`BROADBAND WIRELESS NETWORKS
`
`CDMA/HDR: A Bandwidth-Efficient
`High-Speed Wireless Data Service for
`Nomadic Users
`
`Paul Bender, Peter Black, Matthew Grob, Roberto Padovani, Nagabhushana Sindhushayana, and
`Andrew Viterbi, QUALCOMM, Incorporated
`
`ABSTRACT
`This article presents an approach to providing
`very high-data-rate downstream Internet access by
`nomadic users within the current CDMA physical
`layer architecture. Means for considerably increas-
`ing throughput by optimizing packet data proto-
`cols and by other network and coding techniques
`are presented and supported by simulations and
`laboratory measurements. The network architec-
`ture, based on Internet protocols adapted to the
`mobile environment, is described, followed by a
`brief discussion of economic considerations in
`comparison to cable and DSL services.
`
`INTRODUCTION
`
`The rapid growth and nearly universal coverage
`of industrialized nations and regions by digital
`wireless telephony gives rise to an increasing
`demand for data services as well. While current
`offerings are for data rates equivalent to those
`provided by wireline modems a decade or more
`ago, the gap is closing. Standards are already
`approved and chip sets are available for provid-
`ing data rates above 64 kb/s within this calendar
`year (and century). Just beyond this horizon,
`however, service providers are already planning
`for wireless data rates above 2 Mb/s, approach-
`ing those of wireline, digital subscriber line
`(DSL), and cable. Whether such a wireless ser-
`vice can be made technically and economically
`competitive with wireline and cable is not the
`main issue, although we shall address this
`briefly in the last section. What will drive such
`a service is the demand for rapid low-latency
`availability of Internet access to nomadic users.
`In the next section we describe the characteris-
`tics and perceived needs of this user class. We
`then proceed to explore the characteristics of
`data requirements for speed and latency, after
`which we present a technical system solution
`tailored to these requirements and the charac-
`teristics of a specific implementation as an evo-
`lution of existing CDMA base station and
`
`subscriber terminal architectures. In the final
`section we briefly discuss the economics of such
`a system deployment.
`
`CHARACTERISTICS OF
`NOMADIC USER DATA DEMAND
`In business and the professions, the individual is
`often absent from her or his normal workplace.
`To continue to be productive on the road, both in
`transit and at business or professional meetings,
`connectivity to data at one’s principal workplace
`and more broadly to other databases accessible
`through the Web is essential. Generally, members
`of this nomadic user class demand the same data
`service normally available in their home base.
`Often cited examples of the nature of such ser-
`vices are e-mail retrieval, Web browsing, ordering
`airline tickets, hotel reservations, obtaining stock
`quotes, and report retrieval, in such locations as
`airport lounges, hotel rooms, and meeting places,
`in each case without recourse to the limited or
`interface unfriendly facilities available in such
`places. In fact, one need not necessarily look
`beyond corporate boundaries; professional
`employees often spend nearly as much time in
`company conference rooms as in their own
`offices, and rarely are such rooms equipped with
`the number of ports needed to connect the major-
`ity of participants’ laptops.1
`The nature of such data traffic is decidedly
`asymmetric. A much higher forward (or down-
`link) rate is required from the access point (base
`station) than that generated by the access termi-
`nal (user terminal) in the reverse (uplink) direc-
`tion. Furthermore, just as does the fixed user,
`the nomadic user expects a response to her or
`his request which does not suffer from excessive
`latency. Our goal is to satisfy these needs with
`an evolutional approach which minimizes the
`time and cost for providing such capabilities in
`existing cellular infrastructure and with terminals
`that differ only at digital baseband from existing
`cellular and personal communications systems
`(PCS) handsets.
`
`1 The obvious alternative
`of private campus wireless
`systems will be discussed
`in the last section.
