A great read from RB
The Sound of Silence
Summer Review Issue: Part 2
INSIDE THIS REPORT:
Cover Story: The Sound of Silence:
It's my party…; I can talk if I want to; Sorting silence; CDMA envy; QCOM's transition to 3G
Backpage: Telecosm Table
In the late 1980s, more than ten years ago, Nicholas Negroponte of MIT's Media Lab, made a provocative prophecy.
He said that in the future, video and voice would trade media. Video-then mostly broadcast television that chiefly
traveled through the air-would move to wires (mostly optical). Voice-then overwhelmingly carried in twisted pair
telephone wires-would move to the air.
I immediately endorsed this prediction and dubbed it the "Negroponte Switch." Today it is almost a truism. Despite
all the exceptions that prove the rule (satellite TV and broadband wireless Internet), the overwhelming uptake of
mobile telephony and cable TV and fiber optics makes the Negroponte Switch one of the best technology
predictions ever made.
At the time, Negroponte's call was anything but obvious. Requiring no expensive cables and backhoes, broadcast
TV was ubiquitous, one of the most popular and well-established technologies ever launched. Governments heavily
supported both air TV and twisted pair telephony and still do today. In the breakup of AT&T (T), U.S. regulators
assigned free spectrum rights to the Baby Bells, thus assuring that wireless telephony would be promoted as a
costly mobile supplement rather than as a rival to wireline telephony.
Mobile phones and tethered video both prevailed because they fit the technology paradigm. What governments did
scarcely mattered in the face of the inner logic of video as a broadband spectrum hog that polluted the air but could
fly freely through fiber and cable. The errors of regulators could not prevail against the paradigmatic power of
microchips to enable mobile telephony better in every important way than analog wireline voice. Listen to the
technology and you quickly see that video wants dumb pipes and voice wants clever mobility. What government
wants is interesting in the short run but trivial in the long run.
Today governments and gullible telopolies everywhere lust for GSM (Global System Mobile), the government birthed
and swaddled choice of the European Commission (EC). A Time Division Multiple Access (TDMA) system, it is
maladapted to the Internet's bursty data flows that rarely fit the TDMA timeslots. Beloved of regulated Bellheads,
once even endorsed by the U.S. State Department, and flacked and touted regularly in the technologically and
politically naïve U.S. business press, GSM is the choice of bureaucrats everywhere against the superior U.S.
alternative of Code Division Multiple Access (CDMA). Now Japan's NTT is buying a GSM spearhead in the US
through investment in SBC-BellSouth's (SBC) anti-CDMA wireless entity. Deutche Telekom is also promoting GSM
by buying Voicestream (VSTR), the only major U.S. GSM carrier.
For awhile, the chief rival to the world's telopolies, Regional Bells, and European departments of Post and
Telecommunications (PTTs) was WorldCom (WCOM). It planned to use its Internet prowess to mount an
aggressive challenge to national communications monopolies. But as usual governments rallied to the support of
the incumbent telco establishment. With the help of clueless U.S. Anti Trusters in Washington, EC regulators first
helped block WorldCom's attempt to acquire MCI's Internet unit, on the incredible grounds that the new firm would
dominate the Web. The MCI unit went to the U.K.'s Cable & Wireless. Whew! A U.S. Internet monopoly narrowly
averted. Then EC bureaucrats collaborated with U.S. kin to thwart WorldCom's bid to buy Sprint (FON) and expand
Sprint PCS (PCS), the leading CDMA carrier. Phew, those evil Internet monopolists foiled again! Now in a new
entente of TDMA telopolies, AT&T is rumored to be romancing British Telecom (BTY).
All these companies cozening the U.S. market to promote inferior European technology are political entities, either
largely government owned (NTT and Deutche Telekom) or are regulated political protectorates dominated by
lobbyists rather than engineers (SBC and AT&T). With some 270 million GSM subscribers around the world, GSM
is the world's most impressive success of government industrial policy. It recently won major ground in China
where CDMA was regarded to be the U.S. government choice after Qualcomm (QCOM) executives defensively
showered the Democrats with campaign cash and Irwin Jacobs met publicly with Bill Clinton. GSM now leads
CDMA in China by some 45 million units to one million units.
On the surface, the picture is grim for the paradigm. Even our own Gilder Forum (www.gildertech.com) is fretful.
