MAC Protocols For IEEE 802.11ax: Avoiding Collisions On Dense Networks
MAC Protocols For IEEE 802.11ax: Avoiding Collisions On Dense Networks
MAC Protocols For IEEE 802.11ax: Avoiding Collisions On Dense Networks
11ax:
Avoiding Collisions on Dense Networks
Rafael A. da Silva1 and Michele Nogueira1,2
1
Department of Informatics, Federal University of Paraná, Curitiba, PR, Brazil
2
Electrical and Computer Engineering, Carnegie Mellon University, Pittsburgh, PA, USA
E-mails: [email protected] and [email protected]
arXiv:1611.06609v1 [cs.NI] 20 Nov 2016
Wireless networks have become the main form of Internet access. Statistics
show that the global mobile Internet penetration should exceed 70% until
2019. Wi-Fi is an important player in this change. Founded on IEEE 802.11,
this technology has a crucial impact in how we share broadband access both
avoidance techniques and Medium Access Control protocols have been designed
in order to improve Wi-Fi performance. Analyzing the collision problem,
this work strengthens the claims found in the literature about the low
Wi-Fi performance under dense scenarios. Then, in particular, this article
overviews the MAC protocols used in the IEEE 802.11 standard and discusses
Statistics show that wireless networks have become the main form of Internet access
with an expected global mobile Internet penetration exceeding 70% until 2019 [1]. Wi-Fi
Notice: This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice,
after which this version may no longer be accessible.
1
is an important player in this change. Founded on IEEE 802.11 standard, this technology
has a crucial impact in how we share broadband access both in domestic and corporate
networks. However, over wireless medium, signals can suffer interference and disturbances
from different sources. Further, network devices have to share the communication channel
that, if not well managed, can produce unintelligible signals by concurrent transmissions on
Medium Access Control (MAC) protocols have been proposed to provide efficiency in
the use of the communication channel. Network devices should compete or cooperate to
get access to the channel, and the MAC protocols define how they should behave in case
of the channel is available or busy. In general, those protocols follow a collision avoidance
perspective, allowing broadcasts only when the channel is available. Devices must refrain
from transmitting while the medium is busy. Collision avoidance has been one of the most
important challenges in wireless networks and its complexity has been intensified with the
increasing number of devices and the requirements for Quality of Experience (QoE), mainly
associated with real time applications, such as online video and voice transmissions.
Different works in the literature have indicated the low performance of Wi-Fi networks
under dense scenarios. The elevated number of devices increases the probability of collisions.
Also, the unplanned deployment of Wi-Fi networks arises in overlapping networks operating
in a common frequency and interfering with each other. The popularization of Wi-Fi has
caused the saturation of communication channels, mainly in large urban centers and public
hotspots employing the 2.4GHz free license band. Further, the requirements for QoE raise
the complexity on access control mechanisms once they must ensure the quick transmission
of urgent traffic without producing significant delay.
Hence, in order to cope with these new challenges, the IEEE created the Task Group AX
to improve the 802.11 standard. They aim at increasing throughput and spectrum efficiency
2
of highly dense wireless local area networks (WLANs). Their goal lies in offering a new
amendment to 802.11 until 2019. Currently, a set of works has discussed models and evaluated
methods, technologies and functional requirements to compose this new amendment. Also,
researchers have proposed enhancements to the MAC protocols. Their goal lies in designing
improved protocols considering the particularities of dense environments.
In this context, this work contributes with the literature as follows. It strengthens the claims
about the low Wi-Fi performance under dense scenarios by revisiting the collision problem.
In particular, this article overviews the MAC protocols proposed for the 802.11 standard and
discusses solutions to mitigate collisions. While previous publications about the 802.11ax offer
a broader view [2], [3], our work studies the proposals focusing on collision avoidance. Finally,
This article proceeds as follows. First, we provide a background on current 802.11 standard
and we revisit the problem of collisions related to MAC protocols. Then, we present an
overview about the existing proposals for MAC improvements to 802.11ax amendment. Next,
we discuss challenges, open issues and future directions in this field.
