LYT-HISTORY-May 4/21/11 10:27 AM Page 30
`
`HISTORY OF COMMUNICATIONS
`EDITED BY STEVE WEINSTEIN
`
`LIGHTING UP COPPER
`JOHN M. CIOFFI, STANFORD UNIVERSITY AND ASSIA INC.
`
`ABSTRACT
`
`This invited “History of Communi-
`cations” paper provides a perspective
`on the many contributions and achieve-
`ments to the science of high-speed
`transmission on telephone-line copper
`twisted-pair. This perspective relies
`largely on the author’s 30-year academ-
`ic and industrial DSL working experi-
`ence, but nevertheless attempts to
`include key events and mention key
`individuals in the steady march from
`the kilobits/second of voiceband
`modems to the Gigabits/second of
`today’s copper connections over those
`three-plus decades. Digital Subscriber
`Lines (DSL) and its ancestors are
`emphasized, while some Ethernet con-
`tributions to DSL are also cited.
`
`INTRODUCTION
`There are today more than 1.3 billion
`copper phone-line connections upon
`which the modern world of telecommuni-
`cations inexorably relies, a growing 1/3 of
`them now using DSL. These DSL num-
`bers still steadily increase each year by far
`larger amounts than does use of optical
`fiber, as Fig. 1 illustrates for broadband
`connections on three media. (Indeed
`about 3/4 of the lower FTTx (fiber to the
`x = node/cabinet/building) numbers are
`actually hybrid fiber and DSL connec-
`tions, but are counted inexplicably by the
`source only as “fiber.”) Most modern
`wireless communication relies on cell base
`stations that connect to the remainder of
`the network via copper cables, and indeed
`that wireless dependency upon copper1 is
`projected to explode as wireless “smart-
`phone” data usage stresses spectrum
`resources and increasingly leads to smaller
`(WiFi and/or femto) cells that “backhaul”
`on residential DSL services that today are
`already a growing 70 percent of all broad-
`band connections. With such essential
`copper dependency, perhaps an attempt
`to document historical DSL contributions
`is in order and attempted here.
`The history of copper advance is one of
`incremental steps not unlike silicon-based
`semiconductors’ famous “Moore’s Law”
`where steady evolution of telecommunica-
`tions networks has consistently prevailed
`over revolutionary and costly replacement
`with new, near infinite-bandwidth media,
`
`1 It is often said that “there is no wireless with-
`out wires, but the converse is not true.”
`
`with said steady increase evident the last
`few years in Fig. 1. Figure 2 provides some
`key steps in steady bandwidth advance for
`use of telephony copper twisted pairs.
`Many technologists’ portended death of
`copper has been always answered by unex-
`pected technical advances that squeeze yet
`more bandwidth from copper than many
`ever believed possible. Those advances ini-
`tially came from large telephone company
`research labs, but over the last two decades
`have instead come from small startup
`companies and academic institutions.
`Those few correctly recognized that the
`cost of replacement of billions of connec-
`tions would eventually yield to simple
`increase of speed on the existing copper
`facilities. They were “lighting up copper”
`instead.
`This history will not attempt to
`include further the legacy of voiceband
`modems, whose significant contribution
`to early data communication is well doc-
`umented elsewhere2 (see for instance the
`upcoming paper in this magazine [1]).
`Rather, this history begins with the first
`serious efforts to circumvent the analog
`voiceband. The next section, “Going Dig-
`ital,” recalls some first efforts to expand
`beyond voiceband modems to an all-digi-
`tal telecommunications network. While
`none of these early all-digital-network
`approaches were commercially success-
`ful, they laid a foundation for the DSL
`successes to come. The subsequent “Data
`Com” section also cites early “10base-T”
`and “100base-T” in data communication,
`which were indeed commercially success-
`ful and also contributed to a foundation
`of copper transmission methodology.
`The “Modern Copper Age” section con-
`tinues to the specific Asymmetric Digital
`Subscriber Line (ADSL) technology that
`today dominates worldwide broadband
`connectivity. The next section then pro-
`gresses to the emergence of fiber to
`shorten, but rarely replace completely,
`copper customer connections, leading to
`what is known as “VDSL,” which is the
`telecommunications service providers’
`commercially viable alternative to satisfy
`the growing data bandwidth needs of an
`increasingly connected digital world. No
`history ever should terminate, and indeed
`
`2 Good histories of voiceband modems
`appeared in the September 1984 IEEE JSACs
`(p. 632) by G.D. Forney, R. Gallager et. al., as
`well as the January 1998 IEEE Communica-
`tions Magazine article (p. 16) by K. Pahlavan
`and J. Holsinger.
