Wireless Technology in Industrial Networks

ANDREAS WILLIG, MEMBER, IEEE, KIRSTEN MATHEUS, MEMBER, IEEE, AND

ADAM WOLISZ, SENIOR MEMBER, IEEE

Invited Paper

With the success of wireless technologies in consumer elec- environment can be substantially reduced, thus making plant
tronics, standard wireless technologies are envisioned for the setup and reconfiguration more easy. This is especially im-
deployment in industrial environments as well. Industrial appli- portant in harsh environments where chemicals, vibrations,
cations involving mobile subsystems or just the desire to save
cabling make wireless technologies attractive. Nevertheless, these or moving parts exist that could potentially damage any sort
applications often have stringent requirements on reliability and of cabling. In terms of plant flexibility, stationary systems
timing. In wired environments, timing and reliability are well can be wirelessly coupled to any mobile subsystems or mo-
catered for by fieldbus systems (which are a mature technology bile robots that may exist in order to achieve a connectivity
designed to enable communication between digital controllers and that would otherwise be impossible. Furthermore, the task of
the sensors and actuators interfacing to a physical process). When
wireless links are included, reliability and timing requirements are temporarily accessing any of the machinery in the plant for
significantly more difficult to meet, due to the adverse properties of diagnostic or programming purposes can be greatly simpli-
the radio channels. fied by the use of these wireless technologies.
In this paper, we thus discuss some key issues coming up in wire- Along with the simplification of accessing machinery,
less fieldbus and wireless industrial communication systems: 1) fun-
many industrial applications exist that could benefit from the
damental problems like achieving timely and reliable transmission
despite channel errors; 2) the usage of existing wireless technolo- use of wireless technologies. The localization and tracking of
gies for this specific field of applications; and 3) the creation of unfinished parts, the coordination of autonomous transport
hybrid systems in which wireless stations are included into existing vehicles and mobile robots [1]–[3], as well as applications
wired systems. involving distributed control are all areas in which wireless
Keywords—Bluetooth (BT), fieldbus systems, hybrid systems, technologies could be used in an industrial environment.
IEEE 802.11, IEEE 802.15.4, real-time communications, wireless Many of these industrial applications are served by
technologies.
fieldbus systems [4]–[8] like PROFIBUS [9], [10], WorldFIP
[11], [12] or CAN [13], [14], which are wired. Fieldbus sys-
I. INTRODUCTION tems have been specifically designed for solving automation
or control tasks that rely on the interconnection of digital
The true convenience of being able to connect devices
controllers with other digital controllers as well as sensors
without the use of wires has lead to the unprecedented suc-
and/or actuators (including their underlying physical pro-
cess of wireless technologies in the consumer goods industry.
cesses). The primary goal of these systems is to provide
Based on this success, applications using these technologies
real-time communication services that are both predictable
are beginning to appear in various other settings as well. In
and reliable, i.e., make certain guarantees on eventual
an industrial or factory floor setting, for example, the bene-
delivery of packets and delivery times. Some important
fits of using wireless technologies are manifold. First of all,
characteristics of fieldbus traffic are: 1) presence of cyclic
the cost and time needed for the installation and maintenance
(i.e., recurring) or even periodic traffic (bounded jitter be-
of the large number of cables normally required in such an
tween subsequent packets required), subject to deadlines;
2) presence of important acyclic packets like alarms, which
Manuscript received March 25, 2004; revised March 11, 2005. need to be reliably transmitted with bounded latencies; and
A. Willig is with the Hasso-Plattner-Institute, University of Potsdam,
Potsdam 14482, Germany (e-mail: awillig@ieee.org). 3) most packets are short, on the order of a few bytes. The
K. Matheus is with CARMEQ GmbH, Berlin 10587, Germany (e-mail: protocol architecture of most fieldbus systems covers only
kirsten.matheus@carmeq.com). the physical layer, the data link layer including the medium
A. Wolisz is with the Telecommunication Networks Group (TKN), Tech-
nical University Berlin, Berlin 10587, Germany (e-mail: awo@ieee.org). access control (MAC) sublayer, and the application layer of
Digital Object Identifier 10.1109/JPROC.2005.849717 the OSI reference model.

0018-9219/$20.00 © 2005 IEEE

1130 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

 The obvious benefits of wireless transmission have led to jam the medium, this way challenging reliable and
a number of solutions. These solutions range from voice-ori- timely transmission. On the other hand, ensuring secu-
ented, large-scale cellular networks such as UMTS, to data- rity goals like confidentiality or accountability was not
oriented solutions like wireless LANs (WLANs), wireless the main focus in the design of many fieldbus systems
personal area networks (WPANs) and wireless sensor net- [22], [23]. The recent trend to connect fieldbuses to
works. WLAN systems, like the IEEE 802.11 family of stan- the Internet by means of gateways has led to research
dards [15]–[17], are designed to provide users with high data toward securing the gateway [22], [24], but it is also
rates (tens of megabits per second) over ranges of tens to required to protect a fieldbus against attacks from the
hundreds of meters. These parameters provide the user with inside, for example, by employing proper encryption
untethered access to Ethernet, for example. WPAN systems, and authentication schemes.
such as Bluetooth (BT) [18], [19] and IEEE 802.15.4 [20], • Energy supply and low power operation: In some
[21] have been designed for connecting devices wirelessly fieldbus systems, the same cable can be used for com-
while taking energy efficiency into account. They support munication purposes as well as to supply a station with
medium data rates in the order of hundreds of kilobits per energy. If the cabling were to be dropped completely,
second to a few megabits per second and have ranges on the alternative ways to supply stations with energy would
order of a few meters. Many vendors offer equipment com- have to be found. Some options are wireless energy
pliant to these standards. transmission [25], [26], energy-scavenging methods
Running fieldbus-based applications with wireless [27], or using batteries. For battery-driven stations,
technologies can be especially challenging. Since wire- energy is a scarce resource and should be used eco-
less channels are prone to possible transmission errors nomically. Replacing batteries may be infeasible or
caused by either channel outages (which occur when the can lead to machine downtimes. Several mechanisms
received signal strength drops below a critical threshold) to conserve energy in protocols and applications have
and/or interference, the real-time and reliability require- been developed in the context of wireless (sensor) net-
ments are more likely to be jeopardized than they would be works [28]–[30]. In the design of fieldbus protocols,
over a wired channel. This is one of the key issues to be however, the main concern was real-time commu-
resolved in wireless fieldbus systems, or in general in the nications, not energy efficiency. There are efforts to
usage of wireless technologies in industrial applications, combine both targets [31].
and the focus of this paper. Wireless fieldbus systems and wireless industrial networks
The goal of this paper is to give an overview of the have created interest in both academia and industry. The
problems and issues that arise when considering the use of first publications date back to 1988 [32]. One of the earliest
standardized wireless technologies like IEEE 802.11, BT, or projects, the European Union OLCHFA project, started in
IEEE 802.15.4 in a fieldbus-controlled industrial network. June 1992 with the goal to provide wireless spread-spectrum
The discussion of this topic takes the following structure. transmission for the WorldFIP (formerly just FIP) fieldbus
Section II considers the adverse effect that transmission er- [33], [34]. Today, several companies and consortia are ac-
rors and other properties of the wireless channel have on the tive, for example, the wireless industrial networking alliance
timing and reliability of packet transmissions, irrespective of (WINA).1
the specific wireless technology used. These effects can be
(partially) compensated for by either designing robust and II. FUNDAMENTAL PROBLEMS OF REAL-TIME AND
loss-tolerant applications/control algorithms or by trying FIELDBUS COMMUNICATION
to improve the channel quality when designing a wireless
fieldbus protocol. We focus on taking the latter approach. This section introduces some of the fundamental prop-
Section III introduces the widely used wireless technologies erties of wireless transmission media, without referring to
of interest for use by fieldbus systems in industrial environ- any wireless technology in particular (this is done in the fol-
ments: BT [18], [19], IEEE 802.15.4 [20], [21], and IEEE lowing Section III). Given that there are a number of ma-
802.11 [15]. Given that users will most likely not want to ture and commercially available (wired) fieldbus systems, the
simply throw out their functioning wired fieldbus installa- question is whether there are major difficulties when using
tions in favor of wireless ones, Section IV discusses how them with wireless media. Some examples discussed in this
wireless stations can be integrated into wired fieldbuses to section show that some protocols do indeed have difficul-
create hybrid wired/wireless fieldbus systems. Conclusions ties. One particularly important problem is channel errors;
are provided in Section V. channel errors can cause packets to miss their deadlines, for
Besides the issues discussed in this paper, there are further instance. Accordingly, not only are the consequences of er-
issues to consider in wireless fieldbus systems. Two impor- rors of interest, but also the mechanisms that allow one to
tant ones are the following. deal with them. Some of these mechanisms, which have been
• Security: The wireless medium is an open medium proposed specifically for fieldbus systems, will be discussed.
and without countermeasures, it is easy for an attacker
to eavesdrop, to insert malicious packets, or to simply 1See www.wina.org

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1131

A. Important Properties of Wireless Channels and composite signal at the receiver alternates between
Transceivers constructive and destructive interference, leading
1) Path Loss: The signal strength of a radio signal de- to comparably fast fluctuations in received signal
creases with the distance between a transmitter and a re- strength (time variance). In the case of destructive
ceiver. This decrease is known as path loss. The magnitude interference, the channel is often said to be in a deep
of the path loss depends on several parameters, including the fade, and many errors occur during the decoding of
antenna technology, the frequencies used, and the environ- channel symbols. When the duration of a deep fade
mental conditions that are present. An often-used approxi- spans several consecutive channel symbols, symbol/bit
mation of path loss is the log-distance model. In this model, errors will begin to appear in bursts.
the received signal strength at a distance behaves as • Intersymbol interference (ISI): When the time disper-
for distances larger than a reference sion is large, it can happen that waveforms belonging
distance and a radiated signal strength . The reference to different symbols overlap at the receiver. In the case
distance depends on the antenna technology. The so-called of such ISI, particular effort is necessary to reconstruct
path-loss exponent typically assumes values between two the original symbol.
(free-space path loss) and six [35, Ch. 4], depending on the There are further distortions to wireless waveforms, in-
environment. In factory environments, path loss exponents cluding cochannel interference and adjacent channel inter-
between two and three have been observed [35, Table 4.2], ference from colocated wireless communication systems,
[36], but sometimes values smaller than two can occur as well thermal and man-made noise, as well as Doppler shifts [35],
[37]. [38], [39]. In industrial environments, significant noise as
2) Half-Duplex Operation of Transceivers: Wireless well as distortion of transceiver circuitry can also be cre-
transceivers are not able to transmit and receive simultane- ated by strong motors, static frequency changers, electrical
ously on the same channel because their own signals would discharge devices, and more. Measurements of some key
drown all signals from any other stations. Because of this wireless channel characteristics in industrial environments
fact, most wireless transceivers are half-duplex. They inhibit [36], [37], [40]–[44] have shown that the delay spread can
simultaneous transmit and receive operations while allowing reach values larger than 200 ns. Modulation schemes with
the same circuity to be shared, thus reducing the transceiver symbol rates of several megabauds might then be subject to
complexity. The primary disadvantage of this approach is severe ISI.
the time loss experienced from explicit receive–transmit These phenomena translate into bit errors and packet
turnovers. losses, with possible delays if packets need to be retrans-
3) Physical Layer Overheads: To let the receiver of a mitted. Packet losses occur when, for example, the receiver
packet acquire carrier-/bit-synchronization despite a noisy of a packet fails to acquire carrier synchronization or bit
channel, most wireless systems use extra training sequences synchronization, whereas bit errors refer to errors caused by
of well-known symbols. When the training sequence occurs flipped bits after synchronization has already been success-
at the beginning of a packet, it is called a preamble. For ex- fully achieved [45]. The error characteristics shown by the
ample, the IEEE 802.11 physical layer with direct sequence wireless channel depend on the propagation environment,
spread spectrum (DSSS) requires preambles of 128- s length the chosen modulation scheme, the transmit power, the fre-
[15], transmitted with every packet. quency in use, as well as many other parameters. In general,
Such physical layer overheads tend to be much smaller on however, systems tend to show time-variable and sometimes
wired transmission media. quite high error rates.
4) Channel Errors: A wireless transmitter propagates The following description gives an example of how se-
waveforms into multiple spatial directions at the same time. vere such a situation might be. Measurements in an indus-
These waveforms can be subject to reflections, diffraction, trial environment with an IEEE 802.11b-compliant chipset
or scattering [35]. As a result, multiple copies of the same have shown that short-term bit error rates in the order of
waveform can reach the receiver after following different 10 10 can be reached for a 2-Mb/s quaternary phase
paths with different relative lengths and different travel shift keying (QPSK) modulation [45]. Additionally, there are
times (time dispersion). A common measure for such time minute-long periods where packet loss rates of at least 10%
dispersion is the rms delay spread (root mean square), or (and sometimes up to 80%) have been observed. Also, bit er-
simply delay spread for short.2 The time dispersion has two rors and packet losses are “bursty”, i.e., they occur in clusters
important consequences: with error-free periods (“runs”) between the clusters. Natu-
• With small-scale or multipath fading [35, Ch. 5], rally, these results are particular for the specific chipset and
multiple copies can interfere constructively or destruc- environment, but similar trends have also been observed in
tively at the receiver. If a station (transmitter/receiver) other wireless measurement studies [46]–[49].
or parts of the environment are allowed to move, the
2The
B. Some Consequences of Channel Properties
rms delay spread is obtained from the channel impulse response by
measuring the excess delays and received signal strengths of the second and 1) Consistency Problems: When a system uses the pro-
subsequent received pulses relative to the time instant where the receiver
gets the first pulse. The rms delay spread is the standard deviation of the ducer–distributor–consumer communication model, like the
weighted (by signal strengths) excess delays. WorldFIP [11], [12] fieldbus does, communication is based