During the last fifteen years, 802.11 compatible devices have proliferated in a worldwide
scale. They have been largely deployed because of factors such as cost and simplicity. The
802.11 standard describes wireless communication structures for local range connectivity.
It works with fixed or mobile devices. Since its first release in 1997, several amendments
have been added to it. The Table I lists the main changes on each version. They have been
introduced incrementally. The later versions keep the previous specifications to maintain
backward compatibility. For instance, the current version 802.11ac-2012 supports both
3
802.11n and 802.11a. And the next versions should continue supporting and coexisting with
the current one.
The MAC protocols used in all these versions are essentially the same. They follow a
“listen before talk” approach in which a Carrier Sense is performed to check if the channel
is clear (no other node is transmitting). If it is clear during certain periods of time, the nodes
start their transmissions. These intervals between transmissions are important components
in the coordination of the access mechanism. For instance, devices need to confirm every
reception. Then, they send an acknowledgment (ACK) frame to notify transmitters about the
communication successful. These ACKs must be sent as soon as possible after the reception.
Thus, they send the ACK in the Shortest Inter Frame Space (SIFS). Further, only the receiver
is allowed to transmit its ACK in this interval. And all other types of transmission must wait
intervals longer than SIFS. This reduces the probability of collisions over ACKs. Other types
of frames wait longer Arbitrary Inter Frame Spaces (AIFS). Accordingly to their priorities,
devices define how long they must wait before starting their broadcasts. Finally, the mechanism
adds a random time interval aiming to avoid simultaneous transmissions. This interval is inside
of the Contention Window (CW). It works as a backoff countdown after some Inter Frame
Space. The goal is to provide different access times among nodes with same priority level.
4
BUSY AIFS
CW – Background ( 0 ~ 15 slots )
BUSY AIFS
CW – Best Effort ( 0 ~ 15 slots )
BUSY AIFS
CW – Video ( 0 ~ 7 slots )
802.11b 802.11a/n/ac 802.11g/n
SIFS 10 µs 16 µs 10 µs
BUSY AIFS Slot time 20 µs 9 µs 9 µs
CW Voice
Fig. 1: Transmissions start at random slot within Contention Window for each traffic type.
The random component is efficient to avoid collisions if there are few devices operating.
It works because the CWs are divided into slots. Each device chooses one slot to start its
transmission. If another device starts first, the Carrier Sense detects the busy channel. Then,
the device interrupts its countdown, resuming in the next attempt. However, collisions would
happen if two or more devices choose the same slot in their countdown. In this case, probably
the signal would be unintelligible and the receiver would not send an ACK. When this occurs,
the transmitters doubles their CW size and they try again. It reduces the chance of later
collisions. But, they only observe a failure after the ACK timeout. Hence, this wastes the
entire time transmitting and waiting for the ACK. Further, after a successful transmission,
the CW size returns to its initial value. This causes instability and may provoke the capture
of channel. Some devices can monopolize the channel if they can transmit while others have
their CW expanded.
density increases, even more to smaller CW like that used for voice and video. From top to
the bottom, the first and second tracks in the figure represent the collision probability for these
types of traffic. The CW size for them is expanded only once until it reaches its maximum
size. This results in a high chance of concurrent transmissions with just few nodes. For
instance, the collision probability is higher than 50% with just five nodes sending voice data.
This analysis considers only collisions inherent to the mechanism, not taking into account
anomalies caused by carrier sense fails or hidden nodes. The 802.11 MAC protocol uses a
virtual carrier sense to prevent hidden terminal problem. Control frames are exchanged before
data frames. They reserve the channel in the collision domain. But these frames are sent as
well as conventional data frames. Hence, they suffer the same problems addressed above.
