Joint Reliability of Medium Access Control and Radio Link ...
1584 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 12, DECEMBER 2005
Joint Reliability of Medium Access Control and
Radio Link Protocol in 3G CDMA Systems
Mainak Chatterjee, Giridhar D. Mandyam, Senior Member, IEEE, and Sajal K. Das, Member, IEEE
Abstract—In this paper, we study the reliability as offered jointly by the medium access control (MAC) and the radio link protocol (RLP)
for third generation (3G) code division multiple access (CDMA) standards. The retransmission mechanism supported at the RLP layer
has a considerable amount of delay (80-100 ms) associated with it; hence, it may not be able to support applications with strict delay
requirements. On the other hand, if retransmissions are also performed at a lower layer, such as the MAC layer, then the performance
of CDMA systems can be improved because of the fast retransmissions at the MAC layer. We show how the performance of RLP
improves with respect to three metrics—mean delay, throughput, and RLP recovery—when a finite number of retransmissions is
allowed at the MAC layer. Synthetically generated Web traffic is used as the application, the objects of which are fragmented into
equal-sized RLP frames. We consider soft packet combining support at the receiver, which effectively lowers the frame error rate. We
also consider the possibility of misinterpretations of acknowledgments. Simulation experiments are conducted to verify the
performance as offered jointly by the MAC and the RLP, particularly for cdma2000 standard and wideband CDMA (WCDMA) systems.
The improvement in the TCP throughput is also evaluated.
Index Terms—Radio link protocol, MAC, retransmissions, packet combing, cdma2000, WCDMA.
W ITHthe proliferation and everyday use of the World
Wide Web (WWW) applications, it has become
desirable that such applications also be supported over
error rates (BER), its performance severely degrades .
Any packet loss at the wireless link is interpreted as
congestion by TCP, which then responds to it by reducing
wireless access networks. Following the tremendous suc- the transmission window size, initiating the congestion
cess of wireless voice services, service providers have control mechanism, and resetting the retransmission time
started offering a variety of wireless data services such as . This misinterpretation of channel related losses and
audio, video streaming, file, and Web downloading. These consequent invocation of the congestion control mechanism
wireless data services promise to offer significant revenue by TCP causes an unnecessary reduction in the TCP
growth opportunity for service providers all over the world. throughput. Several schemes have been proposed to
To bring WWW traffic to wireless mobile devices, it is alleviate the effects of noncongestion related losses over
important that a suitable protocol or standard be chosen to wireless links—both for wireless local area networks (see
cater to the growing demands of data services over wireless  and references therein) and cellular networks , ,
channels such that a wide variety of multimedia traffic with , , , . The radio link protocol (RLP) is one such
different quality of service (QoS) requirements can be scheme that has been used for cellular mobile networks for
supported. No matter which wireless data technology or supporting data traffic .
radio interface dominates future wireless networks, the fact
1.1 Radio Link Protocols (RLP)
remains that it will most likely rely on the Internet Protocol
The RLP is generally employed within the data link layer,
(IP) which is currently the most dominant internetworking
i.e., between the physical layer and the TCP layer to help
protocol. As for the transport layer, the transmission control
shield the effect of the loss over wireless links from the
protocol (TCP) is still the major suite for IP and provides
TCP layer , . As shown in Fig. 1, the RLP fragments
reliable end-to-end transmission in wireline networks  in
an upper layer packet (a TCP segment in this case) into
which the channel error rates are extremely low. However, several equal-sized RLP frames before transmitting over the
when TCP is used over a wireless network with high bit wireless channel. (The last frame can be padded if needed.)
A physical layer header is added to the RLP frame before it
. M. Chatterjee is with the Department of Electrical and Computer is transmitted at the physical layer. The fragmentation is
Engineering, University of Central Florida, PO Box 162450 Orlando, FL done to increase the granularity of the transmission. In
32816-2450. E-mail: email@example.com. other words, in case of any error, an RLP frame which is of a
. G.D. Mandyam is with Nokia Research Center, 12278 Scripps Summit
Drive, San Diego, CA 92131. E-mail: firstname.lastname@example.org.
smaller size is affected rather than the entire TCP segment.
. S.K. Das is with the Center for Research in Wireless Mobility and The RLP uses an Automatic Repeat reQuest (ARQ) error
Networking (CReWMaN), Department of Computer Science and En- recovery mechanism to retrieve a lost RLP frame. The
gineering, University of Texas at Arlington, PO Box 19015, Arlington, TX recovery process is initiated by the receiver by requesting a
76019. E-mail: email@example.com.
retransmission of only the missing or erroneous frame. For
Manuscript received 8 July 2004; revised 20 June 2005; accepted 30 June 2005; example, let us consider that a TCP segment is fragmented
published online 14 Oct. 2005.
For information on obtaining reprints of this article, please send e-mail to: into L number of RLP frames which are transmitted one
firstname.lastname@example.org, and reference IEEECS Log Number TC-0232-0704. after another. If, somehow, the ith frame, 1 i L, fails to
0018-9340/05/$20.00 ß 2005 IEEE Published by the IEEE Computer Society
CHATTERJEE ET AL.: JOINT RELIABILITY OF MEDIUM ACCESS CONTROL AND RADIO LINK PROTOCOL IN 3G CDMA SYSTEMS 1585
performance of TCP over 3G networks was evaluated in the
presence of correlated fading channels.
In all these standards and propositions, RLP has been the
only layer below TCP to shield the losses by triggering
retransmissions. Though this single layer of reliability from
RLP shields losses for TCP, there can still be some
performance limitations if the TCP applications have strict
delay requirements. To deal with such demanding applica-
tions or interactive services, it is necessary to incorporate a fast
retransmission mechanism below the RLP. This was done by
incorporating an ARQ mechanism at the MAC layer, thus
providing two layers of retransmission reliability. The benefit
of MAC layer ARQ is that retransmissions can be done very
quickly, without notifying RLP at the upper layer.
1.3 Our Contributions
Fig. 1. Fragmentation of TCP segments into RLP frames.