`
`DSL-based copper-transmission growth
`today is at unprecedented levels and con-
`tinues to expand, so the last full section
`then prognosticates briefly on recent
`“DSM” (Dynamic Spectrum Manage-
`ment) advances shown on the right in
`Fig. 2, as copper-fed customers learn to
`enjoy Gb/s connections to their abodes
`over the next decade.
`
`EARLY TWISTED-PAIR
`DIGITAL TRANSMISSION
`Since Bell’s 1881 invention of the twist-
`ed pair [2], the number of twisted-pair
`telephone connections has steadily
`grown to roughly 1.3 billion worldwide.
`This copper infrastructure is an enor-
`mous asset to the telecommunications
`industry. While history often contends
`others also invented the analog tele-
`phone, no one else claims Bell’s more
`long-lasting copper twisted-pair inven-
`tion.
`
`GOING DIGITAL
`Harry Nyquist’s seminal 1928 work [3]
`motivated conversion from analog to
`digitized voice transmission. Decades
`later, an analog hierarchy of crossbar
`switches consequently transcended to
`digital switching, leveraging the simulta-
`neously evolving semiconductor tech-
`nology that more efficiently processed
`digital bits than analog signals. A voice
`signal can be well represented by 64
`kb/s and reliably regenerated and trans-
`mitted over long distances with almost
`no degradation. Thus, if the telecom-
`munications network core switches
`much more effectively handled digital
`traffic, and analog signals became too
`distorted on long paths between these
`older analog switches, then why use
`analog signals between those switches?
`The reader is also referred to a sur-
`vey by Lechleider written roughly 20
`years ago that outlines contributions to
`that time [3].
`
`T1 Carrier — Bell Laboratories’ Robert
`Aaron [JH1] recognized such digital-
`switch-connect simplification with the
`1962 introduction of “T1” transmission
`technology [4].3 T1 allowed twisted-pair
`
`3 See also the Bell Telephone Laboratories
`“Blue Book” (Transmission Systems for Com-
`munications), 4th Edition, 1971, Western Elec-
`tric Publications.
`
`30
`
`0163-6804/11/$25.00 © 2011 IEEE
`
`IEEE Communications Magazine • May 2011
`
`

`
`LYT-HISTORY-May 4/21/11 10:27 AM Page 31
`
`HISTORY OF COMMUNICATIONS
`
`400M
`in 2012
`
`Many are
`VDSL
`
`DSL
`Cable modem
`FTTx
`
`350
`
`300
`
`250
`
`200
`
`150
`
`100
`
`50
`
`0
`
`Subscribers (in millions)
`
`Q210
`Q309
`Q408
`Q308
`Q108
`Q110
`Q409
`Q209
`Q109
`Q208
`Q407
`Q307
`Figure 1. Worldwide broadband growth (source: Point Topic). Q= 3-month quarter of
`year shown.
`
`ment to create an end-to-end digital
`system. Digital voice transmission in the
`last few miles did not really help voice
`quality, nor did it then have any other
`economic value or driver, but it was too
`much of an elegant challenge for these
`many researchers to ignore.
`Circa 1980, Ralph Wyndrum, Barry
`Bossick, Joe Lechleider and many oth-
`ers at Bell Telephone Laboratories in
`Whippany, New Jersey were trying to
`complete a plan for an all-digital net-
`work. They investigated simple trans-
`mission technologies slightly more
`advanced than T1, and determined that
`up to 160 kb/s of bi-directional trans-
`mission could be achieved over the last
`four to five miles of twisted-pair trans-
`mission, enough for two 64 kb/s voice
`channels, some overhead (16 kb/s), and
`16 kb/s of data (much more than the
`4.8 kb/s voiceband modems achieved in
`those days). Peter Adams in Britain,
`Kazuo Murano in Japan, and others
`also developed similar methods. Origi-
`nally, this was called “Public Switched
`Digital Capacity (PSDC)” but later
`yielded to the name “ISDN” (Integrat-
`ed Services Digital Network). The data
`rates contemplated did not yet antici-
`pate a need for higher-speed services
`and instead focused on ubiquitous digi-
`tal extension of the voice network.
