj (n) > Lj /2Vt instead of Tj (n) > Lj /Vt . distribution of stations may change all the time and could not
Thus, the remaining time of last round for station j can be be controlled. Thus, if we intend to increase the SIR value,
negative. And if Tj (n) < Lj /2Vt , the node will not be we can increase the transmit power difference δ.
included in the next polling access round.
2) Non-conflict graph
B. NON-CONFLICT GRAPH CONSTRUCTION Since an AP polls each mobile station sequentially in our
protocol, it should be guaranteed that any adjacent nodes in
1) SIR map the polling profile do not conflict with each other. That is, the
As described in section III, the SIR value is calculated by AP visits each node in the network exactly once in a polling
Eqs.2 at each station. Then, these SIR records will be upload- round, and the SIR value between two adjacent nodes must
ed to the AP at the beginning of CFP. After the information be greater than a threshold value τ , which guarantees that
collection stage is over, an SIR map, which has been also packets can be transmitted with low packet error rate (PER)
used in [8], [20], is updated at the AP, denoted by MSIR . at a specific data rate. Otherwise, a collision will occur at the
The two-dimension matrix is formally defined as follows: receiving node. To prove the hardness of profile generation
problem, we create a non-conflict graph from the SIR map.
  Note that, the non-conflict graph is a directed graph and SIR
c11 c12 ...
 c21 map is a two-dimension matrix, they are different.
MSIR = c22 ...  (8)
.. ..
 Definition 1: (NON-CONFLICT GRAPH). A non-conflict
..
. . . graph is a directed graph, denoted by Gnc = (Vnc , Enc ),
where Vnc is the collection of mobile stations in the network.
Where each element cij (i 6= j) represents the SIR value at Enc = {eij }, where eij indicates whether there is interference
node j with interference from node i. For example, the SIR from node i to node j at a certain transmission rate. eij = 1 if
map for Fig.1 can be expressed as: signals from AP to node i will not collide with signals from
node i to node j, or else eij = 0. eij can be determined by:
 
30 10.9 20.6 23.1 1 if cij > τ
eij = (10)
 10.9 30 21.7 20.6  0 otherwise
M1 = 
 15.8 21.1 30 20.3 

In a non-conflict graph, the edge between two nodes exists
22.4 22.4 20 30 if there is no collision between them. Two nodes without col-
lision can be adjacent to each other in the polling profile. For
Here, we set cii = 30 since this paper assumes that the PHY- example, the non-conflict graph in Fig.1 can be calculated as
layer can achieve perfect SIC and it means each node can following. Assume that τ = 16.2dB, the minimal data rate
start a symmetric transmission with the AP. will be 12M bps according to Table 1. Let G1 = (V1 , E1 ), in
Theoretical analysis: Here, we would like to analyze the which V1 = {s1 , s2 , s3 , s4 }. For that c1,3 = 20.6 > τ and
SIR value theoretically and discuss the factors that influence c3,1 = 15.8 < τ in M1 , nc1,3 = 1 and nc3,1 = 0, that is, <
the SIR value. Let Ptsta be the transmit power of station i s1 , s3 >∈ E1 and < s3 , s1 >∈ / E1 . Accordingly, we can get
VOLUME 4, 2018 7
LHeader
E1 = {< s1 , s4 >, < s2 , s4 >, < s3 , s4 >, < s1 , s3 >, < O A O
to station
A
Separate Transmission：AP
s2 , s3 >, < s4 , s3 >, < s3 , s2 >, < s4 , s2 >, < s4 , s1 >}. I I time
DT Acknowledgement
And < s1 , s3 > means AP can start polling s3 even if the Concurrent Transmission：AP O O A A
receiving process of s1 has not completed.
FIGURE 8. The special case with only upstream traffic
C. POLLING PROFILE GENERATION PROBLEM

