`Kenneth C. Budka, Hong Jiang, and Steven E. Sommars
`
`Cellular digital packet data (CDPD) is the first wide area wireless data network with
`open interfaces to enter the wireless services market. Operating in the 800-MHz cel-
`lular bands, CDPD offers native support of transmission control protocol/Internet
`protocol (TCP/IP) and connectionless network protocol (CLNP). This paper presents an
`overview of CDPD’s network architecture, its network performance issues, and wire-
`less data applications that take advantage of CDPD’s seamless support of IP and
`CLNP data.
`
`Introduction
`Cellular digital packet data (CDPD) unites two
`dynamic technologies: internetworking and wireless
`communications. Designed as an overlay to typical
`800-MHz analog cellular (Advanced Mobile Phone
`System [AMPS]) networks, CDPD seamlessly supports
`network applications based on Internet protocol (IP)
`or connectionless network protocol (CLNP). Native
`support of these popular networking protocols allows
`mobile data users to run familiar applications and
`facilitates rapid development of new applications that
`take advantage of CDPD’s anytime, anywhere access
`to internets and intranets.
`The CDPD system specification1 was developed in
`the early 1990s by a consortium of U.S. cellular service
`providers, later organized as the CDPD Forum.
`Members of the CDPD Forum sought to create an
`open, nationwide wireless data service. In late 1994,
`CDPD entered commercial service.
`After an initial connection setup procedure
`called registration, CDPD mobile subscribers can
`send and receive data on demand without addi-
`tional connection setup delay. CDPD networks
`were designed to support “pay-by-the-packet” and
`“pay-by-the-byte” billing schemes. These schemes,
`shown in Table I, combined with attractive pricing
`packages, make wireless data a cost-effective option
`for applications that periodically send and receive
`relatively small amounts of data. For such applica-
`tions, the cost of making a circuit-switched wireless
`
`164 Bell LabsTechnical Journal u Summer 1997
`
`call for each transaction or of keeping a call estab-
`lished could be prohibitive.
`The CDPD network builds on the familiar cellular
`network architecture shown in Figure 1. CDPD
`includes specifications for an air link, mobility man-
`agement, accounting, and internetworking.
`
`CDPD Network Architecture
`CDPD networks consist of several major
`components:
`•Subscriber devices,
`•Infrastructure equipment provided by a cellu-
`lar operator, and
`•Network connections to internets and
`intranets, as shown in Figure 2.
`An air link provided by CDPD efficiently supports
`digital data over 800-MHz cellular frequencies, with
`each 30-kHz cellular channel capable of serving multi-
`ple CDPD subscribers simultaneously. A subscriber
`registered with the CDPD network may keep a session
`intact for many hours, regardless of the volume of
`data sent or received.
`CDPD uses different strategies for managing access
`to the forward(data flowing to the mobile user) and
`reverse(data flowing from the mobile user) channels.
`Base stations continuously transmit CDPD’s forward
`air link, sending control and data signals to mobile
`units. The mobile data intermediate system (MD-IS)
`controls the flow of data in the forward direction and
`
`1
`
`SAMSUNG 1033
`
`
`
`sends it serially in the form of link layer frames. A
`mobile data base station (MDBS) then relays the link
`layer frames over the forward air link. The CDPD digi-
`tal sense multiple access control protocol with collision
`detection (DSMA/CD), discussed later in this paper,
`governs transmissions over the reverse air link. This
`Ethernet-like medium access control (MAC) protocol
`arbitrates reverse air link contention.
`CDPD supports both unacknowledged broadcast
`and multicast services. Transmission of a single packet
`over CDPD’s broadcast service efficiently sends the
`packet to all mobiles in a geographic area. CDPD’s mul-
`ticast service transmits messages to a select group of
`mobile end systems (M-ESs) in an area. Multicast
`transmissions follow subscribers in the multicast group
`as they roam. Only one packet is sent over the air for
`all members of a multicast group registered for receipt
`of multicast data on a particular CDPD channel, saving
`air link bandwidth.
`Mobile End System
`The CDPD subscriber device, called a mobile end
`system (M-ES), takes a variety of forms: an integral
`part of a hand-held mobile telephone, a Personal
`Computer Memory Card International Association
`(PCMCIA) card installed in a laptop, or a hardened
`point-of-sale terminal.