`Their biggest challenge was simulta-
`neous bi-directional digital transmission
`on a single twisted pair, which exhibits
`echo of digital signals from a local trans-
`mitter to a co-located opposite-direction
`receiver. Analog voice also had echoes
`that were simply addressed.6 Digital
`transmission could not so simply handle
`echo. Digital transmission required
`either that time-division multiplexing
`(ping-pong), frequency division multi-
`plexing, or digital echo cancellation be
`used to separate the two directions of
`
`transmission. A dispute arose as to
`which multiplexing method was best.
`Echo cancellation had been successful
`in analog voice networks for small inter-
`vals of overlapping speech, but success-
`ful digital transmission now continuously
`required 100 times greater precision,
`but echo cancellation effectively doubles
`the data bandwidth. So researchers
`began to investigate data-driven echo
`cancellation for this subscriber line
`application, which continuous data-driv-
`en echo cancellation had only months
`earlier been demonstrated at that 100
`times greater precision for voiceband
`modems by a young 23-year-old engi-
`neer working in the voiceband modem
`area at Holmdel Bell Telephone Labo-
`ratories (BTL). The Whippany investi-
`gators (coincidentally, sans Lechleider)
`invited the very young designer to a
`meeting that would compare echo can-
`cellation versus ping-pong (frequency
`division had been eliminated) for ISDN.
`They preferred ping-pong, so the meet-
`ing was contentious. Worse yet, after
`explaining the echo cancellation to a
`hostile audience and how it could be
`done, the same young engineer then
`had the audacity to suggest that 160 kb/s
`was too slow, and they really ought to
`consider a much higher speed, enough
`for video at perhaps 1.5 Mb/s, much
`closer to Shannon capacity for a four-
`mile twisted-pair telephone connection,
`at least in the toward-customer direc-
`tion. The laughter was thunderous, and
`the kid was embarrassed beyond belief
`
`6 Although use of echo suppression and even
`some cancellation was well-known for voice sig-
`nals over very long distances (delays) at the
`time, but did not require the same levels of echo
`reduction as were required in simultaneous bi-
`directional data transmission.
`
`transmission lines between switches to
`carry 1.544 Mb/s digitally over roughly
`one mile. While simple to implement in
`the 1960s, T1 used a low-cost early
`transmission line code called “alternate
`mark inversion” (AMI). AMI achieved
`less than a few percent of the famous
`(Claude) Shannon capacity, well known
`to Aaron and others of the times, but
`was simple to implement and sufficient
`for the intended use. T1 repeaters were
`used to reinvigorate digital signals every
`mile between switches, and higher per-
`formance (longer distance) was not
`needed. T1 carried 24 sixty-four
`kilobit/second voice channels, and an
`extra 8 kb/s of signaling/control informa-
`tion.4 Two copper pairs were necessary,
`one for each direction of transmission.
`T1 enabled a digital telecommunications
`network. T1 might perhaps be consid-
`ered the “first DSL,” and was the initial
`step toward lighting copper. A history of
`T1 transmission can be found in a sepa-
`rate Communications Magazine history
`paper by F. T. Andrews that should be
`published before this DSL history, but
`exact publication date to create a refer-
`ence was not available at time of writ-
`ing.
`
`All Digital — PSDC/Integrated Ser-
`vices Digital Network (ISDN) — The
`consequent digital core network and
`proliferation of digital switches left ana-
`log transmission only in the last few
`miles of copper closest to the customer.
`However, these last few miles repre-
`sented over 99 percent of the wired
`connections, and the cost of replacing
`such wired connections could not be
`shared over many customers. Then, as
`now, it made no economic sense to
`entireley replace such “last-mile” wires.5
`Thus, telecommunications engineers
`around the world began to think how
`they might better digitize this last seg-
`
`4 Outside the USA, the equivalent of T1 was
`called E1 and used a very similar technology to
`carry 2.048 Mb/s (or 32, a nice power of 2, voice
`signals). There was no doubt that E1 used a
`more elegant packet layer than T1's prime-num-
`ber 193 bits every 125 microseconds (8 kHz net-
`work clock), but E1 came later and introduced
`nothing more to digital transmission on copper
`than did T1.