1132 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

on unacknowledged broadcasts of data identifiers (by the dis- sion can take multiple token circulations, and packets ar-
tributor), to which the station possessing the identified data riving at master station in the meantime will incur any
item (the producer) broadcasts its actual value.3 All con- corresponding delays. These delay problems are much less
sumers interested in this data can copy the received value into pronounced when nonbursty channels having the same av-
an internal link-layer buffer for their applications. To achieve erage packet error rate are considered. Token passing proto-
spatial consistency among a set of consumers, it is required cols thus serve as an example for the fact that it is often not
that each of these consumers receive the data value. Spatial only the raw presence of bit errors that is important, but also
consistency is required, for example, when distinct con- the characteristics (bursty versus nonbursty) of the errors as
trollers work on the same physical process. Failure to reach well.
spatial consistency might lead to inconsistent control deci- Another problem with token-passing protocols is that the
sions among the controllers. Such spatial inconsistency could station passing a token and its successor in the logical ring
occur, for example, if packets happen to be lost. must be in mutual range. This stipulation cannot always be
As an example, assume that the wireless channel is such guaranteed when stations are mobile. The fieldbus protocols
that for every transmitter–receiver pair a packet is cor- based on token passing have no provisions to deal with mo-
rupted independently with a certain probability . When bility, and appropriate mechanisms have to be added [54].
the producer has received the identifier and broadcasts the 3) Problems for Carrier-Sense Multiple Access (CSMA)-
data value, reaching spatial consistency requires that all Based Protocols: Fieldbus systems like CAN [13], [14] use
consumers receive the data packet, which happens with CSMA-based protocols where collisions are possible. In
probability . As a numerical example: with general, CSMA-based protocols [55] work in a distributed
spatial consistency between consumers is reached fashion, where a station wishing to transmit first needs to
with only 41% probability. sense the transmission medium. If the medium is determined
Another often-found requirement is relative temporal con- to be idle, the station starts to transmit. The many CSMA
sistency. Consider, for example, a set of sensors sampling variants that exist differ in what happens when sensing the
a physical process. To achieve relative temporal consistency, medium busy. In the option chosen for the CAN fieldbus,
all the sensors must sample the process within the same pre- a station wishing to transmit waits for the end of the
specified time window. Some fieldbus systems, such as the ongoing transmission and starts its own packet immediately
PROFIBUS DP [9], [10], use a broadcast-based approach after it. Since another station might do so as well, col-
to promote this kind of consistency. A controller broadcasts lisions can occur. The (wired) CAN protocol is based on
a special control packet called FREEZE, which causes sen- a deterministic mechanism to resolve this contention. This
sors to sample their environment immediately and to buffer mechanism is difficult to use for wireless media. It relies
this value for later retrieval. Under the same assumptions as on a station’s ability to transmit and receive simultaneously
above, reaching relative temporal consistency among sen- on the same channel, which is impossible with half-duplex
sors happens only with probability . wireless transceivers.4
2) Problems for Token-Passing Protocols: Fieldbus sys- Receivers need a minimum received signal strength in
tems like the PROFIBUS [9], [10] rely on distributed token order to successfully decode packets or determine that
passing in order to circulate the right to initiate transmissions another station is currently transmitting (carrier sensing).
between a number of controllers (called master stations in Due to path loss, the minimal signal strength required is not
PROFIBUS). The master stations are arranged in a logical reached once the distance between transmitter and receiver
ring on top of a broadcast medium. It is shown in [50] and becomes too large. Accordingly, carrier-sensing operations
[51] for PROFIBUS and in [52] for the similar IEEE 802.4 may fail, giving rise to the hidden terminal problem [58],
token bus [53] that repeated losses of token packets are a se- from which all CSMA-based protocols suffer. Consider three
vere problem for the stability of the logical ring (ring sta- stations , , and , arranged such that and cannot
bility). When a trial to pass the token from to fails, is sense each other’s transmissions, but station can receive
required to run the next trial immediately. In case of bursty signals from both and (Fig. 1). Station currently
channels, however, it might well happen that the channel be- transmits a packet to . Station wants to do the same,
tween and is currently in a deep fade and will stay there performs carrier sensing, and finds an idle channel because
for some time. This deep fade could potentially render all it is out of ’s range. As a consequence, station starts
successive trials for passing the token useless, causing to transmitting as well, and both and ’s packets collide at
become lost from the ring. A master station which has been .
lost from the ring has no transmit opportunities until it has Several solutions, such as the busy tone solutions [58],
been explicitly reincluded by another master. This reinclu- have been suggested for solving the hidden terminal
3In the producer–distributor–consumer communication model, data is
problem. The most widely used approach today, however,
identified instead of stations. A second common communication model
used in fieldbuses is the master–slave scheme, as implemented, for ex- 4The ultimate goal of CAN’s contention resolution procedure is to ensure
ample, in PROFIBUS [9]. In this scheme communication happens between that the station with the highest priority packet wins contention. Such a fea-
addressable stations. The master needs to know the address of the station ture is implementable on wireless channels, but different mechanisms have
(slave) possessing a data item. It directs a request to the slave, and the slave to be used (for example, [56], [57]). An additional problem to be solved is
has to respond to this immediately. Slaves do not initiate transmissions by that of hidden terminal situations: how to make sure that two stations out of
their own. mutual range agree on which one has the highest priority packet?

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1133

 transmitter is given, are known as open-loop techniques.
One useful approach is the forward error correction (FEC)
coding [59]. In FEC schemes, the transmitter adds redundant
bits to the packet which allow the receiver to correct bit
errors if there are not too many of them. The ratio of user
bits to the overall number of bits after encoding (user bits
plus overhead bits) is called the code rate. Roughly, the
smaller the code rate, the higher the overhead and the better
Fig. 1. Example scenario for hidden terminal problem; circles the error correction capabilities. The rate of uncorrectable
indicate transmission/sensing range. bit errors depends on the channel bit error rate and the
distribution of errors. One example is provided by BT (see
Section III-A). The interference created in situations where
is the RTS/CTS handshake adopted by the IEEE 802.11
multiple piconets overlap leads to “almost binary” channels.
standard [15]. In this scheme, station starts its packet
Such a channel alternates between excellent and very bad
exchange with using a short control packet called re-
states. During excellent states, the overhead of FEC is not
quest-to-send (RTS). Station answers with a clear-to-send
needed, while at the same time the FEC may not be strong
(CTS) packet. Only upon receiving the RTS does station
enough to correct all errors in the bad state and the FEC
continue with sending the data packet. Any other station
overhead is not useful.
receiving an RTS or CTS packet not destined for it must
There are many other options for increasing robustness
remain quiet for the time indicated in the RTS/CTS packets
that do not require feedback (open-loop mechanisms). Take
in order to avoid distortion of ongoing packet exchanges.
for example: 1) using multipath- and interference-resistant
The problem with the RTS/CTS handshake in the context
modulation schemes such as orthogonal frequency division
of fieldbus systems is that the majority of all packets are very
multiplexing (OFDM) [60] or spread spectrum modulation
small. Most packets are on the order of a few bytes, and data
[61], [62]; 2) transmitting a packet not only once but mul-
packets themselves are thus only slightly larger than RTS or
tiple times; or 3) optimizing unit placement and the number
CTS packets. Hence, the RTS/CTS handshake creates sig-
of infrastructure equipment required [such as access points
nificant overhead. Sometimes it may be possible to avoid
(APs)] [63], [64].
hidden terminal situations (and the RTS/CTS handshake) by
Some fieldbus systems, such as PROFIBUS, use retrans-
careful reconsideration of the placement of stations. Another
missions, and can give a transmitter some control of whether
approach is to simply not use RTS/CTS and take the risk of
deadlines are met or not. Protocols where the transmitter
having hidden terminal situations, while reducing its proba-
receives feedback from the receiver and performs retrans-
bility by cutting down the time needed to transmit a packet,
missions when necessary, are commonly called automatic re-
but doing so requires an increase in the bitrate. This approach
peat request (ARQ) protocols [65]. Because of the receiver
is feasible under certain conditions. Industrial applications
giving feedback, they can also be classified as closed-loop
often require only moderate bitrates in the order of hundreds
techniques. Retransmissions are useful when stringent reli-
of kilobits per second or a few megabits per second (CAN
ability requirements have to be met that cannot be reached
has 1 Mb/s as maximum rate). Current wireless technologies,
by open-loop techniques alone (example: important alarm
however, offer bitrates in the order of tens of megabits per
packets). Furthermore, in contrast to open-loop FEC coding,
second (see Section III). Depending on the amount of phys-
ARQ protocols produce overhead only when it is needed to
ical layer overhead (for example, preambles) introduced in
combat errors. During good channel periods, there are no re-
wireless technologies, multiple wireless packets can be trans-
transmissions and the overhead of ARQ protocols is minimal.
mitted in the time needed to transmit one packet on a wired
The available number of retransmissions, however, is natu-
fieldbus.
rally bounded by packet deadlines. All the time the trans-
mitter spends for retransmitting one packet is taken away
C. Mechanisms to Deal With Channel Errors from the deadlines of other packets waiting to transmit. The
Even when problems like hidden terminal situations ei- quest in protocol design is, therefore, to find a good scheme
ther do not occur at all or can be somehow circumvented, that improves the delivery reliability within a given deadline.
the problems created by channel errors remain. Many mecha- In the following section, we discuss selected open-loop
nisms have been developed in the past to make data transmis- and closed-loop protocol mechanisms. These mechanisms
sion over wireless channels more robust against the multitude have been proposed in the context of wireless fieldbus sys-
of possible impairments. It depends on the fieldbus system tems and wireless real-time communications.
and communication model which of these mechanisms are 1) Exploiting Spatial Diversity: As has been explained,
applicable. due to multipath fading, the received signal strength is likely
Fieldbus systems working according to the producer–dis- to change between receivers at different locations. Consider
tributor–consumer model make extensive use of broad- two receivers and , which are the same distance away from
casting. There are no retransmissions, and, in general, the a transmitter. If and are relatively close to each other,
transmitter has no way to verify that deadlines are met. the probability that and experience a deep fade at the
Techniques where no feedback from the receiver to the same time is higher than if their distance is larger than the

1134 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

so-called coherence distance.5 The channel behavior is thus it drops the packet and requests a retransmission (up to a
space dependent, and this spatial diversity can be exploited maximum number of trials per packet). In type-II hybrid
in a number of ways. ARQ, the receiver does not simply drop erroneous packets
Receive diversity is an open-loop technique [66] where the but seeks to use the information contained in these erroneous
receiver is equipped with multiple antennas. The spacing of copies to help in the decoding of further retransmissions.
these antennas should take the anticipated coherence distance A simple example of a type-II ARQ scheme is bit-by-bit
into account. When the signal received by one antenna is in majority voting: once the receiver has received at least three
a deep fade, it can happen that the signal is good enough for erroneous versions of the same packet, it can guess what
proper reception at another antenna if the antenna distance the received packet should be by subsequently applying a
is large enough. The receiver is able to switch between the majority voting procedure to all bits from previous trials.
antennas and can, for example, choose the one giving the In [71], this method is varied by including the deadline and
strongest signal. desired delivery probability (see also [72]) in the choice of
In transmit diversity schemes, the transmitter uses mul- the actual coding schemes. Majority voting is the last resort
tiple antennas. There are transmit diversity schemes working when no correct copy has been received before the deadline.
on the level of individual channel symbols (for example, Other schemes are discussed, for example, in [73] and [74].
[67]). When commercially available wireless network A deadline-aware type-II ARQ scheme (“deadline-depen-
adapters are to be used, transmit diversity schemes working dent coding”) is presented in [75]. The transmitter maps the
on the level of whole packets become more interesting. deadline for a packet and its desired delivery probability to
In the closed-loop scheme discussed in [68], the trans- one of several FEC coding methods, creating a number of
mitter switches transmit antennas only in the case of packet overhead bits for the data. On the occasion that a retransmis-
retransmissions. The first trial of a packet transmission is sion must occur, the transmitter does not repeat the user data
on antenna one, the first retransmission on antenna two, the but rather sends more of the overhead bits. Such an approach
second retransmission on antenna three and so forth, in a is called incremental redundancy.
round-robin fashion. This approach is based on the assump- In the strategy presented in [76], the coding method is
tion that, for bursty channel errors, it is better to switch to a chosen based on how far away the deadline is. The closer
spatially different channel than to retransmit over the same the deadline comes, the stronger the chosen coding method
channel and possibly encounter the same error burst as the (requiring more overhead bits) becomes. This approach in-
one hitting the original packet. cludes, as a “special” coding method, the decision to defer a
Sometimes, for reasons of cost or small form factors, it is retransmission for a while. Such a postponing may avoid sit-
not possible to equip stations with multiple antennas. One uations where energy and bandwidth are wasted when trans-
alternative approach to achieve spatial diversity in the trans- mitting a packet during a deep fade on a bursty channel.
mission process is to let other stations help with retrans- Instead of changing the coding method that is used, the
missions of packets [50]. When station fails to transmit transmitter can also adapt the modulation scheme (see [77]
a packet to station , another station might have picked with a joint consideration for deadlines and energy consump-
up the packet and perform the retransmission on behalf of tion) or increase the transmit power as the deadline comes
. To avoid the required coordination between different pos- closer.
sible helpers , the role of might be confined to 3) An Application Layer Mechanism: Sometimes it may
a dedicated station. Such schemes provide a kind of cooper- not be possible for the lower layers to correct all channel er-
ative diversity [69]. rors. For important asynchronous events, like alarms, this re-
The approaches to exploit spatial diversity in retransmis- striction is intolerable, For periodic data sampling of a slowly
sions have been shown to be effective in reducing the proba- varying, continuous process, one can simply accept occa-
bility of deadlines being missed [50], [68]. They work best, sional losses or try to conceal them. One can, for example, re-
however, when the spatial channels are independent/uncor- place missing samples at the receiver by an estimated value.
related. This is a reasonable assumption to make, when mul- In [78], a scheme is proposed where the receiver estimates
tipath fading is the dominant source of channel errors. When missing values on the basis of Kalman filters. It is demon-
the receiving node is placed close to an interference source, strated that by this technique certain signal classes need only
however, switching transmit antennas would most likely not 5 out of 100 samples to be able to reconstruct the original
be helpful because all spatial channels would be affected. signal with good quality, and can be applied at the applica-
2) Hybrid ARQ Schemes: In hybrid ARQ schemes, tion layer.
retransmissions and error-correction coding (FEC) can be
combined in different ways [70]. In the simple type-I hybrid D. Discussion
ARQ, all packets are FEC encoded and always use the
The wireless channel is a difficult environment. Some of
same code. When the receiver cannot correct all bit errors,
the existing fieldbus protocols face serious degradations in
5This coherence distance depends, among other parameters, on the their achievable deadlines when dealing with bursty channels
antenna type and the propagation environment [35, Sec. 5.8.5], [66]. When (compare the token-passing example from Section II-B2).
multipath components can arrive from every direction, for example, a
spacing between half a wavelength and a full wavelength is sufficient to Other protocols, on the other hand, are not directly im-
achieve diversity gains (compare [66]). plementable at all. One example is the method used in