`The major components of an M-ES include a
`modem/radio and a processor running the CDPD pro-
`tocol stack. CDPD uses Gaussian filtered minimum
`shift keying (GMSK) modulation2 and Reed-Solomon3
`forward error correction (FEC) to provide wireless data
`service in cellular’s typically harsh radio-frequency
`(RF) environment. M-ESs may be full duplex, with
`separate transmitter and receiver sections in the radio,
`or half duplex, with a transmitter and a receiver that
`share major components. Full-duplex mobile units—
`which, unlike half-duplex units, can transmit and
`receive data simultaneously—provide superior
`throughput and service under adverse RF conditions.
`As with all portable wireless devices, power manage-
`ment is an important function. CDPD mobile units
`have a range of maximum transmit power levels from
`0.6 to 3 watts and dynamically change their transmit
`power level to conserve power and reduce interfer-
`ence. An optional sleep mode conserves power by
`
`Panel 1. Abbreviations, Acronyms, and Terms
`AMPS—Advanced Mobile Phone System
`CDMA—code division multiple access
`CDPD—cellular digital packet data
`CLNP—connectionless network protocol
`CM-ES—circuit-switched mobile end system
`CMD-IS—circuit-switched mobile data interme-
`diate system
`CPU—central processing unit
`CS-CDPD—circuit-switched cellular digital
`packet data
`DSMA/CD—digital sense multiple access with
`collision detection
`FEC—forward error correction
`F-ES—fixed end system
`GMSK—Gaussian filtered minimum shift keying
`GPS—global positioning system
`HDML—handheld device markup language
`ICMP—Internet control message protocol
`IP—Internet protocol
`IS—intermediate system
`LAP-D—link access procedures for D (data)
`channel
`MAC—medium access control
`M-ES—mobile end system
`MDBS—mobile data base station
`MD-IS—mobile data intermediate system
`MDLP—mobile data link protocol
`OSI—Open System Interconnection
`PCMCIA—Personal Computer Memory Card
`International Association
`PDU—protocol data unit
`POTS—”plain old telephone service”
`PSTN—public switched telephone network
`PVC—permanent virtual circuit
`RF—radio frequency
`SNDCP—subnetwork dependent convergence
`protocol
`SREJ—selective reject
`TCP—transmission control protocol
`TDMA—time division multiple access
`TIA—Telecommunications Industry Association
`UDP—user datagram protocol
`WAN—wide area network
`
`allowing M-ESs to power down RF circuitry when
`they are not transmitting data and to periodically
`awaken to see if forward data is pending.
`The M-ES uses information broadcast over the
`forward channel to determine when and how to
`
`Bell Labs Technical Journal u Summer 1997
`
`165
`
`2
`
`
`
`Table I. High-level comparison of packet- and circuit-switched data networks.
`Billing
`Network
`Latency for initial
`scheme
`architecture
`transmission (sec)
`≤ 1
`
`Connectionless
`
`Packet data
`(for example, CDPD)
`
`Circuit data
`
`Connection-oriented
`
`5-20
`
`Typical
`pricing
`
`Based on
`data volume
`
`Based on
`connect time
`
`Typical network
`usage
`
`Short, bursty
`transactions
`
`Larger amounts
`of data
`
`Cellular base
`station with CDPD
`
`Mobile telephony
`switching office
`with CDPD switch
`
`Cellular base
`station with CDPD
`
`Packet data
`network
`
`Cellular base
`station with CDPD
`
`CDPD – Cellular digital packet data
`
`Figure 1.
`CDPD as an AMPS network overlay.
`
`search for new RF channels. RF channels may need to
`be changed when:
`• The mobile unit travels between cell sectors,
`• The current RF channel experiences fades or
`interference, or
`• Contention with the AMPS network for RF
`channels triggers channel hopping, as
`described later in this paper.
`Mobile Data Base Station
`The mobile data base station (MDBS) resides at
`the cellular site and typically covers a geographical
`area of 0.5 to 5 km in radius. On the forward channel,
`the MDBS transmits status data about its transmit
`power level, the adjacent sectors’ CDPD channels, the
`decode status of received reverse data, and the activity
`status (busy or idle) of the reverse channel. The MDBS
`
`does not maintain information about registered mobile
`units, nor does it play a direct role in M-ES mobility.
`CDPD often uses the same base station antenna as
`an AMPS cell and the same cellular voice RF plans.