0.8
Collision Probability
0.6
0.4
CW=4−8
CW=8−16
CW=16+
0.2 CW=32+
CW=64+
CW=128+
0
0 10 20 30 40 50 60 70 80
Number of Stations in the Channel
Fig. 2: Collision Probability according CW size (Voice=4∼8 slots and Video=8∼16 slots)
The task group AX has worked on several proposals to improve the 802.11 standard.
They have defined procedures for selecting and evaluating them. They also have specified
deployment scenarios and parameters for testing. These scenarios contain a high number of
6
stations and access points forming dense networks. Thereby, they emulate the problems of high
collision rates. In this context, efficient MAC protocols are critical to improve the throughput
and channel efficiency. The next sections provide an overview of proposed changes to them.
There are many studies about the current 802.11 MAC protocol. Mathematical analyses have
demonstrated its behavior on performance, stability and fairness. Most studies indicate that
larger windows are more suitable for dense networks, and smaller windows are more suitable
for small networks. Thus, researchers came out to Dynamic Adjustment of Contention
Window. Protocols following this approach can adjust the CW size according to the observed
channel conditions. They avoid the access delay caused by a large initial window, as well
as the high collision probability resulted from a small one. The challenge is how to define
the ideal window size. In pursuit of this goal, researchers have proposed protocols based
on medium use estimation. Measurements on Free Time or Channel Congestion can assist
these estimations by each device. Exchanging this information among nodes may improve
synchronization. Further, some protocols suggest a polynomial expansion of CW rather than
Indeed, devices with large initial windows will be penalized if other devices keep their
windows small. The proposal in [5] suggests the Enlargement of Initial CW. The idea lies
in increasing the number of contention slots while reducing the slot time. For instance, in
OFDM, the slot would be reduced from 9µs to 4.5µs and the amount of slots would be
doubled as illustrated in Figure 3. Also, the CWs should be adjusted according to EDCA
rules. They double their size in slots while maintaining the same total duration. Thus, new
7
devices remain compatible with current standard if other intervals based on slot time are
doubled too, e. g. AIFS and CWmin.
The EICW is more relevant for voice and video traffic. It reduces the collision rates as
exposed in our previous work [4]. Further, the new devices would have more opportunities to
send but without silencing the legacy. This should encourage upgrading. However, reducing
slots may have impact on Clear Channel Assessment (CCA). Before transmitting, each device
must check if the channel is not busy. With a reduced slot, they should do this on half
time. Also, imperfections on CCA process can generate collisions. But, the recent hardware
evolution leads to believe that it is possible to perform the CCA with enough accuracy even
in such short time.
Tx, after free channel >= DIFS DIFS Contention Window 144µs
DIFS PIFS 0 1 2 … 5 … 9 … 13 … 17 … 21 … 25 … 29 31
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
SIFS
Busy Channel
In another direction, some proposals suggest reservations based on Time Division Multiple
Access. These TDMA-like protocols have a common CW size among all nodes. And each
device reserves one slot to start its broadcasts. Usually, the first access is made choosing a
random slot. If this transmission is successful, they defer from transmitting in all other slots.
After a CW round they send the next frame in the same slot number. If a slot has been used
for one device, no other can use it. This avoids collisions by reserving slots until they are
liberated when not used. Protocols based in this idea are called Deterministic Backoff, e.
g. ZC, L-BEB or CSMA/ECA and SRB. Usually they have problems with prioritization. But
8
it can be solved allocating more than one slot per device. Unfortunately, this approach fails
if neighbor devices do not cooperate. Legacy devices do not know about reserves. Hence,
they could try to transmit in a reserved slot. This prevents the convergence to a state without
collisions.
