2016 IEEE Symposium on Computers and Communication (ISCC)

Hidden Terminal Management for Uplink Traffic in

Rate-Controlled WiFi Networks
Mustafa Al-Bado∗ , Cigdem Sengul† , Cormac J. Sreenan ∗ , Kenneth N. Brown∗
∗ CTVR,
Department of Computer Science, University College Cork, Ireland
† Nominet, UK

Emails: mustafa.albado@insight-centre.org, cigdem.sengul@nominet.uk, cjs@cs.ucc.ie, k.brown@cs.ucc.ie

Abstract—This paper exposes several problems in managing rate selection performance degrades with the presence of
hidden terminals for uplink traffic in rate-controlled environ- hidden terminals. Probe-based mechanisms typically reduce
ments, and presents solutions to mitigate them. In particular, their rate when packet losses increase, which may lead either
we focus on scenarios, in which, clients are associated with an
access point (AP). The main challenge stems from the negative to rises the likelihood of collisions because of the increase of
interactions between rate-control protocols and hidden terminals. the transmission time if both transmitters have relatively the
To expose the problems, we use a recent channel estimation same SINR at the receiver, or one of the transmitters will
approach (CEA) to differentiate the reason for packet losses mostly win the medium and harshly decrease the medium
into three categories, noise, congestion and hidden terminals. access chances of the other transmitter. Even when the rate
Our testbed and simulation-based experiments show that the
accuracy of hidden terminal estimations using the CEA de- control algorithm uses channel quality measurements, these
grades as MAC-layer ACK frames are sent with relatively high are typically adversely affected by hidden terminals, and hence
transmission rates. To improve the accuracy of the CEA, the do not predict the best rates to match a given SNR level [11].
results demonstrate the necessity and cost of making the AP Finally, selecting the best rate based on SNR does not always
send ACKs based on the minimum ACK rate of all clients. We improve throughput, when losses are purely due to hidden
propose an adaptive scheme that combines both the CEA and
RTS/CTS messages. The proposed scheme increases the overall terminal collisions.
throughput of Minstrel rate-control algorithm by 60% in case To choose the most suitable transmission rate, hidden termi-
of light congested environments. We also proposed a threshold- nal detection mechanisms need to be used. For this reason, we
based adaptive RTS/CTS scheme based on the prior scheme to exploit the CEA proposed in [8], [12]. This approach classifies
handle the highly congested environments. The threshold-based the reason of packet losses as noise, congestion, or hidden
adaptive RTS/CTS scheme improves the overall throughput of the
adaptive RTS/CTS scheme and Minstrel algorithm between 20- terminals. The main idea is to send back-to-back measurement
35%. Finally, we propose and evaluate an opportunistic burst packets using MAC layer fragmentation/TXOP, and use dif-
scheme, which enforce fairness among clients. Simulation results ferent properties of the IEEE 802.11 Distributed Coordination
show that opportunistic bursting outperforms the prior schemes Function (DCF) to decide between different losses. Our paper
and Minstrel algorithm in Jain’s fairness metric (between 0.11 makes three contributions. First, we evaluate the accuracy of
and 0.38) for a realistic given scenario. It also keeps a relatively
high overall throughput. the CEA in typical rate-controlled WiFi environments with
hidden terminals. Second, we show the need and cost, for
I. I NTRODUCTION hidden terminals, of making the AP transmit ACKs to all
The hidden terminal problem is reported to cause significant clients with a rate that matches the client with the lowest data
degradation in communication in several measurement-based rate. Finally, we leverage the CEA and Network Allocation
studies of enterprise and home WiFi networks [13], [11]. To Vector (NAV) to mitigate the effect of hidden terminals in
expose the effects and understand the degree of degradation, rate-controlled WiFi environments. To mitigate the effect of
we focus on hidden terminals for uplink traffic in rate- hidden terminals, we propose three schemes: (1) adaptive
controlled WiFi networks. RTS/CTS, (2) a threshold-based adaptive RTS/CTS, and (3) an
Managing uplink traffic is gaining importance due to the opportunistic bursting algorithm. The results show that using
use of applications such as cloud storage, video gaming, the CEA and adaptive RTS/CTS scheme with the right AP
video conferencing and Internet of Things (IoT). Here, rate setup improves significantly the overall throughput of light
control is crucial to find the most suitable data transmission congested networks. The threshold-based adaptive RTS/CTS
rate based on current conditions [6]. Current rate control uses additional loss thresholds to detect situations where mea-
protocols are either probe-based mechanisms that search for surement frames experience high congestion losses, improving
the best transmission rates using packet loss as an indicator, the performance further. Finally, the opportunistic bursting
or channel-based mechanisms that use channel quality (e.g., scheme enforces fairness among clients by using bursting for
Signal to Noise Ratio - SNR) to predict the best transmission clients have low medium access chances. The results show
rates. Probe-based mechanisms are more commonly supported that the last scheme outperforms the other schemes in fairness
by wireless drivers (e.g., SampleRate [14] and Minstrel [2]) metric with keeping relatively high overall throughput.
compared to channel-based mechanisms. In either case, the The rest of the paper is organized as follows. In Section II,

