A Dynamic Link Adaptation for Multimedia Quality-Based Communications in IEEE_802.11 Wireless Networks

Full text

Page 1

2
A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in
IEEE_802.11 Wireless Networks
Ahmed Riadh Rebai and Mariam Fliss
Wireless Research Group, Texas A&M University at Qatar,
Qatar
1. Introduction
Assuming that the IEEE 802.11 Wireless Local Area Networks (WLANs) are based on a
radio/infrared link, they are more sensitive to the channel variations and connection
ruptures. Therefore the support for multimedia applications over such WLANs becomes
non-convenient due to the compliance failure in term of link rate and transmission delay
performance. Voice and broadband video mobile transmissions (which normally have strict
bounded transmission delay or minimum link rate requirement) entail the design of various
solutions covering different research aspects like service differentiation enhancement (Rebai
et al., 2009), handoff scheme sharpening (Rebai, 2009a, 2009b, 2010) and physical rate
adjustment. The core of this chapter focuses on the last facet concerning the link adaptation
and the Quality of Service (QoS) requirements essential for successful multimedia
communications over Wi-Fi networks. In fact, the efficiency of rate control diagrams is
linked to the fast response for channel variation. The 802.11 physical layers provide multiple
transmission rates (different modulation and coding schemes). The original 802.11 standard
operates at 1 and 2 Mbps (IEEE Std. 802.11, 1999). Three high-speed versions were added to
the original version. The 802.11b supports four physical rates up to 11 Mbps (IEEE Std.
802.11b, 1999). The 802.11a provides eight physical rates up to 54 Mbps (IEEE Std. 802.11a,
1999). The last 802.11g version, maintains 12 physical rates up to 54 Mbps at the 2.4 GHz
band (IEEE Std. 802.11g, 2003). As a result, Mobile Stations (MSs) are able to select the
appropriate link rate depending on the required QoS and instantaneous channel conditions
to enhance the overall system performance. Hence, the implemented link adaptation
algorithm symbolizes a vital fraction to achieve highest transmission capability in WLANs.
“When to decrease and when to increase the transmission rate?” are the two fundamental
matters that we will be faced to when designing a new physical-rate control mechanism.
Many research works focus on tuning channel estimation schemes to better detect when the
channel condition was improved enough to accommodate a higher rate, and then adapt
their transmission rate accordingly (Habetha & de No, 2000; Qiao et al., 2002). However,
those techniques usually entail modifications on the current 802.11 standard. In (del Prado
Pavon & Choi, 2003), authors presented a motivating rate adaptation algorithm based on
channel estimation without any standard adjustment. However, this scheme supposes that
all the transmission failures are due to channel errors and not due to multi-user collisions.

Page 2

Multimedia – A Multidisciplinary Approach to Complex Issues

30
Another way to perform link control is based on local Acknowledgment (Ack) information
for the transmitter station (Qiao & Choi, 2005). Consequently, two new techniques (Chevillat
et al., 2003; Kamerman & Monteban, 1997) where accepted by the standard due to their
efficiency and implementation simplicity. In fact, the source node tries to increase its
transmission rate after successive successful Ack responses, and therefore they do not
involve any change for the 802.11 standard. Moreover and as it was demonstrated by
(Sangman et al., 2011) a fine and excellent physical link adjustment will carry out a quality-
aware and robust routing for mobile multihop ad hoc networks. A good study (Galtier,
2011) was recently addressed regarding the adaptative rate issues in the WLAN
Environment and highlighted the high correlation between the Congestion Window (CW) of
the system, and the rate at which packets are emitted. The given analytical approach opens
the floor and shows that the different mechanisms that have been implemented in the MAC
systems of WLAN cards have strong correlations with other transmission parameters and
therefore have to be redesigned with at least a global understanding of channel access
problems (backoff and collisions) and rate adaptation questions.
In this chapter we propose a new dynamic time-based link adaptation mechanism, called
MAARF (Modified Adaptive Auto Rate Fallback). Beside the transmission frame results, the
new model implements a Round Trip Time (RTT) technique to select adequately an
instantaneous link rate. This proposed model is evaluated with most recent techniques
adopted by the IEEE 802.11 standard: ARF (Auto Rate Fallback) and AARF (Adaptive ARF)
schemes. Thus, we are able to achieve a high performance WLAN transmission.
Consequently, we can extend this approach in various Wi-Fi modes to support multimedia
applications like voice and video tasks.
The rest of the chapter is organized as follows. Section 2 offers a literature survey on related
link-adjustment algorithms and the actual used ones. Section 3 is dedicated to the new
proposed MAARF method and its implementation details. Simulation results will be given
in Section 4 to illustrate the link quality improvement of multimedia transmissions over Wi-
Fi networks and to compare its performance with previous published results (Kamerman &
Monteban, 1997; Lacage et al.,2004). We show how the proposed model outperforms
previous approaches (ARF and AARF) because of its new time-based decision capability in
addition to Ack count feature.
2. Review of the current rate-control approaches
First we recall that the standard IEEE802.11 (IEEE Std. 802.11a, 1999; IEEE Std. 802.11b, 1999;
IEEE Std. 802.11g, 2003) includes various versions a/b/g and allows the use of multiple
physical rates (from 1Mbps to 54Mbps for the 802.11g). Therefore several studies have been
made to develop mechanisms which lead to adapt transmission attempts with the best
physical available rate depending on the estimated channel condition to avoid transmission
failures with Wi-Fi connections. The most important issues that should be taken into
account and are responsible for the design of a reliable rate adaptation mechanism are:
•
The channel condition variation due to a packet transmission error which results to
multiple retransmissions or even a transmission disconnection.
The channel sensitivity against interferences (Angrisani et al., 2011) due to disturbing
incidences, additive random noises, electromagnetic noises, the Doppler effect, an
accidental barrier or natural phenomena.
•