`
`70
`
`0163-6804/00/$10.00 © 2000 IEEE
`
`IEEE Communications Magazine • July 2000
`
`IPR2018-01474
`Apple Inc. EX1004 Page 1
`
`
`
`THE TECHNICAL CHALLENGE
`Little information has been gathered on the exi-
`gencies of nomadic users and the networks to
`serve them, simply because, with the exception of
`a very small percentage of low-speed data ser-
`vice, such networks have not existed. On the
`other hand, with digital cellular networks in place
`for nearly a decade and with large numbers of
`mobile users served for several years, a great deal
`is known about the characteristics of digital wire-
`less networks and their mobile users. A major
`step in the perfection of digital cellular technolo-
`gy was the development and standardization of
`code-division multiple access (CDMA) wireless
`systems and their adoption by the majority of
`North American, Korean, and Japanese carriers
`and manufacturers. Having proven its superiority
`to other access techniques, it is now being imitat-
`ed (sometimes in a modified form) by most of
`the carriers and manufacturers who were the ini-
`tial holdouts and skeptics of its viability.
`CDMA was designed for efficient reverse
`(uplink) and forward (downlink) operations. It
`was initially widely believed that the reverse
`direction in which multiple users access each
`base station, hence representing multiple sources
`of interference with one another, would be the
`capacity limiting (or bottleneck) direction. This
`assumption turned out to be incorrect; the for-
`ward (downlink) was the initial bottleneck for
`three principal reasons:
`• Interference on the reverse link enjoys the
`advantage of the law of large numbers, where-
`by the cumulative interference from multiple
`low-power transmitters tends to be statistical-
`ly stable. The forward link, on the other hand,
`suffers interference from a small number of
`other high-power base stations. This becomes
`particularly serious at the vertices of the
`(imaginary) cellular hexagon where the trans-
`mitting base station and two other interfering
`base stations are equidistant from the intend-
`ed user. This situation is relieved by soft
`handoff, where two or more base stations
`transmit to the user simultaneously.
`• But soft handoff, while greatly diminishing
`interference, which itself increases capacity,
`still overall diminishes the forward link2
`capacity because an additional CDMA car-
`rier must be assigned in the newly added
`base station. Depending on the region of
`(or criterion for) soft handoff, this can
`cause greater or lesser reduction.
`• While on the reverse link, fast and accurate
`power control of multiple users is evidently
`critical to operation and capacity realiza-
`tion, it was initially felt in producing the
`first CDMA standard, cdmaOne (IS-95-A
`[1]), that forward link power control could
`be much slower. This turned out to reduce
`forward link capacity.
`The second and third limiting causes have
`been eliminated or considerably diminished in
`the evolutionary revisions of cdmaOne (IS-95-B
`and CDMA2000). Fast power control is now
`implemented in the forward link, and the region
`of (criterion for) soft handoff has been dimin-
`ished. The first cause (sometimes called the “law
`of small numbers”), remains, however. These
`
`IEEE Communications Magazine • July 2000
`
`improvements have brought the forward (down-
`link) capacity to parity with the reverse (uplink)
`capacity. But for high-speed data, such as down-
`loading from the Internet, this is not enough.
`The downlink demand is likely to be several
`times greater than the uplink. The rest of this
`article deals with new approaches which will fur-
`ther increase the downlink capacity by a factor
`of three to four for data applications only.3
`
`THE TECHNICAL APPROACH TO
`HIGH-SPEED DATA
`Most data applications differ fundamentally
`from speech requirements in two respects
`already noted, traffic asymmetry and tolerance
`to latency. Two-way conversational speech
`requires strict adherence to symmetry; also,
`latencies above 100 ms (which corresponds to
`about 1 kb of data for most speech vocoders) are
`intolerable. For high-speed data downlinked at 1
`Mb/s, for example, 100 ms represents 100 kb or
`12.5 kbytes; furthermore, latencies of 10 s are
`hardly noticeable, and this corresponds to a
`record of 1.25 Mbytes. Thus, smoothing over a
`variety of conditions, which is always advanta-
`geous for capacity, is easily accomplished.
`All communication systems, wired as well as
`wireless, are greatly improved by a combination
`of techniques based on three principles:
`• Channel measurement
`• Channel control
`• Interference suppression and mitigation
`Our approach employs all three. First, on the
`basis of the received common pilot from each
`access point (or base station), each access termi-
`nal (subscriber terminal) can measure the
`received signal-to-noise-plus-interference ratio
`(SNR). The data rate which can be supported to
`each user is proportional to its received SNR.