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1135

the CSMA-based CAN protocol for resolving contention By designing a comparably straightforward system, the de-
between stations. signers of BT intended for it to have widespread use.
The wireless channel should influence the design of in- BT networks are organized into “piconets” in which
dustrial applications. The fault assumptions, for example, are a “master” unit coordinates the traffic to and from up to
different in wireless channels than in wired channels [72]. It seven active “slave” units. The master unit originates the
is likely that errors occur more often on wireless channels request for a connection setup. Within a single piconet,
than on wired ones. Transmission errors on wireless chan- the various slave units can only communicate with each
nels, however, tend to be transient (deep fades end at some other via the master. Nevertheless, every BT unit can be
time and the channel becomes good again), whereas errors on a member of up to four different piconets simultaneously
wired channels are often of permanent nature, due to faulty (though it can be master in only one of them). A formation
cables, connectors, or other hardware components. in which several piconets are interlinked in such a manner
Note that for some applications the rate of residual errors is called a scatternet. Up till now, the role of scatternets is
(after applying any countermeasures) has to be kept at still relatively limited. Some of the specific problems seen
extremely low levels. “Fly-by-wireless” systems in aircrafts, within scatternets are discussed in [87].
for example, might require residual error rates of 10 Piconet traffic is strictly organized into a time-divi-
[72]. New protocol mechanisms or the combination of ex- sion multiple access (TDMA)/duplex scheme [88]. In this
isting protocol mechanisms are needed to achieve such low scheme, the master is only allowed to start transmitting in
levels of residual error rates. Even if these levels become odd-numbered time slots (each slot is 625 s long), while
achievable, it will likely require a tremendous amount of slaves can only respond in even-numbered slots after having
design effort. To reduce this effort, it is helpful to relax the been polled by a master packet.
requirements posed to the communication system. This can Because there is no coordination between different pi-
be achieved, for example, by designing industrial applica- conets, packet collisions may occur if two piconets are
tions that can tolerate a certain percentage of packet losses located near one another. To minimize this collision effect as
or deadline misses. This line of research is pursued in the well as to cope with the fact that frequencies used by other
area of networked control systems [79]–[81]. devices on the radio channel can vary significantly over the
bandwidth of the 2.4-GHz ISM band, every piconet per-
forms a rather fast frequency hopping (FH) scheme over 79
III. REVIEW OF WIRELESS TECHNOLOGIES FOR INDUSTRIAL carries of 1-MHz bandwidth each. The maximum hopping
AUTOMATION frequency of this scheme is set at 1.6 kHz (corresponding to
For the various reasons listed at the beginning of the the slot length of 625 s) and the hop sequence used by each
introduction, wireless technologies might be of advantage individual piconet is derived from the unique address of its
in industrial environments. Due to the general tendency master using a specified algorithm. For each BT packet sent,
toward standardization and the fact that cheap, commer- a new frequency is chosen to send it over. In BT version
cial-of-the-shelf (COTS) wireless technologies are available, 1.2, an adaptive FH scheme (AFH) has been introduced
it seems only logical to investigate these for their suitability which allows for the exclusion of certain carriers once it has
in industrial deployment. Of particular interest for industrial been noted that packet corruption occurs at that carrier’s
environments are technologies that do not require any sort frequency. It must be noted, however, that AFH is used more
of frequency licensing. These technologies include the as a means to improve the performance of a BT piconet in
WPAN technologies such as IEEE 802.15.1/BT and IEEE the presence of other nonhopping systems in the 2.4-GHz
802.15.4/ZigBee as well as the WLAN technologies from ISM band (see also Section III-D) than a way to improve the
the IEEE 802.11 family. performance among coexisting BT piconets.
Despite the advantages a single wireless network might On the physical layer (PHY), data is Gaussian frequency
offer on the factory floor, it will be often required to run shift keying (GFSK) modulated at 1 M/s and transmitted with
multiple WLAN/WPAN networks in parallel in different or a power of 0 dBm (1 mW). With such a transmit power, BT
overlapping regions of the plant. Because of this fact, the co- devices can expect to have up to a nominal range of about
existence of multiple networks of either the same or varying 10 m. BT can also be used with up to 20-dBm transmit power.
types needs to be investigated. We consider how certain com- Transmitting at such high power results in a larger range,
munication patterns typical to those of fieldbus systems can but requires implementing power control to fulfill the sharing
be implemented within such overlapping networks. rules of the ISM band.
On the data link layer, a distinction is made between
asynchronous connectionless (ACL) and synchronous con-
A. BT Technology/IEEE 802.15.1
nection-oriented (SCO) packets: ACL links secure reliable
BT was originally designed as a cable replacement tech- data transmission with an ARQ scheme that initiates the
nology aimed at providing effortless wireless connectivity retransmission of a packet in case the evaluation of the
for consumer devices in an ad hoc fashion [82]–[86]. In order included cyclic redundancy check (CRC) shows inconsis-
to allow for deployment almost worldwide, the BT Special tencies. Six different types of ACL packets exist and can
Interest Group (SIG) placed the technology in the unlicensed occupy either one, three, or five BT time slots, depending
industrial, scientific, and medical (ISM) band at 2.4 GHz. on which type is being used. Three of the ACL packet

1136 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

types include uncoded payloads, while the other three have vantageous to use short packet types. This fact is unfortunate
payloads that are protected by rate-2/3 (code rate) FEC that because short packets are typically used in industrial appli-
uses a shortened Hamming block code of length 15 or 10 cations. In fact, it is very hard to set up a scenario in which
without any interleaving. The uncoded ACL-packet types any other packet type yields a larger throughput than that of
are knowns as DH1, DH2, and DH3, while the three coded DH5 [92].
ones are known as DM1, DM3, and DM5. Using packets These results present a basis on which the suitability of
of type DH5 for data and DH1 for acknowledgment gives using BT technology for specific applications in a specific
the maximum possible (unidirectional) throughput for BT at industrial environment can be investigated.
723 kb/s. Security is supported in BT by the specification of authen-
SCO links, in contrast, support real-time traffic by re- tication and encryption.
serving time slots at periodic intervals. Retransmissions The most recent development for BT is BT version
are not allowed with these types of links, but in BT ver- 2.0 [94]. BT version 2.0 has enhanced data rates using
sion 1.2/2.0, “extended” SCO links have been introduced -DQPSK and 8DPSK modulation schemes in addition to
where a limited number of retransmissions can be made. the traditional GFSK modulation scheme. The transmission
The three different types of SCO packets all have the same rate resulting from these enhancements is about three times
length and require a time of 366 s for transmission. They faster than it was in previous versions of BT.
typically transport 64 kb/s of continuous variable slope
delta (CVSD)-encoded speech [89] and are differentiated by B. IEEE 802.15.4
having either unproteced payloads, rate-2/3 FEC encoded The IEEE 802.15.4 standard [20] was finalized in October
payloads, or rate-1/3 FEC encoded payloads. These packet 2003 and specifies the characteristics of the physical layer
types are known as HV3, HV2, and HV1, respectively. and the MAC layer of a radio networking stack.6 The goal of
The extended SCO link is very flexible, supporting various this standard was to create a very low cost, very low power,
transmission rates. Lost SCO packets can be replaced by an two-way wireless communication solution that meets the
erasure pattern. unique requirements of sensors and control devices [20],
Because of the short range of BT and the small number of [21]. In contrast to BT and IEEE 802.11, IEEE 802.15.4
slaves that are active at any given time, several independent has been specifically developed for use with applications
BT piconets will most likely coexist on a factory floor. With in which a static network exists that has many infrequently
the help of radio network simulations, BT–BT coexistence used devices that transmit only small data packets. Such
results have been presented in [90]–[92]. These simulations applications are exactly what many industrial environments
took traffic, spatial node distribution, fading models, as well would require.
as co- and adjacent channel effects into consideration. Some In order to encourage widespread deployment, IEEE
of the results of these simulations have also been verified in 802.15.4 has been placed in unlicensed frequency bands.
a radio network testbed [93], making them, therefore, more When using the 2.4-GHz ISM band, IEEE 802.15.4 systems
detailed than a purely theoretical approach. can get the same sort of global deployment as BT and IEEE
The results have been obtained for an area of 10 20 m , 802.11b/g systems. The IEEE 802.15.4 standard has also
assuming an average master–slave distance of 2 m. Even
been specified for use in the 868-MHz ISM band in Europe
though a factory floor generally is significantly larger, it is
and in the 915-MHz ISM band in North America. Within
reasonable to assume that the performance depends only on
these bands, DSSS is used in order to comply with the
the node density and not on the absolute numbers of nodes
respective sharing rules of each band as well as to allow
or the area covered. The considered area of 10 20 m sup-
for simple analog circuitry to be used. The maximum data
ports either 30 simultaneous, 1/3 loaded SCO connections
rate of the DSSS is 250 kb/s in a single channel within the
with an average packet loss rate of 1% or 100 (!) simulta-
2.4-GHz band. In total, the 2.4-GHz band accommodates
neous WWW sessions (bursty traffic with an average data
16 such channels. In the 868-MHz ISM band, one channel
rate of 33.2 kb/s each) with a degradation of the aggregate
with a data rate of 20 kb/s is available, whereas in the
throughput of only 5%. With 50 fully functional piconets in
915-MHz band, ten channels of 40 kb/s each can be used.
the area, a maximum aggregate throughput of 18 Mb/s can
Because of various system parameters, especially the MAC
be expected when each individual piconet transmits at an av-
protocol that is in use, the maximum user data rate will most
erage unidirectional data rate of 360 kb/s.
likely be about half of its nominal value, or less. If upper
Furthermore, the results show that in the interference sce-
layers detect a throughput degradation while using a specific
narios the provided FEC methods are unsuitable to handle the
channel within the used frequency band, IEEE 802.15.4 can
almost binary character of the transmission channel. When
scan the frequency band for a channel that promises better
two overlapping piconets operate on different frequencies,
perfomance values and switch to that channel [95] (unless
the signal quality is good, but when they hop to the same fre-
transmitting in the 868-MHz band).
quency, packets can be destroyed beyond recognition. When
using no coding method at all (as is the case with HV3, DH1, 6There is sometimes confusion between IEEE 802.15.4 and ZigBee. The

DH2, and DH3), the power consumed by the network as well ZigBee alliance (see http://www.zigbee.org) is a consortium driven by in-
dustry and research institutions. It finalized the ZigBee specification in De-
as the overall load of the network is reduced. In order to ob- cember 2004 and describes higher layer protocols (networking, application)
tain a good throughput and have low interference, it is disad- that operate on top of IEEE 802.15.4.

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1137

Fig. 2. Superframe structure for IEEE 802.15.4 beaconed mode.

The IEEE 802.15.4 standard differentiates between two increased delay spread found in industrial plants not to cause
different kinds of devices. A full-function device (FFD) can a problem.
become a network coordinator and can work with other FFDs For security purposes, IEEE 802.15.4 provides authenti-
in a peer-to-peer fashion. Reduced-function devices (RFD), cation, encryption, and integrity service. The developer can
on the other hand, are always associated with one of these choose between no security, an access control list, and a
FFDs and are limited to exchanging data with this device 32–128-b Advanced Encryption Standard (AES) encryption
alone. Among RFDs there is no peer-to-peer communication with authentication.
possible. All devices have a 64-b address, but it is possible
for RFDs to obtain a 16-b shorthand address from their co- C. IEEE 802.11 Technologies
ordinator FFD.
IEEE 802.11 is composed of a number of specifications
With respect to the MAC protocol used by the IEEE that primarily define the physical and MAC layers of WLAN
802.15.4 standard, there are two different modes of oper- systems [15]–[17], [44], [97]. Similar to other standards from
ation. In unbeaconed mode, all stations use an unslotted the IEEE 802.x series, the IEEE 802.11 MAC suggest the
CSMA variant. Here, a station initiating transmission of a IEEE 802.2 logical link control (LLC) [98] as a standard
packet does not perform carrier sensing immediately, but interface to higher layers. Since IEEE 802.11 is a WLAN
introduces a random waiting time, called a backoff time. standard, its key intentions are to provide high throughput
Having such a backoff time facilitates the avoidance of and a continuous network connection. Because of the focus
collisions. In beaconed mode (see Fig. 2), the network co- on COTS technologies for wireless connections in industrial
ordinator imposes a superframe structure. The coordinator deployments, only the most common variations and exten-
transmits beacons periodically, choosing one of a number sions of IEEE 802.11 systems will be discussed here. These
of configurable periods between 15.36 ms and 251.65 s. variations and extensions include the general 802.11 MAC,
The remaining superframe starts with the contention-access IEEE 802.11a, IEEE 802.11b, and IEEE 802.11g for phys-
period, in which the RFDs access the medium according to a ical layers, as well as relevant extensions in respect to net-
slotted CSMA-CA variant, which incurs more overhead than work planning and QoS.
the unslotted variant. An optional contention free period is The main parameters of IEEE 802.11 a/b/g are the
available, where the network coordinator allocates guaran- following.
teed time slots (GTSs) to individual RFDs for either uplink
• IEEE 802.11a [16] is placed in 5-GHz bands that
data or downlink data. In addition to these two modes of
are license exempt in Europe (5.15–5.35 GHz and
operation, an inactive period of operation exists. During this
5.47–5.725 GHz) and unlicensed in the United States
period, all nodes including the coordinator in the network
(UNII bands, 5.15–5.35 GHz and 5.725–5.825 GHz).
are put to sleep in order to conserve energy.
Over the whole spectrum, this allows for 21 systems
Data packets are acknowledged and the protocol supports to be running in parallel in Europe and eight in the
retransmissions, but there is no FEC coding. In the beaconed United States [99]. The IEEE 802.11a physical layer
mode, the throughput is smaller than in the unbeaconed (PHY) is based on the multicarrier system orthogonal
mode, in which no beacon frames exist and the unslotted frequency-division multiplexing (OFDM) [60]. Seven
CSMA variant has less overhead. Under the conditions in- modes are defined, ranging from BPSK modulation
vestigated in [96], the maximum user data rate when running with rate-1/2 FEC and a 6-Mb/s data rate to 64-QAM
in the 2.4-GHz ISM band is 38 kb/s with one source, and up modulation with rate-3/4 FEC and a 54-Mb/s data
to 70 kbs when multiple sources are present. These observed rate. The maximum user-visible rates depend on the
data rates are, in fact, quite small. packet sizes transmitted. In the 54-Mb/s mode, the
Similar to BT, IEEE 802.15.4 uses low transmit power transmission of Ethernet packets that are 1500 B long
levels. In addition to this, IEEE 802.15.4 also uses very results in a maximum user rate of about 30 Mb/s,
short symbol rates (up to 62.5 ksymbols/s), allowing the while sending packets with user payloads of just 60 B