Page 3

A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in IEEE_802.11 Wireless Networks

31
•
The packet emissions latency which affects the autonomy of mobile stations in case of
transmission error (since the communication duration is extended and the energy
consumption becomes more important).
The MSs Mobility leads to a distances change, hence, to an appropriate mobility
management protocol over Wi-Fi connections.
•
Depending on the instantaneous channel quality, a rate adjustment will be always needed to
achieve better communication performance with respect for multimedia QoS requirements.
Since WLAN systems that use the IEEE802.11g (IEEE Std. 802.11g, 2003) physical layer offer
multiple data rates ranging from 1 to 54 Mb/s, the link adaptation can be seen as a process
of switching or a dynamic choosing mechanism between different physical data rates
corresponding to the instantaneous channel state. In other words, it aims to select the ‘ideal’
physical rate matching the actual channel condition. The best throughput can be larger or
smaller than the current used one. The adequate rate will be chosen according to the
instantaneous medium conditions. There are two criteria to properly evaluate this
adaptation/adjustment: the first is the channel quality estimation; secondly is the adequate
rate selection.
The estimation practice involves a measurement of the instantaneous channel states
variation within a specific time to be able to predict the matching quality. This creates a
large choice of indicator parameters on the medium condition that may include the
observed Signal to Noise Ratio (SNR), the Bit Error Rate (BER), and the Received Signal
Strength Indicator (RSSI). Those various physical parameters express instantaneous
measurements operated by the 802.11 PHY card after completion of the last transmission.
Regarding the rate selection formula, it entails a first-class exploitation of channel condition
indicators to better predict the medium state and then fit/adjust the suitable physical rate
for the next communication. Consequently, this process will reduce packets’ retransmissions
and the loss rate. Bad channel-quality estimation would result in performance degradation.
Thus, inaccurate assessments resulting from a bad choice of medium state indicators give
rise to inappropriate judgments on the instantaneous conditions and cause deterioration on
the observed performance. Therefore, this estimation is essential to better support
multimedia services and maximize performance and the radio channel utilization.
Accordingly, during packets transmission, the corresponding MS may increase or decrease
the value of its physical rate based on two different approaches:
a.
With the help of accurate channel estimation, the MS will know precisely when the
medium conditions are improved to accommodate a higher data rate, and then adapt its
transmission rate accordingly. However, those techniques (Habetha & de No, 2000;
Qiao et al., 2002) require efforts to implement incompatible changes on the 802.11
standard. Another research work (del Prado Pavon & Choi, 2003) have presented a very
interesting data rate adapting plan based on RSSI measurements and the number of
transmitted frames for an efficient channel assessment without any modification on the
standard. On the other hand, this plan operates under the assumption that all
transmission failures are due to channel errors. Thus, it will not work efficiently in a
multi-user environment where multiple transmissions may fail due to collisions and not
only to the channel quality.