978-1-5090-0679-3/16/$31.00 ©2016 IEEE

 2016 IEEE Symposium on Computers and Communication (ISCC)

we present the necessary background and review the related TABLE I

work. In Section III, we introduce the testbed environment, M ADWIFI DEFAULT ACK RATE PER DATA RATE
and describe the testbed-based simulations. In Section IV Data-rates (Mbps) default ACK-rate (Mbps)
and Section V, we evaluate the hidden terminal detection 6&9 6
12 & 24 12
problems in rate-controlled WLANs and evaluate the different 24, 36, 48 & 54 24
approaches to hidden terminal mitigation respectively. We
conclude the paper in Section VI.
tied to correct identification of the cause of packet losses. If
II. BACKGROUND AND R ELATED W ORK
the channel is experiencing increased noise, stations should
This section presents an overview of Inter-Frame Space change to a lower, more robust rate. If, however, losses are due
(IFS), Fragmentation, transmission opportunity (TXOP) and to congestion-based collisions, selecting a lower rate would
rate control. It also discusses the related work on rate control only increase the time packets spend on the medium, leading
and hidden terminal detection. to even higher congestion. Also, for hidden terminals, reducing
IFS: The time interval between frames is called the IFS. the rate aggravates the problem, providing more opportunities
Different IFSs ensure different priority levels to access the to collide. Nevertheless, in the presence of hidden terminals, it
wireless medium. High priority frames, such as ACKs, are becomes hard to distinguish packet losses due to channel errors
sent using Short IFS (SIFS). Data frames wait DIFS and a and congestion. There are only a few papers [9], [15], [7]
station uses Extended IFS (EIFS), when it determines that that address this problem and they all use adaptive RTS/CTS,
the medium is idle following a reception error. Reception of where RTS/CTS is turned on if congestion is suspected.
an error-free frame during EIFS resynchronizes the frame to An active measurement-based approach for distinguishing
the actual channel idle/busy states. This especially plays an congestion, hidden terminals and noise based losses was
important role in channel access in the case of collisions. proposed in [8], [12]. In this work, congestion-based collisions
Fragmentation: results in a packet divided into multiple occur when two stations simultaneously count down to zero
fragments. These are sent in a burst, separated by an SIFS, using DCF. Therefore, while these collisions only affect data
after contending for the channel only for the first fragment. frames, hidden terminals and noise affect both data and ACK
Any fragment that cannot be sent on time causes all remaining frames. To estimate the probability of losses due to congestion
fragments to be deleted at the receiver. (pc ), hidden terminals (ph ) and noise (pn ), two equal-length
TXOPs: are defined for Enhanced Distributed Channel back-to-back fragments (F1 and F2) are sent using two differ-
Access (EDCA), where the duration of a TXOP determines the ent methods: Standard (S) and Unprotected (U). The Standard
time a sender maintains uninterrupted control of the channel method complies with 802.11: fragments and their ACKs are
(including the time required for ACKs). Hence, multiple sent separated by an SIFS. Hence, F2 is protected by the NAV
frames can be sent within a single TXOP. The transmission of against collision and hidden terminal losses and, F1 is subject
multiple frames are separated again by a SIFS. Later in this to all three types of losses. Therefore, pn is determined by:
section, we explain how frame bursts are used to distinguish
different frame losses. P r(S: F2 success) = 1 − pn = A1 /T1 , (1)
Rate control: While the standard does not define any
where T1 is all F2 frames, A1 of which were successful.
rate control protocols, it does allow the possibility to send
In the Unprotected method, the NAV of F1 only covers the
management and data frames using different modulation and
duration of its ACK, but not F2. Hence, F2 may experience
coding schemes, which achieve different transmission rates.
noise and hidden terminal losses, but not congestion. There-
Given that certain rates can only be achieved under certain
fore:
channel conditions, rate control protocols aim to tune the
transmission rate to optimize throughput. Also, to improve P r(U: F2 success) = (1 − ph )(1 − pn ) = A2 /T2 , (2)
the medium access, the current wireless drivers change the
rate of both data and ACK frames. ACK rates are selected where T2 are all F2 frames using the Unprotected method, of
based on the sender data rate. In this paper, we focus on which A2 were successful. Using (2):
the Madwifi driver, which is commonly used by the research ph = 1 − (T1 · A2 )(A1 · T2 ). (3)
and development community. Table I [7] shows available rates
for IEEE 802.11a and their default ACK-rate in the Madwifi Using pn and ph estimations, pc is calculated as:
driver. We also use Minstrel [2] as the default rate control
P r(F1 success) = (1 − pc )(1 − ph )(1 − pn ) = A0 /T0 , (4)
protocol.
Interaction between Rate Control and Hidden Termi- where To are all F1 frames, of which A0 succeeded. This
nals: The majority of the rate control protocols in the litera- mechanism can also be built using TXOPs as in [8].
ture, such as Minstrel [2], SampleRate [14] and AMRR [10],
use packet loss statistics as an indication of channel quality, III. A IMS AND M ETHODOLOGY
and to update transmission rates. The appropriateness of rate To expose the hidden-terminal problems in rate-controlled
control decisions (i.e., to decrease or increase rate) are directly WiFi networks, we improve the accuracy and employ the
 2016 IEEE Symposium on Computers and Communication (ISCC)