In pFD-MAC protocol, a polling profile is generated to
been shown to be NP-hard. Let us denote TSP as P1 . Then,
ensure the wireless channel is accessed in a contention-free
we show that a special case of PPG problem to minimize the
way. A scheduling approach is required to get a polling
completion time of transmission procedure of a polling round
profile with the objective to minimize the packet transmission
is equivalent to P1 . In fact, when ∆Tij is a fixed value, PPG
time. However, there are some constraints to determine the
problem is equivalent to P1 . Therefore, we only need to find
polling profile. As mentioned in the previous section: 1) if the
a special case in which ∆Tij is fixed.
downloading transmission of current node ends first, the AP
Assume there are N (> 3) stations in the network with on-
can poll the next node even if the receiving process of current
ly one transmission rate Vt . And there is only upstream traffic
polling node is not completed. 2) However, it is forbidden
in the network so that the packet length of upstream traffic Lu
that the node behind the next node is polled. 3) Besides,
is much longer than the packet length of downstream traffic
an appropriate downloading data rate should be selected to
(polling packet) Ld (Lu Ld ). If there is no conflict, ∆Tij
guarantee a low packet error rate (PER) when the packets are
is a fixed value, as shown in Fig.8.
transmitted concurrently.
In order to specify the polling profile generation process ∆Tij = (LACK + LHeader )/Vt + TSIF S (13)
explicitly, some metrics are introduced here. The difference
in completion time between concurrent transmission and Hence, the PPG problem is NP-hard.
separate transmission of two adjacent stations i and j is As the PPG problem is NP-hard, we need to develop a
represented by ∆Tij , which can be calculated as: fast algorithm to get an appropriate polling profile. Especial-
ly, our scheduling mechanism works with the transmission
∆Tij = Tijsep − Tijcon (11) mechanism in parallel. Thus, we hope our scheduling algo-
rithm works in a pipeline way as well. The details of our
where Tijsep means the separate transmission time between scheduling algorithm will be discussed in the next section.
station i and j and Tijcon means the concurrent transmission
time between station i and station j. Tijsep and Tijcon are
D. PROPOSED TRAFFIC-AWARE SCHEDULING
intermediate variables and their values could be calculated
ALGORITHM
by in the scheduling process. Besides, ∆Tij can be positive
pFD-MAC generates the polling profile according to the
or negative. If ∆Tij > 0, it means concurrent transmission is
traffic information and interference information. There are
preferred; otherwise, separate transmission should be better.
three cases need to be considered: 1) the SIR value between
To obtain a polling profile with the minimal packet trans-
two stations is too small to concurrently schedule them.
mission time, the Polling Profile Generation (PPG) problem
Hence, these two stations should be scheduled separately in
is defined. Next, this problem is proved to be NP-hard. Note
which only the symmetric transmission opportunity can be
that, scheduling packets to maximize overall throughput in
used. 2) the SIR value is small, but we can still choose a data
IBFD system is proved to be NP-complete in Janus [8]. But
rate to start a concurrent transmission. 3) the SIR value is big
its transmission mechanism is different with ours.
enough, so the concurrent transmission time is shorter than
Definition 2: (Polling Profile Generation (PPG) Problem).
scheduling them alone.
In an AP-based full-duplex wireless network, a non-conflict
Since the goal of our scheduling algorithm is to obtain
graph, denoted by Gnc = (Vnc , Enc ), is used to represent
a polling profile with the minimum transmission time, we
the interference among mobile stations in the network. And
arrange the nodes registered in the AP step by step with the
a complete graph, denoted by G = (Vnc , E), is constructed
minimum transmission time for each node. Here, a heuristic
to represent the network, where E = {(i, j) : i, j ∈ Vnc and
traffic-aware scheduling algorithm is presented in Algorithm
i 6= j}. Let the cost function from station i to station j be
1, in which θu (i) is the uplink channel time and θd (i) is the
c(i, j): downlink channel time after polling station i. At each step,
−∆Tij if i, j ∈ Enc
c(i, j) = (12) a station is selected and a proper transmission rate is set as
0 if i, j ∈
/ Enc
well. To generate the polling profile, the following situations
The Polling Profile Generation (PPG) problem is to find a will be considered:
path visiting every node in non-conflict graph with a minimal 1) The uplink channel is longer than the downlink. In
cost function c. this situation, the AP can poll the next node while receiving
Lemma 1: The PPG problem is NP-hard. packet of current polling node. We have an opportunity to
Proof: The NP-hard proof of PPG problem is derived start a concurrent transmission. Assume that current polling
from Traveling Salesman Problem (TSP) [27], which has node is node i. Then, ∆Tij for all unscheduled node j
8 VOLUME 4, 2018
Example 10.1109/ACCESS.2018.2841888, IEEE Access
Interference 1
S3
Interference 2
i i th transmission
4
Algorithm 1SAP:: The i mobile station
traffic-aware scheduling
Access Point 3
S1 algorithm O I A O I A
Scheduled Unscheduled D Tcompletion
1: Sµ ← all S4 stations
1
2
need AP to be scheduled Step 0:
O1 A1 O2 O3 O4 A4
2: while Sµ not NULL do 5
S2 I1 I2 A2 I3 A3 I4 A4
Step 1:
3: if θu (i) > θd (i) then O1
I1
A1
I2
O2
A2
O3
I3 A3
O4
I4
A4
A4
4: calculate ∆Tij for each sj in Sµ Step 2:
O1 O4 A1 A4 O2 O3
Oi i th packet from AP Ii i th packet to AP Ai i th Acknowledgement packet
ij } > 0 then
5: if max{∆T I1 I4 A4 I2 A2 I3 A3 +
ij }
6: send s with max{∆T O1 O3 A1 O2 O4 A4
O1 O3 + A1 i O4 O2+A4 I1 I3 A3 I2 A2 I4 A4 +
I1 I3+A3 I4 + A4 I2+A2
7: else (a) PCF O1 O2 A1 O3 O4 A4
I1 I2 A2 I3 A3 I4 A4
8: Scheduling θdO(i)4 = θuO(i) 3 O2 Step 3: -
end ifI4 (b) JanusI1

packet O1 O3 A1 O4 A4 O2
9: I1 I3 A3 I4 A4 I2 A2
10: else Step 4:

O1 O3 A1 O4 O2 A4 O1 O3 A1 O4 O2 A4
I3randomly
A3 +
11: I1 I4
select Aand
4 I2
removeA2 sjT1from SµT2 I1 I3 A3 I4 A4 I2 A2
(c) Our proposed protocol
12: calculate ∆Ti for each si in Sµ with sj FIGURE 9. An example of the heuristic traffic-aware scheduling algorithm for
13: if max{∆Tij } > 0 then network in Fig.1
14: send sj and si with max{∆Tij } concurrently
15: else LHeader to station
TABLE 1. Transmission rates with respect to SIR values
Separate Transmission： O A O A
16: send sjAPalone I I time
DT Acknowledgement SIR (dB) 10 12.3 13.4 16.2 18.3 19.6
17: end
Concurrent if
Transmission： AP O O A A
Rate (Mbps) 3 6 8 12 14 18
18: end if
19: end while
will be selected, but it cannot decide to be sent concurrently
or separately. In step 2, s1 will try to start a concurrent
Beacon: Contention-free period starting packet;
will be calculated. If none CF-P:
CF-End: Contention-free period ending packet;
of them Contention-free Polling packet;
is positive, it means
CF-A: Contention-free Acknowledge packet;
ACK: Acknowledgement packet for Data;
transmission with an unscheduled node. And s2 , s3 and s4
SIFS: Short InterFrame Space
separate transmission is preferred and Contention-free the downlink repetitionchannel
interval can transmit with it concurrently. Note that, O2 will be longer
will be set theSIFSsame asSIFSthe uplinkSIFSchannel.SIFSOtherwise,
Contention SIFS
pick because SIFS
SIFS the upload signal of s1 will interfere the download
upperiod
the unscheduled
Beacon
node with CF-P + Data the maximal CF-P ∆TACK ij and start CF-P +aData signal ACK of s2 ,CF-P so we choose …
a low data rate for O2 . Through
CF-End
Information (to Station 1) (to Station 2) (to Station 1) (to Station 3) (to Station 2) (to Station 4)
AP concurrent transmission at the CF-Aappropriate data rate.
collection + Data ACK CF-A + Data CF-A the comparison ACK of ∆T CF-A for … each unscheduled
time node, s3 will be
(from Station 1) (from Station 1) (from Station 2) (from Station 3) (from Station 3) (from Station 4)
2) The uplink channel is shorter SIFS
than or equal
SIFS
to the
SIFS
selected SIFS
and transmitSIFS
at a proper transmission rate. In Step 3,
NAV downlink. An unscheduled node, called si , will be select- s4CF-P start concurrent transmission with s2 .
Set by CF-P Set by is selected and try Set bytoCF-P
time
ed randomly (stations with upstream packet areSetpreferred) by Beacon
For ∆T > 0, in Step 4, s2 will transmit with s4 concurrently
and transmits a packet at this time. However, it cannot be and the scheduling is over.
determined to transmit node si separately or concurrently
with other nodes.
Di i th packet from AP
Thus, nodeAisii thwill
Ui i th packet to AP
try to transmit with
Acknowledgement packet
V. PERFORMANCE EVALUATION
another
Collecting node sj inACKthe
information unscheduled
period Ti Transmissionnodetimelist saved concurrently, In this section, we analyze the effects of various features of
D1 D3 + A1 D4 D2+A4
andU1 a group of ∆T U3+Aij 3 will be calculated. U4 + A4 If noneUof 2+A2
them is our proposed MAC protocol on network performance. Firstly,
positive, node si will (a) PCFbe sent alone. Otherwise, choose one we compare the performance of pFD-MAC protocol with two
Scheduling D2 ∆T D 3 D4
with the
packet maximal U4 U
and
ij1 U2 send U3
them concurrently at the state-of-the-art centralized MAC protocols, Janus and PCF,
appropriate data rate. (b) Janus in terms of network throughput and packet delay, through
P4+DComplexity
2 P1+D3 P3+D4+A4 In
analysis: P2 the scheduling algorithm, we extensive simulations using a network simulator developed in
U4 U1 U3+A3 U2+A2 T1 T2
need to try to(c)match current
Our proposed protocol polling node with at most n Python [28]. Then, we analyze the efficiency of our schedul-
nodes to find the best next polling node in each step, in which ing mechanism and compare it with Janus’s scheduler under
n represents the number of stations in the network. When we various traffic types. Thirdly, we evaluate the fairness of pFD-
get the next polling node, the AP can start polling the next MAC protocol with respect to access time for each station.
node without having to wait until all the polling orders have Moreover, we investigate the effects of asymmetric traffic
been calculated. Thus, we can say that the time complexity of and inter-node interference on our proposed MAC protocol,
our algorithm is O(n). During the calculation at each step, which can provide some valuable advice to the future design
we mainly need a ∗ n bytes to store the polling profile, b ∗ n of full-duplex MAC protocol. Finally, we analyze the effects
bytes to store the variable vector {∆Tij } and c bytes to store of the duration time of CFP on the network performance,
the other things, where a, b, c are constants. Thus, the space from which we can see that the time of CFP is a compromise
complexity of our algorithm is also O(n). between mobility and network throughput.
Fig.9 shows an example of our scheduling algorithm in
Fig.1, in which O means downstream packet, I means up- A. SIMULATION SETUP
stream packet and A means ACK packet. There are four 1) Network and traffic: In the simulation, random topologies
stations with related information to be scheduled. At first, are used. Stations are randomly distributed and the AP is
in Step 0, we consider each node as transmitting separately, located at the center in an area of 300m ∗ 300m. The default
which is the worst case. In Step 1, s1 with uploading data number of stations is set to 20 and we vary the number
VOLUME 4, 2018 9
TABLE 2. Parameters for PHY and MAC layers B. SIMULATION RESULTS