Page 4

Multimedia – A Multidisciplinary Approach to Complex Issues

32
b.
The alternative way for the link adaptation is to carry out decisions based exclusively on
the information returned by the receiver. In 802.11 WLANs, an acknowledgment (ACK) is
sent by the receiver after the successful data recovery. Only after receiving an ACK frame
the transmitter announces a successful transmission attempt. On the other hand, if an
ACK is either incorrect or not received, the sender presumes a data transmission failure
and reduces its actual data rate down to the next available physical rate-slot. In addition,
the transmitter can increase its transmission rate after assuming a channel condition
enhancement by receiving a specific number of consecutive positive ACKs. These
approaches (Qiao & Choi, 2005; Chevillat et al., 2003) do not require changes on the actual
Fi-Wi standard and are easy to deploy with existing IEEE 802.11 network cards.
Various additional techniques have been proposed in the literature to sharpen the accuracy
of the rate adaptation process and improve the performance of IEEE 802.11 WLANs. The
authors in (Pang et al., 2005) underlined the importance of MAC-layer loss differentiation to
more efficiently utilize the physical link. In fact, since IEEE 802.11 WLANs do not take into
account the loss of frames due to collisions, they have proposed an automatic rate fallback
algorithm that can differentiate between the two types of losses (link errors and collisions
over the wireless link). Moreover it has been shown in (Krishnan & Zakhor, 2010) that an
estimate of the collision probability can be useful to improve the link adaptation in 802.11
networks, and then to increase significantly the overall throughput by up to a factor of five.
In (Xin et al., 2010) the authors presented a practical traffic-aware active link rate adaptation
scheme via power control without degrading the serving rate of existing links. Their basic
idea consists to firstly run an ACK based information exchange to estimate the upper power
bound of the link under adaptation. Then by continuously monitoring the queue length in
the MAC layer, it would be easy to know whether the traffic demand can be met or not. If
not, the emitting power will be increased with respect to the estimated power upper-bound
and will switch to a higher modulation scheme. A similar strategy was presented in
(Junwhan & Jaedoo, 2006) that provides two decisions to estimate the link condition and to
manage both the transmission rate and power.
Several research works (Haratcherev et al., 2005; Shun-Te et al., 2007; Chiapin & Tsungnan,
2008) have implemented a cross-layer link adaptation (CLLA) scheme based on different
factors as: the number of successful transmissions, the number of transmission failures, and
the channel information from the physical layer to determine actual conditions and therefore
to adjust suitably transmission parameters for subsequent medium accesses. As well in (Chen
et al., 2010) a proper-designed cross application-MAC layer broadcast mechanism has been
addressed, in which reliability is provided by the application layer when broadcasting error
corrections and next link rate adaptations (resulting from the MAC layer).
Another approach (Jianhua et al., 2006) has been developed where both packet collisions and
packet corruptions are analytically modeled with the proposed algorithm. The models can
provide insights into the dynamics of the link adaptation algorithms and configuration of
algorithms parameters. On the other hand, in (An-Chih et al., 2009) the authors presented a
joint adaptation of link rate and contention window by firstly considering if a proper backoff
window has been reached. Specifically, if the medium congestion level can be reduced by
imposing a larger backoff window on transmissions, then there may be no need to decrease
the link rate, given that the Signal to Interference-plus-Noise Ratio (SINR) can be sustained.

Page 5

A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in IEEE_802.11 Wireless Networks

33
In the rest of this section we decide to provide details and discuss only the two main
currently-implemented techniques.
2.1 Auto Rate Fallback (ARF)
Auto Rate Fallback (Kamerman & Monteban, 1997) was the first rate-control algorithm
published and quickly adopted/integrated with the Wi-Fi standard. It was designed to
optimize the physical rate adjustment of the second WLANs generation (specifically for
802.11a/g versions which allow multi-hop physical rates). The ARF technique is based
simply on the number of ACKs received by the transmitting MS to determine the next rate
for the next frame transmission. This method does not rely on hidden out-layer information,
such as a physical channel quality measurement like (i.e. the SNR value). Thus, it was easy
to implement and fully compatible with the 802.11 standard. In fact, after a fixed number of
successful transfers equal to 10 or the expiration of a timer T initially launched the ARF
increments the actual physical transmission rate from Ri to a higher rate Ri+1 among those
allocated by the standard. In other words, the ARF mechanism decides to increase the data
rate when it determines that channel conditions have been improved. Unlike other
algorithms reported in the literature, the ARF detection is not based on Physical layer
measurements upon a frame delivery. Basically it simply considers the medium status
improvement by counting the number of consecutive successful transmissions made or the
timer (T) timeout. This timer is defined as the maximum waiting delay which will be
launched by the MS each time it switches between the given data rates. Once this timer ends
without any rate swap the MS will be testing a higher available rate. This practical
implementation is considered as second alternative to adapt the best transmission rate since
it covers the case that medium conditions are excellent and favorable to adopt a higher rate
and the counter of consecutive successful transmissions will never reach the desired value
(10) due to other failures. This case is very common in such wireless networks where a
transmission failure is not only due to an inadequate rate.
In addition, the next transmission must be completed successfully immediately after a rate
increase otherwise the rate will be reduced instantaneously (back to the old smaller value),
and the timer T will be reset. In fact, the mechanism has estimated that the new adopted rate
is not adequate for next network transmissions.
Also after any two consecutive failures the algorithm automatically reduces its actual rate
until it reaches again a number of 10 consecutive ACKs or the expiration of timer T. In this
way, ARF detects the deterioration of the channel quality based on two consecutive failed
transmissions, and chooses to back out to the previous rate. Figure 1 summarizes the
operation of the ARF and shows the corresponding flow diagram.
While ARF increases the transmission rate at a fixed frequency (each 10 consecutive ACKs)
to achieve a higher system throughput, this model has two main drawbacks:
•
Firstly, this process can be costly since a transmission failure (produced by an
unsuitable rate increasing decision made by the ARF mechanism) reduces the overall
throughput. Specifically, for a steady channel status (stable characteristics) ARF will try
periodically to switch to a higher rate by default which leads to unnecessary frame
transmission failure and reduces the algorithm efficiency.

