Content uploaded by Jaime Sanchez Garcia
Author content
All content in this area was uploaded by Jaime Sanchez Garcia on Apr 04, 2014
Content may be subject to copyright.
IEEE Network • November/December 1997
52
synchronous transfer mode (ATM) has been adopt-
ed as the switching and transport infrastructure for
the future broadband integrated services digital
network (B-ISDN). The extension of broadband
services to the wireless environment is being driven mainly by
the proliferation of multimedia portable computers, personal
digital assistants, and personal information assistants [1].
Although there are various competing technologies that may
provide these services, such as third-generation cellular net-
works (e.g., personal communications services, PCS) and wire-
less local area networks (WLANs), wireless ATM (W-ATM)
has the advantage of offering end-to-end multimedia capabili-
ties with guaranteed quality of service (QoS). Furthermore,
ATM can handle both packet mode and synchronous services.
Several issues have to be resolved in order to arrive at a W-
ATM network; one of the most important is the MAC proto-
col, which is tightly related to the actual physical layer (PHY)
adopted.
Generally, the radio link presents problems such as noise,
interference, and limited channel bandwidth (BW). The inter-
ference depends greatly on the cell size covered by the base
station (BS), and the mobility of the mobile terminals (MTs).
The basic wireless architecture considered in this article is
the classical cell,
1
with a BS serving a finite set of MTs by
means of a shared radio channel. The BS is connected to an
ATM switch by means of cable or fiber optics so that it can
access the wired ATM network.
In general, the proposals for the PHY of the W-ATM net-
work can be divided in two broad types:
• Spread spectrum (SS) techniques, where all MTs in a cell
transmit at will using the whole spectrum of the channel
simultaneously. SS techniques may use frequency hopping
or direct sequence; the natural access technique for SS is
code-division multiple access (CDMA).
• Time-division multiplex (TDM), where MTs transmit at spe-
cific times using the total available radio frequency (RF)
spectrum. The access technique in this case is time-division
multiple access (TDMA).
Even though SS techniques have been shown to be very
robust to interference and frequency reuse [2], and are very
appropriate for the digital cellular network, they have a severe
disadvantage at high bit rates. In the case of direct sequence
SS [3] the robustness is directly proportional to the spreading
gain (SG), and the number of MTs that can communicate
simultaneously with the BS increases with the SG. However,
an increase in SG implies the need for more BW.
For example, an SG of 100 at a bit rate of 100 Mb/s would
require at least 10 GHz of BW (assuming a modulation
scheme of 1 b/Hz); this makes CDMA unattractive for broad-
band W-ATM, at least with currently available radio technolo-
gies. On the other hand, if SS is used the MAC protocol is
greatly simplified, since there is no contention involved (all
terminals in a cell may transmit simultaneously). An example
of a W-ATM network prototype that uses frequency-hopping
SS is the SWAN network described by Agrawal
et al. in [4].
Since there is no need to handle access contention, the MAC
protocol focuses mainly on higher-level functions such as
A Survey of MAC Protocols Proposed for
Wireless ATM
Jaime Sánchez, Ralph Martinez, and Michael W. Marcellin, University of Arizona
0890-8044/97/$10.00 © 1997 IEEE
A
A
Abstract
Wireless ATM (W-ATM) networks have been studied extensively in recent years.
Extension of ATM network services to the wireless environment faces many interest-
ing problems. The original ATM network was designed for high-speed, noiseless,
reliable channels. None of these characteristics are applicable to the wireless
channel. One of the most critical aspects of a W-ATM network is the medium
access control (MAC) protocol used by mobile terminals (MTs) to request service
from the base station (BS), which has to consider the quality of service (QoS) of
the specific application. In this article the authors analyze some recently proposed
MAC protocols, particularly those for TDMA systems, and discuss their advantages
and disadvantages.
1
In this article, we use the term “cell” to refer to the area served by a BS.
12
IEEE Network • November/December 1997
53
packet scheduling, admission con-
trol, and handoff mechanism.
Since we are interested in future
broadband ATM networks, which
may offer services up to 150 Mb/s
(when technology advances allow
it), we will focus on the analysis of
proposed TDMA techniques. Actu-
ally, some research projects are
already considering bit rates of 20
Mb/s [5] and even 150 Mb/s [6]
using TDMA.
Most TDMA proposals use the
random access technique called
slotted ALOHA (or a variant of
this protocol) for the dialup process
and reservation of slots for trans-
mission from the MTs to the BS.
This transmission is done in what is known as an uplink (UL)
channel, usually in contention among the MTs. Consequently,
most proposals do not analyze the communication from the
BS to the MTs, called downlink (DL), which is realized by a
TDM technique, since the BS has total control (except for the
noise) of the channel through a scheduler.
Within the TDMA proposals, we will make a further dis-
tinction according to the number of carrier frequencies used
between the BS and the MTs: frequency-division duplex
(FDD), which uses two frequencies, and time-division duplex
(TDD), which uses only one frequency carrier.
The organization of this article is as follows. In the second
section we describe MAC protocols that use the FDD
method, while protocols that use the TDD method are pre-
sented in the third section. After the description of each
protocol we highlight its advantages and disadvantages. In
the fourth section we present a comparison of the main
characteristics of each protocol, and end the article with
some conclusions.
FDD-Based MAC Proposals
T
he idea behind FDD is to have two channels per BS cover-
age area, one for the UL and the other for the DL. Usual-
ly, the UL is used by the MTs for sending request and
information packets, according to some reservation and con-
tention algorithm, while the DL is used by the BS in a sched-
uled mode, for sending acknowledgments (ACKs) and
information packets. Due to the availability of these two chan-
nels, it is possible to have an almost immediate (depending on
distance and bit rate, as discussed later) feedback from the BS
in order to know (at the MT) if a request was successful or if
a collision occurred.
The MAC protocol proposals (listed according to publica-
tion date) considered in this section are: DQRUMA,
PRMA/DA, DSA++, and DTDMA/PR.
Distributed Queuing Request Update Multiple Access
(DQRUMA)
The DQRUMA protocol was proposed by Karol et al. [7].
Description of the Protocol —
This protocol considers a
time-slotted system with no frame reference, where the
request access (RA) and packet transmission (Xmt) chan-
nels are formed on a slot-by-slot basis. The UL stream is
divided in a series of minislots used for requesting access
(RA channel), each one followed by a slot for packet trans-
mission (Xmt channel). If needed, the BS can convert a
Xmt channel into M RA channels, where M depends on the
round-trip propagation delay for a transmission from a
remote MT. Figure 1 shows the basic timing structure of the
DQRUMA protocol.
The downlink (DL) channel consists of a series of mini-
slots for acknowledgment of request accesses, each followed
by a slot for packet transmission. Whenever the BS receives
a successful request from an MT, it immediately sends the
corresponding ACK in the appropriate DL minislot.