measurement-based approach proposed in [8], [12] with adap-

tive RTS/CTS mechanisms. We aim to enable RTS/CTS when
a hidden terminal case is detected by the improved estimator.
We extend the work in [8] by maintaining separate pc , ph
and pn estimates for each rate. In the rest of this section, we
explain how this extension is implemented and evaluated in
both testbed and simulation.

A. Testbed and measurement set-up

Fig. 1. An example scenario that illustrate when the estimator fails.
All the testbed experiments take place in the NITOS [3]
wireless outdoor testbed. We use both Orbit-like nodes as well
as wireless nodes of a higher specification. Orbit-like nodes

0.30
Data−rate:24Mb, ack−rate:24Mp
are PCs equipped with 1 GHz VIA C3 processor, 512 MB Data−rate:24Mb, ack−rate:6Mp
Data−rate:36Mb, ack−rate:24Mp

0.25
Data−rate:36Mb, ack−rate:6Mp
of RAM, 40 GB disk. The higher specification nodes have

0.20
Intel Core 2 Duo P8400 2.26 GHz processor and 2G DDR3
RAM and Solid state drive. All nodes have two Ethernet

0.15
Pn
ports, two 802.11 a/b/g cards, and a Chassis Manager. The

0.10
wireless NIC is Atheros AR5213A [1] using the Madwifi

0.05
driver, and the antennas are multi-band 5dbi, that operate
on both 2.4Ghz & 5Ghz. The operating system is Linux 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

3.2.0 − 32-generic. We use only one of the available wireless Hidden terminal load (Mbps)