In this section, the reported values for the simulation results
Parameter Value Parameter Value
CF-P 28 bytes TCF P (default) 100 ms
are the average of 50 times repeated simulation with random-
CF-A 29 bytes SIFS 10 us ly generated topologies. In addition, to ensure the fairness of
ACK 14 bytes PIFS 20 us the evaluation and without losing the generality, we set the
test time of a round to 200ms for Janus and pFD-MAC. For
TABLE 3. Five different traffic types Janus, it means a round of Janus MAC packet exchanges is
200ms. For pFD-MAC, it means the contention-free period
Traffic Type S1 S2 ... Sn−1 Sn time is 200ms.
TR1 Upstream 100% 100% ... 100% 100%
TR1 Downstream 100% 100% ... 100% 100%
TR2 Upstream 100% 100% ... 100% 100% 1) Throughput and transmission delay
TR2 Downstream 0 0 ... 0 0 Throughput: We first verify the network throughput of pFD-
TR3 Upstream 0 0 ... 0 0 MAC with respect to different number of stations under
TR3 Downstream 100% 100% ... 100% 100%
TR4 Upstream 100% 0 ... 100% 0 different traffic types. And we compare our proposed MAC
TR4 Downstream 0 100% ... 0 100% protocol, Janus and 802.11 PCF as well. Fig.10 shows the
TR5 Upstream 50% 50% ... 50% 50% throughput of three different protocols. For saturated traffic
TR5 Downstream 50% 50% ... 50% 50%
like TR1, pFD-MAC improves throughput by 39% and 69%
over Janus and PCF respectively, when the number of stations
is 40. In addition, the throughput performance of pFD-MAC
from 5 to 100 to evaluate the performance under different increases at first and decreases as the number of stations
traffic. We use five different traffic types in the simulation. As grows while the throughput of Janus decreases almost at first
shown in Table 3. For TR1, both upstream and downstream when the number of stations grows. There are mainly two
are 100%, which means each station in the network always reasons: 1) for pFD-MAC, as the number of stations in the
has a packet to transmit/receive. TR2 has only downstream network increases, the full-duplex communication opportu-
traffic while TR3 only has only upstream traffic. TR2/3 can nities increase. Thus, the network throughput is improved.
be considered as the extreme traffic in our life, for exam- However, since the duration time of CFP is a fixed value in
ple, most end users like to download but not upload (TR2) this test, the more stations in the network, the more overhead
and most sensors prefer to upload data to the controller/AP of collecting interference information at the beginning of
instead of downloading data in the sensor network (TR3). CFP, thus increasing the overall throughput of the network;
TR4 is a typical asymmetric traffic, in which some stations 2) for Janus, as the number of stations increases, the overhead
only have upstream and the others only have downstream. to collect information and scheduling transmission increases.
In TR5, stations and AP have packets to each other with But the performance gain in the packet exchange period will
a probability of 50%. TR5 can be seen as a scenario with not increase because it may reach the maximum value.
random traffic. Moreover, for real traffic on the Internet, it is Under extreme traffic like TR2/3, both pFD-MAC and
shown that the distribution of packet size is not uniform [29]. Janus can only slightly improve the network throughput since
The packet size varies from 64 bytes to 1500 bytes. Thus, only a few concurrent transmission opportunities can be
in our simulation, we test our protocol with different packet utilized. The throughput performance of TR3 is better than
lengths, in which the fraction of large packets (> 1400 bytes) TR2 for pFD-MAC and PCF. It’s easy to understand: 1) for
is 40% and the fraction of small packets (< 100 bytes) is pFD-MAC, the AP can poll the next station when a station is
40% while others account for 20%, just like the statistic data uploading while the AP cannot poll the next station even if
at Chicago exchange point [29]. the uploading process of current node is over; 2) for 802.11
2) PHY and MAC layer model: Most frame structures are PCF, the data ACK for uploading packets can be sent with
the same as the frame structures in 802.11 standard and only the next polling packet together. For asymmetric traffic TR4
minor changes have been made as mentioned above. The and random traffic TR5, pFD-MAC can outperform Janus as
parameters of the physical and MAC layers are given in Table well. The pFD-MAC protocol can obtain better performance
2. We set SIFS to 10us and PIFS to 20us. The duration time in TR5 than TR4 while there is not much difference for Janus
of contention-free period is default set to 100ms, which is a because of the limitation of polling mechanism in pFD-MAC.
typical value for 802.11 PCF and the maximal value is about As for PCF, same performance is achieved under TR4/5 since
400ms. The default transmission rate is set to 18M bps. As the throughput performance is almost determined by the total
mentioned in the previous section, different data rates can be traffic. Moreover, pFD-MAC can obtain the best performance
selected based on different SIR values. Table 1 shows the re- with saturated traffic among different traffic types because
lationship between transmission rates and SIR values, which the overhead of collecting information is relatively minimal
has been measured on the hardware platform WARP [8]. In in this kind of traffic. But for Janus, the performance in
addition, the SIR value is calculated according to Eqs.9 in the random traffic might better than saturated traffic since the
simulation. overhead in scheduling preparation period might be smaller
while the performance gain in exchange period is the same.
10 VOLUME 4, 2018
(a) (b) (c)

1 8 .0
pFDMAC TR2 pFDMAC TR3
26 22
1 7 .5 Janus TR2 Janus TR3
pFDMAC TR1
PCF TR3
Throughput(Mbps)
24 1 7 .0 PCF TR2
Janus TR1 20
PCF TR1 1 6 .5
22
1 6 .0 18 pFDMAC TR4
20 1 5 .5 pFDMAC TR5
Janus TR4
18 1 5 .0 16
Janus TR5
1 4 .5 PCF TR4/5
16
14
1 4 .0
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100
Number of Stations Number of Stations Number of Stations
FIGURE 10. Throughput of different MAC protocols with different number of stations and different traffic types.
(a) (b) (c)