Our main motivation behind this research is to study the
reliability offered jointly by the RLP and the MAC layers of
reach the receiver, the receiver would request for a 3G CDMA standards, particularly for cdma2000 1X-EV-DV
retransmission of that frame. However, the receiver buffers  and wideband CDMA (WCDMA)  systems in
the other frames ð1; 2; Á Á Á ; i À 1; i þ 1; Á Á Á LÞ and, on recep- supporting data traffic. It can be noted that two-layer
tion of the ith frame, reassembles the TCP segment and reliability was not supported in the previous CDMA
delivers it to the upper layer. The recovery from the standards such as IS-99  and cdmaOne . In this
erroneous frames is done before the TCP timer expires so paper, we do not propose any additional reliability, but,
that the TCP throughput remains unaffected. instead, study the performance that can be expected by the
The RLP is usually sufficient to shield the physical layer incorporation of MAC layer retransmissions in 3G systems.
impairment from the TCP. However, RLP might still fail to (Preliminary versions of these results can be found in ,
conceal losses due to two reasons: 1) Applications might .) More specifically:
have strict delay requirements  and 2) the packet loss
rate at the medium access layer (MAC) layer might be high. . We analyze the performance of both cdma2000 and
If the packet loss rate is high, even for elastic data traffic, the WCDMA systems and demonstrate how the incor-
RLP has no way to shield the physical impairment from the poration of MAC layer retransmissions can signifi-
TCP. If real-time traffic is considered, the application is cantly improve the delay performance of the RLP.
often assumed to be loss-tolerant. For example, if packets . We define three performance metrics: mean delay,
throughput, and the fraction of packets recovered by
are dropped in voice communication, human intelligence is
the RLP for a different number of retransmission
able to find the consistency in speech. Similarly, in video
trials at the MAC layer. The performance analysis is
transmission, if packets or frames are lost, the decoder at
done assuming a fading channel and considering a
the receiver conceals the losses by interpolating past and
soft packet combining mechanism at the receiver.
future frames. However, if channel conditions are extremely . We extend the analytical models to incorporate
bad, error concealment becomes almost impossible. In such misinterpretations of acknowledgments (though a
cases, interpolation or interpretation of a signal may not small fraction) at the MAC layer.
only be difficult, but also impossible. Therefore, it is . We conduct simulation experiments to validate the
recommended that as many damaged packets as possible analytical model. It is observed that the delay drops
are salvaged, as quickly as possible. from 80-100 ms to less than 10 ms if the MAC layer is
allowed at least one retransmission attempt, although
1.2 RLPs for CDMA Systems
the decrease in delay is not substantial for more
The performance of RLPs for various code division multiple than two retransmissions. The recovery rate from
access (CDMA) systems have been studied over the years as missing frames at the RLP increases with an
the standard evolved. Performance issues related to TCP increasing frame error rate and also when the
and RLP interaction in the CDMA protocol stack have been misinterpretations of acknowledgments are high.
investigated in . The impact of TCP source activity on the . After studying the improvement in RLP perfor-
call admission control for the cellular CDMA standard IS-95 mance, we study the effect on the TCP throughput
was studied in . Support of data services over the IS-95 with and without the two layers of retransmissions.
physical channels using RLP was proposed in . For IS-99 The rest of the paper is organized as follows: Section 2
(the first IS-95 data standard), the performance evaluation discusses the two-layer retransmission scheme supported by
of TCP over RLP was shown in  and the performance for the 3G CDMA systems. The performance analysis is pre-
circuit mode data services was shown in . The sented in Section 3, where mean delay and the fraction of
performance of TCP over RLP in the cdma2000 system  packets recovered by the RLP are evaluated. Section 4
was shown in . A negative acknowledgment-based presents the simulation model and results with respect to
hybrid ARQ scheme was proposed in . In , the mean delay, throughput, and RLP recovery. In Section 5, we
1586 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 12, DECEMBER 2005
show how misinterpretations of acknowledgments could
possibly affect the performance of the cdma2000 and
WCDMA systems. The improvement in the TCP throughput
is shown in Section 6. Conclusions are drawn in the last
2 MAC RETRANSMISSIONS IN CDMA SYSTEMS
A big challenge in third generation (3G) CDMA systems is
how to handle a wide variety of multimedia services with
different QoS requirements. To meet these demands, mod-
ifications and enhancements are being made to cdma2000 and
WCDMA standards. cdma2000 is being deployed in com-
mercial networks to support voice and data on the same
1.25 MHz carrier . A higher data rate evolution, called
Fig. 2. Two layers of retransmission.
cdma2000 1X-EV-DV, has been standardized by the Tele-
communications Industry Association (TIA) and has been
added to IMT-2000 . As specified by the Third Generation site selection (FCSS), a feature under consideration
Partnership Project (3GPP2) , cdma2000 1X-EV-DV pro- in HSDPA, provides mobility support for best effort
vides integrated voice with simultaneous high-speed packet data services. One way to achieve this is by the
data services such as video, video-conferencing, and other mobile terminal “echoing” information about pack-
multimedia services at speeds of up to 4.8 Mbps. Further- ets it has recently received over the shared channel
during a transition from one base station to another
more, cdma2000 1X-EV-DV is backward compatible with
so as to provide fast synchronization of the packet
cdmaOne and cdma2000 1X, providing wireless operators a
transmission queues between the new and the old
seamless network evolution.
On the other front, high-speed downlink packet access
2. The high-speed downlink shared channel in HSDPA
(HSDPA)  is one of the main propositions for WCDMA
and the forward packet data channel in cdma2000
systems. HSDPA allows instantaneous bit rates up to
1X-EV-DV, which carry payload for best effort data
10 Mbps for best-effort packet data services with certain
users (in both time and code multiplexed modes)
bounds on the delay and capacity. The HSDPA channel is a
might contain several protocol data units (PDUs),
resource that is shared among several users in the mobile
not all of which come from the RLP. In fact, some
communication system. By using a fast scheduler located at
PDUs might come directly from the layer above the
the base station, the HSDPA channel can be assigned to the
RLP. As a result, the MAC-ARQ provides retrans-
user transmitting with the current best channel (i.e., highest
missions quickly in HSDPA, which employs a stop-
data rate). The underlying idea is that only users with good
and-wait hybrid ARQ method. In this method, each
downlink channel conditions will use the HSDPA channels
packet received by the receiver must be acknowl-
and other users will not be allowed to use these channels if
edged on a dedicated feedback channel to the
they experience bad channel conditions. Some of the basic
transmitter. This dedicated feedback takes the form
principles used in HSDPA are fast link adaptation, fast of the reverse acknowledgment indicator with
scheduling, and fast retransmissions of erroneously re- values of þ1 for an ACK, or À1 for a NACK.
ceived packets. However, the receiver does not discard the received
2.1 Fast ARQ at the MAC soft information associated with the incorrectly
received packet. Rather, it buffers the data and
Let us now discuss how the RLP and MAC layers jointly
coherently combines the buffered data with the
provide the enhanced retransmission reliability. Fig. 2 shows
received soft information of the retransmission of the
the working of these two layers at the transmitter and receiver
bad packet . This type of packet combining
ends. A TCP segment reaches the MAC layer after being
provides increased reliability in CDMA systems.
