The Internet Protocol Journal - Volume 8, Number 1

Wireless Data Networks

by Kostas Pentikousis, VTT

Most IPJ readers are familiar with Wirele
ss Local-Area Networks (WLANs; see, for example, IPJ Volume 5, No. 1). Some may even be familiar with recent developments in Wireless Metropolitan-Area Networks (WMANs), such as WiMAX. Although nonproprietary WMAN technologies are still in the standardization phase, the IEEE 802.11 family of protocols has reached maturity and rendered inexpensive (and often free) WLAN access increasingly popular. Both WLANs and WMANs provide high-speed connectivity (in the order of tens of Mbps), but user mobility is restricted. In fact, it is probably more appropriate to talk about “portability” rather than “mobility” [1] when referring to WLANs and WMANs.

Wirele
ss wide-area networks (WWANs), on the other hand, allow full user mobility but at data rates typically in the order of tens of kbps. This will change to some extent when third-generation (3G) cellular networks are fully deployed. Still, 3G deployment is slower than originally anticipated, a development often attributed to the combination of high spectrum license costs, the recent economic downturn, and high equipment costs. As a result, both population and geographical coverage tend to be uneven. For example, in Finland, a forerunner in wireless communications, population coverage is well below the 35-percent level, and geographical coverage is even smaller.

This article introduces several wirele
ss network technologies, perhaps not so widely known, which deserve attention when considering how to provide mobile connectivity to field personnel, introduce machine-to-machine (M2M) communication, or deploy applications that require always-on connectivity. The approach taken in this article is a bit different from the one typically followed in the literature: We focus more on higher-level issues, the information that is essential for application developers, instead of modulation, channel coding, and other lowlevel details. Unlike WLANs and WMANs, none of the networks surveyed provide data rates in the order of tens of Mbps. Nevertheless, successful applications can be built even with stringent bandwidth limitations. For example, online gambling and several gaming applications can be served by really “thin” networks (and possibly “thick” clients).

Cellular Networks
The Global System for Mobile Communications (GSM) specifies a cellular, wide-area, circuit-switched, digital mobile phone network architecture [2]. Circuit-switched networks such as GSM and IS-95, commonly referred to as Code Division Multiple Acce
ss (CDMA) in the United States, can provide wireless data connectivity, cover a large area, and handle mobile host handovers efficiently [3]. Users can transfer data over, say, GSM, by establishing a “dialup” connection [4]. Mobile hosts can roam, even at high speeds, and remain connected throughout.

Communication is full-duplex at a radio data rate of 9.6 kbps or 14.4 kbps in GSM Phase 2+ [5]. User throughput is always smaller than the nominal radio data rate.

While the user is connected using a wirele
ss circuit-switched network, phone calls cannot be initiated or received whether data is being transferred or not. This is not much different from wire-line dialups over basic telephone service. The difference is that a dialup over a Public Switched Telephone Network (PSTN) takes up a resource, namely the wire-line local loop, which is dedicated to a single user, whereas a dialup over a cellular network such as GSM consumes a resource, the radio channel, which is shared among many users. Because of the burstiness that data traffic usually exhibits, circuit switching may lead to inefficient use of the network capacity. Establishing a GSM dialup connection usually takes several seconds, meaning that if the user has a small amount of data to send, a small e-mail message, for example, the overall experience is poor. Moreover, after the connection is established, the channel remains idle between traffic bursts and the allocated bandwidth is wasted. Packet switching is more efficient for bursty data transmission over a shared medium [6].

Another variable that favors packet-switching over circuit-switching, especially over slow wirele
ss networks, is billing. Users of circuit-switched networks are usually charged based on the duration of a connection regardless of the amount of traffic transmitted or received. On the other hand, users of packet-switched networks can be charged based solely on the amount of data transferred—not how long they remain attached to the network. In short, introducing packet switching to wireless networks can lead to better use of network resources and attract more users as data transfers become more economical.