`
`5 A 2010 FCC Report lists the average cost of
`installing fiber (PON, shared connections -
`point-to-point fiber is yet more expensive) as
`$2500/customer. DSL costs is under $100/cus-
`tomer. VDSL systems that mix some fiber and
`copper (and allow a higher DSL speed) cost
`roughly $500/customer.
`
`IEEE Communications Magazine • May 2011
`
`31
`
`

`
`LYT-HISTORY-May 4/21/11 10:27 AM Page 32
`
`HISTORY OF COMMUNICATIONS
`
`DSL evolution timeline
`
`T1
`
`Millions
`
`ISDN
`
`10’s millions
`
`HDSL
`
`ADSL/
`ADSL2+
`
`VDSL
`
`Millions
`
`100’s
`millions
`
`10’s
`millions
`?
`
`?
`
`DSM
`
`Vectored VDSL
`Gigabit DSL
`(phantoms)
`
`1956
`
`1978
`
`1991
`1995
`1993
`
`1997
`
`2001
`
`2004
`
`Figure 2. Estimates of DSL and copper predecessor introductions and volumes of
`deployment.
`
`(particularly when even his own boss
`told him to “shut up and sit down”). But
`that was modern DSL’s birth. I know
`well — that kid was me.
`ISDN activities migrated to the
`American National Standards Institute’s
`“T1D1.3” committee. Standards became
`very necessary with the 1984 ATT
`divestiture, which allowed for equip-
`ment other than that of Western Elec-
`tric to be connected to the network.
`Lechleider, absent from that first
`echo/ping-pong-debate meeting, inde-
`pendently championed echo cancella-
`tion. The debates on the 160 kb/s ISDN
`transmission method raged in that stan-
`dards committee, monitored by a fairly
`young man in his early 30’s from
`Ameritech (one of the seven just-divest-
`ed operating companies) who just hap-
`pened to be on his way to becoming one
`of the most productive standards chair-
`men in telecommunications history, Mr.
`Thomas Starr.7 Under his guidance, a
`compromise proposal for 2B1Q (sug-
`gested by Peter Adams of BT) transmis-
`sion (two bits as one of four levels, 80
`thousand times per second) with echo
`cancellation was driven through stan-
`dards by Starr. ISDN became reality.
`Nonetheless, the Japanese did their own
`
`7 While Starr did not assume the chairmanship
`from a legacy AT&T colleague until 1989, he
`was clearly the leader of this American group
`and later internationally in the DSL area. He
`has been listed recently as one of the 100 most
`influential people in telecommunications, large-
`ly because of his unusual highly respected stan-
`dards-compromising-crafting skills.
`
`ping-pong standard for Japan, while
`Germany did a wider bandwidth ISDN
`standard using three levels instead of
`four — while the American standard
`was adopted internationally. The
`Japanese and German independence
`forced each country on to a special stan-
`dard in each subsequent generation of
`DSL to follow, rather than each profit-
`ing from the volume of the worldwide
`standard, a decision that still today costs
`each country’s operators a premium in
`DSL equipment. The time frame for
`this activity was the mid-to late-1980s.
`While T1 might have been the first
`DSL, ISDN might more realistically be
`considered first because it really did
`connect the subscriber digitally while
`T1 basically did not (usually). As such,
`ISDN formed a foundation for future
`DSLs. Many of the same people
`involved in ISDN, including in particu-
`lar Starr, became DSL advocates and
`experts. However, ISDN was a commer-
`cial failure almost everywhere in the
`world8 — basically, ISDN was too slow
`to offer anything much more than ana-
`log phone service (voiceband modems
`eventually passed ISDN’s 16 kb/s data
`channel, and voice is, well, voice — dig-
`ital or analog). ISDN earned itself the
`well-known substitute acronym “ISDN
`= innovation subscribers didn’t need.”
`
`8 At one point, ISDN connections appear to
`have peaked at about 25M, but they have largely
`yielded to faster ADSL connections (in both
`directions, down and up) everywhere, so there
`are only an estimated millions of them still in
`service as in Fig. 2.