1138 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

 results in a throughput of 2.6 Mb/s [100]. The latter rier-sense mechanism determines the channel to be free, a
throughput value is the one of interest for industrial station may start to transmit. If the channel is sensed busy,
applications, as small packet sizes are dominant in the station awaits the end of the ongoing transmission, where
fieldbus networks. the channel becomes idle again. At this instant, a random
• IEEE 802.11b [17] is a high-rate extension to the backoff timer is started. If the channel becomes busy be-
original IEEE 802.11 DSSS mode and thus uses the fore the timer expired, the timer freezes and restarts once
2.4-GHz ISM band. Although in principle either 11 the channel is idle again. If the timer expires and no other
or 13 different center frequencies can be used for the station has started transmission in the meantime, the station
DSSS (depending on whether you are in the United starts transmission. The contention window from which the
States or in Europe), only three systems can actually backoff values are chosen increases exponentially after each
operate in parallel. In addition to supporting the 1- failed trial to transmit a packet.
and 2-Mb/s modulation rates of the basic IEEE 802.11 The basic CSMA-CA method can be enhanced with an
system, the payload of the IEEE 802.11b PHY allows optional RTS/CTS handshake to avoid hidden terminal sit-
for modulation with 5.5- and 11-Mb/s complementary uations. The user can control whether or not this handshake
code keying (CCK). The maximum user data rates are is used by configuring a threshold for frame sizes. If a frame
7.11 Mb/s in the case of Ethernet packets and 0.75 size exceeds this threshold, then RTS/CTS will be used; oth-
Mb/s in the case of packets with user payloads of 60 erwise, it will not. The second coordination function, point
B in length. coordination function (PCF) [15], is not mandatory and is de-
• IEEE 802.11g is an extension to the IEEE 802.11b signed to provide time-bounded services by means of subdi-
specification and is consequently also placed in the viding time into (variable-length) superframes which in turn
2.4-GHz band. It supports four different physical are subdivided into a contention-free period (CFP) and a con-
layers of which two are mandatory: the PHY that is tention period (CP). Within the CFP, a polling scheme is
identical to IEEE 802.11b and an OFDM PHY that used, while access is regulated according to the DCF during
uses the same modulation and coding combinations the CP. The use of PCF is not very widespread7 and it has a
as IEEE 802.11a. Because of the different frequency reputation of being slightly inefficient [103], [104].
band, the maximum user transmit rates are about The IEEE 802.11 WLAN DCF was designed for “best
26 Mb/s for Ethernet packets and about 2 Mb/s for effort” traffic like file transvers that is neither time critical
packets with user payloads of 60 B when using the nor periodic. The PCF has difficulties to provide periodic
54-Mb/s modulation scheme. services due to superframe stretching (i.e., jitter in the
It can be seen that when transmitting packets that contain start times of superframes) and foreshortened CFP’s [103].
small payloads (as is the case for most fieldbus systems), For better support of timing-sensitive transmissions, there
throughput values are significantly reduced. This reduction is an ongoing activity—IEEE 802.11e—that extends the
is due to the comparably large overhead of IEEE 802.11 basic MAC protocol [105]–[107]. In IEEE 802.11e, the
packets and the different parameters present in the CSMA enhanced DCF (EDCF) is designed as an improvement over
protocol [100]. In contrast to BT or IEEE 802.15.4, IEEE the DCF. The EDCF allows to assign different basic con-
802.11 has been specifically optimized to transmit large data tention window durations according to the priority of a data
files, therefore showing a suboptimal performance when packet, thus providing stochastic service differentiation (data
the majority of data is made up of short control packets. packets with shorter basic contention windows are more
Note that throughput values can decrease even more when probable to obtain channel access than those with longer
additionally using higher layer protocols. Measurements ones). The PCF is improved upon with the introduction of
for the throughput of TCP/IP traffic in Ethernet packets for the hybrid coordination function (HCF) that is not based
the IEEE 802.11b specification, for example, have returned directly on polling, but by which the hybrid coordinator as-
results of a 5-Mb/s maximum throughput [101], [102]. IEEE signs time slots to stations on an explicit basis. The stations
802.11 employs immediate MAC-layer acknowledgment can use these time slots as needed, but are restrained to the
and retransmissions. These acknowledgment packets do not given time window. With this restriction, the possibility to
have to actually contend for the channel; instead, they have achieve periodicity is significantly improved.
a reserved time slot that is only used upon the reception of a Because of the existence of (metal) obstacles and the com-
correctly received packet. parably large transmit power of 20 dBm, the delay spread of
In principle, it is possible to have IEEE 802.11 ad hoc a factory floor has to be looked at if IEEE 802.11 is to be
networks that consist solely of mobile stations (MSs). It is used there. While the delay spread in homes and offices is
more likely, however, that IEEE 802.11 will be used in an assumed to be 50 and 100 ns respectively, it takes on
infrastructure mode, whereby an AP relays all communi- values of 200–300 ns in a factory floor setting (the delay
cations between stations and other networks. To organize spread is much less severe in BT due to its much smaller
the traffic on the radio link, the IEEE 802.11 MAC pro- transmit power and range). In the case of IEEE 802.11b,
vides two coordination functions. The first of these coordina- a conventional RAKE receiver supports (only) about 60-ns
tion functions, the distributed coordination function (DCF),
is mandatory and requires that all stations compete for the 7The PCF is more complex to implement than the DCF and the authors
channel according to a CSMA-CA protocol. When the car- know of no implementation that supports it.

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1139

delay spread in 11 Mb/s mode and 200 ns in 5.5 Mb/s mode one of these modes, the carrier-sense circuitry indicates a
[108]. Nevertheless, more receiver algorithms with suitable busy medium when the received signal energy is above some
robustness have recently come into existence [109], [110]. threshold. It does not matter if the received signal comes from
In case of IEEE 802.11a or g, the situation is better. Be- another IEEE 802.15.4 station, an IEEE 802.11 station, or
cause of the guard interval between channel symbols inherent any other device radiating in the same frequency band. In
in the OFDM technology, delay spreads of several hundred one of the other carrier-sensing modes, it is required that the
nanoseconds can be supported easily without paying atten- received signal be decoded properly, i.e., use the right mod-
tion to the receiver algorithms implemented [108]. ulation scheme. In this mode, an IEEE 802.15.4 station can
When considering the overall network performance and only detect ongoing transmissions from other IEEE 802.15.4
not just the individual link performance (the interference stations and not from, say, an IEEE 802.11 station.
performance is discussed in Section III-D), the number The interference between BT and IEEE 802.11b has been
of publications presenting well-founded results is limited. investigated extensively [101], [102], [113]–[117]. In [118],
The existing ones, however, show that capacity is indeed it was found that IEEE 802.11b requires a carrier-to-interfer-
an issue. The authors in [111] and [112] discuss how the ence ratio (CIR) of about 10 dB to cope with a (narrow-band)
aggregate throughput in a single network decreases with the interference in its (wide) passband. Nevertheless, with IEEE
number of users, due to either hidden or exposed terminal 802.11a becoming more widely spread (especially in indus-
problems8 or due to additional RTS/CTS overhead. With trial environments) and the use of AFH for BT version 1.2,
only ten stations [111] or a hidden node probability of 5% the interference situation is becoming much more relaxed. At
[112], the system throughput is approximately halved (!) in least this is the case for frequency static systems like IEEE
the case of 1500-B payloads (the smaller packet sizes typical 802.11. An IEEE 802.11 system transmits on a fixed set of
in fieldbuses are even less efficient). BT channels. When multiple AFH-enabled BT systems op-
When installing IEEE 802.11a in a cellular fashion, the sit- erate in parallel with an IEEE 802.11 network, all of the BT
uation can be simplified. First of all, there is more bandwidth systems stop using any of the channels occupied by the IEEE
available in the 5-GHz band than in the 2.4 ISM band. Sec- 802.11 network. The BT systems now have to share the re-
ondly, a decentralized channel selection algorithm has been maining channels. Since there are fewer available channels
standardized for IEEE 802.11a in IEEE 802.11h [99], orga- than in the case without an IEEE 802.11 network, the BT
nizing a number of spatially overlapping networks to choose systems create more interference to each other.
nonoverlapping frequency bands. The most relevant param- Along with the introduction of AFH, another coexistence
eter for WLAN frequency planning is the number of mobile improvement has been made to BT. In BT version 1.0/1.1, the
terminals that have to be served. From this parameter, the BT reverse link packet that contained the acknowledgment
optimum number of APs and distance between APs can be was transmitted on a different frequency than that of the for-
determined. ward link packet. In doing so, the loss probability in the case
For security, IEEE 802.11 WLAN’s support several au- of a static interferer was increased because the reverse link
thentication processes which are listed in the specification might hop into the interfered band even if the forward link
(none are mandatory). transmission was successful. With the BT version 1.2 AFH,
forward and reverse link packets are now transmitted on the
D. Coexistence of Wireless Technologies same frequency, and such a situation will not occur anymore.
In the future, it will be standard for multiple wireless tech- If a BT system and an IEEE 802.11b system are so close
nologies to be used in a single environment. This is gen- together that the 20-dB higher transmition power of IEEE
erally not a problem unless the technologies are placed in 802.11b blocks the BT receivers from receiving (when an-
the same frequency band. As the previous sections indicated, tennas are less than 20–30 cm apart), then only methods such
the 2.4-GHz band hosts BT, IEEE 802.15.4, IEEE 802.11b, as a joint scheduler can help to enable a fair coexistence
and possibly other systems. It is thus necessary to investi- [102].
gate the performance of coexisting networks and to introduce Since IEEE 802.15.4 and IEEE 802.11b are frequency
methods for reducing mutual disturbances between them. static and can be manually or automatically made to use
CSMA is, in principle, a mechanism for improving the co- different frequency bands, the authors of [95] have con-
existence of multiple communication systems that use the cluded that IEEE 802.15.4 will have little or no impact on
same frequency band. In fact, the goal of the carrier-sensing IEEE 802.11b as long as some frequency management is
operation is to avoid interfering with ongoing transmissions. deployed. Additionally, typical IEEE 802.15.4 applications
It depends, however, on the actual implementation of the are expected to have a low duty cycle between 0.1% and 1%
carrier-sensing mechanism whether transmissions of other (meaning that stations will transmit nothing for 99.9% or
types of wireless networks can be detected or not. In the 99% of their time, respectively). Stations operating in this
IEEE 802.15.4 standard, for example, the user can choose duty cycle range do not create significant interference to
between different carrier-sensing modes [20, Sec. 6.7.9]. In other networks.

8The hidden terminal problem has been discussed in Section II-B3. In the E. Comparison of Wireless Systems
exposed terminal problem, station A wants to transmit a packet to station C .
Station A senses an interferer B and refrains from transmission, even when In the last few sections, we examined three different
B ’s signals do not reach C . A possible transmission is thus suppressed. systems. All three systems have been designed for use in

1140 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

Table 1
Comparison of BT, IEEE 802.15.4, and IEEE 802.11 Technologies

different scenarios, thus offering various advantages and • In the case of broadcast packets, there is no RTS/CTS
disadvantages over one another depending on their use. The handshake, there are no ACK frames, and no MAC
802.11 systems are suitable for transmitting large amounts layer retransmissions are performed. Accordingly, the
of data. IEEE 802.15.4 is suitable when communication is transmission of responses or the initiation of retrans-
infrequent, small packet sizes are used, and power consump- missions have to be implemented in higher layers.
tion is an issue. BT fills the gap between these two by being From the perspective of the MAC layer, request and
able to transmit at medium data rates with a lower power response packets are equivalent in the sense that the
consumption than IEEE 802.11 (see also Table 1). MAC entity has to contend for the channel in the
Although all three systems can be used on the factory floor, same way for both types of packets. This contention
none of them will be able to run at their nominal perfor- includes the carrier-sensing operation. As a result, in
mance levels because of the adverse radio conditions that the 802.11 DCF case, an acknowledgment or response
will be present. These conditions, such as frequency-selec- packet might be delayed by interference or traffic
tive fading or interference, are especially prevalent if the from colocated systems, causing the round-trip time
system is placed in the 2.4-GHz ISM band. to become randomized. In fieldbus systems such as
Note that the effects of Doppler shifts caused by mobility PROFIBUS [9], [10], on the other hand, response or
have not been considered. The velocity of moving entities acknowledgment packets are sent immediately, and no
on the factory floor is expected to be too small to matter carrier sensing is required. In this type of system, the
(not more than 20 km/h; see [40], based on a user survey). round-trip time is predictable.
Seamless connectivity of moving entities might require • In the case of unicast packets, the RTS/CTS handshake
hand-over algorithms. While for IEEE 802.11 roaming is can be switched off. All stations, however, generate
specified in IEEE 802.11f (though with some performance ACK frames automatically and the MAC layer per-
limits [119]), it is not covered in the specifications of BT forms retransmissions. These ACK frames are not al-
and IEEE 802.15.4 (though not impossible to realize). ways useful. The PROFIBUS offers a mechanism, for
example, where the responder can place answer data
F. Implementation of Fieldbus Services directly into its MAC layer acknowledgment packet.
When wireless fieldbus systems are to be implemented This answer data is fetched from a link-layer buffer,
with one of the COTS technologies discussed in this section, and it becomes the responsibility of the higher layers
the fieldbus services and communication models have to be to write appropriate data to this buffer. Such an imme-
mapped to the services and interfaces offered by these sys- diate acknowledgment carrying data is not possible in
tems. We discuss mapping approaches for two of these tech- IEEE 802.11, however, since the ACK frames are al-
nologies: IEEE 802.11 with DCF, and BT. ways empty. To emulate this behavior, the responder
1) IEEE 802.11 DCF: Even with problems like channel must issue a separate data packet for the response data.
errors or hidden terminal situations taken aside, the IEEE The overall transaction thus includes two extra ACK
802.11 DCF lacks predictability due to its stochastic ac- packets.
cess mechanism with random backoff times and station The way in which the producer–distributor–consumer
contention. communication model ([120]; see also Section II-B1) can
One can eliminate contention by using a contention-free be implemented using the IEEE 802.11 DCF and the IEEE
access mechanism such as polling9 or token-passing on top 802.2 Logical Link Control (LLC) [98] protocol is investi-
of the IEEE 802.11 DCF. In the case of polling, for example, gated in [121]. The LLC is a common link-layer protocol for
this can mean that a station starts to use the DCF access the IEEE 802.x technologies.10 An additional producer–con-
mechanism only after receiving a poll packet. It is impor- sumer protocol (PCP) uses the services of the LLC to
tant to consider whether the fieldbus packets should be em- implement the producer–distributor–consumer mechanism.
bedded into unicast IEEE 802.11 packets or into broadcast To query a data item, the distributor’s PCP entity first
IEEE 802.11 packets: 10Together, the LLC and the different IEEE 802.x technologies cover
the data link layer and the physical layer of the OSI reference model. The
IEEE 802.x technologies provide the physical layer and the MAC sublayer,
9The “standard” solution for polling with IEEE 802.11, PCF, has not whereas the LLC provides the upper parts of the data link layer. It is com-
found widespread deployment. When only DCF equipment is available, posed of mechanisms for error control, connection management, as well as
polling must be implemented separately. others.