Page 6

Multimedia – A Multidisciplinary Approach to Complex Issues

34
•
Secondly, ARF is unable to stabilize the rate variations. In fact if the channel conditions
deteriorate suddenly, the ARF mechanism will be unable to respond fast to these
changes and to suit the current state. It will carry out numerous transmission failures so
that it reaches the desired rate value. Therefore, this algorithm cannot cope with rapid
medium status changes.
Rate Ri
Next Rate
Ri+1> Ri
Rate decrease after:
2 Consecutive transmission failures
Or
1 Transmission failure immediately after a
rate rising
Rate increase after:
10 Consecutive successful transmissions
Or
Timeout (T)
Launch (T)
?
Riand Ri+1: two consecutive rates among physical
rates allowed by the standard
T : Timer (in slot-time)
?
Rate Ri-1
Rate Ri+2

Fig. 1. The ARF flow diagram
2.2 Adaptive Auto Rate Fallback (AARF)
To overcome the given shortcomings, a new approach called Adaptive Auto Rate Fallback
(AARF), was proposed (Lacage et al., 2004). It is based on the communication history and aims
to reduce unnecessary rate variations caused by a misinterpretation of the channel state. Thus,
this method controls the time-making process by using the Binary Exponential Backoff (BEB)
technique (the same used by the CSMA/CD and CSMA/CA access mechanisms).
Therefore, when a packet transmission fails just after a rate increase, a lower rate is chosen
for next transmission attempts. In addition, the number of consecutive successful
transmissions n required for the next rate-switching decision will be multiplied by two (with
a limit of nmax = 50). Similar to the old version in a rate decrease caused by two consecutive
frames transmission errors, this value is reset to nmin = 10. The flow diagram in Figure 2
briefly explains the operation of AARF.
Consequently, this new version dynamically controls the number of positive ACKs needed
for the rate control. Thus, AARF overcomes the old ARF version in case of a long steady
channel conditions by eliminating needless and continuous rate-rising attempts. However it
keeps the same disadvantage of the old implementation in case of rapid changes produced
on the channel state.

Page 7

A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in IEEE_802.11 Wireless Networks

35
Figure 3 illustrates the behavior of both ARF and AARF approaches for a time period equal
to 0.4s needed for 230 data frames. Various physical rates were adopted in this experiment
(1, 2, 5.5 and 11Mbps for 802.11b). During this experimentation, we set channel conditions
supporting the use of the physical rate R3 (5.5Mbps) for data transmission. We note that the
period between two successive attempts is increased using the AARF technique while the
ARF mechanism is trying regularly to increment the current rate to a higher value each ten
successive successful transmissions. For example, within the time interval [0.2s, 0.25s] AARF
doesn’t create any unnecessary rate-switching effort, while ARF carries out three attempts.
Likewise, the AARF algorithm considerably has reduced the number of produced errors due
to bad decisions (3/4 of errors were removed compared to those generated by the ARF
mechanism).
Rate Ri
Next Rate
Ri+1> Ri
Rate decreasing after :
2 Consecutive Transmission Failures
?n = nmin
Or
1 Transmission Failure immediately
after a rate increase
?n = Min( nmax, 2*n )
Rate increasing after :
n Consecutive successful transmissions
Or
Timeout (T)
Launch (T)
?
Riand Ri+1: Two consecutive rates among physical
rates Available with the standard
T : Timer (in slot-time)
nmin= 10; Initial value
nmax= 50; Maximal value
?
?
?
Rate Ri-1
Rate Ri+2