The channel model considers the MTs to be in one of three
states: “empty,” “request,” and “wait to transmit.” When a
mobile registers with a BS, because of a call setup or handoff,
it is assigned a local ID, called a “b-bit access ID.” As soon as
a packet arrives at a MT with its buffer empty, the mobile
sends a Xmt_Req (including its b-bit access ID) to the BS
using the UL request channel (possibly in contention with
other MTs) and changes its state to “request.”
After receiving the Xmt_Req from a MT, the BS updates
the corresponding entry in a request table (RT), which has an
entry for each of the N mobiles contained in the cell, and
sends an ACK to the MT. Each entry in the RT contains the
b-bit access ID and information about whether the MT has
more packets to transmit.
After the reception of the positive ACK to its Xmt_Req,
the MT switches to the wait-to-transmit state and keeps listen-
ing to the DL Xmt_Perm channel, waiting for permission to
transmit from the BS.
When the BS decides to allow transmission by a particular
MT from the RT (according to the current traffic load and a
round-robin policy), it sends a transmit permission
(Xmt_Perm) through the DL Xmt_Perm channel. The corre-
sponding MT, after detecting its own b-bit ID, transmits a
packet in the next slot and switches to either the empty state
■ Figure 1. DQRUMA timing diagram.
Regular slot
Slot converted into
multiple RA channels
Uplink
Time
Request access
(random access)
Packet Xmt channel
Request access
(ramdom access)
Slot i – 1 Slot i Slot i + 1 Slot k Slot k + 1
Piggybacking (PGBK)
req. (contention-free)
Downlink
b-bit ACK of
request access
Xmit_Perm
Packet Xmt channel
b-bit ACK of
request access
b-bit packet transmit permission
(Xmt_Perm) for the next time slot
■ Figure 2. DQRUMA channel model.
Piggy-
backing
Packet
transmit
Request
access
Empty
state
Request
state
0
1
Success
Failure
Wait-to-transmit
state
IEEE Network • November/December 1997
54
(if it has no more packets in the queue) or the wait-to-trans-
mit state (if it has more packets). Figure 2 shows the channel
model for the MT.
Each time an MT transmits an ATM packet, it includes a
piggyback message in case it has more packets to transmit. In
this way, the MT does not have to make a reservation while
its buffer is not empty. This simplifies the protocol and saves
BW.
The authors analyze two possible methods for random
access to the RA channel:
• Dynamic Access Channel Slotted ALOHA with Harmonic
Backoff algorithm (attempt probabilities: 1, 1/2, 1/3,
1/4,...)
• Dynamic Access Channel Binary Stack algorithm, which
takes into account the time-varying number of RA channels
in the DQRUMA protocol
Several results for the normalized delay vs. throughput
(from simulations with both random access algorithms) are
given in the article.
Remarks —
This protocol has the advantage that the MT is
able to receive the ACK to its request packet (from the BS)
almost immediately on a slot-by-slot basis. In case of a colli-
sion, the involved MTs are (quickly) aware of their failure in
getting UL access, and may try retransmission earlier than
with a framing scheme. A nice contribution of this protocol is
the inclusion of a piggyback reservation field, which saves BW
by avoiding further requests when the MT has more packets
to transmit in its queue. This is especially useful for variable
bit rate (VBR) connections.
Another advantage is that DQRUMA uses a minislot pack-
et for access contention. In this way, the probability of receiv-
ing a packet in good shape is larger than when a longer
packet is employed. Furthermore, since the length of a
WATM packet is on the order of six times the length of a
contention packet, the loss of contention packets does not
affect the channel utilization as much as the loss of a regular
packet.
A disadvantage is that DQRUMA does not make any dis-
tinction between VBR and available bit rate (ABR) services,
it treats both as “bursty” traffic. Consequently, it does not
consider any priority handling mechanism.
Although the authors claim that the number of minislots
can be up to 25, it is worth mentioning that they are omitting
the overhead needed at the PHY layer. Considering the aver-
age PHY overhead used by the IS-54, JDC, and Global Sys-
tem for Mobile Communications (GSM) [8] wireless systems,
we get 48 bits of preamble (or synchronization) and 24 bits of
guard time (9 bytes total). Thus, the length of a regular pack-
et would be (53 + 9) = 62 bytes, while the minimum length of
the access packet would be (1 + 9) = 10 bytes, resulting in a
maximum of 6 minislots (62/10).
Packet Reservation Multiple Access with Dynamic
Allocation (PRMA/DA)
This protocol was proposed by Kim and Widjaja in [9], and is
an enhanced version of the PRMA protocol originally pro-
posed by Goodman et al. in [10]. PRMA was designed for
voice and data traffic only, while PRMA/DA considers con-
stant bit rate (CBR), VBR, and data traffic.
Description of the Protocol —
PRMA/DA considers a fixed
length for the UL frame, which is divided into a fixed number
of slots. All slots are the same size. There are four types of
slots (resulting in four subframes), namely, data reservation
slots, VBR reservation slots, CBR reservation slots, and avail-
able slots. The BS is responsible for determining the number
of slots allotted to each type, as well as the number of slots
assigned to each reserving terminal (MT). The DL frame
works in the contention-free TDM format, under total control
of the BS. Figure 3 shows the frame format of PRMA/DA.
After some registration procedure, when an MT has just
become active (i.e., has packets to transmit), it randomly
selects one of the available slots in the UL frame, and trans-
mits one W-ATM packet (possibly in contention with other
MTs). The contention method considered is slotted ALOHA.
The header of the contending packet carries information about
the service type (ST), a statistical parameter called the effec-
tive BW (EB), and other control fields.
■ Figure 3. PRMA/DA frame format.
2 N
a
...
SYNC
PRMA/DA
header
1
Available slots CBR reservation slots VBR reservation slots Data reservation slots
Variable Variable Variable
ATM header
PRMA/DA
trailer (FCS)
48-byte payload
PRMA/DA
header
ATM header
PRMA/DA
trailer (FCS)
48-byte payload
Request packet
PRMA/DA
frame
Wireless packet
SYNC: Synchronization bits
NS: Number of requested slots
FCS: Frame check sequence
ST: Service type
EB: Effective bandwidth
ST EB Control fields
SYNC NS Other control fields
2
...
1 N
c
2
...
1 N
v
N
d
2
...
1
IEEE Network • November/December 1997
55
Depending on the type of traffic
(CBR, VBR, or data), the EB, and
the current availability of slots, the
MAC (at the BS) will assign certain
slot(s) to the succeeding MT in one
of the reservation subframes for
subsequent transmission. The suc-
cessful MT may keep transmitting
W-ATM packets until the end of
the active session without any con-
tention.
For CBR and VBR (real-time)
calls, which have strict timing con-
straints, the unlimited repetition
of contention procedures is worthless, so the protocol con-
siders a parameter called maximum setup time (W
max
). If a
contention procedure lasts for more than W
max
, the call
will be discarded. The setup time for data calls not has no
limitation.