`Network planners may configure the RF channels
`used by CDPD in various ways at the MDBS:
`• Dedicated. One or more 30-kHz channels are
`dedicated to CDPD in each sector of the cell
`site. This is the simplest configuration, but it
`uses the most RF spectrum.
`• Omnidirectional overlay. Each cell broadcasts an
`omnidirectional CDPD signal, overlaid on a
`sectorized AMPS cell. Fewer RF channels are
`used for CDPD, with a corresponding reduc-
`tion in CDPD capacity.
`• Channel hopping. CDPD radios are dedicated to
`
`166
`
`Bell Labs Technical Journal u Summer 1997
`
`3
`
`
`
`Private
`networks
`
`Internet
`
`Data
`link
`
`High-speed
`data link
`
`Router
`
`M-ES
`
`MD-IS
`
`MDBS at
`cell site
`
`CDPD – Cellular digital packet data
`M-ES – Mobile end system
`MDBS – Mobile data base station
`MD-IS – Mobile data intermediate system
`
`Figure 2.
`CDPD network architecture.
`
`each AMPS sector, but CDPD attempts to share
`30-kHz channels with AMPS calls, where
`AMPS calls have priority over data. As
`described later in this paper, channel hopping
`works well when AMPS blocking rates are low
`to moderate.
`At the cost of complicating RF planning, different cell
`configurations can be used in adjacent cell sites, or
`even within the same cell.
`CDPD was designed to coexist with AMPS and to
`share/reuse many components such as power, enclo-
`sures, antennas, and RF amplifiers. Cellular service
`providers have found it easy to add CDPD equipment
`to existing AMPS base stations. CDPD is being used in
`some areas, however, as a standalone data network.
`Mobile Data Intermediate System
`The mobile data intermediate system (MD-IS),
`typically located at the mobile telephony switching
`office, provides:
`• Support for CDPD mobile protocols, including
`transmission of subscriber data. To prevent
`fraud, M-ESs are authenticated as part of the
`registration process and are denied CDPD net-
`work access if they present invalid credentials.
`The MD-IS and M-ES share responsibility for
`ensuring that user data is reliably sent. Data
`flowing between the MD-IS and M-ES is
`
`Other CDPD
`service providers
`
`encrypted to protect it against eavesdropping.
`• Mobility management. Mobiles must be
`tracked as they travel between cells or between
`channels within a single cell.
`• Accounting. The MD-IS records detailed
`accounting data in a standard format.
`• An interservice provider interface. As a mobile
`roams outside its home area, the mobile unit’s
`home MD-IS must cooperate with the serving
`MD-IS. The home MD-IS must determine if
`the mobile is allowed to receive service and
`tunnel forward subscriber traffic from the
`home to serving systems.
`• Connections into wide area networks (WANs).
`The MD-IS is typically connected to one or
`more conventional routers that route sub-
`scriber traffic towards its destination.
`From a network and application viewpoint, CDPD
`is a wireless extension of the Internet. IP-based appli-
`cations usually run on CDPD networks with no modifi-
`cations. End users may find it beneficial to make some
`changes to their applications to improve performance
`and to lower network usage costs, as discussed later, in
`“CDPD Applications.” The CDPD specification supports
`conventional IP (IPv4) and Open System Intercon-
`nection (OSI) mobile devices. Support of the next ver-
`sion of IP (IPv6)4 is planned.
`
`Bell Labs Technical Journal u Summer 1997
`
`167
`
`4
`
`
`
`Home MD-IS
`
`Inter-service
`provider
`interface
`
`Network traffic
`to subscriber
`
`Network
`
`Network traffic
`from subscriber
`
`M-ES
`
`MDBS
`
`Serving MD-IS
`
`MDBS – Mobile data base station
`MD-IS – Mobile data intermediate system
`M-ES – Mobile end system
`
`Figure 3.
`CDPD mobility management.
`
`The MD-IS is the focal point for mobility manage-
`ment, either within a service area or as the subscriber
`roams between CDPD service providers. A CDPD sub-
`scriber must be registered with the CDPD network to
`receive service. After the mobile unit has found an
`appropriate CDPD channel, it registers by sending its
`network credentials, based on shared secrets. These
`credentials validate that the mobile unit is authorized
`to receive CDPD service. In the simplest case, the
`MD-IS has direct access to a subscriber database for
`this authentication.