Fig. 2. The AARF flow diagram
00.05 0.10.15 0.20.25 0.30.35 0.4
1
1.5
2
2.5
3
3.5
4
Time (s)
Rate Ri


ARF
AARF

Fig. 3. ARF and AARF performance evaluation

Page 8

Multimedia – A Multidisciplinary Approach to Complex Issues

36
2.3 Discussion
We have shown in the above study that the actual rate selection algorithms ARF and AARF
do not conduct to an accurate decision when the channel is relatively noisy. Despite the
given transmission enhancements both models still need improvement and refinement since
they cannot react instantly to sudden changes of the channel state. In addition, an interval of
time is needed to reach the maximum throughput in case of ‘ideal’ medium condition. Thus,
these mechanisms do not represent optimal solutions for the physical link adaptation in
noisy and ’ideal’ environments.
Indeed, at a slow channel quality variation AARF is more suitable than ARF as it proceeds
to the elimination of unnecessary rate increases. And thereafter, it decreases greatly the
number of lost packets while relying on already rate exchanges made previously. However,
this improvement is still insufficient since the decision criterion depends only on the nature
of acknowledgments (ACKs), whereas this parameter no longer provides sufficient
information about the instantaneous channel state. As result AARF need a high latency to
reach the maximum throughput. In other words, a negative ACK (or lack of transmission
success) is interpreted only by medium quality deterioration. However, this phenomenon
may be caused by other networks anomalies (destination not reachable, collision occurred
with another data frame, bad CRC, etc.).
It is also observed that when the competing number of stations is high, packet collisions can
largely affected the performance of ARF and make ARF operate with the lowest data rate,
even when no packet corruption occurs. This is in contrast to the existing assumption that
packet collision will not affect the correct operation of ARF and can be ignored in the
evaluation of ARF. Therefore, ARF and AARF can only passively react to the signal quality
experienced at the receiver. In some occasions, we need to actively improve the signal
quality in order to make the transmission rate to meet the traffic demand, even when the
link length is a little large. This enhancement will optimize the overall performance and
typically will demonstrate a practical effectiveness for multimedia transmissions over Wi-Fi
WLANs.
Accordingly, in the next section we propose a new rate adaptation technique to improve the
decision based on instantaneous channel conditions while respecting and still complying
with the 802.11 standard. In addition, the new approach will be compared with those
currently deployed. Simulation results will be also presented to demonstrate the
enhancement of the proposed technique compared to those currently presented. Also
parameters optimization of the new mechanism will be carried out to be then considered
during next scenario simulations.
3. Proposed adaptive rate control technique
The main idea of the proposed method is to introduce a new channel status assessment
parameter which cooperates with the number of ACKs to provide an efficient and accurate
prediction of instantaneous channel conditions and subsequently to improve the actual rate
adjustment mechanism. A logical way to cope with the slow accommodation characteristics
of statistics-based feedback methods is to look for methods that use faster feedback, i.e.,
feedback that quickly provides up-to-date information about the channel status. Such a
feedback — the RTT — has been theoretically discussed in (Rebai et al., 2008), but so far, to

Page 9

A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in IEEE_802.11 Wireless Networks

37
our knowledge, it has not been used in a practical implementation. We use this RTT
measurement in the proposed 802.11 radio to enhance multimedia performance, and also to
provide feedback information about the channel conditions that the MAC layer requires. In
this section, we first define the new parameter which will be required for the system design.
Next, we will describe its implementation and the principle of operation.
3.1 Round Trip Time (RTT)
Reliable transport protocols such as Transport Control Protocol (TCP) (Tourrilhes, 2001)
were initially designed to operate in ‘traditional’ and wired networks where packet losses
are mainly due to congestion. However, wireless networks introduce additional sorts of
errors caused by uncontrolled variations of the medium.
Face to the congestion problems, TCP responds to each loss by invoking congestion control
algorithms such as Slow Start, Congestion Avoidance and Fast Retransmission. These techniques
have been introduced in different versions of the TCP protocol (TCP Reno, TCP Tahoe, etc.)..
These proactive algorithms consist to control the Congestion Window (CW) size based on
observed errors. Another TCP-Vegas version has been proposed by (Kumar & Holtzman,
1998; Mocanu, 2004) and rapidly has been adopted by the TCP protocol since it includes an
innovative solution designed for preventive systems. In fact, it performs a CW size
adjustment based on a fine connection status estimation achieved by a simple measurement
of the TCP segment transmission delay. This delay is called Round Trip Time (RTT) and
represents (as illustrated in Figure 4) the time period between the instant of issuing a TCP
segment by the source noted te and the reception time of the corresponding ACK noted tr.
If the measured RTTs will have larger values, the TCP protocol infers network congestion,
and reacts accordingly by reducing the congestion window size (symbolized by the number
of sent frames and their size). If the values of observed RTTs become smaller, the protocol
concludes an improvement on the medium conditions and that the network is not
overloaded anymore. Therefore, it proceeds dynamically to increment the CW size, and thus
a good operating performance will be achieved based on the new Vegas-version technique.
Time
te
SourceDestination
RTT
RTT = tr - te
Segment
TCP
ACK
tr