How then can we be so sure that the next decade of wireless history will be nearly indistinguishable from the last in
its salient feature: the progressive, inexorable triumph of CDMA and its inventors and most adept practitioners at
Qualcomm?
The argument is necessarily technical. You have to listen to the technology. At its heart you will find the sound of
silence. Virtually all communications theory, all contests for mobile superiority, and the final triumph of CDMA
revolve around the manipulation of silence: adding it, subtracting it, rearranging, condensing, or expanding it.
It's my party ...
To explain, let us revisit the now legendary cocktail party that Qualcomm founder Irwin Jacobs long ago conceived
as an analogy for cellular systems. Several guests have paired off in conversation. But for each listener the sum
total of all the other speakers' "transmissions" is mere noise, interfering with his ability to discern the signal sent by
the one person to whom he is trying to listen. Struggling to be heard, each speaker talks louder and louder, but that
only increases the total noise in the channel—the room—making it even more difficult to hear.
Frantic to save her party, the hostess tries several different solutions. First she disperses her guests into different
rooms, one pair to a room. With no noise at all in the room except for its own conversation, each pair converses
unimpeded, luxuriating in silence. This is frequency division multiple access, the solution used by analog mobile
systems, which dedicates a pair of narrow frequency channels solely to one pair of users in a cell. Since the
frequencies cannot be re-used except by users in cells at least two diameters away, the system also invokes
space division multiplexing.
Alas, while some guests now have a surfeit of silence, many still have none. This is a Manhattan cocktail party, and
there are not enough empty rooms (channels) in the apartment for all conversations. A simple spatial division won't
do. Taking another stab at the problem, the hostess now places three pairs of conversationalists in each room. All
three pairs may converse, but they must take turns. Every 20 seconds, say, one pair gets to talk, the other two
must be silent. Adapting to the problem, the guests find that, when it is their turn, they tend to talk faster than usual.
And while waiting, they search for economical turns of phrase, to encode more meaning in less transmission time.
This is time division multiple access (TDMA), made possible by digitizing the voice signals so we can send them in
rapid bursts, combined with some speech compression codes to save bits, and it triples the number of
conversations the hostess can accommodate.
... I can talk if I want to
But wireless is a paradigm party, the party of the decade. Guests continue to throng in at a rate of millions a week.
Suddenly the solution dawns on the hostess. Her party is ever so cosmopolitan, with Japanese, Koreans, Indians,
and polyglot Chinese and Europeans. So she asks everyone to return to the main room, and speak as softly as
they can manage, but each pair in a different language. Mirabile dictu, everyone can hear and understand the signal
meant for them. A little better if a few guests leave; a little worse if a few more show up, but satisfactorily in any
event. This of course is code division multiple access (CDMA).
Adding silence, taking away noise, the solutions progress on a continuum from the purely physical to the
predominantly logical. In the first, analog solution our only resources are physical—separate the speakers by
space. In the TDMA solution we employ two physical dimensions, and build our isolation chamber out of both space
and time. TDMA plies logic as well, compressing communications to fit better into those slots. But only the final
solution is predominantly logical—assign each speaker a different language, or code. Coding the messages into
different languages provides virtual silence, since listeners easily filter out the sounds of languages they do not
understand, readily identifying them as "noise" rather than signal.
Adding silence in this way is only part of the story. Also crucial is silence removal. Although silence is indepensable
to communications, it is also an arrant waste of bandwidth.
In the core of the Telecosm, defined by the optical network, such waste reveals correct priorities: spend abundant
bandwidth to save expensive processing. Thus optical switches—which like mirrors merely steer or redirect
physical beams of light—drive out electronic switches, which laboriously and expensively process the underlying
packets.
Wireless, though, is an environment of bandwidth constraint rather than bandwidth abundance. Here the logical still
trumps the physical, smart systems are better than dumb ones. Because we cannot expand the physical resource,
as we can in fiber optics simply by laying more fiber, the wireless paradigm favors solutions that use
logic—manifested in electronic processing—most aggressively and adeptly. In short, the winning technology will
exploit most effectively the old microcosmic abundance defined by Moore's Law. In the Telecosm, the Moore's Law
abundance is found less in Intel (INTC) microprocessors fetching instructions and data from memories than in
Digital Signal Processors (DSPs) from companies such as Texas Instruments (TXN). These devices handle real
time streams from analog-to-digital converters made by companies such as Analog Devices (ADI) and National
Semiconductor (NSM).