`As a CDPD subscriber moves from one local AMPS
`sector to another, the M-ES scans for and moves to
`new channels to maintain service. The serving MD-IS
`tracks these handoffs. When the CDPD subscriber
`moves outside his or her home service region, MD-ISs
`from other service providers may be enlisted to main-
`tain CDPD service. An M-ES has a fixednetwork
`address (typically an IP address), but it may receive
`service from any CDPD service provider that operates
`with the subscriber’s home system, as shown in
`Figure 3.
`The home MD-IS of an M-ES manages roaming
`by authenticating the subscriber and sending forward
`data to the MD-IS where the subscriber is currently
`receiving CDPD service. The serving MD-IS provides
`the air link, collects detailed accounting data, and
`
`operates with the subscriber’s home MD-IS.
`
`Circuit-Switched CDPD
`During the early stages of CDPD deployment,
`some AMPS cell sites may not be equipped with
`MDBSs. In addition, CDPD’s usage-based accounting
`may not be cost-effective for applications exchanging
`large amounts of data. Both situations are addressed
`by introducing circuit-switched CDPD (CS-CDPD).5
`Although CDPD’s air link is not used, its mobility
`model and subscriber management are. CS-CDPD
`works using a dedicated connection between the sub-
`scriber and service provider. A CS-CDPD session,
`shown in Figure 4, requires a circuit-switched con-
`nection that includes, for example, a cellular data call
`using an AMPS-specific modem, a land-line public
`switched telephone network (PSTN) call using a con-
`ventional 28.8-kb/s modem, or an integrated services
`digital network (ISDN) data link.
`Today, CDPD service is available in most metro-
`politan areas. When subscribers travel to an area
`without CDPD coverage, they can use the same
`applications, the same IP network address, and typi-
`cally the same modem to establish a CS-CDPD cel-
`lular connection.
`Although CS-CDPD shares much with CDPD,
`such as accounting, encrypted subscriber traffic, and
`
`168 Bell LabsTechnical Journal u Summer 1997
`
`5
`
`
`
`Cellular
`data cell
`
`AMPS
`cell site
`
`Private
`networks
`
`Internet
`
`CM-ES
`
`Land-line
`data call
`(POTS)
`
`AMPS/POTS
`switch
`
`Modem
`bank
`
`CMD-IS
`
`AMPS – Advanced Mobile Phone System
`CDPD – Cellular digital packet data
`CMD-IS – Circuit-switched mobile data intermediate system
`CM-ES – Circuit-switched mobile end system
`POTS – “Plain old telephone service”
`
`High-speed
`data link
`
`Router
`
`Other CDPD
`service providers
`
`Figure 4.
`Circuit-switched CDPD network architecture.
`
`reliability, there are some differences. The circuit-
`switched MD-IS (CMD-IS) shares the same base proto-
`cols with CDPD, but instead of CDPD’s MAC protocol,
`it uses a standard serial data transfer protocol over the
`circuit. A circuit-switched mobile end system (CM-ES)
`may either have an active connection to the CMD-IS
`or may be in a suspended state. When the CMD-IS
`serves a suspended CM-ES, it may call the mobile unit
`to reestablish a connection if it receives traffic destined
`for that CM-ES. Alternatively, the CM-ES may initiate
`the reconnection. CS-CDPD relies on the underlying
`cellular (or wired) network to handle CM-ES mobility
`within a service region and CDPD’s support of roam-
`ing when the CM-ES moves between service areas.
`
`The CDPD Protocol Stack
`The air link is the most valuable resource in nar-
`row bandwidth wireless data networks. As such, the
`CDPD protocol stack was designed for efficient use of
`air link bandwidth. Figure 5shows a high-level pro-
`file of the CDPD protocol stack, including the network
`layer, the subnetwork-dependent convergence proto-
`col (SNDCP), the mobile data link protocol (MDLP),
`the digital sense multiple access with collision detec-
`tion protocol (DSMA-CD), and the physical layer.1
`Network layer
`CDPD networks support both Internet (IP) and
`
`OSI (CLNP) network layer protocols. M-ESs are
`assigned a unique network address by their CDPD net-
`work service provider. M-ESs are anchored at a home
`MD-IS to support roaming.
`Subnetwork-Dependent Convergence Protocol
`Network layer packets can carry fairly long headers.
`CDPD’s SNDCP layer compresses TCP/IP packets using
`Van Jacobsen6 header compression. A similar technique
`compresses CLNP’s verbose network layer packet
`header. As a result, standard 40-octet TCP/IP protocol
`headers are compressed to an average of 3 octets by
`CDPD’s SNDCP layer. CLNP headers with 57 octets are
`replaced with a 1-octet compressed header. Packet pay-
`loads can be further compressed using CDPD’s optional
`V.42bis data compression feature.