Fig. 4. The RTT delay computation

Page 10

Multimedia – A Multidisciplinary Approach to Complex Issues

38
3.2 The RTT parameter integration
An interesting information and immediate channel observation will be deducted after each
data frame transmission by means of RTT measurement and calculation. This feature
represents the innovative part of the new control algorithm to adjust the data rate based on
the channel capacity. A first integration attempt has been presented in (Rebai et al., 2008)
and a rate adaptation design has been proposed. In this chapter, we implement an enhanced
mechanism called Modified Adaptive Auto Rate Fallback (MAARF) which aims to predict the
medium conditions and minimize the unnecessary loss of data. It chooses the appropriate
rate value needed for the next transmission according to the measured RTT value. In fact, it
performs a match between the observed value of RTT and the physical rate selection.
Time
Framei
Source Destination
RTT*
? Date frame
? ACK
? Failed frame (Loss or
error)
ACKi
Tramei+1
ACKi+1
Framei+1
RTOi+1
RTT*

Fig. 5. The date frames transmissions
Furthermore we define two types of RTT. The first variety is the observed value directly
measured from the channel following the frame sending and called instantaneous RTT
denoted by RTT*. The second, denoted by RTTi, is a theoretical value computed based on the
sending rate R and the data frame size. During a successful transmission of a frame i
resulting the receipt of the associated ACKi, a value of RTT* is calculated. We introduce an
associated recovery timer, called "Retransmission Time Out" and noted RTOi, which will
detect receipt/loss of a data frame. Based on this parameter, the transmitter detects the loss
of a frame i in case of no receipt of the corresponding ACKi till the expiration of the RTOi
timer. In this case, the last issued frame i will be retransmitted (see Figure 5).
Similarly, we introduce an interval defined by [
value RTTi corresponding to the rate Ri. If the observed value of RTT* belongs this interval
(i.e. close enough to the theoretical expected value RTTi) the channel conditions will be
considered insignificant and do not require change of the current rate Ri for next
transmissions. In other words, if the value of RTT* belongs the interval [
link quality is assumed stable and therefore it is suitable that the MS will transmit using the
same current rate Ri since the measured RTT* is considered close to the expected RTTi value.
Outside this window, channel changes are presumed:
i
RTT+,
i
RTT−] adjacent to the theoretical
i
RTT+,
i
RTT−] the

Page 11

A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in IEEE_802.11 Wireless Networks

39
•
Improved, if the RTT* value is less than
i
RTT+ since the corresponding ACK frame was
received earlier than expected and therefore channel conditions had got better.
Degraded, if the RTT* value is greater than
•
i
RTT− since the received ACK frame was
delayed and then we assume that the risk of data loss increases.
Time (RTT)
Rate (R)
RTTi
RTTi-1 >RTTi
RTTi+1 < RTTi
Ri
Ri+1 > Ri
Ri-1 <Ri
RTT -
i
The RTTs growth is inversely proportional to the rate values: RTTi
? Ri
= (RTTi+1 +RTTi) / 2
i
= (RTTi-1 + RTTi) / 2
Steady channel statusQuality degradation Quality improvement
RTT +

Fig. 6. The algorithm parameters set
In both cases, the MS must then change its emission rate and adapt it according to these
instantaneous channel state interpretations. We point that
i
RTT+ <
i
RTT− based on the
statement RTTi+1 < RTTi < RTTi-1 since Ri+1 > Ri > Ri-1. The calculation of the parameters
+
i
RTT
and
i
RTT
is associated to the RTTi value as defined in the Equation 1 and 2. values
−

()
1
2
iii
RTTRTTRTT
−
−
=+
(1)

()
1
2
iii
RTTRTTRTT
+
+
=+
(2)
As stated in Figure 6 the
RTT+ will have the center value of the interval [
i
RTT− value will be the middle of the interval [
1
i
RTT−,
i
RTT ], and
similarly,
i
1
i
RTT+,
i
RTT ].
3.3 Modified Adaptive Auto Rate Fallback (MAARF): Principle of operation
Subsequent to each successful frame transmissions, we compare the variation between
instantaneous RTT* and theoretical RTTi values. More specifically, we test if the value of
RTT* has exceeded
ii
decision will be taken after several observations of RTT* samples:
RTT+ and
RTT−bounds or no. However, the according rate adjustment