fragmented at the RLP. For every packet transmitted at the
MAC layer, there is either an ACK or a NACK. This enhances 2.2 n-Phase ARQ
the response time for retransmissions since the RLP is a Both HSDPA and cdma2000 1X-EV-DV use an n-phase stop-
NACK-based protocol. The fast ARQ mechanism (also called and-wait MAC-ARQ . By “n-phase,” we mean that
MAC-ARQ) in HSDPA and cdma2000 1X-EV-DV is impor- multiple ARQ instances are employed in consecutive time
tant to ensure that some performance loss can be recovered. slots (e.g., 5-ms frame durations). For instance, let us assume
There are two reasons why RLP cannot provide the
that three ARQ channels are used as shown in Fig. 3. Then, in
functionality needed for MAC-ARQ:
time slot t, the receiver will receive a packet corresponding to
1. In the process of selecting the base station with the phase 1. In time slot t þ 1, the receiver will receive a packet
strongest signal for the cell selection, the RLP corresponding to phase 2. In time slot t þ 2, the receiver will
terminates at the last network element (e.g., base receive a packet corresponding to phase 3. Again, in time slot
station), resulting in network delays in servicing t þ 3, the receiver will receive a packet corresponding to
retransmission requests at the RLP layer. Fast cell phase 1, and so on. The receiver must keep separate packets
CHATTERJEE ET AL.: JOINT RELIABILITY OF MEDIUM ACCESS CONTROL AND RADIO LINK PROTOCOL IN 3G CDMA SYSTEMS 1587
Fig. 3. An example of 3-phase ARQ.
received from different phases for packet combining and In the following, we derive expressions for DMAC and
packet acknowledgments. However, once any packet for any DRLP . The expression for DRLP will be conditioned to the fact
phase is received correctly, the receiver may deliver the that the MAC layer failed even after the maximum number
packets to the higher layers (e.g., RLP). of allowed retransmissions. Let us first briefly discuss the
underlying channel characteristics and the frame error rates
(FERs) observed due to packet combining.
3 PERFORMANCE MODELING
Let us now calculate the mean delay and the fraction 3.1 Channel Characteristics
(percentage) of packets recovered by the RLP, without In a wireless environment, the characteristics of the radio
considering the time taken for a packet to get scheduled for channel are very important since the channel conditions
transmission at the RLP. We consider the delay suffered by a vary with time and space. There are several factors that
packet before being correctly received by the RLP at the affect the link attenuation. The hindrance may be due to
receiver and then delivered to the upper layer. We assume multipath fading, shadowing, or any other noise source.
that the transmission and propagation delays are negligible. The most commonly used additive white Gaussian noise
That is, if a packet is successfully transmitted at the first (AWGN) channels do not represent the channel conditions
trial, then the delay incurred is zero. This assumption can accurately and, hence, a more realistic view of the channel is
be relaxed by adding an offset (with a statistical fluctuation) necessary for any involved analysis or experimentation.
to the delay value. If N denotes the number of ARQ phases Moreover, the channel model should include the correlated
and T the frame duration, it takes about N Â T ms for the errors, which is most often the case. Therefore, we consider
ACK to reach the transmitter. Only then can the transmitter a 1-path Rayleigh fading model  for the losses, which
remove the packet from its transmission buffer. If a packet generally has the worst behavior among multipath Rayleigh
does not successfully reach the receiver at the first trial, models due to its inherent lack of diversity .
then, on the receipt of a NACK, it will undergo retransmis- 3.2 Packet Combining Mechanism
sion. A delay will be incurred while the MAC attempts its
In packet data services, when the receiver detects an
fast ARQ mechanism to retransmit and recover the missing
erroneous packet in error, it requests a retransmission.
packet. Let the system parameter MAXRETRANS denote
However, the receiver does not have to discard the soft
the maximum number of retransmissions allowed at the
information associated with the incorrectly received packet.
MAC layer. If the packet is not recovered even after Rather, it can buffer the data and coherently combine it with
MAXRETRANS retransmissions by the MAC layer, the RLP the received soft information on the retransmission of the
triggers its own retransmission mechanism. same packet . This is because there is a 3-db gain in the
By definition, zero delay is incurred by a packet which Eb =N0 (Eb is the energy per bit and N0 is the spectral noise
does not undergo any retransmission at RLP or MAC. A density) value if the retransmitted packet is soft-combined
finite delay is incurred only when a packet fails and is with the stored packet under quasistatic channel conditions
consequently retransmitted. The total delay (D) experienced . This is because the retransmitted packet will double
by a packet to successfully reach the receiver can potentially the power of the received signal; hence, the gain is 10 log 2 %
have two components: 1) DMAC , the delay due to the MAC- 3 db. If the packet is not correctly decoded even after the
ARQ mechanism and 2) DRLP , the delay due to the RLP combination, it is retransmitted for the second time (if
retransmissions. If a packet is recovered by the MAC layer, MAXRETRANS ! 2). With the second retransmitted packet,
then the RLP does not need to use its retransmission and, the gain in Eb =No would be 6-db if two copies of the same
hence, DRLP ¼ 0. If the MAC layer fails to recover a packet, packet are retransmitted. That is, with every trial of
then RLP invokes its retransmission and, hence, both the retransmission, there is approximately a gain of 3-db. This
delay components will be nonzero. Thus, the total delay is type of packet combining mechanism provides increased
reliability in CDMA systems. In order to ensure that the
D ¼ DMAC þ DRLP : ð1Þ
receiver does not try to combine packets from one ARQ
1588 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 12, DECEMBER 2005
TABLE 1 TABLE 2
Effective FER Values for cdma2000 1X-EV-DV Effective FER Values for WCDMA
phase with another, it is assumed that the outband the number of ARQ phases and T is the frame duration.
signaling is sent on a forward control channel concurrently Therefore, the mean waiting time for all the packets at the
with each frame that the receiver receives on the forward
MAC layer is DMAC ¼ pNT , where p is the FER of the
wireless channel. The packets which will still fail to reach
3.3 Effective FER the receiver successfully will be recovered by the RLP. The
We consider 1-path Rayleigh channel at 3 Km/h with 1 rate
2 packets being recovered by RLP would have a delay of M Â
8-PSK (phase shift key) modulation in the personal NT ¼ NT since M ¼ 1. This is essentially the RLP delay for
communication system PCS 1800-1900 MHz frequency.
triggering a retransmission. If M ¼ 0, the fraction of packets
According to the simulation results reported in , the
effective FER falls to approximately one-third of its value recovered by RLP is p. Since M ¼ 1, the fraction is given by
for every transmission in a cdma2000 1X-EV-DV system. Cp Â p, assuming the FER value drops by a factor of C (as
The FER values for successive retransmissions as obtained noted earlier), which is different for cdma2000 and
in  are shown in Table 1. Although the reduction in the WCDMA systems. Thus, the mean delay at the RLP is
effective FER could not be generalized, it works well with DRLP ¼ Cp2 Â NT . The total mean delay is given by
Rayleigh fading channels. For example, if the FER is 0.3 for
the initial transmission, then, on successive retransmissions, D ¼ DMAC þ DRLP ¼ pNT þ Cp2 NT : ð2Þ
the effective FER would be 0.095, 0.031, 0.0095, and so on,
decreasing by approximately one-third every time. The The RLP recovery (R) measured by the fraction of packets
corresponding FERs for WCDMA as obtained from  is recovered by RLP is given by
shown in Table 2, for which the decrease in successive FERs
R ¼ Cp2 : ð3Þ
is approximately a little less than half. In general, we can
assume that there is a drop in the FER by a factor of C for The factor C appears due to packet combining.