After considering the investment implications of this part of the paradigm, we return to the party. Consider again our
hostess. Her first solution—each conversational pair got its own room—wasted silence. With each speaker
occupying an entire channel, much of what flows through—some 65 percent in fact-is silence. In typical phone
conversations, each speaker talks about 35 percent of the time, with both parties silent about 30 percent of the
time. And because the analog signals in the communication "look" just like the sound waves they imitate, they
cannot be compressed or accelerated.
Sorting silence
The TDMA solution—three pairs of conversationalists in each room—is digital. But for all its digital compression,
sampling, and rearranging, TDMA does not directly address the wasted silence problem. Dedicating rigid time slots
to each user, TDMA neither knows nor cares whether the user is making good use of allocated bandwidth. Whether
shouting for 911 or whistling Dixie, 65 percent of the time users are dead silent and so are their allocated channels.
Over-provisioning silence and wasting bandwidth to save processing, TDMA represents at best a form of rationing
that uses relatively mindless rules to allocate silence, and thus inevitably wastes it.
Where TDMA wastes silence, CDMA spends logical MIPS to both save and manufacture silence. Under the
guidance of Qualcomm cofounder Andrew Viterbi, who created crucial coding algorithms used in nearly all digital
communications systems, Qualcomm developed the variable rate vocoder. In TDMA or CDMA a vocoder
condenses the 64,000 kbps digitized version of your speech down to between 8 and 13 kbps. But in pauses or
silence, CDMA's variable rate vocoder will output at as little as one-eighth of the full rate. The momentarily silent
user opens up space in the channel for other conversations. The variable vocoder alone accounts for a 250 percent
increase in the capacity of a CDMA cell, but it would be pointless in a TDMA system, which shares time but cannot
share silence.
Thanks to the law of large numbers in CDMA, the salvaged silence is spread out across the shared 1.25
megahertz channel in a variant of statistical multiplexing, the basic economizing principle behind any shared
channel, like an Ethernet or the Internet. As with an Ethernet, there is only a "soft" limit on the number of users:
adding one more will increase the interference in the channel only marginally. Our hostess need not panic if a late
guest shows up at the door.
Even more auspicious for CDMA in the coming era of the wireless web, the hostess need not despair even if
scores of guests decide to deliver speeches with Powerpoint slides or transmit lengthy Postscript files. Because
the CDMA system spreads all the data across all the available spectrum all the time, it can accommodate the
bursty bitstreams characteristic of the Internet. Rather than bandwidth confined to an irretrievable series of
narrowband time slots, CDMA can offer bandwidth-on-demand.
Thus CDMA exploits processing to add silence where it is needed and take away where it would be wasted. But
CDMA's greatest feat and the essence of the system is that it uses processing to turn noise itself into silence.
In any communication channel all the transmission power ends up either as noise or signal, and some of it ends up
as both, because even a well shaped signal is noise to a user trying to receive a different signal. In TDMA, there are
only two possible fixes. Raise the power of the transmission, thus making it more likely that the bits will be
discernible through the noise. But any additional power will show up as additional noise on adjacent channels,
confronting them with the same choice. (The guests at the cocktail party have started to shout.) Or the sender can
add more bits—e.g. in the form of more elaborate error correction—but this will decrease the information rate of the
channel. (The guests at the party have started to repeat themselves.)
CDMA soaks up all this overflow of noise and bits with DSP MOPS (millions of operations per second). The CDMA
transmitter first multiplies the information bits by a pseudo-random noise code and then spreads the resulting
apparently randomized signal across a slice of spectrum more than 100 times the bandwidth of the original signal.
Like a platoon spread out to avoid death from a single grenade, the spread signal cannot be wiped out by noise in
any narrow portion of the channel.
When the signal is spread by a factor of more than 100, its energy is necessarily spread as well…and so is the
energy of the other users of the same spectrum. Collectively the signals of dozens of users have acquired the
essential characteristics of, if not quite white then "gray" Gaussian noise. The result is scores of decibels of
"processing gain": an apparently magical power of hearing a soft sound above a much louder one of similar pitch.