Page 12

Multimedia – A Multidisciplinary Approach to Complex Issues

40
•
As the number of consecutive successful transmissions did not reach a value of n
(required as in AARF algorithm for the next rate-switching decision – it is initially
initialized to nmin) we perform the following tests:
•
If the observed RTT* value is less than
transmissions and the maximum speed (54 Mbps) is not yet acquired, then the
instant RTT* is considered smaller than the RTTi value and rather close to the RTTi+1
one. Subsequently, the MAARF technique switches to a higher bit rate Ri+1 starting
the next attempt (since an improvement of the channel characteristics was
interpreted).
•
If the value of RTT* is greater than
i
attempts and the lower rate (6 Mbps) is not yet reached, this implies that the
instantaneous RTT* value is larger than the expected RTTi and relatively close to
RTTi-1. Thus, MAARF detect an early deterioration of the link quality and therefore
we reduce the current rate Ri-1 for future communications.
•
If the value of RTT* remains between the two theoretical bounds [
RTT+<RTT*<
i
assumes a steady state for subsequent network transmissions).
Similar to the AARF algorithm, when the number of consecutive successful
transmissions reaches the desired value (which can be at any given time 10, 20, 40 or
50), we switch to a higher throughput without consideration of the observed RTT*
values.
i
RTT+(RTT* <
i
RTT+) during h successive
RTT−(RTT* >
i
RTT−) for the last g transmission
i
RTT+,
i
RTT−]
(i.e.
i
RTT−) then the rate will be kept and stay invariant Ri (MAARF
•
Analogically, when a transmission fails (no acknowledgment received within the RTOi
value) MAARF modifies values of the decision-making parameters (n, h and g) as follows:
•
If a transmission error is occurred just after a rate increase, it will be then decremented.
In addition, as shown in Equation 3 the number of successful transmissions n that
should be attained for the next rise will be doubled with a limit value equal to nmax.

()
2
max
* ;
n nn Min
=
(3)
•
If two consecutive errors are detected the MAARF mechanism reduces the current rate,
while resetting the value of successful transmissions to the minimum one (n = nmin) for
the next rising attempt.
The same backoff control technique used for the parameter n adaptation is employed as
well for the parameters h and g adjustment. In fact, these two variables will be
dynamically adapted and will vary between the upper and lower limits to maintain a
rigorous decision to increment/decrement the data rate.
•
When a transmission error occurs just after a rate increase decision caused by an
interpretation of the RTT* value, the current rate will be reduced and the h value (as
shown in Equation 4) will be multiplied by two as the upper limit hmax is not
reached.
•

()
2
max
* ;
h hh Min
=
(4)
•
In other words, during successful transmissions the condition (RTT* <
i
RTT+)
should verified using the new value of h to be able to increment the rate.

Page 13

A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in IEEE_802.11 Wireless Networks

41
•
Likewise, if a transmission error was detected immediately after a rate decrease
decision based on comparisons between RTT* and
i
RTT− values. Then the rate will
be raised to its old value and the responsible g parameter (see Equation 5) will be
doubled if its value does not reach gmax.

()
2
max
*;
gMing g
=
(5)
•
This means that the condition RTT* >
i
RTT−should be established using the new
value of g during subsequent transmissions to be able to decrease again the rate.
Both of the above parameters will also be reset identically to the parameter n after
two consecutive transmission failures as follows:
•
h = hmin ; g = gmin
3.4 The MAARF setting
In this section we detail the new parameters designed for the MAARF algorithm. The IEEE
802.11 standard defines within its 802.11a and 802.11g versions, different physical rate
values which can reach 54Mbps. Thus, we setup:
•
Ri: the current data rate that varies from the following shown values {6, 9, 18, 12, 24, 36,
48, 54}Mbps.
RTT*: is the observed value when sending a frame (measured from the transmission
channel after receiving the corresponding ACK).
RTTi: the theoretical time computed between the frame sending time to the ACK receipt
instant. It reflects the channel occupation and does not include the waiting time to
access the medium by the transmitter. It is given by Equation 6 as follows:
•
•

iem.Frame propagtreat.Receiver
+ t
em.ACKpropag treat.Emitter
+ tRTT =t + t + SIFS + t + t
(6)
•
with, tem is the emission time (Data Frame or ACK), tpropag is the propagation time over
the transmission medium and ttreat is the treatment time of each received frame
In practice, this value will be represented only by the data frame transmission delay as
shown in Equation 7. This approximation is made because of the negligibility of the other
delays compared to the chosen value.