As soon as the contention procedure ends, the BS
notifies the contending terminals whether or not the
channel access was successful. According to the results of
the contention procedure, the BS increases or decreases
(to a minimum of one) the number of available slots for
the next UL frame, according to a particular algorithm.
This is where the protocol receives its “dynamic allocation”
(DA) name.
At the end of each frame, the BS transmits (in broadcast
mode) information about the number of slots in each sub-
frame as well as the number of reservation slots assigned to
each (successful) reserving terminal. Also included is the exact
location of the assigned slots. The header of the UL reserved
slot packet contains synchronization (SYN) bits, a field called
NS that indicates the current number of slots in the MT
buffer, and other control bits. The NS field is used by the
MAC at the BS (along with the statistical traffic parameters,
EB, transmitted by the terminal) to adjust the number of
reservation slots assigned to a particular MT for subsequent
transmissions.
PRMA/DA considers a model with three states to repre-
sent the status of a MT. The states are INACTIVE, CON-
TENDING, and RESERVING. An MT is initially in the
INACTIVE state. When a packet is generated, the MT switch-
es to the CONTENDING state and tries to transmit a W-
ATM packet according to the slotted ALOHA procedure. If
the network access procedure is successful, the MT switches
to the RESERVING state.
The DA algorithm uses four variables for its operation: the
number of available slots (N
a
); the number of slots where col-
lisions occurred (N
c
); the number of successful access slots
(N
s
); and the number of unused slots (N
u
). The basic opera-
tion of the DA algorithm is as follows: assume N
a
= 1, and
assume that after a contention period the BS detects one col-
lision. At this stage, the BS knows that at least two MTs tried
to access the channel. The BS then assigns two available slots
(N
a
= 2) for the contention in the next UL frame. Having N
a
= 2, if the BS detects two collisions (one in each slot), the BS
knows that at least four MTs tried to request service and
assigns N
a
= 4 for the next UL frame.
Now let us assume that after the next contention period the
BS detected one successful access (N
s
= 1), one collision(N
c
= 1), and one unused slot (N
u
= 1). The BS can conclude
that there may be three MTs involved in a collision and
assigns N
a
= 3. In general, if the number of collisions increase,
the BS increases N
a
, while each time a successful access
occurs, the BS decreases N
a
. The operation of the DA algo-
rithm is illustrated in Fig. 4.
Remarks —
This protocol allows the MT to receive the ACK
to its request packet (from the BS) relatively early (at the end
of the access contention period). On the other hand, in case
of a collision the involved MTs must wait until the BS
announces how many available slots will be assigned for the
next UL frame before the MTs can try to gain access again.
This takes more time than for the DQRUMA protocol.
The main contribution of this protocol is the DA algorithm
which helps resolve the contention situation quickly, and
avoids the waste of BW that may occur when there are several
unneeded request slots. However, one drawback is that this
protocol does not use minislots for the access request. Instead,
the first time an MT needs service it transmits a request mes-
sage along with an information packet; so if a collision occurs,
the effect on the throughput may be greater than if a small
request packet had been used. This may not be important in
low-traffic situations, where there may be room for several
available slots, but is definitely a problem for high-traffic situ-
ations.
It is important to point out that the authors do not consider
the overhead of the PHY layer (SYN and guard time) for the
simulation reported in their article. They only consider the
regular (5 bytes) ATM overhead.
Dynamic Slot Assignment (DSA++) Protocol
This protocol was proposed by Petras and Krämling [11]. The
DSA++ protocol relies on a BS that acts as a central MAC
coordinator, giving service to MTs according to the QoS and
instantaneous capacity requirement of each virtual channel
(VC).
Description of the Protocol —
DSA++ uses (for both DL and
UL channels) a variable-length frame structure called a signal-
ing burst. The DL signaling burst (DSB), transmitted in broad-
cast mode, opens a signaling period (frame) of specific length,
which may range from 8 to 15 slots. The DSB contains the fol-
lowing information:
• A reservation message for each UL slot of the signaling
period
• An announcement message for each DL slot of the signal-
ing period
• A feedback message for each random access slot of the pre-
vious signaling period
• Additional signaling messages (e.g., collision resolution,
paging channel, etc.)
To each DL signaling period corresponds a UL signaling
period of the same length. There is an offset between the
starting point of each period to compensate for the round-trip
propagation delay. Figure 5 shows the signaling technique
used by DSA++.
The feedback messages, which are of the ternary type
(empty, success, collision), are used by a fast collision resolu-
tion algorithm for capacity requests transmitted in contention
■ Figure 4. State transition diagram for the number of available slots (at a BS) in PRMA/DA.
<4,0,0>
<2,0,0>
<1,0,1>
<0,1,0>
<1,0,0>
<0,0,1>
<1,1,0>
<0,1,1>
<3,0,0>
<2.0,1>
1 2
<1,1,2>
<0,1,2>
<1,2,0>
<0,2,1> <0,3,1>
<2,1,0>
<1,1,1>
<2,2,0>
<1,2,1>
<0,1,3>
<0,2,2>
<3,1,0>
<2.1,1>
<0,2,0>
3 4
<0,3,0>
<N
s
, N
c
, N
u
>
N
s
: Total number of success slots
N
c
: Total number of collided slots
N
u
: Total number of unused slots
<0,4,0>
6 8
IEEE Network • November/December 1997
56
slots. After the initial registration procedure, the BS allocates
transmission capacity to the MTs (for both the UL and DL
channels) on a slot-by-slot basis. Each slot is the size of an
ATM packet plus SYN overhead.
The assignment of capacity by the scheduler is based on a
priority calculation for each MT (or VC) being served. The
priority is determined according to a set of dynamic parame-
ters (DPs), which includes the number of waiting ATM pack-
ets and their due dates. The DPs are transmitted by each
MT along with each ATM packet (included in the header).
The BS can ask an MT to update the DPs by either polling
or random access, using shorter-length slots (1/4 or 1/8 the
normal size). For this purpose, the BS uses an algorithm
which calculates the number of short slots that must be
available in the next signaling period according to the fol-
lowing parameters:
• Probability of a new packet arrival at each MT in con-
tention mode since the last transmission of their DPs
• Number of MTs in contention mode
• Throughput of the random access procedure
The priorities assigned to the ATM classes of services are
as follows: CBR > VBR > ABR > UBR. For CBR and VBR
classes, the protocol considers a factor called relative urgency
to decide which MT will transmit or receive in the next signal-
ing period.
An important parameter of this protocol is the delay expe-
rienced by the MT in receiving a feedback signal to its request
through a UL random access slot, since the MT needs to
know if its request was successful or not, to try retransmitting
later in the case of collision. To reduce this delay, Petras pro-
poses the use of a short signaling period. Each signaling peri-
od can provide from one to several random access slots, the
maximum given by the size of the DSB.
To further reduce the delay in getting
the feedback, the DSA++ protocol
uses a splitting algorithm for the back-
logged MTs. The authors analyze both
binary and ternary splitting algorithms.