Two-way, packet-switched WWANs permit users to roam freely indoors and outdoors, even at relatively high speeds [7]. Most WWANs employ a cellular architecture to take advantage of frequency reuse and increase capacity while covering a larger area. Furthermore, because the coverage area of a single cell is generally large (cell diameters are typically in the order of dozens of kilometers), mobile hosts do not have to go through frequent and lengthy handovers. Hosts remain connected throughout after they attach to the network, permitting users to receive and transmit data on demand without having to dial up. The following sections survey some of the most widely deployed packet-switched wirele
ss data networks.

Mobitex
Mobitex is the first digital data-only WWAN developed by Eric
sson and Swedish Telecom. Not based on IP, Mobitex was introduced in Sweden in 1986 for emergency communications [8]. It uses a cellular architecture with cell diameters of up to 30 km. Each service area can operate 10–30 channels [9] and each base station is usually allocated 1 to 4 channels. Each channel is composed of a frequency pair: different frequencies are used for the uplink and the downlink.

Communication between the base station and a single mobile host is, neverthele
ss, effectively half-duplex. Although base stations can transmit and receive simultaneously, mobile nodes are unable to do so [10]. The Mobitex Maximum Transmission Unit (MTU) is 545 bytes, with up to 512 bytes of user data. Although the system has undergone several revisions, the raw transfer rate remains only 8 kbps. Effective user throughputs range from 4 kbps (for 125-byte packets) to 4.6 kbps (for 512-byte packets) [11], and round-trip times can be up to 10 seconds.

Mobitex deals with network lapses using a store-and-forward procedure: Packets destined for a mobile node outside the network coverage area are stored while awaiting delivery. When the mobile node reconnects, the stored packets are delivered. Mobitex uses a hierarchical routing architecture that prevents local traffic from being injected into the backbone network. In other words, packets destined for a node in the range of the same base station are switched locally [8]. Besides supporting unicast addre
ssing, Mobitex allows hosts to send one packet to several recipients [10]. According to the Mobitex Association (www.mobitex.org), the technology features “true push functionality,” whereby data can be pushed to both a single mobile node and a predefined group of nodes, a feature that can be very useful when trying to send an urgent message to field personnel. And, because the mobile host does not have to keep querying for pending data, network traffic can be kept to a minimum. All these features can also significantly boost battery life.

According to the Yankee Group, despite the limited data rates, a variety of applications have been developed based on Mobitex, including: burglar and fire alarm systems; paging, interactive me
ssaging, e-mail, form-based applications, and access to databases; telemetry; credit card authorizations; field service; and fleet management. Virtually all of them require small and bursty transfers. Mobitex does not lend itself to large file transfers, e-mail with large attachments, or video transmission. In fact, file transfers of more than 20 KB used to be discouraged [8]. On the other hand, by using a slotted ALOHA [12] variation for channel access, Mobitex can provide message delivery delay guarantees and support hundreds of users within the same cell. Parsa [13] calculated that Mobitex can accommodate 2,000 users per channel, assuming two uplink and two downlink messages per hour. Other networks simply cannot provide tight delay bounds for such a large number of users. For example, the Mobile Data Magazine (No. 1, 2002) reported that a Korean operator launched real-time stock trading and horse gambling mobile applications with great commercial success, by guaranteeing delay bounds notwithstanding the low data rates.

DataTAC
DataTAC (also known as ARDIS in the United States) was developed by Motorola in the mid-1980s. DataTAC is also a non-IP based, widearea, data-only me
ssage-oriented network. A single base station can cover an area exceeding 20 km in diameter [14]. Like Mobitex, communication between the base station and a single DataTAC mobile node is half-duplex, and mobile hosts have to compete to get access to transmit and receive data.

Unlike Mobitex, DataTAC was designed to provide optimal in-building coverage, and it uses a cellular architecture that does not take advantage of frequency reuse. Instead, a single frequency is used, increasing the probability that a packet transmi
ssion is successful (because the same transmission can be picked up by more than one base station), but at the expense of network capacity [8]. Bodsky notes that the U.S. DataTAC operator formerly recommended refraining from transferring files larger than 10 KB.