`This may change continuously, especially for
`mobile users. Thus, over each user’s reverse
`(uplink) channel, the SNR or equivalently the
`supportable data rate value is transmitted to the
`base station. In fact, since typically two or more
`base stations may be simultaneously tracked, the
`user indicates the highest among its received
`SNRs and the identity of the base station from
`which it is receiving it, and this may need to be
`repeated frequently (possibly every slot4). In
`this way the downlink channel is controlled as
`well as measured. Furthermore, by selecting
`only the best base station, in terms of SNR, to
`transmit to the user, interference to users of
`other base stations is reduced. Additionally,
`since data can tolerate considerably more delay
`than voice, error-correcting coding techniques
`which involve greater delay, specifically turbo
`codes, can be employed which will operate well
`at lower Eb/N0, and hence lower SNR and high-
`er interference levels.
`Next, we show how unequal latency, for users
`of disparate SNR levels, can be used to increase
`throughput. Suppose we can separate users into
`N classes according to their SNR levels, and cor-
`responding instantaneous rate levels support-
`able. Thus, user class n can receive slots at rate
`Rn b/s, where n = 1.2..N, and suppose the rela-
`tive frequency of user packets of class n is Pn.
`
`All communication
`systems, wired as
`well as wireless,
`are greatly
`improved by a
`combination of
`techniques based
`on three princi-
`ples: channel
`measurement,
`channel control,
`and interference
`suppression and
`mitigation
`Our approach
`employs all three.
`
`2 Although not the capaci-
`ty of the reverse link,
`which soft handoff actual-
`ly increases.
`
`3 Clearly voice is funda-
`mentally a symmetric ser-
`vice, with stricter latency
`requirements, as we shall
`note below.
`
`4 In speech-oriented
`CDMA, voice frames are
`20 ms long. In the next
`section, we shall establish
`corresponding lengths for
`data, which will be called
`slots. Multiplying Rav of
`(1) by slots per second
`yields throughput in bits
`per second.
`
`71
`
`IPR2018-01474
`Apple Inc. EX1004 Page 2
`
`
`
`slot assigned to class 2. The result would be
`L1/L2 = 8 as required and
`Rav¢¢ = (8FP1R1 + FP2R2)/(FP1 + FP2)
`= 128 kb/s.
`For the general case of N classes and latency
`ratio Lmax/Lmin, it can be shown that the maxi-
`mum achievable throughput, denoted by C, is
`n
`∑
`o
`P
`n
`=
`1
`n
`
`∑ ∑
`
`(
`P L
`min
`n
`+
`1
`
`/
`
`L
`max
`
`)
`
`b / s,
`
`=
`n n
`o
`N
`
`(
`+
`
`/
`P R
`n
`n
`1
`
`)(
`
`L
`min
`
`/
`
`L
`max
`
`)
`
`=
`n n
`o
`
`=
`
`C
`
`n
`∑
`o
`/
`P R
`n
`n
`=
`1
`n
`
`+
`
`Data rate (kb/s) Packet length (bytes) FEC rate (b/sym) Modulation
`
`38.4
`
`76.8
`
`102.6
`
`153.6
`
`204.8
`
`307.2
`
`614.4
`
`921.6
`
`1228.8
`
`1843.2
`
`2457.6
`
`128
`
`128
`
`128
`
`128
`
`128
`
`128
`
`128
`
`192
`
`256
`
`384
`
`512
`
`I Table 1. Various data rates.
`
`1/4
`
`1/4
`
`1/4
`
`1/4
`
`1/4
`
`1/4
`
`1/4
`
`3/8
`
`1/2
`
`1/2
`
`1/2
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`8PSK
`
`16QAM
`
`Suppose slots are assigned one at a time succes-
`sively to each user class. Then the average rate,
`which we define as throughput, is
`N
`
` b / s.
`
`(1)
`
`=∑
`
`n
`
`P Rn n
`1
`
`=
`
`R
`av
`
`This, of course, means that lower-data-rate
`(and SNR) users will have proportionately high-
`er latency. For if B bits are to be transmitted
`altogether for each class, the number of slots
`(and hence time) required for user class n will
`be B/Rn, and hence the latency Ln is inversely
`proportional to Rn.