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1141

broadcasts a packet containing the unique identifier of the • In the current technique, the slave replaces the cyclic
data item. To achieve this broadcasting, the distributor PCP answer data with the event data.
entity requests the unacknowledged connectionless service • In the immediate technique, the slave transmits the
of the LLC. The LLC, in turn, maps this service request to alarm packet by itself, without being polled. Since the
an IEEE 802.11 MAC-layer broadcast and the packet is sent. standard DCF access procedure is used, the slave con-
When the MAC entity of any station receives this packet, tends with ongoing cyclic exchanges.
it passes it on to its own LLC entity, and then on to its PCP Interesting performance metrics for such a system are the
entity. If station happens to be the producer, it gets the mean update time of a cyclically polled data item and the
value of the data item. To send this data item, station then mean alarm latency. This alarm latency can be defined as the
invokes the unacknowledged connectionless LLC service, time between the occurence of an alarm at a slave and recep-
eventually resulting in a MAC layer broadcast. Any station tion of the alarm packet at the master. For a simple channel
receiving the data packet passes it on to its LLC entity and error model that is loosing packets independently with a fixed
then on to its local PCP entity. A station not interested in probability (varied between 0% and 10%), a fixed number
this data item drops it. of stations, and the assumption that at most one slave has an
Since this approach is based on IEEE 802.11 MAC layer outstanding alarm at any time, it can be shown that all three
broadcasts, the data production requests and the packets techniques have similar mean update times. Also, the vari-
carrying the data items are not retransmitted. From the ance of the update times is close together. With respect to
perspective of a consumer who is interested in updates mean alarm latency, the “immediate” technique is much more
of a particular data item, two things are necessary for a advantageous than the “current” technique, which in turn is
successful update: the producer must receive the production better than the “late” technique. The “immediate” technique
request, and the consumer must receive the broadcasted data is dangerous in alarm shower11 situations, since not only do
item. The time between two updates of a data item at a con- the alarms contend with the cyclic traffic, but they also con-
sumer is called the update time. Even when the production tend with each other, causing collisions to be more likely to
requests have a perfectly constant spacing, channel errors occur. These problems would be amplified in hidden terminal
will eventually turn this update time into a random variable. situations.
In master–slave systems like PROFIBUS-DP, there is a With respect to IEEE 802.11, it must be said that many
master cyclically exchanging data with a number of slaves. proposals have been made to modify the physical and/or
Exchanging data means that the master can send data to the MAC layer to better support deterministic or stochastic
slave (for example, data that the slave should output to the real-time services [123]–[125]. Many of these proposals,
physical environment) as well as receive input data from the however, necessitate modifications to existing networking
slave. In addition, a slave might want to issue asynchronous equipment. While the upcoming IEEE 802.11e standard
alarm events (called diagnosis messages in PROFIBUS-DP) [105] offers promising features, further investigation is
from time to time, in order to, for example, notify the master needed to fully exploit the possible mappings of fieldbus
of unusual or dangerous conditions. If the slaves are set to be services to the features offered by this standard. One can
purely passive devices, they can only notify the master about only hope that 802.11e achieves better market penetration
this alarm whenever they are polled, by, for example, setting than PCF has.
a special bit in the answer data packet. 2) BT: Due to its polling-based MAC approach, BT
Mechanisms to implement this behavior are considered in [19], [126], [127] can be an interesting candidate, when
[122] for a monomaster scenario, using IEEE 802.11 DCF its inherent limitations (limited size of a piconet, moderate
along with LLC. For cyclic polling of a slave, the master data rates) are acceptable. The usage of BT in industrial and
uses an acknowledged connectionless LLC service. In this fieldbus applications has been considered in a number of
service, the higher layers on a slave device can prepare an- publications [128]–[131].
swer data and place this into an LLC buffer. When the master In [130], the implementation of the monomaster
sends its data packet, the LLC instance on the slave responds PROFIBUS-DP using the services offered by the BT
with the data stored in this buffer. The LLC data packets of L2CAP (LLC and Adaptation Protocol) is discussed. Both
both the master and the slave devices are mapped to IEEE cyclic polling and delivery of alarm packets are supported.
802.11 unicast packets. Since unicast packets are used, the The L2CAP offers a feature similar to the acknowledged
MAC layer will send an ACK for every received data packet. connectionless LLC service, by allowing the higher layers at
As discussed above, the slave is not able to send the response a slave station to place a piece of data into an L2CAP buffer.
data within the ACK frame, instead the whole channel access When the slave is polled by the master (the poll packet
procedure must be invoked by the slave. can contain data), it answers in the very next time slot and
includes the buffered response data in the answer packet.
The handling of asynchronous alarm packets can be done
As opposed to IEEE 802.11 with DCF, the slaves’ answer
in different ways. In [122], three methods are proposed.
packet does not have to go through any channel access
• In the late technique, the master collects all alarm no-
11An exceptional situation in the physical process or in the operation of
tifications during a cycle and handles them at the end
controllers leads to the generation of alarm messages. These messages may,
of the cycle by separately polling each of the slaves in as an aftereffect, lead to the generation of further alarm messages, and so
need. on. Such a situation is called an alarm shower.

1142 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

mechanism. The proposed implementation for delivery of Wired and wireless segments are coupled through the use
alarm packets is similar to the “current” technique discussed of a single or multiple coupling devices. It is possible for
above. The master polls all of its slaves in a cycle. When a these devices to work on different layers of the OSI reference
poll packet fails, immediate retransmissions are performed. model. The main classes of coupling devices in the realm
Two performance metrics are investigated for this: the mean of fieldbus systems are repeaters (working on the physical
alarm latency and the mean cycle time. Mean cycle time layer), bridges (working on the data link layer), and gateways
can be defined as the time between two successive polls of (application level) [133]–[135].
the same slave. Because of possible retransmissions, the For real-time communications, the delay introduced by
actual cycle time is a random variable. For performance these coupling devices is important. In repeaters and certain
evaluation, it is assumed that the packet error rate between types of bridges, packets are forwarded from one segment to
the master and a slave is constant. When all channels have another with either no modifications to their contents or with
the same packet error rate, both the average cycle time and only minor ones (for example, to translate between different
the average alarm latency increase linearly with the number addressing formats). Here, the forwarding delay can be de-
of slaves. When fixing the number of slaves, the average fined as the time difference between the time instant at which
cycle time and alarm latency grow to infinity as the packet the last bit of a packet is received on the input segment and
error rate increases. the time instant at which the last bit of the packet was trans-
In [131], hybrid configurations consisting of BT-based mitted on the output segment. The forwarding delay depends
monomaster PROFIBUS-DP over the wireless medium, on the type of the device and the actual implementation of
and either monomaster or multimaster Ethernet-based the forwarding operation.
PROFIBUS-DP implementations on the wired medium • In cut-through forwarding, the transmission of signals
are considered, and metrics such as cycle times and alarm on the output segment can start before the packet has
delays are investigated. In [129], a cochannel interference been fully received by the input segment.
model appropriate for multiple overlaid and unsynchronized • In store-and-forward (S+F) forwarding, the coupling
piconets is developed and is used for deriving the packet device starts to generate signals at the output segment
error rate seen by one piconet. This error rate is used to de- at some time after the whole packet has been received.
rive probabilistic bounds on having a data/acknowledgment The origins and consequences of this forwarding delay are
transaction miss a deadline. discussed in the following sections.

A. Cut-Through Forwarding
IV. HYBRID WIRED/WIRELESS FIELDBUS SYSTEMS
Cut-through forwarding is typically the mode of operation
In a hybrid wired/wireless fieldbus system, both wired sta- used by repeaters, but some types of bridges can also be made
tions (with transceivers being attached to a cable) and wire- to work in this mode.
less stations (having wireless transceivers) should be able to 1) Origins of Forwarding Delay: Some of the factors
communicate with each other. which must be taken into consideration when designing
In master–slave protocols, for example, a wired (wireless) cut-through devices include the following [136].
master should be able to send a request to a wireless (wired) • The media can have different frame format require-
slave and get a response back. Often, there are legacy wired ments. In the IEEE 802.11 DSSS physical layer, for
fieldbus systems running in a plant. The addition of wireless example, all frames are prepended with a 128- s-long
stations should not require any modifications to the protocols preamble and a 64- s-long physical header, indepen-
and applications running on these wired stations. dent of any MAC protocol. In the PROFIBUS RS-485
The wireless interconnection of wired stations by means physical layer, there is no preamble and no physical
of encapsulating bridges has been considered in [132]. The layer header. When a packet is to be forwarded from
idea behind such an interconnection is to cut the cabling of a PROFIBUS segment to an IEEE 802.11 DSSS wire-
a wired fieldbus system at some point and insert a wireless less segment, the data bits that are received during pre-
link between the two ends (cable replacement). This cable amble generation time must be buffered.
replacement is useful, for example, for placing a subset of a • The medium speeds may differ. Because of this dif-
fully wired network on a mobile system, such as on a portal ference, buffering is needed in both forwarding direc-
crane or a guided vehicle. Such approaches are not discussed tions. In the case of forwarding from a faster medium
further in this paper. to a slower one, data bits arrive faster than they can
A segment is defined as a set of stations that are attached to be output, and would get lost without buffering. The
a common medium, run the same protocols, agree on trans- case of forwarding from a slower medium to a faster
mission parameters, and thus are able to directly commu- one is different. Most media types require that bits be
nicate. A wired segment consists entirely of wired stations, transmitted without gaps between them. Because of
while a wireless segment consists entirely of wireless sta- this requirement, the coupling device cannot immedi-
tions. When two wireless segments overlap in space, they ately start transmission on its output medium once the
need to be separated by some means, for example, by the use reception of the packet on its input medium has begun.
of nonoverlapping frequencies. Instead, the coupling device has to buffer the incoming

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1143

 bits from the slower medium and start packet transmis- transmission rate fixed at 1.5 Mb/s, the wired-to-wireless
sion to the faster medium at a later time. In the op- forwarding delay decreases as the transmission rate on the
timal case, the time at which this transmission begins wireless side increases. In the reverse direction, however,
is chosen in such a way that the last bit from the slow the wireless-to-wired forwarding delay increases as the
medium will have been received immediately before wireless transmission rate increases. In both cases, the rate
its time for transmission to the fast medium has come. of decrease/increase depends on the packet size. Consider a
• Different systems can require different numbers of bits scenario where two wired stations, both with a 1.5-Mb/s data
to be transmitted when a byte of data is to be sent. In rate, communicate over a single wireless hop (thus having
IEEE 802.11, transmission of a byte requires nothing one wired-to-wireless conversion and one wireless-to-wired
more than transmitting the eight bits of data contained conversion) [136]. In this scenario, the wireless medium
in the byte. In PROFIBUS with an RS-485 medium, on should have a transmission rate between 1.1 and 1.2 Mb/s in
the other hand, three additional bits must be sent along order to achieve a close-to-optimal forwarding delay for all
with each byte so that serial communication can take packet sizes under consideration. Outside this range, larger
place. packets are subject to high forwarding delays.
The forwarding delay incurred by a coupling device depends Cut-through forwarding schemes typically have the
on both the actual combination of these factors as well as on smallest forwarding delays in comparison with S+F or
a number of other parameters, such as the processing delay gateway-based approaches.
or packet size.
To illustrate the influence of these factors, consider for- B. S+F Forwarding
warding a packet from a slower RS-485 PROFIBUS phys- In S+F systems, forwarding devices first receive an entire
ical layer with 1.5-Mb/s data rate (no preambles, no physical packet on the input segment before forwarding it to the
layer header) to a faster 2-Mb/s IEEE 802.11 DSSS phys- output segments. The minimum forwarding delay is thus
ical layer with a total physical layer overhead of given by the time needed to send the packet on the output
s. The PROFIBUS uses 11 b for a byte, the DSSS only segment, plus any additional processing time required by the
eight. When the PROFIBUS packet has a size of B forwarder. When using the numbers from the example given
(110 bits), it remains on the PROFIBUS medium for approx- above, the small packet with B has a forwarding
imately s. Once the coupling device delay of 232 s, and the large packet with B has a
receives the first bit of the transmitted packet, it can start for- forwarding delay of 1.212 ms.
warding the first bit immediately. Due to the physical layer If the forwarding device is a bridge, it might be necessary
overhead and the transmission time needed for 10 8 b at to contend with other stations on the output segment for the
2 Mb/s, the packet leaves the forwarder after 232 s. The right to transmit. When the output segment is a token-passing
forwarding delay is, therefore, s, more than twice ring, for example, the forwarder must receive a token before
the PROFIBUS packet duration. In contrast, when the packet it can transmit the packet any further. This kind of delay is
has a size of B, it takes 1.87 ms on the PROFIBUS, referred to as medium access delay. Frequently, the medium
and only 1.212 ms on the wireless side. If the forwarding access delay is variable. In this case, the forwarding delay
device is able to determine the packet length soon after is also variable, even when packets of the same size are the
reception of the packet on the wired side has begun, it will only ones used.
be able to schedule the start of the packet transmission on The forwarding delays in S+F approaches tend to be,
the wireless side in an optimized way. Specifically, the cou- therefore, larger than for cut-through approaches. The com-
pling device can schedule the start of packet transmission putation of any upper bounds requires knowledge of the
such that the last bit of the packet will have been received upper bounds for the medium access delay.
from the wired medium immediately before it is to be trans-
mitted on the wireless medium. In doing so, the forwarding
C. Consequences of Forwarding Delay and Speed
delay amounts to nothing more than the transmission time
Differences
of a single bit on the wireless medium, namely, 0.5 s. In
this scenario, forwarding works well for large packets, but The presence of forwarding delay and different medium
not so well for small packets. Since fieldbus systems tend to speeds has some important consequences.
primarily use smaller packets, the forwarding delays can be- 1) Handling of Link-Layer Timeouts: In systems using
come quite a serious problem. master–slave communication, a master unit sends a request
A closer investigation of the forwarding delay with packet to a slave unit and expects the response within a cer-
cut-through forwarding is presented in [136] for forwarding tain amount of time (link-layer timeout, or timeout for short).
that occurs between an RS485-based PROFIBUS and In the presence of forwarding devices, such a request–re-
a physical layer developed within the European union sponse exchange takes more time than would be necessary
R-Fieldbus project [40], [137]. This physical layer has a on a single medium. There are two options for setting these
shorter physical overhead than the IEEE 802.11 DSSS. timeouts.
The results, however, can be transferred to other settings • To set it large enough to accommodate the worst case
as well. One interesting result is that the forwarding delay round-trip time between a master and slaves, taking
depends on the forwarding direction. Keeping the wired into account all forwarding delays. The drawback of

1144 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

Fig. 3. Example for queueing delay induced by coupling devices.

this approach is the reduced responsiveness to trans- (but applicable to other systems as well) is the insertion of
mission errors. The detection of a missing response extra idle times between requests [139], [140]. With refer-
(even from a slave within the same segment) takes ence to Fig. 3, the distributor, after receiving response one,
much longer than it would if there were just a single waits for an additional time before issuing request two.12
segment and short timeouts. Due to the probability of 3) Relative Temporal Consistency: Some systems rely on
transmission errors on wireless media, this delay can broadcast packets to achieve relative temporal consistency.
become critical if wireless links are involved. The communication between a set of sensors sampling the
• Select the timeout to be the same as the timeout for physical environment, for example, requires such a consis-
single-segment networks without any forwarders. tency (see Section II-B1). If the sensors were to belong to
When a request issued by station is addressed to different segments, however, the temporal consistency could
station located in a distant segment, this timeout be jeopardized by the fact that the forwarding delay intro-
will be too small to accommodate for the round-trip duces additional jitter between the sampling instants.
time. One option to deal with this problem is to avoid One solution would be to keep all stations synchronized
letting retransmit its request packet. The forwarder to a common time reference and instruct them to sample at
, closest to , responds to ’s request with a special an explicitly specified time. The price to pay, however, is
“response-comes-later” packet [138]. If is located the introduction of a time synchronization protocol (see [29,
on another segment attached to , acquires ’s Ch. 8] for a survey) and the need for hardware clock circuitry
response and forward it on to . No further action of even in small sensors.
is required. If is not located in a segment attached
to , directs its request to another forwarder , D. Repeaters Versus Bridges
gets a “response-comes-later” packet, and the process Repeaters essentially convert the physical representation
continues. A drawback to this approach is that existing of signals from one type of medium to another. They usually
fieldbus protocols typically do not contain such a work in cut-through mode, but S+F is also possible. Since
mechanism. Without this mechanism, existing pro- repeaters operate essentially on the physical layer, there
tocols would have to be modified, and wireless links are no medium access delays. The segments connected by
would no longer be able to be seamlessly integrated a repeater run the same MAC protocol and form a single
into them. network. The protocol stack of the stations does not need to
2) Queueing Effects: The presence of preambles or be changed, except maybe for certain configuration settings
different speeds on the medium can have the effect that such as link-layer timeouts. Repeater-based solutions do,
two identical packets have different transmission times on however, expose wired stations to the errors present in the
different media. Because of this fact, queuing delays are now wireless channel.
going to be present [139], [140]. An example illustrating Bridges operate on the data link layer. Their precise oper-
these delays can be seen in Fig. 3. Assume a producer–dis- ation depends on the similarity between the layers on either
tributor–consumer system. There are two different segment side of them [134], [135]. Bridges often operate in S+F mode
types and . The distributor and the producers for and place restrictions on forwarding. As an example: 1) when
requests Req1 and Req2 are located in segment , the the input packet is corrupted (e.g., checksum failure), for-
producer for request Req3 is located in segment , and warding is suppressed, and 2) when the destination station of
consumers are spread throughout and . It is assumed a packet is known to be in the input segment, no forwarding
that both production requests and producer responses have a is necessary.13 The ability to avoid forwarding for intraseg-
duration of s on segment . On segment these same ment traffic has different benefits.
packets have a duration of s. As seen in the figure, the
forwarding delay (indicated by the dashed lines) increases 12When there are only unicast packets in a master–slave system, there is

over time. The packets the coupling device receives from no need for extra idle times when the forwarding device can filter packets
based on addresses. Referring to the example, the forwarder would not for-
that it needs to forward to will be buffered in a queue. ward the first two requests and responses to segment S .
If the distributor issues a third production request Req3 for 13To achieve this, the bridge has to learn which stations can be found
a variable produced by a station in , the request packet in which segments. Such detection can be done, for example, by snooping
experiences a queueing delay in the forwarding device. packets and exchanging information with other bridges [134]. Additional
mechanisms (spanning tree construction) are needed to avoid forwarding
Similar situations can also occur in master–slave systems loops. When the address fields of a packet are placed at its beginning (which
such as PROFIBUS. A solution developed for such systems is common), a filtering bridge can also work in cut-through mode.