interfaces with IEEE 802.11a[4] on channel 48 as it has the Fig. 2. Probability of noise errors, pn .
lowest interference at the time of running the experiments.
Based on a node’s role in the experiment, one of the following
modes is assigned to its wireless interface: AP for access point Standard and Unprotected methods. First, we present a testbed
, STA for station and monitor mode for a sniffer, which experiment to show that this assumption does not always
captures and stores packets for further analysis. Monitor mode hold in a rate-controlled WLAN and the solution for that.
uses the radiotap header, which records PHY parameters Then, simulation results are presented to show the cost of the
such as the Received Signal Strength (RSS). Both burst and proposed solution.
fast frame modes are disabled. Hence, a node sending
a data frame (not fragments) must wait at least DIFS plus A. Effect of Rate
backoff time before transmission. To illustrate the case when rate-controlled would cause in
We use iperf to generate UDP Unicast traffic between a accurate statistics by the CEA , we use the scenario in Fig 1. In
STA and an AP. All packets are 1940 B, and are fragmented to this scenario, two nodes, Node 1 and Node 2, are connected
two ∼ 1024 B fragments (payload plus headers). Hence, the to the same AP, but hidden from each other. Node 1 has a
second fragments are protected and can be used to calculate high rate connection to the AP, where both data and ACK
the noise error probability in Equation (1). The experiment rates are 24 Mbps. Node 2 has a weaker link to the AP and
results are presented with their 95% confidence intervals from uses the transmission rate of 6 Mbps. In this case, Node 2
6 experiment runs. Each run lasts for 60 s. may not be able to decode the ACKs sent by the AP to Node
1. Hence, even if Node 2 waits for an EIFS, its data frames
B. Simulation settings may collide with the data frames of Node 1, regardless of
We use simulation to understand more complex scenarios whether they are the first or the second packet. This clearly
that have higher node density. Our simulator is ns-3 version breaks the assumption that the second packet is collision free,
3.20. We use the model in [5] to create network topologies and affects the accuracy of the computation of the noise and
that represent the wireless link errors in a real testbed. Instead hidden terminal probabilities (i.e., ph and pn ).
of fragmentation, simulations send UDP or TCP unicast traffic Proposed solution - Lowest ACK Rate: To resolve the
in TXOP mode, as in [8], [12]. problem, we propose that the AP always sends ACKs with
We use the ns3 OnOffApplication as the traffic generator. the minimum ACK rate required by the set of its clients (e.g,
The simulation results are presented with their 95% confidence sending all ACKs at 6 Mbps in Fig 1). The proposed solution,
intervals from 10 simulation runs. Each run lasts 120 seconds. as we will see later, increases the accuracy of the CEA running
on the clients having strong SINR at the receiver. Moreover, it
IV. H IDDEN T ERMINAL D ETECTION saves the other hidden terminals (i.e., with weak SINR at the
The classification of link impairments as described in [8] receiver) from unnecessary transmission (i.e., from wasting a
and [12] is based on the assumption that it is possible to transmission attempt).
control when the second fragment/packet is protected against Experimental results: To set-up the experiment in Fig 1,
hidden terminals. This is the main difference between the we use three test-bed nodes as AP, Node 1 and Node 2. We
 2016 IEEE Symposium on Computers and Communication (ISCC)

calibrate Node 1’s transmission power to achieve a transmis-

sion rate up to 24 or 36 Mbps, while Node 2 is calibrated to

26
Lowest Ack−rate
Default Ack−rate

25
use only the basic transmission-rate (i.e., 6 Mpbs). This way,

24
we emulate the behavior of Minstrel rate control, which picks

Throughput (Mbps)
23
the best-throughput rate. Neither Node 1 nor Node 2 carrier

22
sense each other and hence, they are hidden terminals. Finally,

21
20
we verify that Node 2 can decode ACKs transmitted with the

19
basic transmission rate (i.e., 6Mbps) by the AP and not the

18
higher transmission rates.

17
2 3 4 5
To understand the performance of a hidden terminal with Number of contention nodes

the weaker link, we vary Node 2’s packet sending rate between Fig. 3. The worst case scenario where there are no hidden terminals, but a
0.2 Mbps and 1.8 Mbps, while Node 1 has always a packet slow client.
to send. All packets are sent as fragments using the Standard
method (described in Section II) and hence, second fragment
losses should only occur due to noise. Fig 2 shows pn :

5
Equation (1) measured by Node 1 as Node 2 increases its
traffic load. Note that pn needs to be correctly measured to

Throughput (Mbps)
be able to calculate hidden terminal probability ph . When the

4
Default 802.11a
Adaptive RTS/CTS + low ACK
ACK rate is selected according to Table I, we see a direct Adaptive RTS/CTS + default ACK