every retransmission, where C % 0:33 for cdma2000 1X-EV- Case II (M ¼ 2). Since the ACKs of ðp Â 100Þ% of the
DV (Table 1) and C % 0:45 for WCDMA systems (Table 2). retransmitted packets are yet to arrive, these packets need
3.4 Analysis of Delay and RLP Recovery to be retransmitted again (second MAC retransmission)
Let us now derive expressions for mean delay and the fraction and the effective FER observed so far will be Cp.
of packets recovered by the RLP when MAXRETRANS ¼ M. Therefore, the mean delay associated with the second
Assume that p is the raw FER offered by the physical channel. retransmission is Cp Â pNT . Thus, the mean delay at the
When a packet is transmitted, the MAC usually waits for a MAC is DMAC ¼ pNT þ Cp Â pNT . Since the fraction of
certain number of ARQ phases (say, N) for the ACK. Let the
packets taken care of by RLP is Cp2 when M ¼ 1, the
throughput be defined as the ratio of the number of successful
packets received to the total number of packets undergoing fraction for M ¼ 2 would be Cp Â Cp2 and the delay at the
transmission, including retransmissions. The RLP recovery, R, RLP would be DRLP ¼ ðCpÞ2 p Â 2NT . Hence, the total
is defined as the fraction of packets recovered by the RLP mean delay is obtained as
when the MAC fails to deliver a packet correctly. We will first
derive the expressions for the total delay (D) and RLP D ¼ pNT þ Cp Â pNT þ ðCpÞ2 p Â 2NT : ð4Þ
recovery (R) for one transmission (M ¼ 1) and two transmis- The RLP recovery in this case would be
sions (M ¼ 2) and then generalize them. Since the frame error
rate is finite, the MAC layer will fail with certain probability R ¼ ðCpÞ2 p: ð5Þ
and, hence, the average case delay analysis will have a DRLP
component. Case III (M > 2). Following in a similar manner, we
Case I (M ¼ 1). The waiting time for a packet unsuccess- calculate mean delay at the MAC for any arbitrary M
fully received on the first transmission is N Â T , where N is retransmissions as
CHATTERJEE ET AL.: JOINT RELIABILITY OF MEDIUM ACCESS CONTROL AND RADIO LINK PROTOCOL IN 3G CDMA SYSTEMS 1589
DMAC ¼ ðCpÞiÀ1 pNT
Cp À ðCpÞM
¼ pNT :
1 À Cp
The delay at the RLP is
DRLP ¼ ðCpÞM p Â MNT : ð7Þ
Therefore, the total delay is given by
Cp À ðCpÞM
D¼ pNT þ ðCpÞM p Â MNT : ð8Þ
1 À Cp Fig. 4. Web page traffic scenario.
Additionally, the RLP recovery is given by
the RLP timer. Trlp is a system parameter which is
R ¼ ðCpÞ p: M
ð9Þ negotiated at the time of data connection establishment. A
larger value of Trlp gives more scope for MAC to recover
3.5 Bound on M packets. However, it might delay delivery of segments to
Though the number of retransmissions at the MAC layer is the upper layer (e.g., TCP), thus affecting the delay
a design parameter, it is primarily dictated by the RLP requirements of the data application. So, depending on
timer. The RLP starts a timer when it pushes a packet to the the application requirement and the corresponding TCP
lower (MAC) layer. If a packet fails to reach the receiver by timeout value, the RLP timer is chosen.
the expiration of the timer, the RLP invokes its own
retransmission mechanism to recover the missing packet. 4 SIMULATION STUDY
Therefore, we calculate the maximum number of retrans-
missions that the MAC layer can afford before the RLP To validate the analytical model, we develop a UNIX-based
timer expires and initiates its recovery process. In other simulator, perform extensive simulation experiments, and
words, the delay at the MAC layer must be less than the measure the average delay and the fraction of packets
RLP timer, Trlp . We seek the maximum number of recovered by the RLP for the cdma2000 system. Though
retransmissions allowed at the MAC layer before the RLP multiple RLP frames can be mapped to a physical layer
intervenes. Therefore, the condition to be satisfied is frame by the proper choice of codes from the code-tree ,
we assumed that one fixed-size RLP frame is mapped to a
pNT þ ðCpÞ1 pNT þ ðCpÞ2 pNT þ Á Á Á þ ðCpÞM pNT Trlp ; physical layer frame. The hyper text transport protocol
ð10Þ (HTTP) was used for the data application. We chose such a
model because of its interactive parameters, i.e., the
where M is the number of retransmission trials at the MAC. parameters that also reflect the end-user response. Though
Since we are seeking the maximum number of retransmis- HTTP is considered a non-real-time application, the down-
sions that the MAC can afford, we must find the maximum load time is important to the end user and, hence, some
value of M, Mmax , for which the inequality is satisfied. delay bounds are needed.
Therefore, Mmax is given by
4.1 HTTP Traffic Model
Mmax ¼ maxfm : pNT þ ðCpÞ1 pNT þ ðCpÞ2 pNT þ Á Á Á Instead of investigating the nature of HTTP traffic, we
þ ðCpÞm pNT Trlp g synthetically generate such traffic by using the results
1 À ðCpÞmþ1 obtained in . The basic model of HTTP is shown in Fig. 4
¼ max m : pNT Trlp in which a packet call represents the download of a Web page
1 À Cp
& ' requested by a user. It usually has a main page followed by
Trlp ð1 À CpÞ some embedded objects. A new request (packet call) is
¼ max m : ðCpÞmþ1 ! 1 À
pNT immediately generated after the expiration of the viewing
Trlp ð1 À CpÞ period. The model is similar to an ON/OFF source, where the
¼ max m : m ! logCp 1 À À 1:
pNT ON state represents the activity of a page request and the OFF
ð11Þ state represents a silent period after all objects in that page are
retrieved. The download time of a page follows Weibull
Since Mmax must be an integer, we take the floor function distribution , the mean of which depends on the under-
since the ceiling function will violate (10). Therefore, Mmax lying bandwidth of the wireless channel. We have considered
is given by a data rate of 76.8 Kbps (9,600 bytes/s). Each object (main
" ! # page and embedded objects) of the HTTP traffic is fragmen-
Trlp ð1 À CpÞ
Mmax ¼ logCp 1 À À1 : ð12Þ ted into multiple equal-sized RLP packets. Other statistics
and parameters used to generate the HTTP traffic are shown
Taking the floor function also ensures that the time taken in Table 3. The way each RLP packet is quickly acknowledged
for Mmax retransmissions at the MAC layer does not trigger is discussed next.