.
iem.Frame
i
Frame size
R
RTTt
≈=
(7)
•
RTOi (Retransmission Time Out): is a recovery controlling timer after a frame loss. Its
value is assigned based on RTTi (see Equation 8).

2*
ii
RTORTT
=
(8)
•
i
RTT+ and
i
RTT−: the two decisional parameters (the RTTi borders) which their values
are chosen for each used rate Ri as defined in Equations 1 and 2.

Page 14

Multimedia – A Multidisciplinary Approach to Complex Issues

42
•
•
h: is the rate-increase responsible variable and it belongs to the interval [hmin, hmax] = [4, 16].
g: is the rate-decrement responsible variable and it belongs to the interval [gmin, gmax] =
[2, 8].
n: is the already used parameter by the AARF technique. It represents the number of
successive successful transmissions and belongs to the interval [nmin, nmax] = [10, 50].
•
Finally, we illustrate the detailed MAARF functioning in Figure 7.
Transmission using Rate Ri
Success
?
RTT* < RTT +i
h successive times
?
Data rate increase
to Ri+1
RTT* > RTT -i
g successive times
?
Data rate
decrement toRi-1
ACKreceived
n successive times
?
Data rate increase
to Ri+1
yes
yes
yes
yes
no
no i.e.
RTT +
i < RTT* < RTT -
i
no
2 consecutive
errors
?
rate decrement to
Ri-1and
n=nmin;
h=hmin;
g=gmin;
1 error just after
a rate increase by
RTT* decision
?
rate decrement to
Ri-1and
h=Min(2*h;hmax)
1 error just after
a rate decrease by
RTT* decision
?
rate increase to
Ri+1and
g=Min(2*g;gmax)
noyes
yes
yes
no
no
no
1 error just after
increase
?
rate decrement to
Ri+1and
n=Min(2*n;nmax)
yes
no

Fig. 7. The Transition diagram of the new MAARF algorithm
4. Results and performance evaluation
The algorithms were implemented using the C language on a Unix based operating-system
environment (gcc/terminal MAC) to be then easily integrated into the network simulator.
We conducted various tests using the following configuration:
•
•
•
•
The number of sent frames is 100 frames (approximately 0.5 seconds).
The size of each data frame is equal to the 802.11g minimum frame size (=1200 bytes).
An initial data rate of Ri is 6Mbps (up to 54Mbps).
Failure of an ACK return reflects a transmission failure: packet loss, RTO expired or
error detected by the CRC.
A returned ACK by the receiver indicates a successful transmission only if it is received
before the RTO expiration.
The current value of RTT (RTT*) is read/measured after each ACK reception.
•
•

Page 15

A Dynamic Link Adaptation for Multimedia
Quality-Based Communications in IEEE_802.11 Wireless Networks

43
Several scenarios have been considered to evaluate the performance of the proposed
algorithm compared to the other versions (ARF and AARF).
4.1 Optimization of algorithm parameters
This first experiment is designed to study and optimize the decision-making parameters of
the new algorithm: h (counting the number of successive times in which RTT* <
(reflecting the number of consecutive times that RTT* >
and gmin. In Figure 8, we show the implementation results of different MAARF algorithm
configurations for various parameters values. These results express the chosen physical rate
for each transmitted frame in the network.
i
RTT+) and g
i
RTT−). We discuss the values of hmin
102030405060708090100
6
8
10
12
14
16
18
20
22
24
Frame Number
Rate (Mbps)


MAARF (h=1, g=1)
MAARF (h=4, g=2)
MAARF (h=10, g=4)

Fig. 8. Rate adaptation for different MAARF configurations
We note that by choosing low values of g and h (hmin = gmin = 1), MAARF makes quick
decisions to increment and decrement the physical rate. In fact, it becomes sensitive for
channel variations and adapts sinusoidal regime. On the other hand, by choosing large
initial values of the g and h parameters (hmin=10 and gmin=4) the algorithm does not respond
effectively to significant quality deviations and reacts as AARF. Therefore, we point out that
the best initial and rigorous values of g and h with which MAARF gives the best results are
respectively 4 and 2.
4.2 Test regimes
4.2.1 Unbalanced channel state
We compare now the new scheme against the AARF technique (currently used by the 802.11
WLANs) during unstable channel conditions (random improvement/degradation of the
medium state). In Figure 9, we present the corresponding results graph and we clearly
notice an efficient reaction of the MAARF technique against channel changes. In fact, the
new algorithm detects faster the medium availability by adjusting its physical rate value
starting from the 4th frame, while AARF reacts only from the 10th frame. We also note the