`
`It was going to take at least a Mb/s to
`light up the average consumer with
`excitement; ISDN was too slow and sat-
`isfied no customer need, but it did initi-
`ate DSL expertise.
`DATA COM
`Contributions from the “Ethernet” com-
`munity should not be ignored in a histo-
`ry of telecommunications copper and
`DSL, particularly as data and telecom-
`munication networks have increasingly
`converged together in recent years. Eth-
`ernet originally started via reproduction
`of wireless ALOHA9 networks’ carrier-
`sense and collision avoidance on shared
`coaxial cables, as conceived by Bob Met-
`calfe in 1973 while at Xerox.10 Ether-
`net’s evolution to 10base-T and its
`offspring have proliferated to be used on
`an estimated two billion wired Ethernet
`connections.11 They also provided practi-
`cal motivating proof that higher speeds
`on copper were possible.
`transmission
`Early Ethernet
`“Manchester Encoding” was essentially
`a positive or negative (±1 or one bit)
`single square-wave cycle sent on the
`link roughly 10 million times/second.
`Manchester Encoding is as inefficient
`as the early T1 transmission’s AMI
`code, but enabled cheap 1980s manu-
`facture of Ethernet transceivers. More
`sophisticated Ethernet line codes
`increased user bit rate to 100 Mb/s by
`1995.12 This was an important prece-
`dent to note for future DSL.
`
`MODERN COPPER AGE
`Repeatered T1 connections extended
`the digital network closer to business
`customer’s locations, facilitating multi-
`(Continued on page 34)
`
`9 ALOHA protocols were introduced in the late
`1960's by N. Abramson of the University of
`Hawaii, as in “The ALOHA System - Another
`Alternative for Computer Communications','
`Proc. 1970 Fall Joint Computer Conference
`AFIPS Press, see also IEEE Communications
`Magazine, August 2009, for Abramson's history
`“The Aloha Net: Surfing for Wireless Data.”
`
`10 The best Metcalfe reference is his 1973 Har-
`vard dissertation, reproduced in “Packet Com-
`munication”, MIT Project MAC Technical
`Report MAC TR-114, December, 1973 , but the
`work was done at Xerox Parc.
`
`11 This estimate comes from Jag Bolaria of Lin-
`ley Marketing Group.
`
`12 A good reference on this is the IEEE 802.3-
`1995 Ethernet standard.
`
`32
`
`IEEE Communications Magazine • May 2011
`
`

`
`LYT-HISTORY-May 4/21/11 10:27 AM Page 34
`
`HISTORY OF COMMUNICATIONS
`
`(Continued from page 32)
`
`ple digital phone connections. However,
`T1’s one-mile inter-repeater spacing
`was too short. Following ISDN stan-
`dardization success, Lechleider (then at
`Bellcore) proposed [3] that each of T1’s
`two twisted pairs instead use ISDN’s
`improved echo-cancelled 2B1Q line
`code to double T1 repeater spacing to
`two miles. The proposed 2B1Q system
`basically transmitted bi-directionally
`two bits 400,000 times a second (as one
`of four levels each time) for 800 kb/s on
`each pair, yielding a 1.6 Mb/s total.
`Lechleider called this “High-Speed Dig-
`ital Subscriber Line” (HDSL). By 1991,
`HDSL’s 2B1Q was essentially standard-
`ized in the USA after a competition
`with other proposals (see the section on
`“Line Code Wars” below). Tom Starr
`again skillfully guided a consensus
`HDSL standard/report in the American
`DSL standards group, then operating
`under the revised name ANSI T1E1.4.
`Europe’s “E1” T1-equivalents use 2.048
`Mb/s at roughly 20 percent higher sam-
`pling rates, but also re-used the 2B1Q
`line code.13 Unlike ISDN, HDSL’s high-
`er speeds increase the (imperfectly)
`twisted pair’s radiation. These HDSLs
`thus also sense radiated energy from
`one another, a phenomenon known as
`“crosstalk.” Crosstalk noise thus limited
`HDSL’s signal-to-noise ratio and conse-
`quently HDSL’s data rates, but at least
`1.6 Mb/s (800 kb/s on each line) could
`reliably traverse two miles. The largest
`crosstalk occurs between signals travel-
`ing in opposite directions, where the
`near-end large transmit signal crosstalks
`at a high level into the attenuated sig-
`nal coming from the far-end. This is
`called NEXT14 in copper transmission.