1590 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 12, DECEMBER 2005
TABLE 3 TABLE 4
Statistics for HTTP Simulation Parameters
4.2 Fast ARQ at the MAC packets are recovered and the third retransmission is
We emphasize that the analysis is flexible enough to handle hardly required. Fig. 7 shows the throughput of the
newer advanced packet data access methods for both system which is the combined throughput due to MAC
upstream and downstream traffic. Though the analysis and RLP. It can be seen that the throughput decreases as
does not specifically follow 3GPP and 3GPP2 assumptions, the FER increases. This is obvious because larger FER
the analysis can be readily adapted to new physical layer damages more packets during transmission and, thus, the
setups that use RLP and hybrid-ARQ. For example, the number of retransmitted packets is also high. Fig. 8
authors in  describe the latest 1X-EV-DV evolution presents the efficiency with which the RLP recovers the
(enhanced reverse link) with frame durations as low as missing packets. If the MAC-ARQ mechanism is turned
10 ms. Though the frame duration can vary depending on off, effectively implying MAXRETRANS = 0, all the
the modulation and coding schemes, the largest frame missing packets are recovered by the RLP. When
duration is set to 20 ms. MAXRETRANS = 1, there is a considerable drop in the
We consider a configuration of the cdma2000 1X-EV-DV RLP recovery because the MAC does most of the recovery
system where the transmitter transmits one RLP packet in with just one retransmission. However, there is not much
each 5 ms physical layer frame and waits for the ACK. If the improvement if the maximum number of such retransmis-
ACK does not arrive in 20 ms (equal to four ARQ phases, sions at the MAC layer is more than 1. With MAXRETRANS
which is specific to cdma2000) , then the frame is = 2 or 3, the RLP recovery is almost zero because virtually
retransmitted immediately. It can be seen from Fig. 5 that all the packets are recovered with three retransmissions.
frames with sequence numbers 0; 1; 2; 3; Á Á Á are being
transmitted. Frames 0 and 2 are undergoing retransmission 5 MISINTERPRETATION OF ACK/NACK
because of nonreceipt of ACKs. The ACK timers of these
packets are again reinitialized. In a practical scenario, such So far, we have not considered the probability that the ACK
or NACK packets might get corrupted. In reality, there is a
as the downlink, it is not necessary that the receiver deliver
small probability that some of the ACKs and NACKs from
the ACK/NACK precisely at the slot boundaries, each of
the MAC-ARQ will be misinterpreted , . The
which is 5 ms. The actual physical layer boundary is more
decoder at the transmitter might interpret a NACK as an
precise. This is due to the fact that, normally, when using
ACK and therefore would not transmit the packet, assum-
coherent receivers in the reverse link, the base station
ing correct reception. The RLP at the receiver will detect the
suffers from some processing delay. The allowed number of
missing packet and trigger its own retransmission mechan-
retransmission trials is varied between 1 and 3. If a packet is
ism to recover the packet. On the other hand, if an ACK is
not successfully received or combined at the receiver even
decoded as a NACK, then a retransmission will be triggered
after the maximum number of MAC retransmissions, then
by the MAC if the number of retransmissions has not
the RLP retransmission is triggered. Other parameters used
reached the maximum limit (i.e., MAXRETRANS). If the
for the simulation are listed in Table 4.
number of allowable retransmissions at the MAC is
4.3 Analytical and Simulation Results exhausted and a packet has not been ACKed, then the
Fig. 6 shows the total delay (D) for various MAXRE- RLP will initiate the recovery process. No matter which
TRANS values. Not much difference is observed between layer does the recovery, it results in unnecessary (duplicate)
MAXRETRANS = 2 and MAXRETRANS = 3. This is due retransmissions. It can be noted that any kind of wrong
to the fact that, after two transmissions, most of the interpretation at the MAC can only be detected by the RLP
and, thus, the reliability of the RLP cannot be ignored. This
is where the real benefit and importance of RLP is
manifested. The percentage of error recovery by the RLP
depends on the percentage of misinterpretations, which we
call the “falseACK.”
5.1 Analytical Modifications
Due to the presence of the falseACKs, we need to modify
the analytical expressions for D and R derived in
Fig. 5. Number of ARQ phases (N) = 4 timing diagram. Section 3.4. Let f denote the percentage of falseACKs. The
CHATTERJEE ET AL.: JOINT RELIABILITY OF MEDIUM ACCESS CONTROL AND RADIO LINK PROTOCOL IN 3G CDMA SYSTEMS 1591
Fig. 6. Mean delay.
Fig. 7. Throughput.
fraction of packets not reaching the RLP at the first trial is p pðCpÞ2 ð1 À fÞ2 . Therefore, the additional fraction of packets
simply because p is the FER and one packet gets mapped to recovered by the RLP in this case is
one physical layer frame. Out of these, a fraction f will be
AddM¼2 ¼ fp þ fpðCpÞð1 À fÞ þ pðCpÞ2 ð1 À fÞ2 : ð14Þ
misinterpreted as ACK and the fraction ð1 À fÞ will under-
go retransmission at the MAC layer, leading to an FER of In general, the additional fraction of packets recovered
Cp. The fraction of packets not correctly reaching after the by RLP is derived as
first retransmission (if MAXRETRANS = 1) is pðCpÞð1 À fÞ. X
Therefore, for M ¼ 1, due to the misinterpretations, the RLP AddM ¼ fpðCpÞi ð1 À fÞi þ pðCpÞM ð1 À fÞM : ð15Þ
recovers an additional fraction of packets given by
The summation term corresponds to the misinterpretations
AddM¼1 ¼ fp þ pðCpÞð1 À fÞ: ð13Þ
and the second term corresponds to the fraction of packets
If M ¼ 2, the fraction of packets misinterpreted on failing to reach the receiver even after MAXRETRANS
the second retransmission is fpðCpÞð1 À fÞ. The fraction retransmissions. Therefore, the RLP recovery expression in
of packets not delivered correctly by the MAC is (9) is modified as
1592 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 12, DECEMBER 2005
Fig. 8. RLP recovery.