60 pFDMAC TR1 40 pFDMAC TR2 pFDMAC TR4
35
50 Janus TR1 35 pFDMAC TR3 pFDMAC TR5
PCF TR1 Janus TR2/3 30 Janus TR4/5
40 30 PCF TR2/3 25 PCF TR4/5
Delay(ms)
25
30 20
20
15 15
20
10 10
10 5
5
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100
Number of Stations Number of Stations Number of Stations
FIGURE 11. Delay of different MAC protocols with different number of stations and different traffic types.
Delay is another important metric to measure the net- MAC. For saturated traffic TR1, the pFD-MAC protocol
work performance. One MAC protocol may have a large can reduce the average packet delay by 48% and 33%, as
throughput while suffering a long delay. In the simulation, compared to Janus and PCF with the number of stations is
considering that the overhead of collecting information and 40, respectively. With increasing the number of stations, the
the scheduling preparation (only Janus) is suffered by each delay time increases for all three protocols. Among them,
packet, the overhead time will be counted in the total delay packet delay of PCF increases faster than the other two
time. Thus, the delay time of pFD-MAC will be the overhead protocols, pFD-MAC and Janus, since the delay time largely
of collecting interference information and the average time depends on the average transmission time of an access round
of an access round in CFP. As for Janus, the delay time is in this case. When it comes to unsaturated traffic like TR4/5,
calculated as the overhead of scheduling preparation period the delay time of PCF is approximately to pFD-MAC because
and ACK period pluses the average time of an access round in the overhead time of collecting information is relatively high.
the packet exchange period. And we use the duration time of Especially, for extreme traffic TR2/3, the delay time of pFD-
an access round to present the packet delay time of PCF. The MAC is higher than PCF when the number of stations is
overhead time of Janus is 2530us to schedule 36 packets in a large since the overhead time of collecting information starts
round and the per-packet overhead is 69us [8]. And it is hard dominating the overall delay time. Moreover, there is a gap
to be optimized because the scheduling process and packet between pFD-MAC and Janus with respect to the delay time
transmission procedure of Janus work in serial mode. In the under different traffic types, which can be considered as the
simulation, we set the average scheduling time for a packet overhead of packet scheduling. In our proposed protocol,
is 25us since we think scheduling process can be accelerated the scheduling process works with the packet transmission
by more powerful hardware. And the per-packet overhead is process in parallel so that the delay time is largely decreased.
about 30us for Janus in the simulation. In other words, we can say that our protocol has better
scalability than Janus.
Fig.11 shows the delay time of different protocols with
different traffic types. From this figure, we can see that our All in all, our proposed MAC protocol can achieve good
proposed protocol has the shortest delay time under different network throughput performance while maintaining low
traffic types while achieves the highest throughput as shown transmission delay. Moreover, as network size grows up,
in Fig.10. And this mainly benefits from the parallelism be- the performance of pFD-MAC is steady and shows good
tween scheduling process and transmission process in pFD- scalability.
VOLUME 4, 2018 11
(a) TR1 (b) TR2/3 (a) TR1 (b) TR2/3

28
75 19 pFDMAC 16.5
18 26 PCF pFDMAC TR2
Throughput(Mbps)
70 19.72% 16.0
Scheduler Gain(%)
24 pFDMAC TR3
65 17 PCF 4.05%
pFDMAC 16 22 15.5
60 Janus
15 pFDMAC TR2 20 42.53%
15.0
55 pFDMAC TR3 9.16%
14 Janus TR2 18 14.5
50
13 Janus TR3
45 16 14.0
(c) TR4 (d) TR5 (c) TR4 (d) TR5
70 67.5 22 pFDMAC pFDMAC
65 65.0 21 PCF 22 PCF
Throughput(Mbps)
Scheduler Gain(%)
60 62.5 20 24.89%
60.0 19 35.48% 20
55 18
57.5
50 55.0 17 18
45 52.5 16 16 27.73%
pFDMAC 50.0 pFDMAC 15 12.66%
40 Janus Janus
47.5 14 14
20 40 60 80 100 20 40 60 80 100 20 10 0 10 20 30 20 10 0 10 20 30
Number of stations Number of stations The value of (dB) The value of (dB)
FIGURE 12. The efficiency comparison of scheduling mechanism between FIGURE 13. The throughput with different inter-node interference and
pFD-MAC and Janus under different traffic type, in which we ignore the different traffic types in which the inter-node interference is determined by the
overhead of information collection and scheduling. value of δ according to Eqs.9.
2) Efficiency of the proposed polling-based scheduling so the scheduler gain of Janus is large than pFD-MAC’s; 2)
mechanism the transmission of ACK packet is counted in our scheduling
To verify the efficiency of scheduling mechanism of our mechanism while the ACK period of Janus is not included
proposed protocol, we compare our polling-based schedul- in the simulation. For TR3, the scheduling gain should be the
ing mechanism with Janus’s scheduling mechanism. In the same between pFD-MAC and Janus, the gap between them is
evaluation, the overhead of information collection and trans- the overhead of the transmission time of ACK packet in pFD-
mission scheduling is ignored. In order to ensure the integrity MAC. Moreover, as shown in Fig.12, the performance of our
of our protocol, the ACK has not been deleted in our protocol scheduling mechanism is more steady than Janus since the
while the ACK period is ignored in Janus’. In addition, the scheduling policy in Janus largely depends on the network
duration time of a round is set to 100ms. As shown in topology. And there is a big difference in the inter-node
Fig.12, our polling-based scheduling mechanism can largely interference in different topologies. On the other hand, to
outperform Janus’ in saturated traffic TR1. The reason why illustrate the performance of the scheduling mechanism in
our scheduler can outperform Janus’s is that the optimization pFD-MAC, we also compare the network throughput with
policy is different. The optimization policy of Janus is trying and without the scheduling algorithm. Since the scheduling
to find the optimal full-duplex connection to decrease the mechanism mainly utilizes the asymmetric transmission in
transmission time by matching the maximum LF value [8]. different traffic, the performance gain of our scheduling al-
However, the optimization policy of our protocol is to mini- gorithm is equivalent to the performance gain of asymmetric
mize the transmission time as much as possible. For the same transmission, which is shown in Fig.13. And we will discuss
reason, as the number of stations increases, the performance it together with the effects of inter-node interference in the
gain of our polling-based scheduling mechanism increases next subsection.
as well since there will be more full-duplex transmission
opportunities to start asymmetric transmission as the number 3) Effects of inter-node interference and asymmetric traffic
of stations increases. However, the performance of Janus’s on network performance
scheduler will decreases since the scheduling policy of Janus To study the effects of inter-node interference and asymmet-
is that matching the concurrent transmission with the maxi- ric traffic on network performance, we evaluate the network
mum value of LF while our scheduling policy is to minimize throughput with different power difference under different
the transmission time. Obviously, our scheduling mechanism traffic types. According to Eqs.9, we can regulate the inter-
is more efficient in saturated traffic. node interference (SIR value) by changing the value of the
For unsaturated traffic like TR2/3/4/5, the performance power difference δ. The throughput will increase as we
gain of our proposed scheduling mechanism is slightly worse increase the value of δ. However, the inter-node interference
than Janus’s. There are two main reasons: 1) the polling- only limits the asymmetric transmission between different
based mechanism limits the use of asymmetric transmission. stations, the symmetric transmission will not be influenced.
Especially when it comes to TR2 and TR4, each upstream Thus, the effect of inter-node interference varies with dif-
packet cannot be transmitted until they are polled by the ferent traffic. The result is shown in Fig.13, in which we
AP. However, Janus can schedule the packet in a free way think there is no asymmetric transmission can be utilized
12 VOLUME 4, 2018
TABLE 4. Fairness under different traffic types
Traffic type TR1 TR2 TR3 TR4 TR5