`After the header and payload are compressed,
`packets are segmented into 128-byte-long protocol
`data units (PDUs), which are encrypted for transfer
`between the M-ES and MD-IS. Encryption keys are
`exchanged between the M-ES and MD-IS using the
`Diffie-Hellman public key encryption algorithm.7 New
`keys are generated periodically.
`Mobile Data Link Protocol
`CDPD’s MDLP is similar to link access procedures
`for D (data) channel (LAP-D),8,9 a popular link proto-
`col. A number of modifications have been made to
`
`Bell LabsTechnical Journal u Summer 1997 169
`
`6
`
`
`
`IP or
`CLNP
`
`IP or
`CLNP
`
`IP or
`CLNP
`
`Transport
`
`Network
`
`Transport
`
`Network
`
`IP or
`CLNP
`
`Data
`link
`
`MDLP
`
`CDPD
`MAC
`
`Physical
`
`SNDCP
`
`MDLP
`
`Frame relay PVCs
`
`RF
`channel
`
`Subnetwork
`services
`
`Subnetwork
`services
`
`Subnetwork
`services
`
`Subnetwork
`services
`
`M-ES
`
`MDBS
`
`Frame
`relay switch
`
`MD-IS
`
`IS
`
`F-ES
`
`CDPD – Cellular digital packet data
`CLNP – Connectionless network protocol
`F-ES – Fixed end system
`IP – Internet protocol
`IS – Intermediate system
`M-ES – Mobile end system
`MAC – Medium access control
`
`MDBS – Mobile data base station
`MD-IS – Mobile data intermediate system
`MDLP – Mobile data link protocol
`PVC – Permanent virtual circuit
`RF – Radio frequency
`SNDCP – Subnetwork dependent
` convergence protocol
`
`© CDPD Forum, Inc.
`
`Figure 5.
`Profile of the CDPD protocol stack.
`
`tailor the link layer for the wireless environment, how-
`ever. Unnecessary link layer retransmissions waste air
`link bandwidth. For bandwidth efficiency, MDLP selec-
`tively requests retransmission of lost packets using
`selective reject (SREJ) packets. MDLP also supports
`multicast and broadcast addressing over its unacknowl-
`edged data service. Only one copy of a link layer frame
`is sent over the forward link to all M-ESs in a multicast
`or broadcast group, saving air link bandwidth.
`A major benefit of terminating MDLP at the
`MD-IS is efficient management of handoffs. By moni-
`toring the MDLP addresses of link layer frames
`received over each CDPD channel stream it serves, an
`MD-IS passively detects M-ES handoffs and updates
`internal routing tables. MDLP frames that may have
`been lost during a handoff are retransmitted using
`MDLP’s reliable data service.
`
`Digital Sense Multiple Access with Collision Detection
`Protocol
`Access to the CDPD reverse link is governed by
`CDPD’s DSMA/CD protocol, discussed in “Tuning
`CDPD’s Reverse Link MAC Protocol.” The protocol
`parameters share a number of similarities with the
`MAC protocol used by Ethernet. The DSMA/CD
`parameters are configurable, allowing CDPD
`service providers to tune the reverse link for
`desired performance.
`Physical layer
`As an AMPS overlay network, CDPD uses the
`same 30-kHz channels as AMPS. The physical air link
`bit stream is transmitted using GMSK2 at a raw data
`rate of 19.2 kb/s. Data sent over the air link is pro-
`tected using a (63, 47) Reed Solomon code3 and is
`transferred in a series of physical layer blocks. The
`
`170
`
`Bell Labs Technical Journal u Summer 1997
`
`7
`
`
`
`Table II. Accumulation of overhead in the CDPD protocol stack.
`
`Packet length
`
`After header and
`data compression
`
`After segmentation
`
`After framing
`
`After blocking
`
`Bytes
`
`N + h
`
`N + h
`128
`
`N + h
`128
`
`8
`
`+ (N + h)
`
`+ (N + h)
`
`8
`
`8
`
`N + h
`128
`
`N + h
`128
`
`+ (N + h)
`
`+ (N + h)
`
`420
`282
`
`385
`282
`
`Cumulative bytes for 1,000
`bytes of network layer data
`
`1,000
`
`1,008
`
`1,064
`
`Forward channel: 1,585Forward channel:
`
`Reverse channel: 1,453Reverse channel:
`
`coding scheme is well-suited to combating the burst
`errors common on the air link that result from
`Rayleigh fading and other channel impairments.