Although neither Mobitex nor DataTAC provides native IP support, middleware can take care of protocol translation and allow unmodified, off-the-shelf applications to communicate. The maximum Data-TAC me
ssage size is 2048 bytes [15], but the maximum over-the-air packet size depends on the link layer. For rural areas the maximum radio data rate is 4.8 kbps, and the maximum over-the-air packet size is 256 bytes. In metropolitan areas, the radio data rate is 19.2 kbps and the maximum packet size is 512 bytes [16]; end-user throughput does not exceed 10 kbps on average. Traditionally, DataTAC was used for dispatching and law enforcement applications. The Worldwide Wireless Data Network Operators Group (www.datatac.com) reports that DataTAC networks are also used for two-way messaging, wireless email, telemetry, access to corporate databases, and package tracking by courier carriers.

CDPD
Cellular Digital Packet Data (CDPD) was designed by IBM and McCaw Cellular Communications in the early 1990s to take advantage of channels that do not carry voice traffic in the Advanced Mobile Phone Service (AMPS), the first-generation analog cellular network [17]. Data channels are allocated dynamically, sharing the network capacity with AMPS voice traffic, which is quite different from Mobitex and DataTAC. This, for example, might mean that data can be transmitted and received only when phone calls do not consume all available capacity. One could argue that CDPD considers data traffic le
ss important than voice. However, the standard allows network operators to specifically assign channels to data traffic only. In theory, deployment can be more economical than it is for other WWANs because CDPD takes advantage of existing AMPS infrastructure and does not require licensing new spectrum. Original projections anticipated that as CDPD gained popularity—and AMPS became obsolete—more CDPD dedicated channels would be allocated. With time, CDPD would have taken over the existing AMPS bandwidth, effectively becoming a data-only WWAN.

CDPD is based on a Carrier Sense Multiple Acce
ss (CSMA) variant called Digital Sense Multiple Access [14] and transparently provides IP services, constituting a great advantage. CDPD allows for an MTU of 2048 bytes. However, one has to account for the TCP/User Datagram Protocol (UDP) and IP headers that are used to encapsulate the application payload before sending it over the CDPD network and also for the fact that CDPD user data is transmitted in much smaller blocks. Although the CDPD raw data rate is 19.2 kbps, the effective throughput is in the order of 10 kbps and response times have been reported to be in the order of 4 seconds [18].

GPRS
The General Packet Radio Service (GPRS) is overlaid on a GSM network in a fashion similar to the way CDPD is embedded in AMPS: Voice and data traffic share the same bandwidth and network infrastructure [14]. In other words, GPRS is an add-on to GSM networks, and it requires certain hardware and software upgrades and introduces packet switching to a circuit-switched architecture. GSM voice traffic is oblivious to the presence of GPRS data traffic. Similar to CDPD, GPRS is designed to appear as a regular IP subnetwork both to hosts attached over the air interface and to hosts outside the GPRS network.

The GPRS standard was finalized by the European Telecommunications Standards Institute (ETSI) in late 1997 as part of GSM Phase 2+ [5]. It is regarded as a transitional technology toward 3G networks [19], and is commonly referred to as 2.5G. One of its main advantages is that the same device can be used to transmit and receive data, and initiate and accept phone calls. GPRS defines three cla
sses with respect to simultaneous usage of voice and data. Class A mobile hosts can transmit and receive voice and data at the same time. Class B hosts can transmit and receive either voice or data but not both simultaneously. Finally, class C hosts have the user manually select if the host should be attached to the GSM (voice) or GPRS (data) network. When compared to Mobitex, DataTAC, and CDPD, GRPS class A devices can have simultaneous access to a packet-switched and circuit-switched network. Of course, GSM-only devices do not have this capability either, as mentioned earlier.