Abandoning the attempt to power past competing users, CDMA lowers the energy of all signals to the minimum
needed to reach the receiver. Because the receiver has the matching noise code—which is inverted to delete the
noise through destructive interference—the message can be extracted from the background drone. What would
have been a cacophony of competing voices now appears as a low murmur, like the celestial hum of a Gregorian
choir against which the soloist stands in vivid relief.
Transforming noise into a form of silence, CDMA also redeems "multipath" signals arriving by different routes, and
thus at slightly different times. Normally multipath is irredeemable noise to a time slotted system, but CDMA uses
an early invention of optical sage Paul Green, the "rake receiver," to combine the three best multipath signals into
an amplified CDMA stream. On the sector edges, the rake receiver even permits the soft shifting of users from a
crowded cell (route 101 at rush hour) to an underused neighbor (Los Altos Hills)—thus effectively shrinking and
expanding cells in response to the movement of traffic.
By substituting the transmission of logic for the transmission of power and by binding up the energy that remains
into a wider and more redundant logical web, CDMA ensures that a greater portion of that energy is actually used to
deliver information, the definition of spectral efficiency.
CDMA envy
Every other wireless system is now emulating these CDMA devices, doing more processing for better power
control and more bits per hertz (an EDGE effort) or more processing to do magic with antennas (an NTT DoCoMo
focus) or more processing for better error correction and turbo codes (CDMA 2000 uses them as a critical
ingredient in its doubling of voice capacity). Because all of these technologies are being advanced by ingenious
engineers, at any given moment performance data will incite a flurry of claims about the latest heroic tweak, or the
newest, best, vaporware ever, or never. This is a trillion dollar argument and the room can get very noisy.
The value of a paradigm is to keep one from losing the signal amidst the competitive noises. And in wireless the
paradigm is "more logic and less power." However noisy the argument, CDMA's triumph is readily discernable in
the simple fact that for the Third Generation (3G) systems of the Internet, TDMA, which fails the paradigmatic test,
has virtually abandoned the field. With the lonely and inexplicable exception of AT&T's EDGE, every seriously
proposed 3G system is based on CDMA.
The evolving WCDMA standard is being written in an apparent attempt to be as different from Qualcomm's CDMA
as possible, while still deriving its advantages, so the Europeans could retain or the Japanese attain world
leadership in wireless. The Japanese government in particular has been pressing NTT DoCoMo to ensure the
global 3G standard would come from Japan. To that end, NTT has been developing WCDMA since the early 1990s.
In theory WCDMA enhances the advantages of CDMA by expanding fourfold the spread (or so called "chipping
rate") in its "spread spectrum." Thus, each signal is spread more widely by a higher chipping rate, each bit is lower
power, rendering the resulting background noise even more Gaussian (a smoother more random appearing hum
against which to search out the signal). Since more users are sharing a single channel, the statistical multiplexing
advantages of variable rate vocoding are enhanced.
The law of large numbers, however, suggests that at some point "large" becomes "too large" and returns beyond
this point diminish. Before settling on a 1.25 MHz spread almost a decade ago, Qualcomm sage Klein Gilhousen
studied a range of spectrum widths and chipping rates for expanding the signal. He found that an eightfold increase,
from a chipping rate of 16 to 128, produced a 20 percentage point increase in spectral efficiency. By increasing the
rate another eightfold to 1024 over 10 MHz, however, the WCDMA proposal of the time would yield less than half
that increase, about 9 percent. The current WCDMA proposal would yield about 6 percent greater spectral
efficiency.
In the real world of limited spectrum, the right channel width must reconcile the advantages of channel spread with
the need for channel flexibility. More channels are easier to swap from voice to data and back again to
accommodate shifting demand during the course of the day or to offer bandwidth on demand.
While gaining paltry and diminishing returns in spectral efficiency, the wider spread incurs real and rising costs in
chip set complexity and handset power. Operating across a wider spectrum, a WCDMA rake receiver must
process three times as many signal elements in the 156 microseconds available, requiring more rake components,
each operating three times as fast. The result is more complex and expensive chips, more silicon area, more
power consumption, and ultimately, perhaps, lower performance.
Nevertheless, in principle, WCDMA has at least debatable merits. In practice, the standards process has become
an exercise in product design by committee, with each member desperate to inseminate the elephant with his own
intellectual property, producing the usual lumpy patchwork of a patent bearing camel. Extending the theory that if
CDMA is good, more is better, WCDMA adjusts power levels 1600 times a second rather than CDMA's 800. But
power control bits, because they cannot be coded for error protection (there is no time to decode them) must be
sent at much higher power than the rest of the signal, boosting interference, and sweeping the design well beyond
the point of diminishing returns.