`Table IIshows the amount of overhead added by
`the CDPD protocol stack for a network layer packet
`containing a compressed header and payload a total of
`(N+h) bytes long. As Table II shows, nearly one-third
`of the 19.2 kb/s throughput of the air link physical
`layer is spent on error correction coding redundancy
`and, in the case of the forward channel, the addition
`of channel bits and collision feedback bits. Packets sent
`over the forward channel carry more overhead than
`reverse packets because of the in-band control bits sent
`over the forward channel to mediate access to the
`reverse channel.
`Typical maximum network layer throughputs
`without V.42bis enabled are roughly 13.2 kb/s
`(reverse) and 12.1 kb/s (forward). Headers of user
`datagram protocol (UDP) and Internet control message
`protocol (ICMP) packets are not compressed by
`CDPD’s SNDCP layer. Packet payloads that have been
`either encrypted or already compressed by an applica-
`tion may not be further compressed by SNDCP’s
`V.42bis algorithm. Higher maximum throughputs are
`attainable when packet headers and payloads are read-
`ily compressible.
`
`Channel Hopping
`AMPS networks with three-sectored cells and a
`reuse factor of seven—a popular configuration for
`AMPS networks in North America and other parts of
`
`the world—are typically equipped with 10 to 25 chan-
`nels per sector. To offer tolerable call-blocking proba-
`bilities, the channels in these sectors must be used
`with moderation.
`When operating in the MDBS channel-hopping
`mode, described earlier, an MDBS “borrows” an idle
`AMPS channel to transfer data to and from M-ESs. If
`an AMPS call starts using the channel being bor-
`rowed by the MDBS, the MDBS quickly moves the
`affected channel stream to another idle AMPS chan-
`nel. If no channel is available, the CDPD channel
`“blacks out” until an AMPS channel becomes idle.
`To help M-ESs relocate the CDPD channel after a
`channel hop, the MDBS periodically broadcasts
`messages to inform M-ESs of AMPS channels that
`are likely to be used after preemption by an AMPS
`call. If data applications and users can tolerate occa-
`sional blackouts, the CDPD channel-hopping fea-
`ture can squeeze additional revenue from idle
`AMPS channels for AMPS service providers.
`Operating in this parasitic mode does not guaran-
`tee that an MDBS will be able to find an idle AMPS
`channel. In sectors with heavy AMPS call loads, chan-
`nel hops and blackouts can be frequent, adversely
`affecting CDPD network performance. As a result, in
`some cells channel hopping may not offer adequate
`performance for delay-sensitive applications.
`An M/G/c/c queuing system10 can analytically
`quantify idle AMPS capacity and determine how
`AMPS call loads, AMPS holding times, and the num-
`ber of AMPS channels per sector influence the CDPD
`
`Bell LabsTechnical Journal u Summer 1997 171
`
`8
`
`
`
`10
`
`25
`20
`15
`Number of AMPS channels per sector
`
`30
`
`1% call blocking
`
`3% call blocking
`
`5% call blocking
`
`10% call blocking
`
`20% call blocking
`
`10
`
`9 8 7 6 5 4 3 2 1 0
`
`1.4
`
`1.2
`
`1.0
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`0
`
`AMPS channels per sector
`Average number of idle
`
`idle period length
`
`Normalized average AMPS channel
`
`(a)
`
`(b)
`
`10
`
`15
`20
`25
`30
`Number of AMPS channels per sector
`
`35
`
`AMPS – Advanced Mobile Phone System
`
`Figure 6.
`(a) Average number of idle AMPS channels per sector as a function of AMPS channels per sector. (b) Average duration of
`AMPS channel idle periods.
`
`where B(c,a)denotes the Erlang blocking formula for
`an M/G/c/c system with offered load a
`c
`a
`=
`
`.
`
`(2)
`
`(cid:230)Ł(cid:231) (cid:246)ł(cid:247)
`
`j
`
`!
`
`a j
`
`(,)
`Bca
`
`c
`
`=
`
`0
`
`j
`
`Under the assumption that each channel receives an
`equal fraction of the AMPS call load, we may also cal-
`culate T, the average length of time an AMPS channel
`is idle, where
`
`(,))
`Bca
`(,))
`Bca
`
`.