R ¼ ðCpÞM p þ AddM : ð16Þ percentage is very small and is usually around 2 percent
Clearly, the number of packets recovered by RLP , so we varied f from 1 percent to 5 percent. Fig. 9 shows
increases with misinterpretations. The delay expression that the mean delay due to simulation as well as analysis
(from (8)) for MAXRETRANS = 2. We observe that the delay
obtained in (8) will modify to
is more as compared to Fig. 6, where f ¼ 0. Fig. 10 shows
M the percentage of RLP recovery for MAXRETRANS = 2 and
D¼ ðCpÞiÀ1 pNT þ ðAddM Þ Â MNT : ð17Þ this percentage is more than that for MAXRETRANS = 1.
This is because almost all the packets are recovered with
Note that, if f ¼ 0, (16) and (17) will reduce to (9) and (8), two retransmissions at the MAC layer and the RLP mostly
respectively. recovers the ones that were misinterpreted. It is expected
that the packet recovery by RLP will be more at larger
5.2 Results with falseACK values of FER and higher percentage of ACK misinterpreta-
We conducted simulation experiments by incorporating a tions. However, we did not consider FER greater than
certain percentage of ACK misinterpretations. This 30 percent because, at that high FER, there will be other
Fig. 9. Mean delay (MAXRETRANS = 2).
CHATTERJEE ET AL.: JOINT RELIABILITY OF MEDIUM ACCESS CONTROL AND RADIO LINK PROTOCOL IN 3G CDMA SYSTEMS 1593
Fig. 10. RLP recovery (MAXRETRANS = 2).
Fig. 11. Throughput (MAXRETRANS = 2).
severe consequences with maintaining the link. Fig. 11 misinterpretations was varied from f ¼ 1% to f ¼ 5%.
shows the degradation in the system performance in terms Fig. 12 depicts the mean delay for MAXRETRANS = 2.
of throughput when M ¼ 2. Fig. 13 shows the degradation in the system performance in
Comparing the simulation and analytical results, we terms of the throughput for MAXRETRANS = 2. The
observe that they are in good agreement for FER < 0:2. percentage of RLP recovery for MAXRETRANS = 2 is
However, they differ by less than 1 percent for FER > 0:2. One
2 shown in Fig. 14. The RLP recovers more packets with
possible reason might be the assumption on the FER for MAXRETRANS = 1 than with MAXRETRANS = 2. This
successive retransmissions. The FERs used for simulation
happens because, as explained earlier, almost all the packets
(see Table 1) do not strictly fall by 1 for cdma2000 systems,
3 are recovered with two retransmissions at the MAC layer.
which was assumed for obtaining the analytical results.
The RLP mostly recovers the ones which were misinter-
5.3 WCDMA Performance preted. It can be expected that the RLP recovery will be
Simulation experiments were also conducted for WCDMA more for even larger FER values. The value of C ¼ 0:45 for
systems, using the same parameters as those used for WCDMA (see Table 2) can be used for analytical purpose.
cdma2000 simulations, except that the WCDMA frame To compare the performance of cdma2000 1X-EV-DV and
duration was 2 ms (3 Â 1.67 ms). The percentage of WCDMA systems, we summarize in Table 5 the mean delay,
1594 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 12, DECEMBER 2005
Fig. 12. Mean delay (MAXRETRANS = 2).
Fig. 13. Throughput (MAXRETRANS = 2).
RLP recovery, and throughput of these two standards for segment is lost. Thus, the TCP segment loss probability,
MAXRETRANS = 2 and misinterpretation fraction f ¼ 5%. T CPloss , is given by
T CPloss ¼ 1 À ð1 À pÞL ; ð18Þ
6 TCP THROUGHPUT ENHANCEMENT
where p is the frame error probability at the physical layer.
With the two layers of retransmission, we evaluate the If, however, we assume an underlying RLP (1, 2, 3) in
throughput improvement at the TCP layer. Recall from operation, i.e., one retransmission at the first trial, two
Section 1.1 that TCP segments are fragmented into multiple retransmissions at the second, and three retransmissions at
RLP frames. If we assume that a TCP segment of size T bytes the third trial, then the effective frame loss probability at the
is fragmented into equal-sized RLP frames, then the RLP layer is p7 . Hence, with the RLP layer, the TCP segment
number of such frames is L ¼ dRe, where R is the payload loss probability is modified to
of each RLP frame. The actual size of the RLP frame would L
be R plus some header information. For a TCP segment to T CPloss ¼ 1 À ð1 À p7 Þ : ð19Þ
be reassembled successfully, all the L frames must be Furthermore, if we assume that the MAC layer is also in
received correctly. If one or more RLP frames fail, the TCP operation, allowing only one retransmission attempt (i.e.,
CHATTERJEE ET AL.: JOINT RELIABILITY OF MEDIUM ACCESS CONTROL AND RADIO LINK PROTOCOL IN 3G CDMA SYSTEMS 1595
Fig. 14. RLC Recovery (MAXRETRANS = 2).
Performance of cdma2000 1X-EV-DV and WCDMA
MAXRETRANS = 1) to recover any lost or corrupted RLP assumed to be 0.2 seconds, T0 ¼ 2 Â RT T , and b ¼ 2, as per
frame, then the effective frame loss probability at the RLP the traces obtained in . MSS is assumed to be 1,460
layer is ðp7 Þ2 and (19) is modified to bytes. RLP frames were assumed to be of constant size of
L 80 bytes. The transmit window size for the TCP is assumed
T CPloss ¼ 1 À ð1 À p14 Þ : ð20Þ to grow without limit. The improvement in TCP throughput
We will assume that the TCP throughput is given by : is due to two reasons. First, the fragmentation of the TCP
segments into RLP frames prevents the entire TCP segment
ST CP ¼ from being retransmitted, if lost. Second, due to faster
pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ À pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
MSS Á; ð21Þ retransmissions and soft combining, the effective frame loss
2bT CP 3bT CP
loss þT min 1;3 loss ÞT CP 2
RT T loss ð1þ32T CPloss
3 0 8 probability is drastically reduced.
where MSS is the maximum segment size, RT T is the
round trip time for the TCP ACKs, T0 is the TCP 7 CONCLUSIONS
retransmission timer, and b is a system constant. T0 is
evaluated as an exponentially moving average of the As the demand for real-time wireless data services
instantaneous RT T s. increases, more efficient and faster protocols have to be
Substituting the value of T CPloss obtained from (18), (19), designed for timely delivery of data packets. In this paper,
and (20), we can obtain the TCP throughput when 1) both we demonstrated how cdma2000 1X-EV-DV and WCDMA
RLP and MAC are off, 2) RLP is on and MAC is off, and promise to be such protocols which can deliver real-time
3) both RLP and MAC are on, respectively. Fig. 15 shows data over the wireless medium. This is made possible with
the relative performance of these three cases. The RT T is the help of the joint reliability offered by both the MAC and
1596 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 12, DECEMBER 2005
Fig. 15. TCP throughput.