GSM uses a combination of Frequency Division Multiple Acce
ss (FDMA) and Time Division Multiple Access (TDMA) for channel allocation, as explained in detail in [5]. In short, each frequency channel carries eight TDMA channels. Each of these channels is essentially a time slot in a TDMA frame. Thus, any GSM frequency channel can carry up to eight circuit-switched connections with each slot reserved for a single connection (read voice call). In GPRS, each slot is treated as a shared resource and any mobile host can use it to transmit or receive data. In addition, a mobile host can be allocated more than one of the eight available slots in the same TDMA frame. In other words, GPRS can multiplex different traffic sources in one channel and allocate several channels to the same traffic source.

GPRS defines four different channel coding schemes [20], namely CS1, CS2, CS3, and CS4, with radio data rates 8.8 kbps, 13.3 kbps, 15.6 kbps, and 21.4 kbps, respectively. CS1 is the most “conservative” (includes more error correction bits) and is used for signaling packets and when poor channel conditions prevail. CS4 is the most “optimistic” (includes minimal error correction bits), and, a
ssuming excellent channel conditions, allows operators to advertise a maximum radio data rate of 171.2 kbps per 200-kHz frequency channel (or TDMA frame).

In practice, CS4 is rarely used because it can lead to frequent retransmi
ssions of lost packets and overall network underperformance. CS3 is commonly used, providing 124.8 kbps per frequency channel. Because a mobile host can be allocated multiple slots, user throughputs can range between 40 and 60 kbps. Mobile hosts typically use an MTU of 1500 bytes.

Communication between the base station and any given mobile host is full-duplex but can be asymmetric; that is, the downlink and uplink capacities need not be the same. The GSM A
ssociation has defined 12 multislot classes for GPRS. Each class is associated with a maximum number of uplink and downlink slots that can be allocated to a single mobile host. The slot allocation is usually written as M + N, where M is the maximum number of downlink slots and N is the maximum number of uplink slots. For example, class 1 is “1 + 1” (one downlink slot plus one uplink slot); class 2 is “2 + 1”; . . . ; and class 12 is “4 + 4” (four downlink and four uplink slots). In addition, each multislot class has an active slot constraint: A mobile host cannot use more than K active slots simultaneously. Given the number of slots and the channel coding scheme, one can calculate the peak rate. For example, for a class 12 device the sum of the physical downlink and uplink rates cannot exceed 124.8 kbps, if CS3 is used. However, the active slot constraint limits this rate even further. In the case of a class 12 mobile node, K = 5, that is, only “4 + 1”, “3 + 2”, “2 + 3”, or “1 + 4” slots can be used simultaneously. See www.gsmworld.com

EDGE and Beyond
Enhanced Data for GSM Evolution (EDGE), also known as Enhanced GPRS, builds on the changes introduced by GPRS to GSM. EDGE e
ssentially increases the radio data rates by using a more efficient modulation scheme [21], namely 8-Phase Shift Keying (8-PSK) instead of the Gaussian Minimum Shift Keying (GMSK) used by both GSM and GPRS. EDGE defines nine modulation coding schemes named MCS1 to MCS9. MCS1 to MCS4 use GMSK with radio data rates similar to the four GPRS coding schemes. The real throughput improvements come from MCS6 (29.6 kbps per slot) through MCS9 (59.2 kbps per slot). The data rate usually associated with EDGE is a (shared) 384 kbps. This corresponds to using MCS7 for all 8 TDMA slots. Higher data rates are theoretically possible (up to 473 kbps using MCS9) but are not commonly deployed.

EDGE improves not only on the high end of data rates but also on the low end [22]. First, the greater diversity of coding schemes permits an EDGE network to choose the most appropriate one depending on channel conditions. Changing coding schemes is dynamic. Second, EDGE supports packet resegmentation: Packets that failed to be transmitted succe
ssfully can be resegmented and retransmitted using a more “conservative” coding scheme.

Table 1 summarizes the main high-level features for the WWANs surveyed.