`Suppose, on the other hand, that we require
`all users to have essentially the same latency5
`irrespective of the Rn they can support. Then as
`each user class is served, it will be allocated a
`number of slots inversely proportional to its rate.
`Let Fn be the number of slots allocated to class
`n, where Fn = k/Rn, k being a constant. In this
`case, the average rate or throughput is
`
` b / s.
`
`(2)
`
`′ =
`R
`av
`
`=
`
`1
`
`N
`∑
`P R F
`n n n
`=
`1
`n
`N
`N
`∑
`∑
`/
`P R
`P F
`n
`n
`n n
`=
`=
`1
`1
`n
`n
`In this case, however, the latency of all user
`classes will be the same (assuming the total
`number of bits K to be large and thus ignoring
`edge effects).
`To assess the cost in throughput for equalizing
`latency, consider the extreme case of only two
`user classes, each equally probable (P1 = P2 =
`1/2) but capable of supporting very disparate rates
`R1 = 16 kb/s, R2 = 64R1 = 1,024 kb/s. Then in
`the first case, Rav = 520 kb/s. but L1/L2 = 64. In
`the second case, L1 = L2, but R¢av = 31.51 kb/s.
`To see that there is a more rational allocation
`which is less “unfair” than a latency ratio of 64,
`and still achieves a better throughput than R¢av,
`consider a compromise which guarantees that
`the highest latency is no more than, for example,
`8 times the lowest latency. Then in the second
`case, we would assign 8 slots to class 1 for every
`
`5 This is the case for voice.
`The only difference is that
`in speech, transmitter
`power levels are controlled
`to equalize received power,
`while here time, in terms
`of frames, is controlled to
`equalize energies.
`
`72
`
`(3)
`where R1 < R2 … RN and no is such that Rn £ C
`for all n £ no, while Rn > C for all n > no.
`Surprisingly, with this maximizing strategy,
`each user’s latency is either Lmax (for those for
`which Rn < C) or Lmin (for those for which Rn >
`C). To determine the maximum throughput it is
`necessary to have a histogram of the achievable
`rates for users of the wireless network in question.
`This will be discussed in the next section. Also, as
`we shall find there, practical numerology consider-
`ations may require us to deviate from this strict
`bimodal latency allocation, although the ratio
`Lmax/Lmin will remain as the principal constraint.
`
`IMPLEMENTATION OF
`HIGH-DATA-RATE CODE-DIVISION
`MULTIPLE ACCESS
`
`In the last section we discussed the key factors
`and parameters of a wireless system designed to
`optimize the transport of packet data. In the fol-
`lowing we will describe such a system design,
`beginning first with a description of the air inter-
`face, to continue in the next section with a
`description of the network architecture. The
`design leverages in many ways the lessons
`learned from the development and operation of
`CDMA IS-95 networks, but makes no compro-
`mises in optimizing the air interface for data ser-
`vices. Furthermore, a compelling economic
`argument can be made for a design that can
`reuse large portions (to be exact, all but the
`baseband signal processing elements) of compo-
`nents and designs already implemented in IS-95
`products, both in the access terminals and access
`points (APs).
`Due to the highly asymmetric nature of the
`service offered, we will focus most of our atten-
`tion on the downlink. In the IS-95 downlink, a
`multitude of low-data-rate channels are multi-
`plexed together (with transmissions made
`orthogonal in the code domain) and share the
`available base station transmitted power with
`some form of power control. This is an optimal
`choice for many low-rate channels sharing a
`common bandwidth. The situation becomes less
`optimal when a low number of high-rate users
`share the channel. The inefficiencies increase
`further when the same bandwidth is shared
`between low-rate voice and high-rate data users,
`since their requirements are vastly different, as
`discussed previously. It should be noted that
`
`IEEE Communications Magazine • July 2000
`
`N
`
`+
`
`IPR2018-01474
`Apple Inc. EX1004 Page 3
`
`
`
`increasing the bandwidth available for transmis-
`sion cannot help in this regard if the data rate of
`the users is increased proportionally as well.