`
`(3)
`
`T
`
`( 1
`
`
`= -
`
`c a
`
`( 1
`a
`
`--
`
`Figure 6shows the average number of idle AMPS
`channels per sector (Figure 6a) and average idle period
`
`channel-hopping feature. The model can also help
`determine when it is appropriate to use the CDPD
`channel-hopping feature.11
`Assume that in a sector with cAMPS channels,
`AMPS calls—that is, calls that originate in the sector
`and calls that are handed off to a sector—are gener-
`ated in accordance with a Poisson process with nor-
`malized rate acalls/unit. The length of time an AMPS
`call holds onto an AMPS channel is assumed to be
`generally distributed with unit mean. If we use argu-
`ments from renewal theory, the number of idle chan-
`nels per sector, nidle, can be expressed as
`=-
`
`-(
`(,)),1
`n
`ca
`Bca
`idle
`
`(1)
`
`172 Bell LabsTechnical Journal u Summer 1997
`
`9
`
`(cid:229)
`
`
`CDPD channel
`active (Ta)
`
`CDPD channel
`blacked out (Tb)
`
`Time
`
`No idle AMPS
`channels
`available
`
`Channel hops
`
`AMPS – Advanced Mobile Phone System
`CDPD – Cellular digital packet data
`
`Figure 7.
`CDPD channel active and blackout periods.
`
`per channel (Figure 6b) for a variety of systems of
`interest. Normalization assumes AMPS channel hold-
`ing time of one unit. At AMPS call blocking rates of 3
`to 5%, which are typical call blocking rates for mature
`networks, a fairly large number of AMPS channels are
`idle, on average. Furthermore, at a typical AMPS
`channel holding time of 90 seconds, AMPS idle peri-
`ods are fairly long, implying that a parasitic data net-
`work should have ample time to detect and use idle
`AMPS channels.
`Using the CDPD channel-hopping feature does
`not come without cost. As depicted in Figure 7, peri-
`ods of time may occur when all AMPS channels are
`occupied by AMPS calls and the CDPD channel will
`not be available. If these periods are short and infre-
`quent, some data applications may be able to tolerate
`the disruptions.
`With the further assumption that AMPS channel-
`holding times are exponentially distributed, the
`Laplace transforms of the distributions of active and
`blackout periods can be calculated.12,13 For the case of
`an AMPS sector equipped with one CDPD channel
`stream, the mean length of the active periods Ta can be
`expressed as
`
`= -1
`(,)
`Bca
`(,)
`cBca
`The mean length of blackout periods Tb for the case of
`one CDPD channel stream per sector is independent of
`the AMPS call load and can be expressed as
`1
`=.
`c
`
`T
`
`a
`
`.
`
`T
`
`b
`
`(4)
`
`(5)
`
`Figure 8shows the mean length of active (Figure
`8a) and blackout (Figure 8b) periods for several sys-
`tems of interest for sectors equipped with one CDPD
`channel. (The length of a blackout period in this con-
`figuration is independent of AMPS call loads.)
`Note from Figure 8a that as AMPS call blocking
`periods increase, active period lengths shorten. As a
`result, channel streams spend a greater fraction of time
`“blacked out.” Determining whether the CDPD chan-
`nel-hopping feature provides adequate service
`depends on the delay sensitivity of the applications
`running on a CDPD network and the rates charged for
`CDPD service. For AMPS blocking rates of 5% or less
`in AMPS sectors with fewer than 25 channels, the
`CDPD channel-hopping feature may be a viable alter-
`native to an AMPS channel dedicated to CDPD. At
`higher blocking rates or when applications cannot tol-
`erate periodic blackouts, CDPD must be deployed on
`dedicated channels.
`
`Tuning the CDPD Reverse Link MAC Protocol
`The delay and throughput performance of the
`CDPD air link will dominate the performance observed
`by most applications running on a CDPD network. The
`air link will likely be the lowest throughput leg on a
`packet’s journey between an M-ES and the wireline
`network. In addition, link layer retransmissions
`between an M-ES and an MD-IS may be needed
`to recover from air link transmission errors, adding
`further delay.