RLP layers. The MAC layer retransmissions can recover  M. Chatterjee, G.D. Mandyam, and S.K. Das, “Fast ARQ in High
Speed Downlink Packet Access for WCDMA Systems,” Proc.
missing or damaged packets very quickly without notifying European Wireless, pp 451-457, Feb. 2002.
the upper layer (e.g., RLP) and thus reduce the delay from  M. Chatterjee, S.K. Das, and G.D. Mandyam, “MAC Layer
about 100 ms to less than 10 ms. Due to the soft packet Retransmissions in 1XTREME,” Proc. 13th IEEE Int’l Symp.
Personal, Indoor and Mobile Radio Comm. (PIMRC), vol. 3, pp. 1452-
combining mechanism, the effective FER experienced by a 1456, 2002.
packet is lowered for every successive retransmission. We  H.K. Choi and J.O. Limb, “A Behavioral Model of Web Traffic,”
analyzed the system performance in terms of delay and the Proc. Int’l Conf. Network Protocols (ICNP), pp. 327-334, 1999.
fraction of packets recovered by the RLP for a finite number  S. Choi, Y. Choi, and I. Lee, “IEEE 802. 11 MAC-Level FEC with
Retransmission Combining,” IEEE Trans. Wireless Comm., Jan.
of retransmissions at the MAC layer. The analysis is 2005.
corroborated with simulation experiments using syntheti-  Y.-J. Choi and S. Bahk, “Scheduling for VoIP Service in cdma2000
cally generated HTTP traffic. We also considered misinter- 1x EV-DO,” Proc. IEEE Int’l Conf. Comm. (ICC) 2004, vol. 3,
pp. 1495-1499, 2004.
pretations of the ACKs at the MAC layer. Results
 A. Chockalingam and G. Bao, “Performance of TCP/RLP Protocol
demonstrated that the fast retransmission mechanism at Stack on Correlated Fading DS-CDMA Wireless Links,” IEEE
the MAC layer enhances the overall TCP throughput. Trans. Vehicular Technology, vol. 49, no. 1, pp. 28-33, Jan. 2000.
 D.E. Comer, Internetworking with TCP/IP, vol. 1. Prentice-Hall,
ACKNOWLEDGMENTS  R.T. Derryberry and Z. Pi, “Reverse High-Speed Packet Data
Physical Layer Enhancements in cdma2000 1xEV-DV,” IEEE
The authors are grateful to the anonymous referees for Comm. Magazine, pp. 41-47, Apr. 2005.
insightful comments which greatly helped them improve  M. Dottling, J. Michel, and B. Raaf, “Hybrid ARQ and Adaptive
the quality of the paper. Modulation and Coding Schemes for High Speed Downlink
Packet Access,” Proc. 13th IEEE Int’l Symp. Personal, Indoor, and
Mobile Radio Comm. (PIMRC), vol. 3, pp. 1073-1077, 2002.
REFERENCES  H. Elaarag, “Improving TCP Performance over Mobile Net-
works,” ACM Computing Surveys, vol. 34, no. 3, pp. 357-374, Sept.
 http://www.3gpp.org, 2005. 2002.
 http://www.3gpp2org, 2005.  F. Frederiksen and T. Kolding, “Performance Modeling of
 http://www.itu.org, 2005. WCDMA/HSDPA Transmission/HARQ Schemes,” Proc. IEEE
 TIA/EIA/IS-2000-2, “Physical Layer Standard for cdma2000 Vehicular Technology Conf. (VTC), pp. 472-476, Fall 2002.
Spread Spectrum Systems,” Mar. 2000.
 A.S. Joshi, M.N. Umesh, A. Kumar, T. Mukhopadhyay, K. Natesh,
 TIA/EIA/IS-95-B, “Mobile Station—Base Station Compatibility
S. Sen, and A. Arunachalam, “Performance Evaluation of TCP
Standard for Dual Mode Spread Spectrum Systems,” 1999.
over Radio Link Protocol in TIA/EIA/IS-99 Environment,” Proc.
 TIA/EIA/IS-707-A.2, “Data Service Options for Spread Spectrum
IEEE Int’l Conf. Personal Wireless Comm., pp. 216-220, 1999.
Systems: Radio Link Protocol,” Mar. 1999.
 A. Bakre and B.R. Badrinath, “I-TCP: Indirect TCP for Mobile  J.M. Harris and M. Airy, “Analytical Model for Radio Link
Hosts,” Proc. 15th Int’l Conf. Distributed Computing Systems, pp. 136- Protocol for IS-95 CDMA Systems,” Proc. IEEE 51st Vehicular
143, May 1995. Technology Conf. (VTC), vol. 3, pp. 2434-2438, Spring 2000.
 G. Bao, “Performance Evaluation of TCP/RLP Protocol Stack over  C. Geßner, R. Kohn, J. Schniedenharn, and A. Sitte, “Layer 2 and
CDMA Wireless Links,” ACM Wireless Networks J., vol. 2, pp. 229- Layer 3 of UTRA-TDD,” Proc. IEEE 51st Vehicular Technology Conf.
237, 1996. (VTC), vol. 2, pp. 1181-1185, Spring 2000.
 R. Caceres and L. Iftode, “Improving the Performance of Reliable  H. Kaaranen, A. Ahtiainen, L. Laitinen, S.K. Naghian, and V.
Transport Protocols in Mobile Computing Environments,” IEEE J. Niemi, UMTS Networks: Architecture, Mobility, and Services. Wiley
Selected Areas in Comm., vol. 13, pp. 850-857, June 1995. & Sons, 2001.