Another way to reserve the channel access is allocating time intervals rather slots. As
proposed in TDuCSMA [6] a service provider could assign a certain percentage of the
bandwidth for each node or user. For instance, a Time Cycle (TC) of 100ms can be
divided in Time Frames (TF) of 10ms. Each TF is allocated for one station. During each
allocated period, the selected node can transmit data without contending. It uses the shortest
AIFS while the others continue applying the normal contention. The decision of access the
channel is maintained in CSMA/CA. The mechanism is implemented on top of IEEE 802.11
architecture. This proposal can guarantee a Constant Bit Rate for demanding users. However,
synchronization and coordination are open issues yet. Coordinators in neighboring networks
need to avoid overlaps. And it is not clear how much the protocol depends on clock precision.
This proposal was tested for audio, video and elastic data applications achieving good results.
B. Discussion
We have analysed the main key enhancements proposed in the literature and reviewed
in the previous subsection. Using the Network Simulator, version 3, we analyze collision
rates per total of received packets and throughput comparing results from the following
protocols: Extending the Initial Contention windows (EICW), time slots allocation without
contention (TDuCSMA) and deterministic backoff (CSMA/ECA). In Figure 4, we observe that
all protocols have successful reduced collisions in relation to the 802.11n standard baseline,
over which 802.11ax is established. However, this reduction does not necessarily yields an
increase in throughput. The simulations followed a scenario with stations sharing the same
channel. They sent UDP data streams to their access points ensuing a generated traffic like the
9
models which represent the typical predominant web traffic, but above the channel capacity.
We have applied the same parameters used in the EICW evaluation in [4] and we added the
CSMA/ECA as available in [7] and TDuCSMA as described in [6]. In each sample, we vary
the number of stations to measure the performance of each protocol as the scenario becomes
denser. Further, we also combine the protocols applying them in pairs and together.
For the Collision Rate, CSMA/ECA achieves the best results individually. The reduction
in collision is a little bit better if ECA is combined with EICW. But TDuCSMA degrades
performance when combined with ECA. TDu changes the determined slot defined by ECA
on every time frame, forcing a new convergence process. In the other hand, when EICW
and TDu are simultaneously applied, they amplify their collision reduction. Intuitively, we
can assume that fewer collisions will result in a better throughput caused by less waste in
channel time. However, the reduction in collisions observed in ECA has no corresponding
in throughput increasing. The problem is the timeout in the queues. While waiting their turn
on ECA, the packages reach the limit time delaying before an opportunity to transmit. This
shows that a collision-free protocol does not always deliver the best performance.
0.6
30
Collisions / Total Rx Packets
0.5
25
0.4
Mbit/sec
802.11 baseline
EICW 20
0.3 CSMA/ECA
TDuCSMA 15
0.2 EICW+ECA
ECA+TDu
0.1 10
EICW+TDu
EICW+ECA+TDu
0 5
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Number of Stations in the Channel Number of Stations in the Channel
(a) (b)
Although the observed results may vary according to the volume and type of traffic, we
notice that the EICW combined with TDuCSMA have gained the highest throughput in major
evaluated scenarios, including triple play. But, researches on these proposals are incipient in
the context of TGax. There is a lack of studies or benchmarks comparing them. This is an
opportunity for future researches. Probably, experiments on these MAC enhancements will
TGax group has discussed other improvements like spatial efficiency and interference
mitigation in order to reduce collisions and improve performance. Since 802.11n, MIMO has
been used to send many streams at same frequency band. This method divides the channel
into parallel subchannels. While each subchannel has lower speed, the parallelism results in
higher aggregated speed. MIMO-OFDM uses multiple antennas turning their signals through
right-angles (orthogonal). Hence, they can be easily decoded even stacked closer together.
At first, MIMO has been employed to increase the throughput between one transmitter and
one receiver. Since 802.11ac, it has been employed to provide simultaneous transmission to
many receivers. In this case, every subchannel carries data for one receiver but only one
can transmit them at time. Thus, the channel access method is the same as a single access.