27
IJain 1 1 1 0.99 0.99
26
Throughput(Mbps)
25
when δ = −20dB and there are plenty opportunities to start TCFP=20ms
asymmetric transmission when δ = 30dB. Moreover, we can 24 TCFP=50ms
say that the performance gain by utilizing asymmetric trans- TCFP=100ms
mission benefits from our proposed polling-based scheduling 23 TCFP=200ms
TCFP=400ms
algorithm.
20 40 60 80 100
For TR1, the asynchronous characteristic is determined Number of Stations
by the difference in upstream and downstream traffic in
FIGURE 14. Throughput with respect to the duration time of contention-free
the symmetric transmission. And the throughput can be im- period under different network sizes
proved by 19.72% with asymmetric transmission while the
throughput can be improved by 42.53% with the symmetric
transmission compared with PCF. For TR2, there is only a the time. Yet rigorous time allocation obtains perfect fairness,
few opportunity to utilize asymmetric transmissions so that as shown in Table.4. The minor imperfect fairness in TR4/5
only 4% throughput improvement is obtained, but we can could be the mechanism that we allow a station slightly
still improve the network throughput by 9.16% with sym- overuse its deficit. To get a better network performance, we
metric transmissions. For TR3, only symmetric transmission can relax the deficit allocation. For example, again, we have
can be used and no concurrent transmission between two 10 stations in the network and TCF P = 100ms, then the
different stations. When it comes to the asymmetric traffic access time for each station can be 1.2 ∗ 100ms/10 = 12ms.
like TR4, our proposed traffic-aware scheduling algorithm Thus, more packets could be transmitted and the network
can increase the network throughput by 35.48% and only performance will be increased as well. However, it could
12.66% throughput improvement benefits from symmetric introduce unfairness to some extent but it is very small. After
transmissions. With the random traffic like TR5, symmet- all, we give each node the same access opportunity, and the
ric and asymmetric transmissions bring almost the same packet length distribution for each node is the same.
throughput improvement to the network.
From the perspective of system design, the symmetric 5) Effect of TCF P on network performance
transmission determines the lower bound of the network As mentioned in the above section, the SIR map may only
performance and the asymmetric transmission determines the be updated at the beginning of CFP. The longer the duration
upper bound of the network performance. In addition, the time of CFP is, the smaller the overhead time of collecting
full-duplex station supports symmetric transmission and the the interference information. However, the real SIR map
half-duplex station only support asymmetric transmission in is changing all the time as stations move. Fig.14 shows
the AP-based network. Thus, if the inter-node interference the throughput with different TCF P under traffic type TR6.
is strong, we should deploy more full-duplex stations in the When the number of stations in the network increases, the
network; if the inter-node interference is weak, we might throughput with a small TCF P decreases faster than a big one
use more half-duplex stations in the network to exploit the since the overhead time to collect interference information
asymmetric transmission in the network. Of course, the AP is the same. With increasing TCF P , the throughput will be
must support full-duplex communication to start a symmet- improved as expected. Actually, we can choose the value
ric/asymmetric transmission. of TCF P according to specific scenario. If the stations in a
network move slow, a long CFP can be selected. And if the
4) Fairness in our protocol stations in a network move fast, a short CFP should be set.
This part verifies the fairness of our proposed pFDMAC However, if stations move fast while the number of stations
protocol. As described in Section IV-A, we use the channel is large in the network, we should set a small value of TCF P .
access time of a station as the fairness metric in the wireless However, if we still collect the interference information
system. And the fairness is measured by Jain index [30]. The among all stations, the overhead time of collecting these
deficit time Tbase for each station is allocated according to interference information may be 100% of the contention-free
the duration time of CFP and the number of stations in the period. At this time, we can only poll part of the overall
wireless network. For example, we have 10 stations in the stations registered in the AP so we only need to collect part
network and TCF P = 100ms, then the access time for each of the inter-node interference in the network. And this is one
station will be 100ms/10 = 10ms. However, the channel of the future work to improve our proposed MAC protocol as
access time includes the upload time and download time for well. On the other hand, we can give up collecting the inter-
a station, 10ms usually is not enough to transmit all packets node interference information and only utilize the symmetric
because the AP cannot maintain concurrent transmissions all transmission which is the last choice.
VOLUME 4, 2018 13
VI. CONCLUSIONS AND FUTURE WORK [14] M. Hirzallah, W. Afifi, and M. Krunz, “Full-duplex spectrum sensing and
This paper presents a polling-based traffic-aware MAC pro- fairness mechanisms for wi-fi/lte-u coexistence,” in Global Communica-
tions Conference, 2017, pp. 1–6.
tocol called pFD-MAC for AP-based full-duplex wireless [15] J. Hu, Y. Liao, L. Song, and Z. Han, “Fairness-throughput tradeoff in full-
networks. pFD-MAC uses a polling scheme to transmit pack- duplex wifi networks,” in Global Communications Conference, 2017, pp.
ets according to a polling profile. And a heuristic traffic- 1–6.
[16] N. Piscataway, “Wireless lan medium access control (mac) and physical
aware algorithm is developed to generate the polling profile layer (phy) specifications,” IEEE D3, pp. C1–1184, 1996.
in order to fully exploit full-duplex opportunities. Extensive [17] B. Han, J. Li, J. Su, M. Guo, and B. Zhao, “Secrecy capacity optimization
simulations reveal that the parallelism between scheduling via cooperative relaying and jamming for wanets,” IEEE Transactions on