Table 1: WWAN Characteristics

 

Transmit/Receive

Radio Data Rate

User Throughput

MTU

Mobitex

Half duplex

8.0 kbps

<4.6 kbps

512 B

DataTAC

Half duplex

19.2 kbps

<10 kbps

2048 B*

CDPD

Full duplex

19.2 kbps

<10 kbps

2048 B

GPRS

Full duplex

<171 kbps

40–60 kbps

1500 B

EDGE

Full duplex

<473 kbps

50–60 kbps

1500 B

* Typically 512 B

Discussion and Trends
Among the WWANs presented, Mobitex and GPRS can be singled out as the most widely deployed; they also have enjoyed significant gains in the number of users and traffic volume in recent years. The popularity of enterprise wirele
ss e-mail (due in part to the success of the Research in Motion BlackBerry devices) allowed Mobitex and DataTAC operators to revive their business models briefly. Worldwide, however, GSM dwarfs all other technologies: There are more than 1 billion GSM subscribers compared to the 1 million Mobitex users. DataTAC enjoys an even smaller user base. Even if a small percentage of GSM subscribers use GPRS and EDGE, the potential market for wireless applications is tremendous. On the other hand, subscribers who do not take advantage of GPRS or EDGE do use the inexpensive, (two-way) Short Message Service (SMS), which is built in GSM. Two-way messaging was available for many years but was certainly popularized by less affluent and younger GSM users in the late 1990s. SMS is now commonplace, and in many countries it is more popular than e-mail. Dedicated data-only networks such as Mobitex have to look elsewhere for their niche.

For some, Mobitex, let alone DataTAC and CDPD, is virtually moribund. In the United States, for example, Cingular sold its Mobitex network and is investing heavily on GRPS and EDGE. DataTAC and CDPD are phased out by service providers in the United States in favor of newer technologies. Low-speed packet radio is considered lackluster and is not popular with younger crowds. After all, narrowband WWANs had their chance and failed to attract large numbers of subscribers. Recent pricing trends, too, reveal a heavy operator push in favor of GRPS and EDGE. In Finland, for example, 100 MB over GPRS costs le
ss than 18 euros (approximately $24). Compare that to the $30–50 that 1 MB of traffic costs over Mobitex. Service and product popularity create economies of scale that cannot be ignored.

Nonethele
ss, open standards, an explicit focus on business applications with Quality-of-Service (QoS) guarantees in service response times, and narrowband M2M communication may well keep Mobitex going for years to come. Besides, bundling Mobitex with a wireless network that features fast and inexpensive connectivity, for example, WLAN or Bluetooth, might be promising: Large downloads and software updates can be done over the high-speed wireless network and critical messages can always reach the user through the WWAN.

Bundling several functions in a single handheld device is, after all, a major trend in the industry. Vendors scramble to integrate Personal Information Managers (PIMs), voice and data communications, as well as entertainment features (digital camera, games, or digital music players) in a single product. This is quite different from earlier mobile devices, which tended to be either single-purpose or tied to a particular set of applications. Even the BlackBerry devices still work, to some extent, in a closed architecture. Enterprise e-mail systems need to be supported by and integrated with BlackBerry servers in order to be acce
ssible over the WWAN. Yet, one of the main objectives in 2.5G and 3G is to allow mobile users to use standard Internet protocols on a mobile radio network at significantly higher bit rates than other systems. In particular, GPRS was designed with certain office applications in mind and can support consumer and enterprise mobile communications alike, without being tied to any given platform or application servers. I expect that functionality bundling and 2.5G and 3G WWANs will allow for more open systems and will expedite the transformation of WWAN operators from integrated application providers to wireless ISPs.

For Further Reading
[1] Charles Perkins, Mobile IP Design Principles and Practices, ISBN 0201634694, Addison-Wesley, 1998.

[2] Joachim Tisal, GSM Cellular Radio Telephony, ISBN 0471968269, John Wiley & Sons, 1998.