`Therefore, a first fundamental design choice
`is to separate the services, that is, low-rate data
`(voice being the primary service in this category)
`from high-rate data services, by using possibly
`adjacent but nonoverlapping spectrum alloca-
`tions. To summarize, a better system is one that
`uses an IS-95 or cdma2000-1X RF carrier to
`carry voice and a separate high-data-rate (HDR)
`RF carrier to deliver high-rate packet bursts.
`With a dedicated RF carrier, the HDR down-
`link takes on a different form than that of the
`IS-95 designs. As shown in Fig. 1, the downlink
`packet transmissions are time multiplexed and
`transmitted at the full power available to the
`AP, but with data rates and slot lengths that vary
`according to the user channel conditions. Fur-
`thermore, when users’ queues are empty, the
`only transmissions from the AP are those of
`short pilot bursts and periodic transmissions of
`control information, effectively eliminating inter-
`ference from idling sectors.
`The pilot bursts provide the access terminals
`with means to accurately and rapidly estimate
`the channel conditions. Among other parame-
`ters, the access terminal estimates the received
`Ec/Nt of all resolvable multipath components
`and forms a prediction of the effective received6
`SNR. The value of the SNR is then mapped to a
`value representing the maximum data rate such
`a SNR can support for a given level of error per-
`formance. This channel state information, in the
`form of a data rate request, is then fed back to
`the AP via the reverse link data rate request
`channel (DRC) and updated as fast as every 1.67
`ms, as shown in Fig. 2. The reverse link data
`request is a 4-bit value that maps the predicted
`SNR into one of the data rate modes of Table 1.
`In addition, the access terminal requests trans-
`mission from only one sector (that with the high-
`est received SNR) among those comprising the
`active set. Here the definition of active set is
`identical to that for IS-95 systems, but unlike IS-
`95, only one sector transmits to any specific
`access terminal at any given time.
`The main coding and modulation parameters
`are summarized in Table 1.
`The forward error correcting (FEC) scheme
`employs serial concatenated coding and iterative
`decoding, with puncturing for some of the higher
`code rates [2].
`
`1.67 ms
`
`Power
`
`Transmission
`
`Idle
`
`User 1
`
`User 2
`
`User 3
`
`2 slots
`
`1 slot
`
`2 slots
`
`Time
`
`1.67 ms
`
`Pilot
`bursts
`
`User
`signature
`
`Power
`control
`
`Pilot
`
`Power
`control
`
`I Figure 1. An access point transmission diagram.
`
`Following the encoder, these traditional sig-
`nal processing steps are applied: symbol repeti-
`tion is performed on the lower-data-rate modes;
`scrambling, channel interleaving, and the appro-
`priate modulation is applied to obtain a constant
`modulation rate of 1.2288 MHz for all modes.
`The in-phase and quadrature channels are then
`each demultiplexed into 16 streams, each at 76.8
`kHz, and 16-ary orthogonal covers are applied to
`each stream. The resulting signal, obtained by
`adding the 16 data streams, is then spread by
`quadrature pseudonoise (PN) sequences, ban-
`dlimited and upconverted. The resulting RF sig-
`nal has the same characteristics as an IS-95
`signal, thus allowing the reuse of all analog and
`RF designs developed for IS-95 base stations,
`including the power amplifiers, and the receiver
`designs for subscriber terminals.
`Table 2 summarizes the SNR required to
`achieve a 1 percent packet error rate (PER).
`Note that at the lower rates this corresponds
`to Eb/N0 ª 2.5 dB, a result of using iterative
`decoding techniques on serial concatenated
`codes, while for the two highest rates, Eb/N0
`increases considerably because 8-phase shift key-
`ing (PSK) modulation and 16-quadrature ampli-
`tude modulation (QAM) are employed. These
`were obtained both by bit-exact simulation and
`
`(a)
`
`(b)
`
`(Estimate data rate)
`
`(Request data rate)
`
`(TX at requested rate)
`
`Pilot-DRC
`
`Pilot-DRC
`
`Pilot-DRC
`
`I Figure 2. A channel estimation and data request channel timing diagram: a) access terminal receive; b)
`access terminal transmit.
`
`IEEE Communications Magazine • July 2000
`
`6 Ec represents the
`received signal energy den-
`sity and Nt represents the
`total nonorthogonal single
`sided noise density. Nt
`comprises intercell inter-
`ference, thermal noise,
`and possibly nonorthogo-
`nal intracell interference.