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1145

 • Intrasegment traffic can be handled with the lowest F. Case Study: PROFIBUS
possible response times and small timeouts. In addi-
tion, when multiple unacknowledged packets need to The PROFIBUS is a particularly interesting example for
be transmitted in a series within the same segment, no studying hybrid network architectures, since a variety of dif-
extra idle times have to be inserted. ferent approaches have been developed for this system. Some
• Intrasegment traffic can be handled simultaneously in of these approaches were developed in the context of the
different segments, increasing the overall traffic ca- European Union R-Fieldbus project [40], [137]. In the fol-
pacity. In order to exploit this increase in capacity, the lowing sections, these approaches are briefly described.
allocation of stations to various segments should take 1) Single-Logical-Ring Solution: All wired and wireless
the anticipated traffic into account. segments are coupled by repeaters into a single broadcast do-
• As far as intrasegment traffic is concerned, errors on main. All stations run the PROFIBUS token-passing protocol
wireless media can be confined completely to a wire- and are organized into a single logical token-passing ring.
less segment. The worst case response times for this solution clearly de-
pend on the number of media boundaries that need to be
crossed [143] as well as their respective forwarding delays
E. Gateway-Based Approaches
[136], [140]. The idle times needed to avoid queueing in
Sometimes it may not be possible to use any forwarding repeaters due to different medium speeds have been inves-
device. In WorldFIP, for example, with its producer–dis- tigated in [140]. The influence on the setting of link-layer
tributor–consumer model, the producer of a data item must timeouts (called slot time in PROFIBUS) is considered in
respond to a distributor’s generation request within 70 bit [138].
periods [141], where 1 bit period is equal to the inverse of In PROFIBUS, the slot time serves not only as a
the data rate. If the distributor’s wired segment runs at a link-layer timeout, but also plays a role in the detection of
speed of 2.5 Mb/s, only 28 s are available for the request to lost tokens. When the token owner crashes, the transmis-
be answered. Asuming the producer is a wireless station in a sion medium remains silent, since no other station currently
neighboring wireless segment, two forwarding steps would has the right to initiate transmissions. To detect this, a sta-
be needed, including all the forwarding delay, preambles, tion is required to listen to the medium. If the medium is
and so forth. Going through these steps would take much idle for longer than the duration [9],
longer than 28 s when commercially available equipment station concludes token loss, generates a new token, and
is used. starts transmissions immediately. Increased slot times thus
In gateway-based approaches, different segments are typi- lead to decreased responsiveness to token losses. Assignment
cally coupled by a dedicated application running in a gateway of station addresses with small can reduce this problem.
station. This application connects to different segments, each 2) Multiple Logical Ring (MLR) Solution: As an al-
running a separate fieldbus protocol stack. The dedicated ap- ternative to the single logical ring solution adopted in the
plication interacts with the application layer services of these R-Fieldbus project, the approach to use MLRs has also been
different protocol stacks to allow communication between investigated [138], [144], [145]. The stations are grouped
segments to take place. into segments, also referred to as domains.
Consider the case where two segments and , coupled There are different kinds of coupling stations between seg-
by a gateway, use a fieldbus protocol based on a master–slave ments of different types. A brouter (bridge/router) is a master
scheme. When master (located in ) issues a request ad- station for two separate token-passing rings on two different
dressed to slave (located in ), the gateway can intercept segments. In contrast, a repeater can be used, for example,
’s request and become a master to itself. The gateway for coupling a segment consisting of only wireless slave sta-
then receives ’s response and generates a response packet tions to a wired segment that has master stations or to connect
for . Such an approach, however, requires large timeout multiple segments into a single logical ring.
values for the link-layer of master . The disadvantages of Different strategies can be applied for allocating stations to
having such a large timeout value were discussed in Sec- segments and for allocating segments to logical rings. In the
tion IV-C. domain-driven MLR approach, there is a separate logical ring
An alternative to this approach is to let the gateway for each segment. Segments are coupled through brouters. If
generate the response to the masters request from a cache, a wireless master station is not reachable or loses the token,
thus acting as a proxy. This approach is applicable in then only its own logical ring is affected, and the other rings
master–slave and in producer–(distributor)–consumer sys- remain functional. It may, however, still not be acceptable to
tems. The method presented in [142] for WorldFIP uses distort the operation of other wireless master stations in the
a wireless-to-wired gateway that serves as a central base segment of station . In the wireless-master-driven MLR ap-
station for the wireless part. The MAC protocol on the proach, each wireless master station is therefore allocated its
wireless side is based on a TDMA scheme. The base station own segment and runs its own token-passing ring. Such a seg-
is responsible for caching all process variables produced ment is coupled to wired segments through its own brouter.
by wireless stations (acting as a consumer here), and for Finally, in the domain-group-driven MLR, multiple segments
becoming a producer for these variables on the wired part. can be allocated into a single ring. The goal of the allocation
The same is done in the other direction. is to minimize interring traffic. Segments belonging to the

1146 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

same ring are coupled through repeaters, whereas brouters 3) Virtual Ring Extension Solution: A coupling archi-
are used to couple segments belonging to different rings. tecture for coupling a wireless segment to a token-passing
The MLR approach shares advantages with bridge-based PROFIBUS segment was proposed in [146]. This architec-
solutions (see Section IV-D). Specifically, token losses and ture is mainly useful where “wireless last-hop” segments are
ring instabilities induced by repeated losses of tokens are to be attached to a wired backbone segment. The coupling
confined to a single logical ring. device is bridge-like, running two different protocols on the
In the MLR solution, link-layer timeouts are configured wireless and wired sides. Specifically, on the wired side the
for good responsiveness in intraring transactions. Interring PROFIBUS token-passing protocol is used, whereas on the
transactions are handled by other means. Two proposals are wireless side a polling-based MAC is employed.
as follows. The coupling device uses cut-through forwarding for in-
• The “response-comes-later” approach (see also Sec- tersegment data and response frames. All other packets orig-
tion IV-C) has been proposed in [138]. inating from the wired segments, especially the token packet
• The proposal in [144] is based on the presence of and ring-maintenance packets, are not forwarded. Instead,
different types of responses/acknowledgment. The the coupling device acts as the wired segment on behalf of
receiver of a request has the freedom to send an ac- the wireless stations. Specifically, it pretends to accept and
knowledgment packet without any data. It is therefore transmit token frames, it participates in ring maintenance and
not necessary for the master (initiator) to immediately so forth.
retransmit this request a number of times. Instead, the 4) Application Layer Gateway Solutions: Examples of
initiator can postpone the next transmission trial to its the application layer gateway approach are described in [147]
next token holding time. and [148]. In [147], a PROFIBUS network is augmented with
Suppose that station is the initiator, station the wireless stations by means of a protocol converter. This pro-
intended responder, the neighboring brouter to , tocol converter acts as a PROFIBUS master on the wired part
and the neighboring brouter to . Station sends and as a virtual master on the wireless part. On the wire-
a request to station ’s address. Brouter intercepts less side, an application layer instance runs a polling protocol
this request and sends an empty response to station . that utilizes IP datagrams on top of IEEE 802.11 with DCF
Brouter then encapsulates ’s request packet into (virtual polling algorithm). Running this protocol effectively
a packet belonging to a specific protocol running be- eliminates the need for contention between wireless stations,
tween brouters, and directs this packet to brouter . as well as the presence of collisions between them. The pro-
Brouter decapsulates the request, sends it to sta- tocol converter puts all the packets that are destined for a cer-
tion , acquires ’s response, encapsulates ’s re- tain (wired or wireless) station into a separate message queue
sponse, and then forwards the encapsulated response to and transmits them when appropriate (defined by the length
brouter . Brouter then buffers this response and of the polling cycle and the allowable service time per sta-
sends it to station the next time it repeats the same tion).
request. This approach requires no change to station
and ’s protocol stacks, but requires the brouters
V. CONCLUSION
to perform routing procedures. Brouter must now
know that station can be found behind brouter . Selected issues related to wireless fieldbus systems have
A disadvantage of this approach is that the time needed been discussed in this paper. Wireless technologies can
for to get the final answer is not only influenced bring many benefits to industrial applications, one of them
by medium access delays (a brouter must wait for the being the ability to reduce machine setup times by avoiding
token), but also influenced by queueing delays in the cabling. The market offers mature wireless solutions, such
brouters. The queueing delays can occur because of as the IEEE 802.11 standard, the IEEE 802.15.4 standard, or
cross traffic passing through the brouters and . BT. So far, however, wireless technologies have not gained
This cross traffic could, for example, originate from widespread acceptance on the factory floor. One reason
other rings. for this lack of acceptance is the difficulty in achieving
If brouters forward without using additional protocol mech- the timely and successful transmission of packets over
anisms, deadlock scenarios can occur [140, Ch. 4]. Consider error-prone wireless channels. With the design of suitable
two logical PROFIBUS token passing rings, and , cou- protocol mechanisms and transmission schemes, along with
pled by a brouter . A master station acquires the token the careful combination of these schemes, important steps
in ring and sends a request to a slave station in . At toward increasing the acceptance of wireless technologies
almost the same time, a master station in ring captures for industrial applications can be made.
the token and sends a request to a slave in . The brouter The approach based on “hardening” the protocol stack can
cannot forward master station ’s request packet to ring benefit from relaxing user requirements and making applica-
because master station has the token and is waiting for an tions more tolerant against errors. In fact, a key observation
answer. Similarly, brouter cannot forward master station from the field of wireless sensor networks [29] is that the joint
’s request to ring . design of applications (here: controllers) and the networking

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1147

stack, along with careful cross-layer design within the net- [9] General Purpose Field Communication System, EN 50170, Volume
working stack itself, is more likely to give better results than 2: PROFIBUS, Union Technique de l’Electricité, 1996.
[10] U. Jecht, W. Stripf, and P. Wenzel, “Profibus—Open solutions for
designing each element in isolation. the world of automation,” in The Industrial Information Technology
There are many research opportunities in the fields of Handbook, R. Zurawski, Ed. Boca Raton, FL: CRC, 2005.
wireless fieldbus systems and wireless industrial commu- [11] General Purpose Field Communication System, EN 50170, Volume
3: WorldFIP, Union Technique de l’Electricité, 1996.
nications. One such opportunity involves the search for [12] J.-P. Thomesse, “The WorldFIP fieldbus,” in The Industrial Infor-
new protocol mechanisms to improve real-time capabilities. mation Technology Handbook, R. Zurawski, Ed. Boca Raton, FL:
A key component in the design and evaluation of such CRC, 2005.
[13] ISO Standard 11 898—Road Vehicle—Interchange of Digital Infor-
mechanisms is the formulation of appropriate performance mation—Controller Area Network (CAN) for High-Speed Commu-
measures, benchmark applications, and wireless channel nication, 1993.
models that have been adapted to industrial environments. [14] G. Cena and A. Valenzano, “Operating principles and features of
CAN,” in The Industrial Information Technology Handbook, R. Zu-
Another opportunity involves the assessment of the many rawski, Ed. Boca Raton, FL: CRC Press, 2005.
emerging wireless technologies (ultrawideband, MIMO [15] Information Technology—Telecommunications and Information
techniques, smart antennas, wireless ad hoc and sensor net- Exchange Between Systems—Local and Metropolitan Area Net-
works—Specific Requirements—Part 11: Wireless LAN Medium
works) from both a technological and a market perspective Access Control (MAC) and Physical Layer (PHY) Specifications,
in terms of their potential use in industrial applications. Yet 1999.
another research opportunity concerns a trend in fieldbus [16] IEEE Standard for Telecommunications and Information Exchange
Between Systems—LAN/MAN Specific Requirements—Part 11:
systems to carry multimedia and TCP traffic in addition to Wireless Medium Access Control (MAC) and Physical Layer (PHY)
control traffic. As a consequence, there is a need for wire- Specifications: High Speed Physical Layer in the 5 GHz Band, 1999.
less-adapted protocol support for these data types, which [17] IEEE Standard for Information Technology—Telecommunica-
tions and Information Exchange Between Systems—Local and
would not degrade the quality of service rendered to the Metropolitan Networks—Specific Requirements—Part 11: Wireless
control traffic. From a practical perspective, plant engineers LAN Medium Access Control (MAC) and Physical Layer (PHY)
need software tools for planning, configuration, and main- Specifications: Higher Speed Physical Layer (PHY) Extension in
the 2.4 Ghz Band, 1999.
tenance of wireless industrial networks. One component of [18] Specification of the Bluetooth System, Version 1.1, Dec. 1999.
such a software suite would need to determine the placement [19] J. C. Haartsen, “The Bluetooth radio system,” IEEE Pers. Commun.,
of wireless stations and coupling devices. An optimization vol. 7, no. 1, pp. 28–36, Feb. 2000.
[20] IEEE Standard for Information Technology—Telecommunica-
goal might be to minimize the installation costs while sat- tions and Information Exchange Between Systems—Local and
isfying the real-time requirements of individual stations. Metropolitan Area Networks—Specific Requirements—Part 15.4:
Further areas of research include security, mobility support, Wireless Medium Access Control (MAC) and Physical Layer (PHY)
Specifications for Low Rate Wireless Personal Area Networks
and the joint consideration of real-time transmission and (LR-WPAN’s), Oct. 2003.
energy efficiency. [21] E. Callaway, P. Gorday, L. Hester, J. A. Gutierrez, M. Naeve, B.
Heile, and V. Bahl, “Home networking with IEEE 802.15.4: A devel-
oping standard for low-rate wireless personal area networks,” IEEE
ACKNOWLEDGMENT Commun. Mag., vol. 40, no. 8, pp. 70–77, Aug. 2002.
[22] C. Schwaiger and T. Sauter, “Security strategies for field area net-
The authors would like to thank the reviewers and the ed- works,” in Proc. IEEE 2002 28th Annu. Conf. Industrial Electronics
itor for their helpful comments. The authors would also like Society (IECON), 2002, pp. 2915–2920.
to thank K. Klues from TKN for helping to proofread the [23] A. Treytl, T. Sauter, and C. Schwaiger, “Security measures for in-
dustrial fieldbus systems—State of the art and solutions for IP-based
paper. approaches,” in Proc. 2004 IEEE Int. Workshop Factory Communi-
cation Systems (WFCS), pp. 201–209.
[24] C. Schwaiger and T. Sauter, “A secure architecture for fieldbus/in-
REFERENCES
ternet gateways,” in Proc. 8th IEEE Int. Conf. Emerging Technolo-
[1] P. E. Rybski, S. E. Stoeter, M. Gini, D. F. Hougen, and N. P. Pa- gies and Factory Automation (ETFA), 2001, pp. 279–285.
panikolopoulos, “Performance of a distributed robotic system using [25] J. Hirai, T.-W. Kim, and A. Kawamura, “Practical study on wireless
shared communications channels,” IEEE Trans. Robot. Autom., vol. transmission of power and information for autonomous decentral-
18, no. 5, pp. 713–727, Oct. 2002. ized manufacturing system,” IEEE Trans. Ind. Electron., vol. 46, no.
[2] V. K. Kongezos and C. R. Allen, “Wireless communication between 2, pp. 349–359, Apr. 1999.
A.G.V.’s (autonomous guided vehicle) and the industrial network [26] D. Dzung, C. Apneseth, G. Scheible, and W. Zimmermann, “Wire-
C.A.N. (controller area network),” in Proc. IEEE Int. Conf. Robotics less sensor communication and powering system for real-time in-
and Automation, 2002, pp. 434–437. dustrial applications,” presented at the 2002 IEEE Workshop Factory
[3] J. A. Janet, W. J. Wiseman, R. D. Michelli, A. L. Walker, and S. Communication Systems (WFCS 2002), Västerås, Sweden.
M. Scoggins, “Using control networks for distributed robotic sys- [27] S. Roundy, D. Steingart, L. Frechette, P. Wright, and J. Rabaey,
tems,” in Proc. IEEE Int. Conf. Robotics and Automation, 1999, pp. “Power sources for wireless sensor networks,” presented at the Wire-
1138–1143. less Sensor Networks 1st Eur. Workshop (EWSN 2004), Berlin, Ger-
[4] J. R. Pimentel, Communication Networks for Manufacturing. En- many.
glewood Cliffs, NJ: Prentice-Hall, 1990. [28] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci,
[5] N. P. Mahalik, Ed., Fieldbus Technology—Industrial Network “Wireless sensor networks: A survey,” Comput. Netw., vol. 38, pp.
Standards for Real-Time Distributed Control. Berlin, Germany: 393–422, 2002.
Springer, 2003. [29] H. Karl and A. Willig, Architectures and Protocols for Wireless
[6] P. Pleineveaux and J.-D. Decotignie, “Time critical communication Sensor Networks. Chichester, U.K.: Wiley, 2005.
networks: Field buses,” IEEE Network, vol. 2, no. 3, pp. 55–63, May [30] A. J. Goldsmith and S. B. Wicker, “Design challenges for energy-
1988. constrained ad hoc wireless networks,” Wireless Commun., vol. 9,
[7] J.-D. Decotignie and P. Pleineveaux, “A survey on industrial commu- pp. 8–27, Aug. 2002.
nication networks,” Ann. Telecomm., vol. 48, no. 9, p. 435ff, 1993. [31] J. A. Stankovic, T. F. Abdelzaher, C. Lu, L. Sha, and J. C. Hou,
[8] R. Zurawski, Ed., The Industrial Information Technology Hand- “Real-time communication and coordination in embedded sensor
book. Boca Raton, FL: CRC, 2005. networks,” Proc. IEEE, vol. 91, no. 7, pp. 1002–1022, Jul. 2003.