`M-ESs also experience air link delay as they wait
`to transmit packets over the reverse air link. To control
`access to the reverse air link, CDPD uses a MAC proto-
`col similar to the one employed by Ethernet. CDPD’s
`reverse link MAC protocol has a number of tunable
`parameters that strongly influence the throughput and
`delay performance of the reverse link. Proper tuning
`of the reverse link is important because, during periods
`of reverse air link congestion, small, delay-sensitive
`control messages, such as those sent by an M-ES dur-
`ing registration, must still reach the MD-IS with rela-
`tively small delay.
`As Figure 9shows, the reverse air link is a slotted
`channel consisting of a series of microslots, each 60
`bits long. The forward channel carries a stream of con-
`
`Bell LabsTechnical Journal u Summer 1997 173
`
`10
`
`
`
`10
`
`25
`20
`15
`Number of AMPS channels per sector
`
`30
`
`1% call blocking
`
`3% call blocking
`
`5% call blocking
`
`10% call blocking
`
`20% call blocking
`
`10
`
`9 8 7 6 5 4 3 2 1 0
`
`0.10
`0.09
`0.08
`0.07
`0.06
`0.05
`0.04
`0.03
`0.02
`0.01
`0
`
`period duration (Ta)
`
`Average normalized active
`
`blackkout period duration* (Tb)
`
`Average normalized
`
`(a)
`
`(b)
`
`10
`
`12
`
`14
`
`16
`
`18
`
`20
`
`22
`
`24
`
`26
`
`28
`
`30
`
`Number of AMPS channels per sector
`
`* Independent of AMPS call loads.
`AMPS – Advanced Mobile Phone System
`
`Figure 8a.
`Average normalized active period duration (Ta)for sectors equipped with one CDPD channel.
`Figure 8b.
`Average CDPD channel blackout period duration (Tb)for sectors equipped with one CDPD channel.
`
`trol bits that inform M-ESs of the busy/idle status of
`each reverse air link microslot, as well as a regular bit
`pattern that allows an M-ES to easily determine
`microslot boundaries.
`M-ESs listen for idle microslots during a series of
`transmission attempts. If an M-ES determines that the
`microslot is idle, it transfers data to the MDBS in short
`bursts, as shown in Figure 10. Each reverse channel
`burst begins with a dotting sequence, which helps the
`MDBS easily detect reverse link transmissions, fol-
`lowed by a reverse channel synchronization word.
`After transmission of this 60-bit preamble, the M-ES
`
`transmits an integral number of physical layer blocks.
`Each reverse link block is 385 bits long and carries
`roughly 32 bytes of uncompressed network layer data.
`As soon as the MDBS detects a dotting sequence on
`the reverse air link, it sets the reverse channel
`busy/idle status flags to busy and signals other M-ESs
`to refrain from sending data over the reverse link
`while the burst is being transmitted. Block decode
`status bits carried over the forward channel acknowl-
`edge each physical layer block that the MDBS receives
`and correctly decodes. If the MDBS determines that
`the first block in a burst is in error, it assumes that the
`
`174 Bell LabsTechnical Journal u Summer 1997
`
`11
`
`
`
`Forward channel
`(Base station to M-ES)
`
`Microslot busy/idle and block decode
`status bits interleaved in forward bit stream
`
`1 Microslot = 60 bits
`
`IB B B B B B B B B B B B B B B B B B B B B
`
`I I I I I I I I I I I I I I I I I I I I I I I I I I
`
`Reverse channel
`(M-ES to base station)
`
`FEC
`
`Block 2
`
`Data
`
`FEC
`
`Block 1
`
`Data
`
`Dotting sequence
`and synch word
`
`M-ES receives block
`decode status for block 1
`
`FEC – Forward error correction
`M-ES – Mobile end system
`
`Figure 9.
`Slotted structure of the reverse air link.
`
`error is due to a collision and informs the M-ESs to
`stop transmitting their bursts and to reschedule their
`transmission attempts.
`Full-duplex M-ESs are capable of sending more
`than one block in a reverse link burst. The reverse link
`MAC parameter MAX_BLOCKScontrols the maximum
`number of blocks that a full-duplex M-ES may send in
`a single burst. If the M-ES receives any block in the
`burst in error, it halts transmission and retransmits the
`last link layer frame it believes the MDBS has not
`received. Half-duplex M-ESs, however, can only send
`one block at a time; as soon as they send a block over
`the reverse air link, they quickly tune to the forward
`air link to determine whether the block was correctly
`received at the MDBS.
`Because half-duplex M-ESs send only one block
`per transmission attempt, they e