relation between noise probability, pn and the load of Node 2

3
after a threshold load (0.6 Mbps for 24 Mbps and 0.8 Mbps
for 36 Mbps). Above 1 Mbps, pn goes up to ≈ 0.25 when
1 2 3 4
Node 1 uses 24 Mbps transmission rate and AP uses 12 Mbps Number of Hidden Terminal Nodes
for ACKs. When Node 1 switches to 36 Mbps, and the AP,
Fig. 4. Slow hidden terminals competing against a node with high SNR but
consequently, to 24Mbps for ACKs, we observe a lower pn . low traffic
This is because faster transmissions have a better chance to
avoid collisions.
Fig 2 also shows pn with the proposed solution of using varying SNR and insufficient measurements for hidden termi-
the lowest ACK rate. In this case, pn does not increase with nal detection. We use the Minstrel [2] rate control algorithm.
the hidden terminal load, and is a proper measure of the noise
A. Clients with Varying SNR
probability.
We simulate an asymmetric interference scenario with 2-5
B. Cost of the Lowest ACK Rate Solution clients and a single AP. One of the clients has always a high
SNR but has a low traffic demand (e.g., can transmit up to
Using the lowest used ACK rate for all clients may lead to
54 Mbps, but sends with a packet rate of 0.5 Mbps). All the
a reduction of the overall network throughput. To understand
other clients have weak links to the AP (i.e., can support only
the cost, we go back to the scenario presented in Section IV-A.
6 Mbps) but they are always backlogged. The traffic is UDP
We consider the worst case scenario when there are no hidden
with 1024 B packet size.
terminal clients but one of the clients can only send to the AP
Fig. 4 shows the overall throughput for three algorithms:
with the lowest data rate (i.e., 6 Mbps for 802.11a). The other
default 802.11a, adaptive RTS/CTS using the CEA with the
clients can send to the AP with the highest rate (i.e., 54 Mbps
default ACK rate and the CEA with the lowest ACK rate. In
for 802.11a). Fig 3 shows that the total throughput reduction
this scenario, Adaptive RTS/CTS with the default ACK rate
as the number of high-rate clients increase. In this case, the
does not bring much advantage. However, using the lowest
throughput reduction varies between 3% to 13%.
ACK rate fixes the hidden terminal detection, and spares
unnecessary transmission for the slow hidden terminal clients.
V. H IDDEN T ERMINAL M ITIGATION
Particularly, the slow rate clients read the NAV in the ACK.
Next, we evaluate the performance of different approaches thus, they save their CW of needlessly increase, which has a
for hidden terminal mitigation. We first propose to combine direct affect on their throughput.
the CEA with adaptive RTS/CTS. Here, we turn RTS/CTS on
when the hidden terminal probability is higher than a certain B. Insufficient Measurements
threshold (10% in our evaluation based on simulation results). When the channel is highly congested, it may be difficult
RTS/CTS is turned off when the hidden-terminal probability to collect enough statistics to determine the cause of packet
drops below this threshold. losses. Hidden terminal detection is affected because the CEA
All evaluations are based on simulation and consider various relies on the loss statistics of the second packet to differentiate
scenarios, which affect channel utilization. Specifically, we between hidden terminal and noise losses. However, when
simulate challenging scenarios such as hidden terminals with there is high congestion, the first packet has a higher loss
 2016 IEEE Symposium on Computers and Communication (ISCC)

Algorithm 1 Opportunistic Burst Scheduling

1: timer, T : Expiry time of last transmitted packet

10 11 12 13
Default 802.11a
Adaptive RTS/CTS + loss thresh
Adaptive RTS/CTS + low ACK
Adaptive RTS/CTS + default ACK 2: C: # of ACKs received from the AP
3: Bs : Current burst size
Throughput (Mbps)