Remarks —
An advantage of this proto-
col is broadcast of the information that
defines the next signaling period in a
single DL burst (UL slot assignment,
DL slot assignment, and ACKs for pre-
vious packets or access requests). This
releases all other slots in the DL signal-
ing period, allowing the BS to imple-
ment an MT power control algorithm if
needed.
This advantage must be weighed
against the risk of losing the broad-
cast packet, which means that a
whole signaling period would be lost.
Otherwise, the loss of a control pack-
et addressed to a specific MT would
not affect the throughput as much.
Even though there are no collisions
in the DL channel, since the BS is
the only one that transmits in the rel-
evant cell, a packet may be lost due
to interference from other cells (co-
channel interference) or a deep fade.
Another advantage of this proto-
col is that it allows a UL slot to be
divided into up to four short slots to
be used for access request in con-
tention mode. This feature, along
with the use of the Priority Splitting algorithm, helps to resolve
the access contention quickly and efficiently.
Dynamic TDMA with Piggybacked Reservation
(DTDMA/PR)
This protocol was proposed by Qiu et al. [12]. DTDMA/PR is
an extension of the DTDMA protocol originally proposed by
Raychaudhuri and Wilson to be used in an ATM-based wire-
less personal communications network (PCN) [13].
Description of the Protocol
— DTDMA/PR considers a fixed-
length frame, with minislots for reservation (in the UL chan-
nel) and ATM-packet-size slots for transmission of user
information. The UL frame is divided into three subframes,
the first for reservation minislots, the second for long-term
reservable slots, and the third for short-term reservable slots.
The boundary between the long-term and short-term reserv-
able subframes is movable according to the traffic volume
being handled.
This protocol considers three types of traffic, with all pack-
ets having the same length:
• CBR packets, generated periodically during the active peri-
od (the protocol considers some sort of voice activation
detector).
• VBR packets, which are also generated periodically during
the active period, but in groups of different size. The num-
ber of packets in each group is a random variable with cer-
tain probability distribution function.
• ABR packets, generated in bursts.
The frame length is chosen according to the voice
encoder/decodeer (codec) characteristics, with the intention of
■ Figure 5. The signaling scheme of the DSA++ protocol.
Downlink signaling period
Downlink
signaling
burst
Downlink
Uplink
Random access
channels
Offset
N – 1 Slot N N + 1 N + 2 N + 3 N + 4
Time
Feedback
...
Feedback
Reservations
Announcements
Uplink signaling period
...
■ Figure 6. The frame structure of the DTDMA/PR protocol.
Reservation
minislots Long-term reservable subframe
Wireless frame
Movable
boundary
Short-term
reservable
subframe
Type
(CBR or VBR or ABR) Number of extra slots
ID
Other
info
No. of slots
required
ID
Other
info
Piggy-
back
Data load
IEEE Network • November/December 1997
57
facilitating the provision of CBR ser-
vices. Otherwise, the slot size is consid-
ered equal to the size of an ATM packet
(53 * 8 + overhead bits).
The operation of the DTDMA/PR
protocol is as follows (assuming the
MTs have already been registered at the
BS): when a CBR (VBR or ABR)
source generates a new active period, it
will randomly choose a mini-slot at the
beginning of the next frame and will
transmit a reservation packet (perhaps in contention with
other MTs). At the end of the reservation period, the BS will
send a broadcast message containing the IDs of the MTs that
were successful in making a reservation, the number of slots
assigned to each MT, and the slot positions assigned (in either
the long-term or short-term reservable subframe). Figure 6
shows the frame structure for the DTDMA/PR protocol.
Reservations for CBR and VBR can happen only in the
long-term subframe, while reservations for ABR can happen
only in the short-term one. The CBR and VBR terminals
that were successful in accessing the BS can keep using the
same time slots as long as they have more packets to send
(within the same active period), while ABR terminals have
to release the assigned slot as soon as they finish transmit-
ting a packet.
Both CBR and VBR services are considered delay-sensi-
tive, so the corresponding packets myst be discarded if they
exceed some time limit. Meanwhile, the ABR traffic is consid-
ered delay-insensitive, and the ABR packets can be buffered
until the BS assigns a slot for their transmission.
Since the authors assume no channel errors, the reasons for
reservation failures are either reservation packet collisions or
excessive traffic. They also assume that there is no propaga-
tion delay involved in the reception of the ACKs to the reser-
vation requests, and no processing time is required by the BS
in responding to those requests.
A mechanism to guarantee service to VBR traffic is includ-
ed. This consists of sending a piggyback reservation message
along with the VBR packet when the number of assigned slots
is less than the number of packets generated in the current
slot. The slots eventually assigned by the BS in response to a
piggyback request must be released at the end of the current
frame. To keep the unused slots at the end of the frame, a
mechanism of slot reordering and reassignment has been
implemented.
Remarks —
One advantage of this protocol is that it considers
the use of a piggyback reservation (previously proposed in the
DQRUMA protocol), which is especially useful for VBR VCs.
Another advantage is the use of minislots for reservation so
that if an access packet collides, there is not much waste of
BW.
A contribution of this protocol is the consideration of two
types of reservable subframes with movable boundary: the
long-term subframe, meant to be used by CBR and VBR con-
nections, and the short-term subframe (whose slots have to be
released after a one-packet transmission), aimed at ABR con-
nections. This may be used to provide more BW to CBR and
VBR services according to the current needs.
A drawback of this protocol is that it does not take into
account the different constraints of the QoS involved in
each VC. Therefore, in case of two request packets colliding
from MTs with different delay constraints, the probability of
retransmission in the next frame is the same for both. Hence,
the probability of successfully accessing the BS is the same for
all MTs since there are no priorities involved.
TDD Protocols
T
hese protocols use only one carrier frequency to communi-
cate both ways, from the BS to MTs and from MTs to the
BS. They save some hardware in the MTs since both the
transmitter and receiver operate at the same frequency, but
generally add extra delay due to the turnover between trans-
mitter and receiver modes. The protocols analyzed in this cat-
egory, again in order of publication, are MASCARA,
PRMA/ATDD, and DTDMA/TDD.
Mobile Access Scheme Based on Contention and
Reservation for ATM (MASCARA)
This protocol was proposed by Bauchot et al. [14] as the MAC
protocol for the W-ATM Network Demonstrator (WAND)
project being developed with the support of the European
Community (EC).
Description of the Protocol —
The MASCARA protocol oper-
ates in a hierarchical mode by means of a master scheduler
(MS) in the BS (called access point) and a slave scheduler at
each MT. The DL traffic is transmitted in TDM mode, while
the UL packets are transmitted in a mix of reservation and
contention modes.
MASCARA is based on a variable-length timeframe (TF),
which consists of two subframes, one for the UL channels
and the other for the DL channels. The DL subframe is
divided into two periods, the frame header (FH) and down
periods. The UL subframe is also divided into two periods,
up period and contention. All the periods are of variable
length, and all are further subdivided into a variable number
of time slots. The structure of the MASCARA frame is
shown in Fig. 7.