CHATTERJEE ET AL.: JOINT RELIABILITY OF MEDIUM ACCESS CONTROL AND RADIO LINK PROTOCOL IN 3G CDMA SYSTEMS 1597
 F. Khan, S. Kumar, K. Medepalli, and S. Nanda, “TCP Giridhar D. Mandyam received the BSEE
Performance over cdma2000 RLP,” Proc. IEEE 51st Vehicular degree (magna cum laude) from Southern
Technology Conf. (VTC), vol. 1, pp. 41-45, Spring 2000. Methodist University (SMU) in 1989, the MSEE
 W.C.Y. Lee, Mobile Cellular Telecommunications. McGraw-Hill, degree from the University of Southern Califor-
1995. nia (USC) in 1993, and the PhD degree from the
 H. Lin and S.K. Das, “Performance Study of Link Layer and MAC University of New Mexico in 1996. He is the
Layer Protocols in Supporting TCP in 3G CDMA Systems,” IEEE director of the Radio Systems Group in the
Trans. Mobile Computing, vol. 4, no. 5, pp. 489-501, Sept./Oct. 2005. Radio Communications Laboratory of Nokia
 H. Lin and S.K. Das, “Performance Study of CDMA Systems with Research Center and the head of the Nokia
RLP and MAC Layer Retransmissions,” Proc. IEEE Int’l Symp. Research Center, San Diego. At SMU, he was a
Modeling, Analysis, and Simulation of Computer and Telecomm. University Scholar and Hyer Society Scholar. He held positions at
Systems (MASCOTS), pp. 313-320, Oct. 2002. Rockwell International, Qualcomm Inc., and Texas Instruments before
 P. Luukkanen, Z. Rong, and L. Ma, “Performance of 1XTREME joining Nokia Research Center (NRC) in Dallas in 1998. In 2002, he
System for Mixed Voice and Data,” Proc. IEEE Int’l Conf. Comm. became the director of the Radio Systems Group at NRC’s Radio
(ICC), vol. 5, pp. 1411-1415, 2001. Communications Laboratory. In 2004, he became the first head of the
 J. Padhye, V. Firoiu, D. Towsley, and J. Kurose, “Modeling TCP newest Nokia Research Center division in San Diego, California. While
Throughput: A Simple Model and Its Empirical Validation,” Proc. at NRC, he continued his participation in CDMA standardization work,
ACM SIGCOMM pp. 303-314, 1998. which he started while at Texas Instruments. He was a contributor to the
 S. Parkvall, E. Dahlman, P. Frenger, P. Beming, and M. Persson, development of the cdma2000 wireless standard. More recently, he has
“The Evolution of WCDMA towards Higher Speed Downlink led efforts in building experimental radios and participated in research
Packet Data Access,” Proc. IEEE 53rd Vehicular Technology Conf. into “Beyond 3G” technologies. He is the inventor or coinventor of 10
(VTC), vol. 3, pp. 2287-2291, 2001. issued US patents. He has also published more than 50 conference and
 M. van der Schaar, S. Krishnamachari, S. Choi, and X. Xu, journal papers and four book chapters. In addition, he was a guest editor
“Adaptive Cross-Layer Protection Strategies for Robust Scalable for a special issue of the Eurasip Journal on Applied Signal Processing
Video Transmission over 802. 11 WLANs,” IEEE J. Selected Areas in entitled “3G Wireless Communications and Beyond.” He is a coauthor of
Comm., vol. 21, no. 10, pp. 1752-1763, Dec. 2003. the book Third-Generation CDMA Systems for Enhanced Data Services
 S.K. Sen, J. Jawanda, K. Basu, N.K. Kakani, and S.K. Das, “TCP (Academic Press, 2002). Dr. Mandyam is a senior member of the IEEE.
Source Activity and Its Impact on Call Admission Control in
CDMA Voice/Data Network,” Proc. Fourth ACM/IEEE Int’l Conf. Sajal K. Das received the BTech degree in 1983
Mobile Computing and Networking (MobiCom) 1998, pp. 276-283, from Calcutta University, the MS degree in 1984
1998. from the Indian Institute of Science, Bangalore,
 S. Souissi and S.B. Wicker, “A Diversity Combining DS/CDMA and the PhD degree in 1988 from the University
System with Convolutional Encoding and Viterbi Decoding,” of Central Florida, Orlando, all in computer
IEEE Trans. Vehicular Technology, vol. 44, no. 2, pp. 304-312, May science. He is a professor of computer science
1995. and engineering and also the founding director
 Y.C. Tseng and C.M. Chao, “Code Placement and Replacement of the Center for Research in Wireless Mobility
Strategies for Wideband CDMA OVSF Code Tree Management,” and Networking (CReWMaN) at the University of
IEEE Trans. Mobile Computing, vol. 1, no. 4, pp. 293-302, Oct.-Dec. Texas at Arlington (UTA). His current research
2002. interests include resource and mobility management in wireless and
 M.N. Umesh, A.S. Joshi, A. Kumar, and T. Mukhopadhyay, “A sensor networks, mobile and pervasive computing, wireless multimedia
NAK Based Hybrid Type II ARQ Scheme for cdmaOne/cdma2000 and QoS provisioning, mobile Internet protocols, distributed processing,
Systems,” Proc. IEEE Vehicular Technology Conf. (VTC), vol. 5, and grid computing. He has published more than 350 research papers,
pp. 2596-2600, Fall 1999. directed numerous funded projects, and holds five US patents in
 Y. Xiao, H. Li, and S. Choi, “Protection and Guarantee for Voice wireless mobile networks. He received the Best Paper Award from ACM
and Video Traffic in IEEE 802.11e Wireless LANs,” Proc. IEEE MobiCom ’99, ICOIN ’01, ACM MSWIM ’00, and ACM/IEEE PADS ’97.
INFOCOM, vol. 3, pp. 2152-2162, Mar. 2004. He was also a recipient of UTA’s Outstanding Faculty Research Award
 A.J. Viterbi, CDMA: Principles of Spread Spectrum Communication. in Computer Science (2001 and 2003), College of Engineering
Addison-Wesley, 1995. Excellence in Research Award (2003), and University Award for
 X. Zhaoji and B. Sebire, “Impact of ACK/NACK Signalling Errors Distinguished Record of Research (2005). He is the coauthor of the
on High Speed Uplink Packet Access (HSUPA),” Proc. IEEE Int’l book Smart Environments: Technology, Protocols and Applications
Conf. Comm. (ICC), May 2005. (John Wiley, 2005). He is the editor-in-chief of the Pervasive and Mobile
Computing journal and serves on the editorial boards of five international
Mainak Chatterjee received the PhD degree journals, including IEEE Transactions on Mobile Computing, IEEE
from the Department of Computer Science and Transactions on Parallel and Distributed Systems, and ACM/Springer
Engineering at the University of Texas at Wireless Networks. He has served as general chair of IEEE WoWMoM
Arlington in 2002. Prior to that, he completed ’05, IWDC ’04, IEEE PerCom ’04, CIT ’03, and IEEE MASCOTS ’02;
the BSc degree in physics (Hons) from the general vice chair of IEEE PerCom ’03, ACM MobiCom ’00, and HiPC
University of Calcutta in 1994 and the ME degree ’00-01; program chair of IWDC ’02 and WoWMoM ’98-99; TPC vice
in electrical communication engineering from the chair of CIT ’05 and ICPADS ’02; and as TPC member of numerous
Indian Institute of Science, Bangalore, in 1998. IEEE and ACM conferences. He is the vice chair of two IEEE technical
He is currently an assistant professor in the committees (TCPP and TCCC) and a member of the advisory boards of
Department of Electrical and Computer Engi- several cutting-edge companies. He is a member of the IEEE and the
neering at the University of Central Florida. His research interests IEEE Computer Society.
include economic issues in wireless networks, applied game theory,
resource management and quality-of-service provisioning, ad hoc and
sensor networks, CDMA data networking, and link layer protocols. He
serves on the executive and technical program committee of several
international conferences. . For more information on this or any other computing topic,
please visit our Digital Library at www.computer.org/publications/dlib.