4: Bmin : Minimum burst size

9
8

5: Bmax : Maximum burst size

7

6: K: # of packets to add to/remove from Bs

6
5

7: if T.IsExpired = F ALSE then

4

8: C ←C +1
3

1 2 3
Number of Hidden Terminal Nodes
4
9: if C < Bs − 1 then
10: Send packetC+1 with NAV covers packetC+2
Fig. 5. High congestion leads to insufficient measurements. This is detected
using two thresholds, which improves throughput. 11: else
12: if C = Bs − 1 then
TABLE II 13: Send packetC+1 with NAV ← 0
T WO LOSS RATIOS ND A DAPTIVE RTS/CTS 14: else
15: if C = Bs then
L1/2 > T hratio L1 > T hloss Case Outcome
True True High congestion RTS/CTS on 16: Bs ← max (1, Bs − K)
True False Low traffic RTS/CTS off 17: end if
False True Possibly HTs RTS/CTS off 18: end if
False False Possibly HTs RTS/CTS off
19: if B < Bmin then
20: Disable RTS messages
probability. In this case, TXOP is not reserved for second 21: end if
packets. Therefore, there will not be enough second packets, 22: C←0
degrading the accuracy of estimations. 23: end if
To address this problem, we introduce two new thresholds 24: else
to turn on RTS/CTS: T hratio and T hloss . T hratio defines 25: if C = Bs − 1 then
a threshold for the ratio of first packets to all received 26: B ← min(B + K, Bmax )
packets, L1/2 . Normally, this ratio should be 0.5, and we 27: end if
select T hratio = 0.7 for our evaluations. T hloss is the loss 28: Request medium using RTS message
ratio of first packet, L1 . Again, we select this threshold as 29: end if
0.7. Table II lists all the possible cases based on the value of
these ratios. The RTS/CTS is turned on when both ratios are
above thresholds. We simulate the scenario in the previous enabled if the used rate requires them to have a successful
section, but in this case the client with high SNR is also transmission even in the existence of another rates that can be
always backlogged, creating high congestion. Fig. 5 depicts used without RTS messages. Receiving the ACK, the client
the aggregate throughput. We observe that the lowest ACK sends the next packet within SIFS. This way, clients burst as
rate does not help much with estimations and hence, adaptive long as they have packets and they do not exceed the burst
RTS/CTS performance is not affected by ACK rate. Here, the size (Lines 7-13 in Algorithm 1).
problem is the lack of enough measurements to detect hidden In Algorithm 1, the burst size is dynamic and determined
terminals. Using the thresholds, this problem is mitigated, based on the results of transmitting the last packet in the burst.
resulting in overall increase in throughput between 20-35%. The burst size is always kept between two thresholds: Bmin
and Bmax to strike a balance between clients to access the
C. Opportunistic Burst Scheduling for Hidden Terminal Mit- medium. The prior role is to enforce fairness by allocating
igation extended time for clients that suffering severe hidden terminal
Although adaptive RTS/CTS improves performance, it does situation (Lines 15-17 and 25-27 in Algorithm 1). Here, if
incur additional overhead. Next, we propose an opportunistic the last packet in the burst (i.e, the unguarded packet) is
burst algorithm that amortizes the cost of RST/CTS messages successfully transmitted the burst size of the next transmission
by allocating more transmission time to clients with experi- will be K packets shorter until it reaches Bmin , then the RTS
encing poor channel conditions when they capture the channel. messages will be disabled. While if the last packet is not
The opportunistic burst algorithm, presented in Algorithm 1, transmitted successfully, the next burst transmission will be
is called at each Acknowledgement (ACK) reception after extended by K packets until Bs reaches Bmax .
enabling RTS massages and allocates additional TXOP based For more realistic evaluation, we simulate the indoor testbed
on the burst size Bs except the last packet in the burst. link profiles published in [5]. The testbed consists of 9 nodes.
To improve fairness, RTS messages are enabled only if the Node 4, which is reachable from the other nodes, is selected
transmitter cannot leverage from capture affect [5]. This is as the AP. Table III shows the number of hidden terminals for
different than adaptive RTS/CTS, where RTS messages are each node and the average SNR of this node at the AP. This is
 2016 IEEE Symposium on Computers and Communication (ISCC)

TABLE III VI. C ONCLUSION

H IDDEN TERMINALS IN TESTBED
The hidden terminal problem is well-known in WLANs.
Node # of hidden terminals SNR @ AP Current approaches to detect hidden terminals are typically
1 3, 5, 7, 8, and 9 −86 tested under simple scenarios without taking into account rate
2 8 and 9 −74
3 1, 5, 6, 7, 8 and 9 −76 control. In today’s WLANs, rate control of both data and ACK
5 1, 3, 6, 7 and 8 −54 frames is an automatic feature. Our results show that hidden
6 3 and 5 −85 terminal detection accuracy significantly degrades in rate-
7 1, 3 and 5 −76
8 1, 2, 3 and 5 −77
controlled environments. We show hidden terminal detection
9 1, 2 and 3 −79 can be improved by selecting ACK rates appropriately. Using
the improved hidden terminal detection, it becomes possible
to mitigate hidden terminals using adaptive RTS/CTS, and our
opportunistic bursting algorithm, by allowing slower hidden
14

terminals to have a better chance to access to the medium.