`A significant fraction of telephone
`lines, however, have lengths greater
`than two miles. Something more was
`needed for digital connectivity to the
`residential customer (above ISDN’s too
`slow 160 kb/s). Lechleider proposed
`asymmetric transmission rates in non-
`overlapping upstream/downstream fre-
`quency bands
`to avoid NEXT.
`Lechleider’s asymmetry then avoided
`NEXT. Thus was born a basic concept
`of ADSL (where A = asymmetric).
`At this point, after getting a Ph.D. and
`
`13 In some cases 3 American-speed lines operat-
`ing each at 800kb/s were used for a total of
`2.4Mb/s with an extra 352kb/s of overhead.
`
`working on disk drives at IBM for a while,
`I returned to Stanford as faculty in 1986,
`heard of ADSL and was elated, and
`found a way to meet Joe Lechleider,
`recalling my earlier “1980 laughed-out”
`meeting (at which he was coincidentally
`not present). While we differed in age by
`more than three decades, our early con-
`versations and meetings were very excit-
`ing. Lechleider explained that digital
`applications (video to customer or infor-
`mation to customer) were likely to be
`asymmetric, thus more frequencies should
`be allocated to the downstream (to cus-
`tomer) direction than the upstream (from
`customer). If the upstream frequencies
`were limited, then the NEXT would also
`be limited, and all the rest of the frequen-
`cy band was then NEXT-free. My earlier
`1.6 Mb/s at four miles calculation had a
`legitimate supporter in “Uncle Joe.”
`Lechleider introduced me to Belcore’s
`Dave Waring, who successfully managed
`Bellcore’s DSL efforts thereafter for
`decades (and still does today) as well as
`re-acquainted me with Ken Kerpez, who
`made numerous contributions in model-
`ing and fair testing of DSL over the years,
`despite having a limited budget with
`which to work for years.
`There was a hand-off at this point. Joe
`was nearing retirement age, and knew he
`would stay only to finish the HDSL work.
`He found Bellcore money to finance the
`first years of Stanford research in ADSL,
`which the National Science Foundation
`also matched. There was just barely
`enough to develop a design and to proto-
`type. Some interesting findings emerged
`from this effort — basically, only a multi-
`carrier transmission approach could
`achieve the data rates, and indeed if fiber
`were used in part of the network (say
`within two miles of the customer, or even
`several HDSLs to the two-mile point),
`the last two miles could sustain 6 Mb/s in
`ADSL. This was big — 6 Mb/s is a lot of
`data (even today) and certainly enough
`for good video (and perhaps a few simul-
`taneous videos). ADSL had legs. Uncle
`Joe sent one last check as Bellcore also
`reduced funding overall of the area, told
`me it was up to me, and retired. Little did
`I know that there was a gauntlet to run of
`unpredictably epic proportion before
`DSL could really light up.15
`LINE CODE WARS
`While it would seem HDSL was a logical
`simple extension of ISDN in terms of
`transmission, even such a simple extension
`
`14 Near-End Crosstalk. X =Cross. FEXT or
`Far-End CrossTalk is between signals traveling
`in the same direction on adjacent twisted pairs.
`
`15 I still wonder today if Joe knew well of that
`gauntlet before handing off with a smile on his
`face.
`
`did not go unchallenged. The remnants of
`that challenge plagued DSL advance for
`years. AT&T Information Systems16 pro-
`posed carrier-less amplitude/phase modu-
`lation17 (CAP) for HDSL. CAP is QAM
`with a minor simplification that causes the
`carrier frequency and symbol clocks to be
`exactly synchronized. CAP demonstrated
`a slight improvement in recorded inde-
`pendent laboratory tests in a 1991 T1E1.4
`investigation. Nonetheless, HDSL stayed
`with ISDN’s known 2B1Q transmission
`line code. The CAP proponents were dis-
`appointed about this HDSL decision, and
`believed CAP then deserved the next
`standard (ADSL).
`CAP was proposed for ADSL, but
`its use there would have been fatally
`flawed (as would have been also 2B1Q).