`
`73
`
`IPR2018-01474
`Apple Inc. EX1004 Page 4
`
`
`
`corroborated by laboratory measurements with a
`complete RF link.
`At this point we are able to estimate the maxi-
`mum achievable throughput per sector as dis-
`cussed in the previous section. Figure 3a shows a
`graph of the cumulative distribution function of
`the SNR for a typical embedded sector of a large
`three-sector network deployed with a frequency
`reuse of one. In particular, the SNR values are
`those of the best serving sector and representative
`of a uniform distribution of users across the cover-
`age area. From the results of Fig. 3a and the
`
`–10
`
`–5
`
`0
`
`5
`
`10
`
`15
`
`Ec/Nt (dB)
`
`knowledge of the SNR required to support a given
`data rate (Table 1), it is straightforward to derive
`the histogram of data rates achievable in such an
`embedded sector. The result is shown in Fig. 3b
`where the SNRs used in the calculation are those
`of Table 1 with an additional 2 dB of margin to
`account for various losses. Finally, Fig. 3c shows
`the realized throughput per sector per 1.25 MHz
`of bandwidth versus the parameter Lmax/Lmin.
`Note that the throughput is doubled for a latency
`ratio Lmax/Lmin = 8.
`
`NETWORK ARCHITECTURE
`Since the radio link has been designed to pro-
`vide efficient access to packet data networks, it
`is natural to turn to the most ubiquitous packet
`data network — the Internet — when selecting
`the network architecture. Adopting Internet pro-
`tocols in the communication between the access
`terminal and the access network allows users to
`access the widest variety of information and ser-
`vices, including e-mail, private intranets, and the
`World Wide Web. Furthermore, the selection of
`Internet protocols in the design of the access
`network allows the access network equipment to
`take advantage of the ever decreasing costs and
`increasing performance of Internet equipment.
`First, we examine the communication link
`between the access terminal and the access net-
`work. Figure 4a shows the protocol stack used in
`such a link.
`In order to carry traffic between the user and
`the network, we need to select a network-layer
`protocol. We chose the Internet Protocol (IP) [3]
`because it is the network-layer protocol of the
`Internet. The Internet carries its network-layer
`protocol over a variety of transports. For exam-
`ple, asynchronous transfer mode (ATM) often
`carries Internet traffic on the Internet backbone,
`Ethernet often carries Internet traffic on local
`area networks (LANs), and the Point-to-Point
`Protocol (PPP) [4] often carries Internet traffic
`over dialup connections. We chose PPP for the
`following reasons. First, PPP is widely supported.
`Moreover, PPP allows the transport of a variety
`of network-layer protocols, supports methods for
`
`76.8
`
`102.4
`
`153.6
`
`204.8
`
`307.2
`614.4
`921.6 1228.8 1873.0 2457.0
`Data rate (kb/s)
`
`Data rate (kb/s)
`
`38.4
`
`76.8
`
`102.6
`
`153.6
`
`204.8
`
`307.2
`
`614.4
`
`921.6
`
`1228.8
`
`1843.2
`
`2457.6
`
`Ec/Nt (dB)
`
`–12.5
`
`–9.5
`
`–8.5
`
`–6.5
`
`–5.7
`
`–4.0
`
`–1.0
`
`1.3
`
`3.0
`
`7.2
`
`9.5
`
`I Table 2. SNR for a 1 percent packet error rate.
`
`1.0E+00
`
`9.0E-01
`8.0E-01
`
`7.0E-01
`
`6.0E-01
`
`5.0E-01
`4.0E-01
`
`3.0E-01
`
`2.0E-01
`1.0E-01
`
`0.0E+00
`
`(a)
`
`cdf
`
`0.3
`
`0.25
`
`0.2
`
`0.15
`
`0.1
`
`0.05
`
`0
`
`(b)
`
`Throughput (kb/s)
`
`Probability
`
`800
`
`750
`
`700
`
`650
`
`600
`550
`
`500
`
`450
`
`400
`350
`
`300
`
`1
`
`2
`
`3
`
`4
`5
`Latency ratio
`
`(c)
`
`6
`
`7
`
`8
`
`I Figure 3. a) Ec/Nt distribution for a typical embedded sector in a three-sector
`network with universal frequency reuse in each cell; b) data rate histogram;
`c) sector throughput vs. latency ratio Lmax/Lmin.