The TF always begins with an FH, which is used by the BS
to broadcast to all MTs a descriptor of the current TF (includ-
ing the lengths of each period), the results of the contention
procedures from the previous frame, and the slot allocation
for each active MT. The MTs use the UL contention slots to
transmit reservation requests (for subsequent frames) or some
control information. Since most of the traffic can be predicted
by the BS, use of the contention-based slots is reduced to a
minimum.
Each of the periods within a frame has a variable length
that depends on the instantaneous traffic in the wireless
link. The periods that operate in reservation mode may col-
lapse to zero slots. The contention period is always main-
tained to at least some minimum number of slots, since an
MT may ask for registration at any time by sending a con-
trol packet.
For handling the transmission process, MASCARA defines
the concept of a “cell train,” which is a sequence of ATM
packets belonging to one MT, ranging from 1 to n, with a
common header. The length of a time slot as well as the
length of the MPDU header are defined as the length of an
ATM packet (53 bytes). The MPDU header includes the
■Figure 7. MASCARA timeframe structure.
Variable-length timeframe
Variable
boundary
Variable
boundary
Time
Broadcast
Reservation-based
traffic
Contention-based
traffic
From MT to
BS
From BS to
MT
Variable
boundary
FH period Downlink period Uplink period Contention period
IEEE Network • November/December 1997
58
SYN bits needed at the PHY and the specific bits of the
header. The payload of a cell train is called a MASCARA pro-
tocol data unit (MPDU). Figure 8 shows the structure of the
“cell train.”
The MS takes into account the service class of the current
ATM VCs, the negotiated QoS, the amount of traffic, and the
number of reservation requests to determine the type and vol-
ume of traffic that will be transmitted in the following frame.
This latter information is kept in a slot map which specifies
the size of the three periods (downlink, uplink, and con-
tention), as well as the assignment of time slots (in the current
TF) to each involved MT.
The BS broadcasts the slot map within the FH at the
beginning of each TF. With the aid of this slot map, each
MT can determine if it will be allowed to either receive or
transmit MPDUs in the current frame. This mechanism
allows the MTs to perform some power-saving procedure,
such as entering a “sleeping” mode when there is no traffic
scheduled for it.
The MS uses an algorithm called Priority Regulated Allo-
cation Delay-Oriented Scheduling (PRADOS) to schedule
transmissions over the radio interface. This algorithm is based
on the priority class, the agreed characteristics, and the delay
constraints of each active connection. Passas et al. describe
this algorithm [15].
PRADOS introduces priorities for each connection accord-
ing to its service class as specified in Table 1 (a higher num-
ber indicates higher priority).
PRADOS combines priorities with a leaky bucket traffic
regulator (LBTR). The LBTR uses a token pool that is intro-
duced for each connection. The generation of tokens happens
at a fixed rate equal to the mean ATM packet rate of each
VC. The size of the pool is equal to the maximum number of
ATM packets that can be transmitted with a rate greater than
the declared mean. Starting from priority 5 and ending with
priority 2, the scheduler satisfies requests for UL and DL as
long as tokens are available. For every slot allocated to a con-
nection, a token is removed from the corresponding pool.
Remarks —
The introduction of the concept of a cell train is a
good contribution of this article, since it provides variable
capacity to MTs in multiples of slots that have the standard
size of one ATM packet (53 bytes). Although not totally
defined, the authors consider a wireless data link control
(WDLC) sublayer to take care of erroneous or lost ATM
packets. The WDLC technique will depend on the constraints
imposed by the QoS parameters of the different services.
Another important contribution of this protocol is the pro-
posed PRADOS algorithm for the master scheduler, since by
deciding on the allocation of slots on a frame-by-frame basis
helps in the fulfillment of the negotiated QoS parameters for
each connection.
A disadvantage of this protocol is that the size of the access
request packet, which may be in contention with other MTs, is
large (equivalent to two ATM packets, one for SYN and over-
head, the other for control informa-
tion). In the case of high traffic, where
the probability of collision increases,
this could result in an important reduc-
tion of throughput.
Another drawback of this proposal is
that the authors have not yet decided
on the contention procedure (they are
investigating slotted ALOHA and the
Stack algorithm). However, they do not
mention anything about the optimiza-
tion of the time it will take to access the UL channel, which is
critical for CBR services.
Last but not least important is that a variable-length frame
introduces an extra difficulty in assigning capacity to MTs with
CBR services. Assuming the case of a voice (64 kb/s) call, if
the frame length (in milliseconds) is less than the time to fill
an ATM packet (~6 ms), there may be frames where no slots
need be assigned. Otherwise, if the frame length is longer
than 6 ms, it might be necessary to assign more than one slot
in a frame for this call.
Packet Reservation Multiple Access with Adaptive
Time-Division Duplex (PRMA/ATDD) Protocol
This protocol was proposed by Priscoli for the MEDIAN sys-
tem [16]. The MEDIAN system is a project being developed
in Europe as part of the ACTS program, with partial support
of the EC. The aim of the project is to develop a high-speed
(~150 Mb/s) wireless network compatible with the B-ISDN
ATM network, to operate in an indoor environment using the
60 GHz band.
Description of the Protocol —
The PHY of the MEDIAN sys-
tem is based on the orthogonal frequency-division multiplex-
ing (OFDM) technique combined with a TDD scheme.
Implementation of the protocol relies mostly on software
which is concentrated in the BS.
The TDD frame is of constant length equal to 64 slots,
although the UL and DL subframes are of variable length.
The DL subframe (DLS) occupies the first part of the frame,
and the UL subframe (ULS) occupies the second portion. The
basic information unit in PRMA/ATDD is called an extended
cell (or slot for simplicity), which consists of an ATM packet
(53 bytes) plus the overhead needed for wireless transmission,
resulting in a total of 1024 bits. Figure 9 shows the frame
structure of this protocol.
The first slot of the frame is used for synchronization, while
the second slot is intended to carry broadcast information.
The other 62 slots are used to carry information packets from
the BS to MTs in the DLS, and from MTs to the BS in the
ULS. The broadcast information includes:
• The number of slots in the DLS (which defines the number
of slots in the ULS)
■ Figure 8. MPDU structure relative to time slots in MASCARA.
MPDU body: cell trains of 3, 4, 2, and 1 ATM packets
PHY and MPDU headers
Time slot
1st MPDU
3rd MPDU 4th MPDU
Time
2nd MPDU
■ Table 1. Priority numbers assigned to the services in MASCARA.