12

VII. ACKNOWLEDGEMENT
Throughput (Mbps)
10

This publication has emanated from research conducted

9
8

with the financial support of Science Foundation Ireland (SFI)

7

and is co-funded under the European Regional Development

6

Opportunistic burst
5

802.11a (RTS/CTS enabled) Fund under Grant Numbers 10/CE/i1853 and 13/RC/2077.
4

Adaptive RTS/CTS + loss thresh

Default 802.11a
3

50 75 100
Load per Client (MB)
125 150 R EFERENCES
[1] Atheros. http://www.wifi-stock.com/file/cm9 user manual.pdf.
Fig. 6. Aggregate TCP traffic throughput in simulated testbed links
[2] minstrel specification. http://madwifi-project.org/browser/madwifi/trunk/
athrate/minstrel/minstrel.txt
[3] Nitos: Network implementation testbed laboratory using open source
platforms.
a complex scenario with varying number of hidden terminals [4] IEEE standard for information technology-telecommunications and
for each client. In the evaluation, we simulate a file upload, information exchange between systems-local and metropolitan area
with file-sizes varying as 50, 75, 100, 125 and 150 MB. The networks-specific requirements - Part 11: wireless lan medium access
control (MAC) and physical layer (PHY) specifications. IEEE Std
clients choose a random number between 0 and 75 s to start 802.11-2007 (Revision of IEEE Std 802.11-1999), Jun 2007.
a file upload. Here, we set K = 1, Bmin = 3 and Bmax = 6. [5] M. Al-Bado, C. Sengul, and R. Merz. What details are needed for
Figs. 6 and 7 show the aggregate throughput achieved, wireless simulations? A study of a site-specific indoor wireless model.
In Proceedings of the INFOCOM, pages 289–297, Mar. 2012.
and Jain’s fairness among different flows, when all clients [6] S. Biaz and S. Wu. Rate adaptation algorithms for ieee 802.11 networks:
are using TCP traffic. The results show that increasing the A survey and comparison. In ISCC 2008. IEEE Symposium on, pages
upload file size above 70 MB leads to a slight decrease of the 130–136, July 2008.
[7] X. Chen, P. Gangwal, and D. Qiao. RAM: rate adaptation in mobile
overall throughput when the opportunistic bursting is used in environments. IEEE Transactions on Mobile Computing, 11(3):464–477,
comparison with default 802.11a and threshold-based adaptive March 2012.
RTS/CTS. However, Jain’s fairness metric is considerably [8] D. Giustiniano, D. Malone, D. J. Leith, and K. Papagiannaki. Measuring
transmission opportunities in 802.11 links. IEEE/ACM Transactions on
higher than any of the other models (between 0.11-0.18 for Networking, 18(5):1516–1529, Oct. 2010.
threshold-based adaptive RTS/CTS and between 0.15-0.36 for [9] J. Kim, S. Kim, S. Choi, and D. Qiao. Cara: Collision-aware rate adap-
default 802.11a, see Fig 7). The results suggest that using RTS tation for ieee 802.11 wlans. In Proceedings of the IEEE INFOCOM,
pages 1–11, April 2006.
messages and bursting for severe condition hidden terminal [10] M. Lacage, M. H. Manshaei, and T. Turletti. Ieee 802.11 rate adaptation:
nodes reduces slightly the overall throughput but affectively A practical approach. In Proceedings of the ACM MSWiM ’04, pages
improves the fairness between clients. 126–134, New York, NY, USA, 2004. ACM.
[11] K. LaCurts and H. Balakrishnan. Measurement and analysis of real-
world 802.11 mesh networks. In Proceedings of the ACM IMC ’10,
pages 123–136, 2010.
[12] D. Leith and D. Malone. Field measurements of 802.11 collision, noise
and hidden-node loss rates. In Proceedings of the 8th International
0.9

Symposium on Modeling and Optimization in Mobile, Ad Hoc and

Wireless Networks (WiOpt), pages 412–417, May 2010.
0.8

[13] J. Manweiler, P. Franklin, and R. Choudhury. Rxip: Monitoring

the health of home wireless networks. In Proceedings of the IEEE
0.7
Fairness

INFOCOM, pages 558–566, March 2012.

[14] R. T. Morris, J. C. Bicket, and J. C. Bicket. Bit-rate selection in wireless
0.6

networks. Technical report, Master?s thesis, MIT, 2005.

[15] S. H. Y. Wong, H. Yang, S. Lu, and V. Bharghavan. Robust rate
0.5

Opportunistic burst
802.11a (RTS/CTS enabled)
Adaptive RTS/CTS + loss thresh
adaptation for 802.11 wireless networks. In Proceedings of the ACM
MobiCom ’06, pages 146–157, 2006.
0.4

Default 802.11a

50 75 100 125 150

Load per Client (MB)

Fig. 7. Fairness of TCP traffic in a simulated testbed