`
`74
`
`IEEE Communications Magazine • July 2000
`
`IPR2018-01474
`Apple Inc. EX1004 Page 5
`
`
`
`differing quality of service (QoS) requirements,
`and also supports methods for authentication.
`Lastly, PPP has low overhead, an important fea-
`ture for a wireless transport.
`It is well known that the Internet carries dif-
`ferent types of traffic with different QoS require-
`ments. Some traffic, such as Transmission Control
`Protocol (TCP) [5] traffic, tends to be more sensi-
`tive to errors and less sensitive to delay. Other
`traffic, such as Real-Time Transport Protocol
`(RTP) [6] traffic, tends to be more sensitive to
`delay and less sensitive to errors. In order to sup-
`port these differing QoS constraints over a single
`physical link between two Internet nodes (e.g.,
`routers or personal computers), many nodes
`insert traffic with different QoS requirements into
`different queues. Then, by servicing the queues
`based on the different QoS requirements, the
`node attempts to provide the QoS desired by the
`different types of traffic. For PPP sessions, multi-
`ple queues over a single physical link are support-
`ed using the PPP Multilink Protocol (MP) [7]. In
`this configuration each queue is carried by a dif-
`ferent PPP link. This feature allows PPP to sup-
`port differing QoS requirements. For instance, in
`the example shown in Fig. 4a the system has
`negotiated three PPP links.
`Since radio link bandwidth is a limited
`resource, we should consider protocol overhead
`when choosing a protocol that will be carried
`over the radio link. PPP has been designed to
`minimize its own protocol overhead. In addi-
`tion, it supports the compression of network-
`layer protocol headers such as TCP and
`IP/UDP/RTP header compression, further
`reducing the overhead of carrying user traffic
`over radio links.
`It is typical to operate HDR with a received
`signal-to-noise ratio that results in a physical-
`layer PER of approximately 1 percent. This
`error rate is significantly higher than the error
`rate seen on most wireline networks. Since most
`network protocols and most network applica-
`tions were designed assuming wireline error
`rates, the wireless link error rate needs to be
`reduced. The most straightforward method of
`reducing the error rate is for access terminals to
`operate at a higher signal-to-noise ratio regime.
`However, the increase in the signal-to-noise
`ratio required to reach wireline error rates
`results is a substantial decrease in overall
`throughput. A more efficient method for
`decreasing the error rate of the wireless link is
`obtained by implementing a form of automatic
`repeat request (ARQ). HDR implements a neg-
`ative acknowledgment (NACK)-based radio link
`protocol (RLP) whereby incorrectly received
`blocks of data are detected and then retransmit-
`ted. This allows PPP and the higher layers to
`operate at an error rate regime similar to that
`experienced in wireline networks.
`As shown in Fig. 4a, each PPP link may be
`carried by a separate RLP stream. In this specific
`example the system has negotiated three separate
`RLP streams to carry the three PPP links. This
`introduces the flexibility of allowing for finer con-
`trol of the QoS. For instance, depending on the
`QoS requirement, different transmit scheduling
`policies with differing priorities may be imple-
`mented on some PPP streams. Additionally,
`
`LCP
`
`SCP
`
`Signaling
`link layer
`
`PPP link
`
`RLP
`data
`
`IP
`
`Multilink PPP
`PPP link
`
`RLP
`data
`
`PPP link
`
`RLP
`data
`
`(stream 0)
`
`(stream 1)
`
`(stream 2)
`
`(stream 3)
`
`Framing layer
`Physical layer
`
`I Figure 4a. The air interface protocol stack — an example.
`
`RLPs with different effective error rates may be
`used on other PPP links. The framing layer shown
`in Fig. 4a is responsible for multiplexing the sep-
`arate RLP streams into one physical layer.