Nevertheless, 802.11ax probably adds multiple transmitters in MIMO. This demands changes
in MAC protocols to support multiple accesses. They should allow more than one device
adopt a scheduled coordinated method. Coordinated means devices should exchange control
frames to reserve the channel. And scheduled means the AP should trigger the concurrent
transmissions. To reduce overhead, the requests for an uplink may be inserted into a normal
frame. The stations send their common frames just adding one bit into frame header. It
11
informs they have more data to be sent [8]. Then, the AP groups these stations and triggers
the simultaneous access. It sends a control frame setting which stations can transmit. This
frame must contain information about frequency offset, spatial stream, tone, length and so
on. This allows the stations to start their transmissions in a synchronized way as illustrated in
Figure 5. Once the control frame reserves the channel, the other devices are silenced avoiding
collisions. However, the AP access procedure is not clear yet. The trigger frame could be
sent after a SIFS or follow EDCA rules or other method as PCF.
Another promising technique for multi user transmissions is OFDMA. It greatly improved
performance of LTE and WiMax networks where it is already in use. OFDMA assigns different
OFDM subchannels for each user. By that, transmitters can use their assigned subchannels to
transmit without collide with others. But, in this case, only the carrier locking does not imply
that transmissions are not allowed [9]. It makes difficult to check if the channel is clear or busy
in CCA. This is another reason in favor of AP-coordinated MAC protocols. Since 802.11n,
the APs have to get Channel State Information (CSI) from stations for downlink MIMO.
This provides a better view of their domain. The current CSI feedback may be improved to
add more data about PHY layer, buffers and collisions. The coordinators could make better
decisions with this information. For instance, if APs know the type of collisions then they
could distinguish between normal or anomalous collisions. They could apply the appropriate
12
protection if collisions are caused by hidden terminals. The challenge is provide CSI feedback
without introduce much overhead.
Although simultaneous transmissions may cause interference, they can be used to increase
the channel efficiency when implemented in a controlled way. Signal Interference Canceling
(SIC) techniques have been used to provide In-Band Full-Duplex communication. Usually,
devices cannot receive while transmitting in the same channel. It happens because their
own strong signal interferes on other signals from distant sources. But, SIC removes this
interference taking advantage of devices know their own signal. It results in their peer’s
signals and enables communication. And this method does not have costs with extra band to
provide FD. Evaluation results have shown until 300% of throughput enhancement in TGax
proposed scenarios [10].
But, IBFD introduces new challenges to MAC protocols. For instance, other devices
detect a collision when two near devices use IBFD. Thus, they apply Extended Inter Frame
Spaces (EIFS) much longer than normal IFS. Further, the current RTS-CTS control frames
cannot properly reserve the channel for IBFD. The frames could have different lengths. Then
the device sending the RTS cannot stand the other frame length. And the frames are not
transmitted exactly at same time. Hence, the mechanism must be adjusted to (a) exchange
ACK frames, (b) solve the hidden node problem in asymmetric data traffic and (c) avoid
starve other nodes. However, discussions about IBFD are still incipient in TGax forums.
More details about the MAC issues for IBFD systems can be found on the survey in [10].
B. Other improvements
The recent 802.11ac amendment provides MSDU and MPDU aggregation. Frame aggre-
gation decreases MAC header, IFS and other extra overhead. Further, the contention time is
13
incurred only once in the transmission of the large frame instead of many small packets. But,
the probability of successful transmission is higher for smaller frames due wireless errors.
An Adaptive Block Frame Length is discussed to set the ideal length [11]. In another line,
some proposals suggest adding some management functions to the AP. They could determine
priorities based on STA characteristics giving more opportunities for faster devices [12].
Or they could cooperate. An Access Controller can coordinate them in managed deployed
networks, e. g. campus or conference environments [13]. Wifox, ADWISER and MegaMIMO
are practical examples of dense networks operating with a control entity. Further, APs can
share Beacon frames among neighbour BSS. This could reduce network overhead with control
frames. And beacon frames could be prioritized to avoid collisions.