`The billion telephone lines exhibit wide
`variation (varying linear transfer char-
`acteristics and highly variable and time-
`variant noise spectra), which when
`stretched close to Shannon limits, forc-
`ing a highly variable best transmission
`bandwidth. The optimum Shannon
`spectrum often has a different
`on/off/on/off/.... /off nature for realistic
`DSL channels. On/off/on/off means the
`optimum transmitted spectrum places
`energy in separated spectrum segments,
`Basic transmission theory shows that a
`single carrier can never achieve the per-
`formance of the on/off/..../off spectrum
`(at least one carrier for each “on” band
`is necessary [5]). This effect is amplified
`in practice because of realistic code
`implementation and a 6 dB margin
`required for unforeseen line impulse
`noise changes. The Stanford work had
`studied this problem for years and the
`conclusion was irrefutable: multiple car-
`riers were necessary, or the industry
`could forget 1.6 Mb/s at four miles and
`6 Mb/s at two miles, or essentially DSL
`would have failed.18
`(Continued on page 36)
`
`16 This group later became Lucent, and a DSL
`modem portion of ATT then was spun off as
`ATT Paradyne.
`
`17 This is a slight simplification of basic
`quadrature-phase modulation that exploits the
`symbol clock and carrier frequency can be syn-
`chronized in DSL transmission since there is no
`intermediate independent carrier adjustment in
`a twisted-pair transmission channel. Most pre-
`fer to simply call both CAP and QAM as
`“QAM” since the difference is trivial.
`
`18 Some examples further illustrate the theory
`simply in the textbook reference at website
`http://www.stanford.edu/group/cioffi/ee379c/ ,
`which extends reference [5].
`
`34
`
`IEEE Communications Magazine • May 2011
`
`

`
`LYT-HISTORY-May 4/21/11 10:27 AM Page 36
`
`HISTORY OF COMMUNICATIONS
`
`A DIGRESSION INTO BASIC DSL TRANSMISSION THEORY
`
`This papers’ reviewers encouraged inclusion of some transmission theory here to
`expound upon this point. There are those who confuse a result that applies only
`to voiceband modems with application of its conclusion to DSL. Figure 3 will
`help illustrate this “multi-bowl” point. Reference [5] first made it general,
`expounding on some earlier results of Price, Kalet, Zervos, Salz and others that
`appear19 in the references of [5]. Simply stated for the purposes of this histori-
`cal article, many authors make an infinite signal-to-noise-ratio (SNR(f)) approxi-
`mation 1+SNR(f) ≈ SNR(f) in transmission analysis, and this approximation
`holds well over the entire used bandwidth of a voiceband modem. However, in
`DSL, SNR(f) is often zero in Shannon’s famous water-filling spectra shown in
`Fig. 3. The assumption 1+0 ≈ 0 is not accurate in all the unused bands (and
`near their edges). If one follows the assumption in these earlier works, it is
`equivalent to assuming that there is infinite energy (water) available to be
`poured from above, causing all of the Shannon bowls to overflow into one
`another, and thus into one very large single carrier (for which a DFE would be
`optimum with QAM or CAP). It is also important to note that assumptions
`equivalent to “generalized Nyquist bands” made by these same authors are
`described as DFEs in those articles but a review of filter-realization theory and
`Paley-Wiener criteria will reveal that discontiguous bands in that theory MUST
`be implemented with multiple carriers. So the authors call it single carrier, but it
`is really a number of carriers equal to the number of bands. However, the
`amount of water (energy) needed to force a single carrier and thus single band
`on most DSL channels greatly exceeds that available, and thus the equivalence
`of multi-carrier and single-carrier does not hold. Reference [5] also shows that if
`codes less than capacity achieving are used with any gap (or margin) above 0 dB
`(DSL uses a 9.8 dB gap to capacity with 6 dB of it left for time-varying noise
`effects), that the difference between single-carrier and multicarrier rapidly mag-
`nifies. This effect was in plain evidence in the so-called DSL Olympics test
`results that are mentioned later in this history.
`Further, it is not possible to design a single DFE that corresponds to the
`same three spectra, as the fundamental assumptions behind DFE realization
`then no longer apply (essentially the filters blow up). Correct DFE theory inter-
`pretation is that three separate DFEs are necessary for the situation in Fig. 3
`(see Reference [5]).