`In addition to user traffic, the HDR radio
`link must support the transport of signaling
`messages. The model for the transport of sig-
`naling streams is based on PPP. Signaling is
`partitioned into two basic types: the Link Con-
`trol Protocol (LCP) and Stream Control Proto-
`col (SCP). Similar to the PPP LCP, the LCP is
`used to negotiate radio link protocols and
`options at the start of the session and to con-
`trol the radio link during the session. For exam-
`ple, the LCP is used at the start of the session
`to negotiate the link layer authentication type
`that will be used for the duration of the session.
`Similar to the PPP Network Control Protocol
`(NCP), the SCP is used to carry stream-specific
`signaling messages. For example, SCP is used
`to transmit the RLP NACKs upon detection of
`missing RLP data.
`In the remainder of this section we discuss in
`which elements of the network the various layers
`of the protocols may be implemented. First we
`will describe the implementation on the user
`side of the air interface, to be followed by a brief
`description of the network side.
`On the user side of the air interface reside
`two basic functional elements: the access termi-
`nal and the computer. These elements may
`reside in two devices, as in the case of a wireless
`HDR modem connected to a portable computer,
`or may be combined into a single device such as
`a wireless personal digital assistant (PDA). In
`the latter case, the device must implement the
`entire protocol stack, while in the former the
`protocol stack implementation may be parti-
`tioned in two ways.
`In the first partitioning method, the access
`terminal implements the entire protocol stack.
`This partitioning is sometimes referred to as the
`network model. When using this partitioning, the
`access terminal and computer may physically be
`connected over Ethernet, through a PCMCIA
`interface, or over the Universal Serial Bus
`(USB). Figure 4b shows the layering endpoints
`of the network model.
`In the second partitioning method, the access
`terminal implements the entire protocol stack with
`the exception of PPP and everything above PPP,
`
`IEEE Communications Magazine • July 2000
`
`75
`
`IPR2018-01474
`Apple Inc. EX1004 Page 6
`
`
`
`A first
`fundamental
`design choice is
`to separate the
`services, that is,
`low-rate data
`(voice being the
`primary service in
`this category)
`from high-rate
`data services,
`by using possibly
`adjacent but
`non-overlapping
`spectrum
`allocations.
`
`Network side
`
`Transceiver
`
`Controller
`
`Network
`access
`server
`
`User side
`
`Computer
`
`Access
`terminal
`
`IP
`
`IP
`
`PPP
`
`SCP
`
`LCP
`
`Signaling framing layer
`
`RLP data
`
`Framing layer
`
`Physical layer
`
`I Figure 4b. Air interface protocol endpoints — the network model.
`
`while the computer implements PPP and all pro-
`tocols above PPP. This partitioning is sometimes
`referred to as the relay model. In the relay model,
`the access terminal and computer may be physical-
`ly connected via RS-232 or USB. Figure 4c shows
`the layering endpoints of the relay model.
`The network side of the air interface is mod-
`eled on the traditional Internet network access
`server (NAS) [8]. There are three basic function-
`al elements: the transceiver, the controller, and
`the NAS. The transceiver implements the physi-
`cal layer. The controller implements the framing
`layer, the RLP data layer, the signaling link
`layer, and all layers above the signaling link
`layer. The three functional blocks communicate
`over IP using open interfaces. The NAS imple-
`ments the PPP layer and all layers above PPP.
`Figures 4b and 4c show the layering endpoints
`for the partitioning.
`In our design of the access network, the access
`point implements the transceiver, the controller,
`and the NAS functional elements. The network
`interface implements the protocols and interfaces
`needed to connect the access point to an IP net-
`
`work and a backhaul network. Since the transceiv-
`er, controller, and NAS communicate over IP
`using open interfaces, there is no strict require-
`ment for all the elements to be located in the
`access point. For example, in a more traditional
`cellular implementation, one might choose to
`centralize the controller and the NAS.
`Only the transceiver and controller are spe-
`cific to the radio link. The NAS and network
`interface are standard equipment used by today’s
`Internet service providers (ISPs). By using an
`interface such as the widely supported Layer
`Two Tunneling Protocol (L2TP) between the
`NAS and the modem pool controller, it is possi-
`ble to use this standard ISP equipment for many
`applications.
`With the exception of other access points, the
`access point communicates with all elements in
`the access network using widely deployed Inter-
`net protocols. In addition, th