Dynamic Sensitivity Controls have been discussed to achieve CSMA/CA optimizations [14].
They regulate the relative size of contention domains which affects throughput. Simulation
results have shown gains in managed networks as well in dense residential apartment buildings.
Adaptive configuration of CCA threshold (or receiver sensitivity) can reduce the network
operating range. This decreases the collision domain area. But DSC are dependent to scenario
setup and path loss parameters. Collisions increase when using higher CCA levels. And spatial
reuse is suppressed when CCA level is lower. Thus, MAC Protocol should adapt the backoff
procedure according CCA level. For instance, the backoff countdown would continue even if
channel is busy since the signal is weak enough to interfere. Beyond this, the backoff process
can be migrated to signaling through OFDM subcarriers [15]. Back2f and WiFi-BA protocols
are examples on this type of access mechanism. This would reduce the contention time and
avoid collisions. But these proposals require additional hardware such as SIC structures used
in IBFD. However, they could emerge for 802.11ax if they afford backward compatibility and
V. C ONCLUSION
High efficient wireless networks have been a topic of interest to researchers from academia
and industry. Since it was constituted, TGax group works on improve throughput and spectrum
efficiency in 802.11 standard. They have discussed several proposals to solve network
problems in dense environments. Although there is still a long time for its completion,
the amendment drafts suggest some improvements. This article gave an overview of these
the rising number of users. Further, this article also gave trends leading to the near future
implementation. The current themes are classified and discussed. There are many proposals
on the agenda. However, this article contributed to a basis for foreseeing the next generation
Wi-Fi networks.
[10] D. Kim, H. Lee, and D. Hong, “A Survey of In-Band Full-Duplex Transmission: From the Perspective of PHY and
MAC Layers,” IEEE Communications Surveys & Tutorials, vol. 17, no. 4, pp. 2017–2046, Fourthquarter 2015.
[11] N. Zhang, G. Kang, and B. Zhu, “Discussions on adaptive frame length in MAC based on block ACK,” IEEE, doc: IEEE
802.11-14/0613, Tech. Rep., 05 2014, uRL: https://mentor.ieee.org/802.11/dcn/14/11-14-0613-00-00ax-discussions-on-
adaptive-frame-length-in-mac-based-on-block-ack.pptx Last access: 08/26/2016.
[12] K. Yunoki and Y. Misawa, “Access Control Enhancement,” IEEE, doc: IEEE 802.11-13/1349r0, Tech. Rep., 11
2013, uRL: https://mentor.ieee.org/802.11/dcn/13/11-13-1349-00-0hew-access-control-enhancement.pptx Last access:
08/18/2016.
[13] Y. Li, Z. Luo, M. Yang, and B. Sun, “Feasibility of Coordinated Transmission for HEW,” IEEE, doc: IEEE
802.11-13/1157r3, Tech. Rep., 09 2013, uRL: https://mentor.ieee.org/802.11/dcn/13/11-13-1157-03-0hew-feasibility-
of-coordination-transmission-for-hew.ppt Last access: 08/26/2016.
[14] G. Smith, “CCA Proposal DSC Text,” IEEE, doc: IEEE 802.11-16/0310r2, Tech. Rep., 07 2016, uRL:
https://mentor.ieee.org/802.11/dcn/16/11-16-0310-02-00ax-dsc-proposed-text.docx Last access: 08/26/2016.
[15] T. Nguyen, L. Lanante, H. Ochi, T. Uwai, and Y. Nagao, “Uplink multi-user MAC protocol for 11ax,” IEEE, doc:
IEEE 802.11-14/0589r0, Tech. Rep., 05 2014, uRL: https://mentor.ieee.org/802.11/dcn/14/11-14-0598-00-00ax-uplink-
multi-user-mac-protocol-for-11ax.pptx Last access: 08/26/2016.