5 CBR (constant bit rate)
4 rt-VBR (real-time variable bit rate)
3 nrt-VBR (non-real-time variable bit rate)
2 ABR (available bit rate)
1 UBR (unspecified bit rate)
Priority number Service class
IEEE Network • November/December 1997
59
• The assignment of each slot for
the MTs
• Signaling related to PRMA
parameters and call setup or ter-
mination
A radio virtual call identifier
(RVCI) is used to relate the slots
with their associated calls in the
broadcast packet. This way, there
is a unique mapping between the
RVCI and the VC identifier
(VCI)/virtual path identifier (VPI)
inside the wireless system. The
RVCI field needs only 5 bits to
address the maximum number of
calls (30) handled by a BS.
The BS functional architecture consists of uplink and
downlink buffers, serial-to-parallel and parallel-to-serial con-
verters, and several network entities (NE). There are four
NEs that are directly related to the MAC layer.
Static List Handler — The SLH determines the call static
parameters to be stored in the static list (SL), which is updat-
ed only at call setup. Each record of the SL refers to a call in
progress, and includes the next four fields: RVCI,
∆
max-up
,
∆
max-down
, and R
max-up
. RVCI is the identifier of the call to
which the record refers, ∆
max-up
and ∆
max-down
are the UL and
DL maximum delays allowed by the VC, and R
max-up
is the
UL maximum bit rate.
Dynamic List Handler — The DLH controls a list of records,
each one containing information about a specific ATM packet
waiting in the BS buffer (or in an MT buffer), to be transmit-
ted in the air interface. Each record has the following fields:
UD, RVCI, and t
max
. UD is a bit that indicates whether the
ATM packet to which the record refers is a UL or DL packet.
The parameter t
max
defines the last time at which the packet
must be transmitted in order to avoid its loss due to excessive
waiting time.
When a DL packet arrives at the BS, it is stored in the DL
buffer. The BS deduces the RVCI value from the VCI/VPI
field in the header, and records this value in the DLH. The
DLH uses the information contained in the SLH to compute
t
max
. After this, the BS inserts the corresponding new DL
record in the dynamic list (the DL in DLH).
When a UL packet arrives at the BS, it is momentarily
placed in the UL buffer in order to deduce the VCI/VPI
from the RVCI field. After the DLH receives this VCI/VPI,
it uses the information stored in the SLH to construct the
appropriate record that will be added to the DL. The DLH
receives from the PRMA parameter computer (PPC) the
number of slots that will be available, termed S
av
(i), for con-
tention in the next UL frame. The DLH then assigns the
remaining Sav(i)-62 useful slots to the packets that occupy
the corresponding records in the DL, as long as they have
not expired.
Broadcast Packet Generator — This entity generates the
broadcast packets, according to the information received
from the DLH (DL frame duration, UL/DL slot assign-
ments, and PRMA state transitions), from the PPC, and
from the BS signaling handler (call setup/termination relat-
ed signaling).
PPC — This entity uses the data provided by the DLH to
compute, at each frame, the most appropriate values for the
permission probabilities related to the various transport ser-
vices (CBR, VBR or ABR), and to the number of available
slots per frame. These computations take into account both
the instantaneous traffic level and the QoS requirements of
currently active connections.
Figure 10 shows the functional architecture of the
PRMA/ATDD protocol that runs in the BS; the figure also
includes a brief description of the signals handled.
Remarks —
A contribution of this protocol is the idea of using
two list handlers (LHs) for the MAC protocol in the BS: the
static LH (with parameters related to the whole duration of
the call) and the dynamic LH (with parameters related to the
current packets in the buffer). The DLH helps decide which
Mts will be served in the next frame according to the expira-
tion time of the buffered packets. These lists are used to
assign priorities to ATM packets according to their expiration
risks.
An advantage of this protocol is the use of a fixed-length
frame, which facilitates the provision of CBR services by
assigning a fixed number of slots in each frame. Nevertheless,
the UL (and correspondingly, the DL) subframe varies
dynamically, in response to the traffic demand.
Similar to PRMA/DA, a disadvantage of this protocol is the
use of full-size slots instead of minislots for access request.
This is inherited from the original PRMA protocol, which was
not designed for the W-ATM network.
Dynamic TDMA with Time-Division Duplex Protocol
(DTDMA/TDD)
Dynamic TDMA was proposed by Raychaudhuri et al. [17], as
part of the WATMnet prototype system developed at C&C
Research in NJ. More detail on the MAC protocol parame-
ters is given by Xie et al. in [18].
Description of the Protocol —
This protocol is based on a
TDMA/TDD structure with a fixed-length frame. The DL
subframe is handled in simple TDM format, transmitted in a
single burst. It consists of two parts. The first part contains
control and feedback (ACK) signals, while the second part is
used for data transmission from BS to MTs.
The UL subframe is handled in a dynamic format and
divided into four slot groups: a group for request (mini) slots,
which uses slotted ALOHA; a dynamic allocation group,
which carries ABR and/or UBR traffic; a fixed and shared
allocation group, which carries VBR traffic; and a fixed allo-
cation group for CBR traffic. Figure 11 shows the dynamic
TDMA/TDD frame format.
Even though the total frame length is fixed, the boundary
between DL and UL subframes varies gradually, according to
■ Figure 9. PRMA/ATDD frame format.
Available
Available
ith broadcast packet (i + 1)th broadcast packet
Null and sync.
symbols
Null and sync.
symbols
Time
ith downlink
frame
ith uplink
frame
(i + 1)th downlink
frame
(i + 1)th uplink
frame
ith TDD frame (i + 1)th TDD frame
Slot
number
Assigned
to call
0 1
a
2
a
3
b
4
a
0 1
c
2
c
3
a
4
b
31
b
32
b
33
b
34
b
31
a
32
a
33
b
34
b
59 60
b
61
c
62 59 60
b
61
c
62
IEEE Network • November/December 1997
60
the traffic experienced by the network. Inside the subframes,
the boundaries between the different slot groups are also
movable.
For the UL subframe, the R-B control packets contain 5
bytes of information plus 2 bytes of CRC. The data packets
contain the original 53 byte ATM packet with 2 bytes from
the Header substituted by a wireless Header, plus 2 bytes for
CRC. Each one of the two types of packets needs an addition-
al modem preamble of 16 bytes for synchronization and
equalization purposes.
For the DL subframe, only one modem preamble is needed
at the beginning of the subframe, since the transmission is
done only by the BS in a broadcast mode. A single W-ATM
ACK message can acknowledge up to 20 packets.
When an MT has packets to transmit, it sends a request
through a control slot, possibly in contention with other MTs.
At the beginning of the next DL subframe, the BS transmits
the slot allocation information along with the ACKs and other
control information.
For the case of CBR VCs, the allocation is done once per
session. For the case of ABR/UBR VCs, it is performed on a
burst-by-burst basis with dynamic reservation of slots from the
ABR/UBR group and from the unused CBR or VBR slots.
For VBR VCs, the allocation is accomplished on a fixed
shared basis, with some slots assigned for the duration of an
active period, plus some extra slot(s) assigned according to a
usage parameter control (UPC)-based statistical multiplexing
algorithm.