1148 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

 [32] A. Lessard and M. Gerla, “Wireless communication in the automated [55] L. Kleinrock and F. A. Tobagi, “Packet switching in radio channels:
factory environment,” IEEE Network, vol. 2, no. 3, pp. 64–69, May Part I—Carrier sense multiple access modes and their throughput-
1988. delay characteristics,” IEEE Trans. Commun., vol. COM-23, no. 12,
[33] D. A. Roberts, “‘OLCHFA’: A distributed time-critical fieldbus,” pp. 1400–1416, Dec. 1975.
in Proc. IEE Colloq. Safety Critical Distributed Systems, 1993, pp. [56] A. Kutlu, H. Ekiz, and E. T. Powner, “Performance analysis of MAC
6/1–6/3. protocols for wireless control area network,” in Proc. Int. Symp. Par-
[34] I. Izikowitz and M. Solvie, “Industrial needs for time-critical wire- allel Architectures, Algorithms and Networks, 1996, pp. 494–499.
less communication & wireless data transmission and application [57] A. Kutlu, H. Ekiz, M. D. Baba, and E. T. Powner, “Implementation
layer support for time critical communication,” presented at the of “comb” based wireless access method for control area network,”
Euro-Arch’93 Conf., Munich, Germany. in Proc. 11th Int. Symp. Computer and Information Science, 1996,
[35] T. S. Rappaport, Wireless Communications—Principles and Prac- pp. 565–573.
tice. Upper Saddle River, NJ: Prentice-Hall, 2002. [58] F. A. Tobagi and L. Kleinrock, “Packet switching in radio chan-
[36] , “Characterization of UHF multipath radio channels in fac- nels: Part II—The hidden terminal problem in carrier sense mul-
tory buildings,” IEEE Trans. Antennas Propag., vol. 37, no. 8, pp. tiple-access and the busy-tone solution,” IEEE Trans. Commun., vol.
1058–1069, Aug. 1989. COM-23, no. 12, pp. 1417–1433, Dec. 1975.
[37] , “Indoor radio communications for factories of the future,” [59] S. Lin and D. J. Costello, Error Control Coding, 2nd ed. Engle-
IEEE Commun. Mag., vol. 27, no. 5, pp. 15–24, May 1989. wood Cliffs, NJ: Prentice-Hall, 2004.
[38] J. K. Cavers, Mobile Channel Characteristics. Boston, MA: [60] R. van Nee and R. Prasad, OFDM for Wireless Multimedia Commu-
Kluwer, 2000. nications. Boston, MA: Artech House, 2000.
[39] A. Neskovic, N. Neskovic, and G. Paunovic. (2000) Modern [61] S. Glisic and B. Vucetic, Spread Spectrum CDMA Systems for Wire-
approaches in modeling of mobile radio systems propagation less Communications. Boston, MA: Artech House, 1997.
environment. IEEE Commun. Surveys Tuts. [Online]. Available: [62] L. B. Milstein and M. K. Simon, “Spread spectrum communica-
http://www.comsoc.org/livepubs/surveys tions,” in The Communications Handbook, J. D. Gibson, Ed. Boca
[40] J. Hähniche and L. Rauchhaupt, “Radio communication in au- Raton, FL: CRC/IEEE Press, 1996, pp. 199–212.
tomation systems: The R-fieldbus approach,” in Proc. 2000 IEEE [63] K.-S. Tang, K.-F. Man, and S. Kwong, “Wireless communication
Int. Workshop Factory Communication Systems (WFCS 2000), pp. network design in IC factory,” IEEE Trans. Ind. Electron., vol. 48,
319–326. no. 2, pp. 452–459, Apr. 2001.
[41] C. Koulamas, A. Lekkas, G. Papandopoulos, G. Kalivas, and S. [64] D. Stamatelos and A. Ephremides, “Spectral efficiency and optimal
Koubias, “Delay performance of radio physical layer technologies base placement for indoor wireless networks,” IEEE J. Sel. Areas
as candidates for wireless extensions to industrial networks,” in Commun., vol. 14, no. 4, pp. 651–661, May 1996.
Proc. IEEE Int. Conf. Emerging Technologies and Factory Automa- [65] S. Lin, D. J. Costello, and M. J. Miller, “Automatic-repeat-request
tion (ETFA ’01), pp. 133–142. error-control schemes,” IEEE Commun. Mag., vol. 22, no. 12, pp.
[42] T. S. Rappaport, S. Y. Seidel, and K. Takamizawa, “Statistical 5–17, Dec. 1984.
channel impulse response models for factory and open plan building [66] A. Paulraj, “Diversity techniques,” in The Communications Hand-
radio communication system design,” IEEE Trans. Commun., vol. book, J. D. Gibson, Ed. Boca Raton, FL: CRC/IEEE Press, 1996,
39, no. 5, pp. 794–807, May 1991. pp. 213–223.
[43] D. Hampicke, A. Richter, A. Schneider, G. Sommerkorn, R. Thomä, [67] S. M. Alamouti, “A simple transmit diversity technique for wireless
and U. Trautwein, “Characterization of the directional mobile radio communications,” IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp.
channel in industrial scenarios, based on wide-band propagation 1451–1458, Oct. 1998.
[68] A. Willig, “Exploiting redundancy concepts to increase transmission
measurements,” in Proc. IEEE Vehicular Technology Conf., 1999,
reliability in wireless industrial LANs,” IEEE Trans. Ind. Electron.,
pp. 3358–3362.
submitted for publication.
[44] B. O’Hara and A. Petrick, IEEE 802.11 Handbook—A Designer’s
[69] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, “Cooperative diver-
Companion. New York: IEEE Press, 1999.
sity in wireless networks: Efficient protocols and outage behavior,”
[45] A. Willig, M. Kubisch, C. Hoene, and A. Wolisz, “Measurements of
IEEE Trans. Inf. Theory, vol. 50, no. 12, pp. 3062–3080, Dec. 2004.
a wireless link in an industrial environment using an IEEE 802.11-
[70] H. Liu, H. Ma, M. E. Zarki, and S. Gupta, “Error control schemes
compliant physical layer,” IEEE Trans. Ind. Electron., vol. 49, no. 6,
for networks: An overview,” MONET—Mobile Netw. Appl., vol. 2,
pp. 1265–1282, Dec. 2002.
no. 2, pp. 167–182, 1997.
[46] Funbus-Projektkonsortium. (2000, Oct.) Das Verbundprojekt Draht- [71] E. Uhlemann, P.-A. Wiberg, T. M. Aulin, and L. K. Rasmussen,
lose Feldbusse im Produktionsumfeld (Funbus)—Abschlußbericht. “Deadline-dependent coding—A framework for wireless real-time
[Online]. Available: http://www.softing.de/d/NEWS/Funbusbericht. communication,” in Proc. Int. Conf. Real-Time Computing Systems
pdf and Applications, 2000, pp. 135–142.
[47] D. A. Eckhardt and P. Steenkiste, “A trace-based evaluation [72] P.-A. Wiberg and U. Bilstrup, “Wireless technology in in-
of adaptive error correction for a wireless local area network,” dustry—Applications and user scenarios,” in Proc. IEEE Int.
MONET—Mobile Networks and Applications, vol. 4, pp. 273–287, Conf. Emerging Technologies and Factory Automation (ETFA ’01),
1999. pp. 123–133.
[48] G. T. Nguyen, R. H. Katz, B. Noble, and M. Satyanarayanan, “A [73] X. Wang and M. T. Orchard, “On reducing the rate of retransmission
trace-based approach for modeling wireless channel behavior,” pre- in time-varying channels,” IEEE Trans. Commun., vol. 51, no. 6, pp.
sented at the Winter Simulation Conf., Coronado, CA, 1996. 900–910, Jun. 2003.
[49] D. Duchamp and N. Reynolds, “Measured performance of wireless [74] S. Kallel, “Analysis of a type II hybrid ARQ scheme with code com-
LAN,” presented at the 17th Conf. Local Computer Networks, Min- bining,” IEEE Trans. Commun., vol. 38, no. 8, pp. 1133–1137, Aug.
neapolis, MN, 1992. 1990.
[50] A. Willig, “Polling-based MAC protocols for improving real-time [75] E. Uhlemann, “Adaptive concatenated coding for wireless real-time
performance in a wireless PROFIBUS,” IEEE Trans. Ind. Electron., communications,” Ph.D. dissertation, School Inf. Sci., Comput.
vol. 50, no. 4, pp. 806–817, Aug. 2003. Elect. Eng., Halmstad University, Halmstad, Sweden, Sep. 2004.
[51] A. Willig and A. Wolisz, “Ring stability of the PROFIBUS token- [76] M. Elaoud and P. Ramanathan, “Adaptive use of error-correcting
passing protocol over error-prone links,” IEEE Trans. Ind. Electron., codes for real-time communication in wireless networks,” presented
vol. 48, no. 5, pp. 1025–1033, Oct. 2001. at the INFOCOM 1998, San Francisco, CA.
[52] H. ju Moon, H. S. Park, S. C. Ahn, and W. H. Kwon, “Performance [77] C. Schurgers, V. Raghunathan, and M. B. Srivastava, “Power man-
degradation of the IEEE 802.4 token bus network in a noisy environ- agement for energy-aware communication systems,” ACM Trans.
ment,” Comput. Commun., vol. 21, pp. 547–557, 1998. Embedded Comput. Syst., vol. 2, pp. 431–447, Aug. 2003.
[53] Information Processing Systems—Local Area Networks—Part 4: [78] A. Honarbacht and A. Kummert, “WSDP: Efficient, yet reliable,
Token-Passing Bus Access Method and Physical Layer Specifica- transmission of real-time sensor data over wireless networks,” pre-
tions, Aug. 1990. sented at the Wireless Sensor Networks 1st Eur. Workshop (EWSN
[54] N. Malpani, Y. Chen, N. Vaidya, and J. Welch, “Distributed token cir- 2004), Berlin, Germany.
culation in mobile ad hoc networks,” IEEE Trans. Mobile Comput., [79] “Special issue on networked control systems,” IEEE Trans. Autom.
vol. 4, no. 2, pp. 154–165, Mar.–Apr. 2005. Control, vol. 49, no. 9, pp. 1421–1603, Sep. 2004.