Parallel and Distributed Systems, vol. 26, no. 4, pp. 1117–1128, 2015.
mechanism and packet transmission significantly improve the [18] B. Han, J. Li, J. Su, and J. Cao, “Self-supported cooperative networking
network performance in term of throughput and transmission for emergency services in multi-hop wireless networks,” IEEE Journal on
delay. More importantly, we analyze many factors influence Selected Areas in Communications, vol. 30, no. 2, pp. 450–457, 2012.
[19] Q. Liu, X. Wang, B. Han, X. Wang, and X. Zhou, “Access delay of cog-
the design of full-duplex MAC protocol which can provide nitive radio networks based on asynchronous channel-hopping rendezvous
valuable advice in future work. In order to make pFD-MAC and csma/ca mac,” IEEE Transactions on Vehicular Technology, vol. 64,
practical, there are still many works to do. And we intend no. 3, pp. 1105–1119, 2015.
[20] A. Tang and X. Wang, “A-duplex: Medium access control for efficient
to evaluate our protocol on WARP platform in our future coexistence between full-duplex and half-duplex communications,” IEEE
work. Besides, we also want to extend our protocol to ad-hoc Transactions on Wireless Communications, vol. 14, no. 10, pp. 5871–
wireless networks because it is now only available in an AP- 5885, 2015.
[21] S. Goyal, P. Liu, O. Gurbuz, and E. Erkip, “A distributed mac protocol
based wireless network. How to mitigate pFD-MAC to use for full duplex radio,” in Signals, Systems and Computers, 2013 Asilomar
other emerging full-duplex techniques in the physical layer, Conference on, 2013, pp. 788–792.
such as full-duplex multi-input multi-output (MIMO) [31], [22] Y. Liao, K. Bian, L. Song, and Z. Han, “Full-duplex mac protocol design
and analysis,” IEEE Communications Letters, vol. 19, no. 7, pp. 1185–
remains an open problem. 1188, 2015.
[23] L. Song, Y. Liao, K. Bian, L. Song, and Z. Han, “Cross-layer protocol
design for csma/cd in full-duplex wifi networks,” IEEE Communications
REFERENCES Letters, vol. 20, no. 4, pp. 792–795, 2016.
[1] D. Kim, H. Lee, and D. Hong, “A survey of in-band full-duplex trans- [24] X. Xie and X. Zhang, “Does full-duplex double the capacity of wireless
mission: From the perspective of phy and mac layers,” Communications networks?” Proceedings - IEEE INFOCOM, pp. 253–261, 2014.
Surveys & Tutorials IEEE, vol. 17, no. 4, pp. 2017–2046, 2015. [25] W. Cheng, X. Zhang, and H. Zhang, “Rts/fcts mechanism based full-
[2] W. Yu, J. Cao, X. Zhou, X. Wang, K. C. C. Chan, A. T. S. Chan, and H. V. duplex mac protocol for wireless networks,” in GLOBECOM Workshops,
Leong, “A high-throughput mac protocol for wireless ad hoc networks,” 2014, pp. 5017–5022.
IEEE Transactions on Wireless Communications, vol. 7, no. 1, pp. 135– [26] M. Shreedhar and G. Varghese, “Efficient fair queueing using deficit round
145, 2008. robin,” in Conference on Applications, Technologies, Architectures, and
[3] S. Lv, W. Zhuang, X. Wang, and X. Zhou, “Link scheduling in wireless Protocols for Computer Communication, 1995, pp. 231–242.
networks with successive interference cancellation,” Computer Networks, [27] T. H. Cormen, Introduction to algorithms. MIT press, 2009.
vol. 55, no. 13, pp. 2929–2941, 2011. [28] A centralized MAC protocol simulator, GitHub, December 2017, http-
[4] D. Bharadia, E. Mcmilin, and S. Katti, “Full duplex radios,” in ACM s://github.com/liusongee/MAC.git.
SIGCOMM 2013 Conference on SIGCOMM, 2013, pp. 375–386. [29] Packet size distribution comparison between Internet links in 1998
[5] M. Fukumoto and M. Bandai, “Mimo full-duplex wireless: Node archi- and 2008, Center for Applied Internet Data Analysis, June 2010,
tecture and medium access control protocol,” in Seventh International http://www.caida.org/research/traffic-analysis.
Conference on Mobile Computing and Ubiquitous NETWORKING, 2014, [30] R. Jain, D. Chiu, and W. Hawe, “A quantitative measure of fairness
pp. 76–77. and discrimination for resource allocation in shared computer systems,”
[6] L. Chen, F. Wu, J. Xu, K. Srinivasan, and N. Shroff, “Bipass: Enabling Computer Science, vol. cs.ni/9809099, 1984.
end-to-end full duplex,” in The International Conference, 2017, pp. 114– [31] D. Bharadia and S. Katti, “Full duplex mimo radios,” in Usenix Conference
126. on Networked Systems Design and Implementation, 2014, pp. 359–372.
[7] W. Choi, J. Park, Y. Kim, A. Sabharwal, and H. Lim, “Design and imple-
mentation of a full-duplex pipelined mac protocol for multihop wireless
networks,” IEEE Access, vol. PP, no. 99, pp. 1–1, 2017.
[8] J. Y. Kim, O. Mashayekhi, H. Qu, M. Kazandjieva, and P. Levis, “Janus: A
novel mac protocol for full duplex radio,” CSTR, vol. 2, no. 7, p. 23, 2013.
[9] W. Cheng, X. Zhang, and H. Zhang, “Full-duplex spectrum-sensing and
mac-protocol for multichannel nontime-slotted cognitive radio networks,”
IEEE Journal on Selected Areas in Communications, vol. 33, no. 5, pp.
820–831, 2015.
[10] J. I. Choi, M. Jain, K. Srinivasan, P. Levis, and S. Katti, “Achieving single SONG LIU received the B.S. degree in electrical
channel, full duplex wireless communication,” in International Conference and computer engineering in 2013 from Shang-
on Mobile Computing and Networking, MOBICOM 2010, Chicago, Illi- hai Jiao Tong University (SJTU), Shanghai China
nois, Usa, September, 2010, pp. 1–12. and the M.S. degree in computer science in 2015
[11] N. Singh, D. Gunawardena, A. Proutiere, B. Radunovi, H. V. Balan, and from National University of Defense Technology
P. Key, “Efficient and fair mac for wireless networks with self-interference (NUDT), Changsha, Hunan, China. He is currently
cancellation,” in International Symposium on Modeling and Optimization pursuing the Ph.D degree in National University
in Mobile, Ad Hoc and Wireless Networks, 2011, pp. 94–101. of Defense Technology. His research interests in-
[12] A. Sahai, G. Patel, and A. Sabharwal, “Pushing the limits of full-duplex: clude wireless network, full-duplex communica-
Design and real-time implementation,” arXiv preprint arXiv:1107.0607, tions and optimization theory.
2011.
[13] M. Jain, J. I. Choi, T. Kim, D. Bharadia, S. Seth, K. Srinivasan, P. Levis,
S. Katti, and P. Sinha, “Practical, real-time, full duplex wireless,” in
International Conference on Mobile Computing and NETWORKING,
2011, pp. 301–312.
14 VOLUME 4, 2018
BIAO HAN received the BE and master’s degrees