In case more than one slot is needed for ABR/UBR VCs,
contiguous slots are assigned to reduce overhead. VBR and
CBR calls can be blocked, while ABR/UBR calls are always
accepted subject to appropriate rate flow control.
Functionally, the MAC protocol can be divided into two
components: supervisory MAC (S-MAC), and core MAC (C-
MAC). The S-MAC at the BS performs channel scheduling
for both the UL and DL channels for all services (CBR, VBR,
and ABR). Also, the S-MAC builds a schedule table based on
the relevant QoS parameters. Finally, the S-MAC takes care
of call admission control.
The C-MAC serves as the interface between the data link
control and PHY. According to the schedule table supplied
by the S-MAC, the C-MAC multiplexes and demultiplexes
the packets for transmission into the wireless channel for
each VC.
Remarks —
One advantage of this
protocol is the division of a frame
into small (8-byte) slots, which allows
it to use one slot for random access
transmission of control packets (from
MT to BS), and to assign several slots
(seven for an ATM packet) to the dif-
ferent services (CBR, VBR, ABR),
according to the traffic demand and
QoS involved.
Another advantage is that it
includes a data link control (DLC)
layer which handles some functions
that complement the tasks of the
MAC protocol. For example, to
resolve the delay constraint imposed
by the CBR services, the DLC uses a
■ Figure 10. BS functional architecture in PRMA/ATDD.
9
ATM switch
adapter
Buffer
handler
1
3
4
Downlink
buffer
Uplink buffer
Signal no.
1
2
3
4
5
6
7
8–12
13
14
15
16
17
18
19
Basic information carried by the signal
ATM packets coming from the ATM network
ATM packets forwarded to the ATM network
VPI/VCIs of the arrived DL ATM packets
Buffer addresses of the DL ATM packets to be forwarded to the BB layer
VPI/VCIs of the arrived UL ATM packets
RVCIs of the arrived UL/DL ATM packets
RVCIs of the DL ATM packets to be retrieved from the DL buffer
Signaling information
Dynamic list related information
Static list related information
Number of available slots per frame
Permission probabilities
UL/DL slot assignments; PRMA state transitions (ACK/NACK)
DL frame duration
Broadcast ATM packets
ATM
interworking
layer
Mac layer
Mac layer
Baseband
layer
PRMA parameter
computer
Dynamic list
handler
State list
handler
Broadcast cell
generator
ATM cell-to-
bitstream converter
Bitstream-to-ATM
cell converter
ATM signaling
handler
2
6
7
8
10
11
14
12
13
15
16
1
19
17
18
2
5
ATM
node
■ Figure 11. Dynamic TDMA/TDD MAC frame format.
Modem
preamble
B-R
ACK
B-R control
Frame
header
Wireless header
Wireless header
Modem
preamble
TDM
downlink
D-TDMA
uplink
ATM header
(compressed)
ATM header
(compressed)
Fixed +
shared
allocation
Fixed
allocation
CRC CRC
R-B control
(S-ALOHA) ABR/UBR VBR CBR
Dynamic
allocation
8-byte
contr.
48-byte payload 48-byte payload
CRC
IEEE Network • November/December 1997
61
first-in first-out (FIFO) buffer to ensure that ATM packet jit-
ter will be kept under acceptable limits. Another useful idea
introduced is to let the DLC handle retransmission of erro-
neous CBR packets by using ABR channels without disturbing
continuous transmission in the CBR reserved channel.
This protocol may be upgraded by adding an algorithm to
handle retransmission of reservation packets (which are han-
dled with the traditional slotted-ALOHA protocol), to give
priority to MTs requesting service for CBR or rt-VBR traffic.
Conclusions
S
everal recently proposed MAC protocols for W-ATM were
reviewed. It appears that the protocols which use FDD in
the PHY can deal with the access contention procedures more
quickly. However, the use of only one carrier frequency (as in
the protocols that use TDD) can be advantageous in some sit-
uations where frequencies are scarce.
When considering the noise, fading, and high interference
characteristics of most wireless environments, it is important
to contemplate a method for fast collision resolution in the
random access stage. In this context it is convenient to use
small slots (minislots) for the random access channel(s) so
that collisions do not produce significant throughput degrada-
tion.
It also appears to be convenient to handle the ACKs to the
access requests (from the BS to the MTs) on a slot-by-slot
basis. For the protocols where the BS sends the result of the
contention procedure on a frame-by-frame basis, the MTs
involved in a random access process have to wait until the
next frame to know the result of the contention process; after
that (in case of failure) they may try another access. This may
represent a waste of BW, especially in medium traffic situa-
tions (where there may be some free slots). It may also repre-
sent a waste of time, which may be critical in the case of an
MT that just handed off and is using a CBR (or VBR) delay-
sensitive service.
The handling of ACKs on a slot-by-slot basis is easier to
implement in an FDD system than in a TDD one, which may
indicate an advantage of FDD over TDD. However, this
advantage must be weighed against the disadvantage that
FDD requires two carrier frequencies.
An advantage of TDD over FDD is that when the DL traf-
fic is bigger than the UL traffic, TDD uses BW more effi-
ciently by allocating most of the slots to the DL subframe.
DL traffic can be larger than UL traffic when several MTs are
downloading huge files or receiving video on demand.
It is also important to consider the QoS constraints for the
different services (CBR, rt-VBR, nrt-VBR, ABR, and UBR),
which must be reflected in different priorities assigned to the
request packets in the collision resolution algorithm.
To summarize the advantages and disadvantages, in Table 2
we present an overall comparison of some of the characteris-
tics of the protocols reviewed. Some of the entries are based
on subjective judgments rather than formal analysis. Relative
communication complexity is based on the load imposed on
the MT radio equipment in order to get synchronized and
keep track of the slots. Channel utilization efficiency takes
into account the size of the access contention slot.
Finally, a more realistic evaluation of the performance of
the proposed MACs would need to take into account the spe-
cific impairments of the RF channel, which to our knowledge
has only been done in the DTDMA/TDD protocol. To par-
tially compensate for these impairments, some form of for-
■Table 2. Comparison of the protocols.
Physical layer type FDD FDD FDD FDD TDD TDD TDD
Frame type No frame Fixed Variable Fixed Variable Fixed Fixed
Frame size — 6 ms 8–15 slots 16 ms — 64 slots 2 ms
Size of access Fraction of Same as 1/4 of ATM Fraction of 2 ATM packets 128 bytes 1/7 of ATM
contention slot ATM pkt. ATM pkt. packet ATM pkt. packet
QoS support VBR BW-related: CBR, VBR, ABR CBR, VBR, ABR CBR, rt-VBR nrt- Delay-related CBR, VBR, ABR
voice,video,data VBR, ABR, UBR
Relative algorithm Low Medium Medium Low High High High
complexity
Relative communi- High Low Medium Low High Medium Medium
cation complexity
Channel utilization High Medium High High Medium Medium High
efficiency
Channel impair- Not considered Not considered Not considered Not considered Not considered Not considered Log-normal
ments analysis fading
Control overhead Low Medium High Medium High High Medium
Random access S-ALOHA, bin. S-ALOHA Splitting alg. S-ALOHA Not defined yet S-ALOHA S-ALOHA
technique stack (2 and 3) (p
prob
= q)
Relative complexity High Low Medium Low Medium Low Low
of providing CBR
services
Call admission No Yes No No Yes No No
control
DQRUMA PRMA/DA DSA++ DTDMA/PR MASCARA PRMA/ATDD DTDMA/TDD
IEEE Network • November/December 1997
62
ward error correction should be used, at least in the access
and control signals.