`
`Infinite SNR
`(1 band)
`
`Finite SNR
`(3 bands)
`
`Figure 3. Plot of Shannon’s water-filling. The curve is NSR(f), the inverse of
`SNR(f), and energy/water is poured into the curve to lie at a constant level,
`with three “bowls” illustrated. This is the optimum spectra for best perfor-
`mance, and a single-carrier cannot achieve optimum performance unless the
`SNR is infinite. Use of the same three disjoint spectra in a single decision feed-
`back equalizer causes a violation of basic filter design (causes an unrealizable
`filtering effect) and the assumptions underlying decision-feedback theory no
`longer apply and the decision-feedback system cannot be realized (instead
`three are needed, one for each band). The difference between a single-carrier
`system and multicarrier system is magnified, when as in DSL, the capacity gap
`is nonzero (minimum of 6 dB in DSL to account for time-varying noises).
`
`(Continued from page 34)
`
`But the CAP supporters wanted their
`standard. I made a considerable effort to
`talk to the various interests to explain
`CAP’s fundamental-flaw for ADSL.
`However, I was not successful. Broadcom
`was formed as a spin-out of HDSL sup-
`plier Pairgain, where UCLA Professor
`Henry Samueli was CTO and had done a
`very good job implementing the first
`HDSL 2B1Q transmission chips that
`were used in the above-mentioned HDSL
`laboratory tests. He was joined in the
`Broadcom spin-out by Pairgain’s VP of
`Engineering Henry Nicholas. It was more
`expedient for them to adjust their HDSL
`design to CAP’s close cousin QAM (thus
`avoiding AT&T patents on CAP) and
`rapidly market a QAM ADSL chip. They
`thus proposed a “compromise” of doing
`QAM (close enough to CAP), argued
`similarly to AT&T Information Sys-
`tems/Paradyne. This placated somewhat
`the CAP supporters (although they really
`wanted CAP) and they formed some-
`thing of an anti-DMT alliance. The
`Broadcom founders were astute busi-
`nessmen, and they knew full well they
`(Broadcom) could get a chip to market
`faster than AT&T-Microelectronics (the
`chip partner of Paradyne and AT&T-IS,
`now known as Agere) and thus uniquely
`capitalize on ADSL’s potentially enor-
`mous market of one billion customers
`worldwide. All efforts to convince them
`to use multiple carriers aborted because
`Broadcom would lose a time-to-market
`advantage, and that economic incentive
`blurred the ability to see the technical
`argument that a single QAM carrier
`would fail from a transmission stand-
`point. I had failed to convince anyone
`that the right transmission strategy was
`multicarrier (Uncle Joe understood, but
`he had retired), except for some excep-
`tionally talented Stanford students, Jacky
`Chow, Jim Aslanis, and Peter Chow, and
`some very experienced friends from mul-
`ticarrier-voiceband-modem manufacturer
`Telebit (a consulting job for me) CTO
`John Bingham and Mark Flowers.
`Together, that latter group became
`Amati Communications Corporation,
`which was founded in June 1991 to
`design and manufacturer a multicarrer
`ADSL modem. With less than 10
`employees, and funding from Nortel’s
`American marketing group19 (Nortel’s
`(Continued on page 38)
`
`19 A special thanks is still due today to Northern
`Telecom Marketing VP Stephen Fleming, now
`at Georgia Tech, for his faith and funding of
`that early effort.
`
`36
`
`IEEE Communications Magazine • May 2011
`
`

`
`LYT-HISTORY-May 4/21/11 10:27 AM Page 38
`
`HISTORY OF COMMUNICATIONS
`
`(Continued from page 36)
`
`Canadian Bell Northern Research
`group wanted to do QAM and was
`aligned with Broadcom), Amati pro-
`posed Discrete MultiTone (DMT), a
`form of multicarrier that heavily favors
`digital low-cost implementation, with
`bit-swapping (a method to adapt con-
`tinuously to the unique changes in
`noise on each, and across all the bil-
`lion DSL connections. Broadcom pro-
`posed QAM (inside a system from
`Reliance Comm/Tech). AT&T Infor-
`mation Systems/Paradyne proposed
`CAP. Tom Starr, despite the unpopu-
`lar nature of his decision, insisted on
`due process and laboratory testing of
`all three befor