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1149

 [80] X. Liu and A. Goldsmith, “Wireless communication tradeoffs in dis- [105] Draft Supplement to Part 11: Wireless LAN Medium Access Control
tributed control,” in Proc. 42nd IEEE Conf. Decision and Control, (MAC) and Physical Layer (PHY) Specifications, Medium Access
2003, pp. 688–694. Control (MAC) Enhancements for Quality of Service (QoS), Nov.
[81] , “Kalman filtering with partial observation losses,” presented 2002.
at the IEEE Conf. Decision and Control, Atlantis, Paradise Island, [106] A. Grilo and M. Nunes, “Performance evaluation of IEEE 802.11e,”
Bahamas, 2004. in Proc. 2002 IEEE Int. Symp. Personal, Indoor and Mobile Radio
[82] (2005) Bluetooth. [Online]. Available: http://www.bluetooth.com Communications (PIMRC), vol. 1, pp. 511–517.
[83] J. Haartsen, “Bluetooth—The universal radio interface for ad hoc, [107] P. Garg, R. Doshi, R. Greene, M. Baker, M. Malek, and X. Cheng,
wireless connectivity,” Ericsson Rev., no. 3, pp. 110–117, 1998. “Using IEEE 802.11e MAC for QoS over wireless,” in Proc. 2003
[84] J. Bray and C. F. Sturman, Bluetooth: Connect Without Cables. Ea- IEEE Int. Performance, Computing, and Communications Conf., pp.
glewood Cliffs, NJ: Prentice-Hall, 2000. 537–542.
[85] C. Bisdikian, “An overview of the Bluetooth wireless technology,” [108] R. van Nee, G. Awater, M. Morikura, H. Takanashi, M. Webster,
IEEE Commun. Mag., vol. 39, no. 12, pp. 86–94, Dec. 2001. and K. W. Halford, “New high-rate wireless LAN standards,” IEEE
[86] Bluetooth 1.2 Core Specification, Nov. 2003. Commun. Mag., vol. 37, no. 12, pp. 82–88, Dec. 1999.
[87] G. Miklòs, A. Ràcz, Z. Turànyi, A. Valkò, and P. Johansson, “Perfor- [109] M. V. Clark, K. K. Leung, B. McNair, and Z. Kostic, “Outdoor IEEE
mance aspects of Bluetooth scatternet formation,” in Proc. 1st ACM 802.11 cellular networks: Radio link performance,” in Proc. IEEE
Int. Symp. Mobile Ad Hoc Networking and Computing, 2000, pp. Int. Conf. Commun. (ICC), vol. 1, 2002, pp. 512–516.
147–148. [110] K. K. Leung, B. McNair, L. J. Cimini, and J. H. Winters, “Outdoor
[88] A. C. V. Gummalla and J. O. Limb. (2000) Wireless medium access IEEE 802.11 cellular networks: MAC protocol design and perfor-
control protocols. IEEE Commun. Surveys Tuts. [Online]. Available: mance,” presented at the IEEE Int. Conf. Communications (ICC),
http://www.comsoc.org/pubs/surveys New York, 2002.
[89] J. Greenfkes and K. Riemen, “Code modulation with digitally con- [111] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed
trolled companding for speech transmission,” Philips Tech. Rev., pp. coordination function,” IEEE J. Sel. Areas Commun., vol. 18, no. 3,
335–353, 1970. pp. 535–547, Mar. 2000.
[90] S. Zürbes, W. Stahl, K. Matheus, and J. Haartsen, “Radio network [112] S. Sadalgi. (2000, May) A performance analysis of the basic access
performance of bluetooth,” presented at the IEEE Int. Conf. Com- IEEE 802.11 wireless LAN MAC protocol (CSMA/CA). [Online].
munication (ICC), New Orleans, LA, 2000. Available: http://paul.rutgers.edu/~sadalgi/network.pdf
[91] S. Zürbes, “Considerations on link and system throughput of blue- [113] G. Ennis, “Impact of Bluetooth on 802.11 direct sequence,” IEEE,
tooth networks,” in Proc. IEEE Int. Symp. Personal, Indoor and Mo- Tech. Rep. IEEE 802.11-98/319, Sep. 1998.
[114] J. Zyren, “Extension of Bluetooth and 802.11 direct sequence inter-
bile Radio Commun. (PIMRC), 2000, pp. 1315–1319.
ference model,” IEEE, Tech. Rep. IEEE 802.11-98/378, Nov. 1998.
[92] K. Matheus, S. Zürbes, R. Taori, and S. Magnusson, “Fundamental
[115] N. Golmie, R. van Dyck, and A. Soltanian, “Bluetooth and 802.11b
properties of ad hoc networks like Bluetooth: A radio network per-
Interference: Simulation Model and System Results,” IEEE, Tech.
spective,” presented at the IEEE Vehicular Technology Conf. (VTC),
Rep. IEEE802.15-01/195R0, Apr. 2001.
Orlando, FL, 2003.
[116] D. Fumolari, “Link performance of an embedded bluetooth personal
[93] K. Matheus and S. Magnusson, “Bluetooth radio network perfor-
area network,” presented at the IEEE Int. Conf. Communications
mance: Measurement results and simulation models,” presented at
(ICC), Helsinki, Finland, 2001.
the Int. Workshop Wireless Ad Hoc Networking (IWWAN), Oulu, [117] I. Howitt, “IEEE 802.11 and Bluetooth coexistence analysis method-
Finland, 2004. ology,” in Proc. IEEE Vehicular Technology Conf. (VTC), vol. 2,
[94] Bluetooth 2.0 Core Specification cdr., Nov. 2004. 2001, pp. 1114–1118.
[95] I. Howitt and J. A. Gutierrez, “IEEE 802.15.4 low rate—Wireless [118] I. Howitt, V. Mitter, and J. Gutierrez, “Empirical study for IEEE
personal area network coexistence issues,” in Proc. Wireless Com- 802.11 and Bluetooth interoperability,” in Proc. IEEE Vehicular
munications and Networking Conference 2003 (WCNC 2003), pp. Technology Conf. (VTC), vol. 2, 2001, pp. 1109–1113.
1481–1486. [119] H. Velayos and G. Karlsson, “Techniques to reduce IEEE 802.11b
[96] G. Lu, B. Krishnamachari, and C. S. Raghavendra, “Performance MAC layer hand-over time,” presented at the IEEE Int. Conf. Com-
evaluation of the IEEE 802.15.4 MAC for low-rate low-power wire- munications (ICC), Paris, France, 2004.
less networks,” in Proc. 2004 IEEE Int. Conf. Performance, Com- [120] J.-D. Decotignie, “Which network for which application,” in
puting, and Communications, pp. 701–706. The Industrial Information Technology Handbook, R. Zurawski,
[97] Part 11: Wireless LAN Medium Access Control (MAC) and Physical Ed. Boca Raton, FL: CRC, 2005.
Layer (PHY) Specifications, Amendment 4: Further Higher Data [121] D. Miorandi and S. Vitturi, “Performance analysis of producer/con-
Rate Extension in the 2.4 GHz Band, ANSI/IEEE Std 802.11, Jun. sumer protocols over IEEE 802.11 wireless links,” presented at the
2003. Proc. IEEE Workshop Factory Communication Systems (WFCS),
[98] International Standard ISO/IEC 8802-2:1998: Information Tech- Vienna, Austria, 2004.
nology—Telecommunications and Information Exchange Between [122] , “Analysis of master–slave protocols for real-time industrial
Systems—Local and Metropolitan Area Networks—Specific Re- communications over IEEE 802.11 WLANs,” presented at the IEEE
quirements—Part 2: Logical Link Control, 1998. Industrial Informatics Conf. (INDIN ’04), Berlin, Germany.
[99] Part 11: Wireless LAN Medium Access Control (MAC) and Physical [123] R. O. Baldwin, “Improving the real-time performance of a wireless
Layer (PHY) Specifications, Amendment 5: Spectrum and Transmit local area network,” Ph.D. dissertation, Faculty Elect. Eng., Virginia
Power Management Extensions in the 5 GHz Band in Europe, Oct. Polytechnic Inst. State Univ., , Blacksburg, Jun. 1999.
2003. [124] J. L. Sobrinho and A. S. Krishnakumar, “Real-time traffic over the
[100] K. Matheus, “Wireless local and wireless personal area network IEEE 802.11 medium access control layer,” Bell Labs Tech. J., vol.
technologies for industrial deployment,” in The Industrial Commu- 1, no. 2, pp. 172–187, 1996.
nication Technology Handbook, R. Zurawski, Ed. Boca Raton, [125] G.-S. Ahn, A. T. Campbell, A. Veres, and L.-H. Sun, “Swan: Service
FL: CRC, 2004. differentiation in stateless wireless ad hoc networks,” presented at
[101] Mobilian. (2001) Wi-Fi (802.11b) and Bluetooth: An examination of the INFOCOM 2002, New York.
coexistence approaches. [Online]. Available: http://www.mobilian. [126] (1999) Specification of the Bluetooth System. Bluetooth Consor-
com tium. [Online]. Available: http://www.bluetooth.org
[102] K. Matheus and S. Zürbes, “Co-existence of Bluetooth and IEEE [127] C. de Morais Cordeiro, D. P. Agrawal, and D. H. Sadok, “Interfer-
802.11b WLANs: Results from a radio network testbed,” in Proc. ence modeling and performance of Bluetooth MAC protocol,” IEEE
IEEE Int. Symp. Personal, Indoor and Mobile Radio Communica- Trans. Wireless Commun., vol. 2, no. 6, pp. 1240–1246, Nov. 2003.
tions (PIMRC), vol. 1, 2002, pp. 151–155. [128] U. Bilstrup and P.-A. Wiberg, “Bluetooth in industrial environment,”
[103] M. A. Visser and M. E. Zarki, “Voice and data transmission over in Proc. 2000 IEEE Workshop Factory Communication Systems
an 802.11 wireless network,” in Proc. IEEE Personal, Indoor and (WFCS 2000), pp. 239–246.
Mobile Radio Conf. (PIMRC) ’95, pp. 648–652. [129] A. El-Hoiydi and J.-D. Decotignie, “Soft deadline bounds for
[104] M. Veeraraghavan, N. Cocker, and T. Moors, “Support of voice ser- two-way transactions in Bluetooth piconets under co-channel inter-
vices in IEEE 802.11 wireless LANs,” presented at the INFOCOM ference,” in Proc. 8th IEEE Int. Conf. Emerging Technologies and
2001, Anchorage, AK. Factory Automation (ETFA), 2001, pp. 143–150.

1150 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 6, JUNE 2005

[130] D. Miorandi and S. Vitturi, “A wireless extension of PROFIBUS [148] S. Lee, K. C. Lee, M. H. Lee, and F. Harashima, “Integration of
DP based on the Bluetooth radio system,” J. Ad Hoc Netw., to be mobile vehicles for automated material handling using profibus and
published. IEEE 802.11 networks,” IEEE Trans. Ind. Electron., vol. 49, no. 3,
[131] , “Hybrid wired/wireless implementations of profibus DP: pp. 693–701, Jun. 2002.
A feasibility study based on ethernet and bluetooth,” Comput.
Commun., vol. 27, pp. 946–960, Jun. 2004.
[132] S. Cavalieri and D. Panno, “On the integration of fieldbus traffic
within IEEE 802.11 wireless LAN,” presented at the 1997 IEEE
Int. Workshop Factory Communication Systems (WFCS’97),
Barcelona, Spain. Andreas Willig (Member, IEEE) received the
[133] A. S. Tanenbaum, Computer Networks, 3rd ed. Englewood Cliffs, diploma degree in computer science from the
NJ: Prentice-Hall, 1997. University of Bremen, Bremen, Germany, in
[134] R. Perlman, Interconnections Second Edition—Bridges, Routers, 1994 and the Dr.-Ing. degree in electrical engi-
Switches and Internetworking Protocols. Reading, MA: Ad- neering from the Technical University Berlin,
dison-Wesley, 1999. Berlin, Germany, in 2002.
[135] J.-D. Decotignie, “Interconnection of wireline and wireless field- He is with the Hasso-Plattner-Institute at
busses,” in The Industrial Information Technology Handbook, R. Zu- University of Potsdam, Potsdam, Germany. His
rawski, Ed. Boca Raton, FL: CRC, 2005. research interests include wireless networks,
[136] C. Koulamas, S. Koubias, and G. Papadopoulos, “Using cut-through fieldbus and real-time systems, ad hoc and sensor
forwarding to retain the real-time properties of profibus over hybrid networks, all with specific focus on protocol
wired/wireless architectures,” IEEE Trans. Ind. Electron., vol. 51, design and performance aspects.
no. 6, pp. 1208–1217, Dec. 2004.
[137] L. Rauchhaupt, “System and device architecture of a radio-based
fieldbus—The RFieldbus system,” presented at the 4th IEEE
Workshop Factory Communication Systems 2002 (WFCS 2002),
Västerås, Sweden.
[138] L. Ferreira, M. Alves, and E. Tovar, “Hybrid wired/wireless profibus
networks supported by bridges/routers,” in Proc. 2002 IEEE Work- Kirsten Matheus (Member, IEEE) received the
shop Factory Communication Systems (WFCS 2002), pp. 193–202. Diploma degree in electrical engineering from
[139] M. Alves, E. Tovar, F. Vasques, G. Hammer, and K. Röther, “Real- the Technical University of Aachen, Aachen,
time communications over hybrid wired/wireless PROFIBUS-based Germany, in 1995 and the Dr.-Ing. degree in
networks,” in Proc. 14th Euromicro Conf. Real-Time Systems, 2002, electrical engineering from the University of
pp. 142–151. Bremen, Bremen, Germany, in 1998.
[140] M. F. Alves, “Real-Time Communications Over Hybrid From 1998 to 2003, she worked for the
Wired/Wireless PROFIBUS-Based Networks,” PhD dissertation, Ericsson Eurolab Research department on the
Faculty of Engineering, Univ. Porto, Porto, Portugal, Feb. 2003. standardization of HIPERLAN/2 and Bluetooth
[141] P. Morel, A. Croisier, and J.-D. Decotignie, “Requirements for wire- technologies. She is currently working as a devel-
less extensions of a FIP fieldbus,” in Proc. IEEE Int. Conf. Emerging oper with CARMEQ GmbH, Berlin, Germany,
Technologies and Factory Automation (ETFA ’96), pp. 116–122. a Volkswagen subsidiary. Since 2003, she has worked on various aspects
[142] P. Morel and A. Croisier, “A wireless gateway for fieldbus,” in Proc. concerning wireless technologies in the automotive industry. Her research
6th Int. Symp. Personal, Indoor and Mobile Radio Communications interest is the optimization of radio network capacities in uncoordinated ad
(PIMRC ’95), vol. 1, pp. 105–109. hoc networks.
[143] M. Alves, E. Tovar, and F. Vasques, “Evaluating the duration of mes-
sage transactions in broadcast wired/wireless fieldbus networks,” in
Proc. 27th Annu. Conf. IEEE Industrial Electronics Soc. (IECON
’01), pp. 243–248.
[144] L. Ferreira, E. Tovar, and M. Alves, “Enabling-inter-domain transac-
tions in bridge-based hybrid wired/wireless PROFIBUS networks,” Adam Wolisz (Senior Member, IEEE) received
in Proc. IEEE Int. Conf. Emerging Technologies and Factory Au- the Dr.—Ing. degree from the Silesian Technical
tomation (ETFA ’03), pp. 15–22. University, Gliwice, Poland, in 1983.
[145] L. Ferreira and E. Tovar, “Timing analysis of a multiple logical ring He is currently a Professor of Electrical Engi-
wired/wireless PROFIBUS network,” in Proc. 2004 IEEE Int. Work- neering and Computer Science at the Technical
shop Factory Communication Systems (WFCS), pp. 81–90. University Berlin, Berlin, Germany, where he
[146] A. Willig, “An architecture for wireless extension of PROFIBUS,” in is directing the Telecommunication Networks
Proc. 29th Annu. Conf. IEEE Industrial Electronics Society (IECON Group (TKN). He is also a Member of the Senior
’03), vol. 3, pp. 2369–2375. Board of GMD Fokus, being especially in charge
[147] K. C. Lee and S. Lee, “Integrated network of PROFIBUS-DP of the Competence Centers GLONE and TIP.
and IEEE 802.11 wireless LAN with hard real-time requirement,” His research interests are in architectures and
in Proc. IEEE 2001 Int. Symp. Industrial Electronics, vol. 3, pp. protocols of communication networks as well as protocol engineering with
1484–1489. impact on performance and QoS aspects.

WILLIG et al.: WIRELESS TECHNOLOGY IN INDUSTRIAL NETWORKS 1151