References
[1] D. Raychaudhuri, “Wireless ATM Networks: Architecture, System Design and
Prototyping,”
IEEE Pers. Commun.
, Aug. 1996, pp. 42–49.
[2] A. J. Viterbi, “The Orthogonal-Random Waveform Dichotomy for Digital Mobile
Personal Communication,”
IEEE Pers. Commun.
, 1st qtr., 1994, pp. 18–24.
[3] T. S. Rappaport,
Wireless Communications Principles and Practice
, Prentice
Hall, 1996, pp. 274–84.
[4] P. Agrawal
et al.
, “SWAN: A Mobile Multimedia Wireless Network,”
IEEE
Pers. Commun.
, Apr. 1996, pp. 18–33.
[5] J. Mikkonen and J. Kruys, “The Magic WAND: A Wireless ATM Access System,”
Proc. ACTS Mobile Summit ’96
, Granada, Spain, Nov. 1996, pp. 535–42.
[6] S. Zeisberg
et al.
, “Channel Coding for Wireless ATM using OFDM,”
Proc.
ACTS Mobile Summit ’96
, Granada, Spain, Nov. 1996, pp. 23–29.
[7] M. J. Karol, Z. Liu, and K. Y. Eng, “Distributed-Queuing Request Update
Multiple Access (DQRUMA) for Wireless Packet (ATM) Networks,”
Proc. IEEE
INFOCOM
’95, pp. 1224–31.
[8] R. C. V. Macario,
Cellular Radio, Principles and Design
, 2nd ed., London:
Macmillan, 1997, pp. 190–240.
[9] J. G. Kim and I. Widjaja, “PRMA/DA: A New Media Access Control Proto-
col for Wireless ATM,”
Proc. ICC ’96
, Dallas, TX, June 1996 pp. 1–19.
[10] D. J. Goodman et al., “Packet Reservation Multiple Access for Local Wireless
Communications,”
IEEE Trans. Commun.
, vol. 37, no. 8, Aug. 1989, pp. 885–90.
[11] D. Petras and A. Krämling, “MAC Protocol with Polling and Fast Collision
Resolution for an ATM Air Interface,”
IEEE ATM Wksp.
, San Francisco, CA,
Aug. 1996.
[12] X. Qiu, V. O. K. Li, and J.-H. Ju, “A Multiple Access Scheme for Multime-
dia Traffic in Wireless ATM,”
J.Special Topics in Mobile Networks and
Appls. (MONET)
, vol. 1, no. 3, Dec. 1996, pp. 259–72.
[13] D. Raychaudhuri and N. D. Wilson, “ATM-Based Transport Architecture for
Multiservices Wireless Personal Communication Network,”
IEEE JSAC
, vol.
12, no. 8, Oct. 1994, pp. 1401–14.
[14] F. Bauchot
et al.
, “MASCARA, a MAC Protocol for Wireless ATM,”
Proc.
ACTS Mobile Summit ’96
, Granada, Spain, Nov. 1996, pp. 17–22.
[15] N. Passas
et al.
, “MAC Protocol and Traffic Scheduling for Wireless ATM Net-
works,” proposed for publication in
ACM Mobile Networks and Appls. J.,
1996.
[16] F. D. Priscoli, “Medium Access Control for the MEDIAN System,”
Proc.
ACTS Mobile Summit ’96
, Granada, Spain, Nov. 1996, pp. 1–8.
[17] D. Raychaudhuri
et al.
, “WATMnet: A Prototype Wireless ATM System for
Multimedia Personal Communication,”
IEEE JSAC
, vol. 15, no. 1, Jan. 1997,
pp. 83–95.
[18] H. Xie
et al.
, “Data Link Control Protocols for Wireless ATM Access Chan-
nels,”
Proc. of ICUPC ’95
, Tokyo, Japan, Nov. 1995, pp. 1–5.
Biographies
JAIME SÁNCHEZ [A] (jsanchez@ece.arizona.edu) received an M.Sc. in telecommuni-
cations and electronics from CICESE Research Center in Ensenada, Mexico, in
1979, and a B.Sc. in communications and electronics from the National Polytech-
nic Institute (IPN) at Mexico D.F. in 1976. Since 1979, Mr. SÖnchez has been at
CICESE Research Center as an associate researcher, teaching graduate courses
and leading projects related to digital telephony and ISDN. He led the group
that won first place in the Third Annual National Contest in Telecommunications
sponsored by Ericsson-México in 1988, with the project “Prototype of a Digital
PABX." He is pursuing his doctoral degree in EE, major in communications, at
George Washington University’s Virginia campus. He is currently in the ECE
Department at the University of Arizona as a visiting scholar. His research inter-
ests include broadband networks, wireless communications, and ATM protocols.
R
ALPH MARTINEZ (martinez@ece.arizona.edu) is an associate professor in the Elec-
trical and Computer Engineering Department with joint appointments in the Radi-
ology and Biomedical Engineering Departments. He has been at the University of
Arizona since 1982. Before then, he spent 14 years in industry as a researcher
in computer system design and applications, specializing in distributed process-
ing architectures and internet gateways for computer networks. At the Naval
Ocean Systems Center (1974–1979), he was responsible for applications of new
VLSI devices to naval systems. At General Dynamics Electronics Division
(1979–1982), he was the system architect for the design of the Global Positioning
System, Phase II, and was branch head for an R&D group in local area network
protocol development and applications to new business areas. Since joining the
Electrical and Computer Engineering Department, he has been involved in
research in interoperable global information systems, internetworking, picture
archiving and communications systems, and multimedia telemedicine systems.
M
ICHAEL W. MARCELLIN (mwm@ece.arizona.edu) graduated summa cum laude with
the B.Sc. degree in electrical engineering from San Diego State University in 1983,
where he was named the most outstanding student in the College of Engineering.
He received the M.S. and Ph.D. degrees in electrical engineering from Texas A&M
University in 1985 and 1987, respectively. He joined the Department of Electri-
cal and Computer Engineering at the University of Arizona in 1988, where he is
currently an associate professor. His research interests include digital communi-
cation and data storage systems, data compression, and signal processing. He is
a member of Tau Beta Pi, Eta Kappa Nu, and Phi Kappa Phi. He is a 1992
recipient of the National Science Foundation Young Investigator Award, and a
corecipient of the 1993 IEEE Signal Processing Society Senior Award.