[3] Tero Ojanpera and Ramjee Prasad, WCDMA: Towards IP Mobility and Mobile Internet, ISBN B000066OB4, Artech House, 2001.

[4] R. Ludwig, B. Rathonyi, A. Konrad, K. Oden, and A. Joseph, “Multi-layer tracing of TCP over a Reliable Wirele
ss Link,” presented at ACM SIGMETRICS 1999.

[5] C. Bettstetter, H.-J. Vogel, and J. Eberspacher, “GSM phase 2+ General Packet Radio Service GPRS: Architecture, Protocols, and Air Interface,” IEEE Communications Surveys & Tutorials, Vol. 2(3), pp. 2–14, 1999.

[6] Larry L. Peterson and Bruce S. Davie, Computer Networks: A Systems Approach, 3rd ed., ISBN 155860832X, Morgan-Kaufmann, 2003.

[7] Rudi Bekkers and Jan Smits, Mobile Telecommunications: Standards, Regulation, and Applications, ISBN 0890068062, Artech House, 1999.

[8] Ira Brodsky, Wirele
ss: The Revolution in Personal Telecommunications, ISBN 089006717, Artech House, 1995.

[9] Nathan J. Muller, Wirele
ss Data Networking, ISBN 0890067538, Artech House, 1995.

[10] A. K. Salkintzis and C. Chamzas, “Mobile Packet Data Technology: An insight into Mobitex Architecture,” IEEE Personal Communications, Vol. 4(1), pp. 10–18, 1997.

[11] M. S. Taylor, M. Banan, W. Waung, and M. Taylor, Internetwork Mobility: The CDPD Approach, ISBN 0132096935, Prentice-Hall, 1996.

[12] Andrew S. Tanenbaum, Computer Networks, 4th ed., ISBN 0130661023, Pearson Education, 2003.

[13] K. Parsa, “The Mobitex packet-switched Radio Data System,” presented at IEEE PIMRC ’92, 1992.

[14] Sami Tabbane, Handbook of Mobile Radio Networks, ISBN 1580530095, Artech House, 2000.

[15] J. Rodriguez, W. Schollenberger, M. Anzib, and B. Widyarso, Mobile Computing: The eNetwork Wirele
ss Solution, ISBN 0738412856, IBM Redbooks, 1999.

[16] Research in Motion, “Developer’s guide for BlackBerry and RIM Wirele
ss Handhelds—Radio API (DataTAC) Version 2.1,” 2001.

[17] John Agosta and Travis Ru
ssel, CDPD: Cellular Digital Packet Data Standards and Technology, ISBN 0070006008, McGraw-Hill, 1996.

[18] P. Sinha, N. Venkitaraman, T. Nandagopal, R. Sivakumar, and V. Bharghavan, “A Wirele
ss Transmission Control Protocol for CDPD,” presented at IEEE WCNC ’99, 1999.

[19] A. K. Salkintzis, “A survey of mobile data networks,” IEEE Communications Surveys & Tutorials, Vol. 2(3), pp. 2–18, 1999.

[20] L. F. Chang, “Wirele
ss Internet—Networking Aspect,” in Wireless Communication Technologies, New Multimedia Systems, N. Morinaga, R. Kuhno, and S. Sampei, Eds., Kluwer Academic Publishers, 2000, pp. 215–244.

[21] Behrouz A. Forouzan, Data Communications and Networking, 2nd ed. Update, ISBN 0072822945, McGraw-Hill, 2002.

[22] Alexander J. Huber and Josef F. Huber, UMTS and Mobile Computing, ISBN B000089CJ3, Artech House, 2002.

KOSTAS PENTIKOUSIS, PhD, studied computer science at Aristotle University of The
ssaloniki and Stony Brook University. He is an ERCIM Fellow at VTT, The Technical Research Center of Finland, and currently resides in Oulu, Finland. For more about his research and publications visit: www.cs.stonybrook.edu/~kostas. The best way to reach him is via skype. E-mail: kostas@cs.sunysb.edu