in computer science in 2007 and 2009 respectively,
both from National University of Defense Tech-
nology (NUDT), China, where he is currently an
assistant professor. He received the PhD degree
from University of Tsukuba, Japan, in 2013. From
January 2012 to April 2012, he has been a visiting
scholar in Department of ECE at University of
Florida. His research interests are in software de-
fined networking (SDN), wireless communication
and network security. He is a member of the IEEE.
WEI PENG was born in Sichuan, China in 1973.

He received the M.S and the Ph.D degrees in
Computer Science from the National University
of Defense Technology, China, in 1997 and 2000
respectively. From 2001 to 2002, he was a research
assistant at the school of Computer from the Na-
tional University of Defense Technology, China,
then a research fellow from 2003. His major in-
terests are internet routing, network security, and
mobile wireless networks.
VOLUME 4, 2018 15

Extraction of The Rice Leaf Disease Image Based On BP Neural Network

Încărcat de

Informații document

Descriere originală:

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Extraction of The Rice Leaf Disease Image Based On BP Neural Network

Încărcat de

Drepturi de autor:

Formate disponibile

This article has been accepted for publication in a future issue of this journal, but has not been

A Polling-based Traffic-aware MAC

I. INTRODUCTION gation and fairness need to be well investigated at the MAC

(AP → S2,3,4 and S1,4 → AP ) in half mode. And it is

SIFS SIFS SIFS SI

traffic information Beacon

traffic information. Station 1 SIR1 Station n SIRn

Beacon: CFP starting packet; CF-P: Contention-free Polling packet; SIFS: S

SIFS packet,SIFSthe polled station should return an acknowledgment Traffic A heuristic

C. POLLING PROFILE GENERATION PROBLEM

end ifI4 (b) JanusI1

10: else Step 4:

TABLE 2. Parameters for PHY and MAC layers B. SIMULATION RESULTS

(a) (b) (c)

(a) (b) (c)

(a) TR1 (b) TR2/3 (a) TR1 (b) TR2/3

TABLE 4. Fairness under different traffic types

Traffic type TR1 TR2 TR3 TR4 TR5

BIAO HAN received the BE and master’s degrees

WEI PENG was born in Sichuan, China in 1973.

S-ar putea să vă placă și