Bad as this all sounds, NTT DoCoMo is pledged to have a nominal WCDMA network up and running somewhere by
late next spring, though at the moment it promises only 64 kbps data, the rate of Qualcomm's IS95B, a 2G system
available in Japan today from DDI-IDO. Qualcomm will have a WCDMA demo-chip available by the end of this year,
produced in consultation with one or two likely future customers, such as Japan Phone.
Happily, the standard's worst mistakes are unlikely to survive. The Japanese already plan to ignore such "features"
as the "compressed mode" for GSM compatibility. According to John Brewer, of Vincio Group, the Europeans are
unlikely to do WCDMA at all: "WCDMA deployment dates stretch out by a year every six months." CDMA is hard.
Stonewalled out of effective participation in WCDMA standards meetings so far, expect to see Qualcomm's
influence over the evolving standard grow. The spin-off of the chip company should improve matters even further
as it separates Qualcomm the supportive vendor from Qualcomm the royalty collector.
For similar reasons, expect current Korean CDMA One operators to stay with CDMA 2000 and HDR rather than
move to WCDMA, despite all blandishments of NTT. WCDMA is going to have a long gestation and a painful
delivery. CDMA 2000 is already here.
Neither the first phase of CDMA 2000, which will provide data rates at up to 144 kbps, plus doubling voice capacity,
nor the second phase which incorporates HDR with data up to 2.4 Mbps (expect announcements over the next
year pushing that number up several fold) require dramatic upgrades to the network. As one Qualcomm engineer
summed it up: the same hardware, the same waveform, the same channels, the same cell geometries, pin
compatible chips, and forward and backward compatible handsets.
QCOM's transition to 3G
The new mobile chip, the MSM 5000 series, available to operators in sample quantities this month, brings capacity
and features that will be particularly beloved by the folks at Wingcast, the Ford (F)-Qualcomm joint venture. It plans
navigational, safety, security, scheduling, and even entertainment services in your car. Included are improved GPS
reception, enhanced graphics support for, e.g. your Wingcast Maps, MP3 for music by phone, Bluetooth, and
enhanced voice recognition for hands-free operation to limit what our friend Mark Mills calls unplanned
car-telephone pole interactions.
Part of Qualcomm's strategy for seamless transition to 3G, the added features should prompt most handset
manufacturers to use the new chip as soon as it ships this year. Thus, when CDMA 2000 becomes generally
available, millions of phones equipped to benefit will already be in subscriber's hands.
The HDR chip strategy will be similar. The current sample chip—the first true 3G chip—will never be released.
Instead the HDR design will be incorporated on the next generation of 2000 chips, once again pin compatible with
previous versions, and once again significantly before HDR services are offered nationwide.
Optimized for Internet Protocols, HDR is a pure CDMA system on the reverse link (mobile phone to base station).
The forward link, where data rates up to 2.4 Mbps kick in, adds a dynamic time division element that stands static
TDMA on its head. Every 1.67 milliseconds, using CDMA's power control and channel monitoring abilities, HDR
determines which current mobile user can get the best reception from the base station and uses all available
capacity, up to 2.4 Mbps in current versions, to execute that user's download request. Because of the volatility of
mobile reception, the best and worst users may swap status in milliseconds. By favoring the best user most of the
time, average throughput across the channel will be much higher than if the system rotated, TDMA like, through
rigid pre-assigned time slots. By clearing the current best user's downloads out of the queue ASAP, everyone's
waits are reduced. The net result, is to ramp up effective average throughput which is far more important than the
theoretical maximums usually cited when comparing 3G data systems.
Both GPRS, the generation 2.5 transition for GSM networks on the way to WCDMA, and EDGE, the 3G data
solution for some TDMA networks, will be hard pressed to compete with the CDMA 2000/HDR combination.
GPRS claims to offer data rates up to 115 kbps. But this requires bonding together eight adjacent power hungry (as
much as 20x the power consumption of CDMA) channels, and pumping as much as 5W through the handset, not
likely to have positive effects on battery life, the phone, or the facial complexion of the user. In practice GPRS data
rates will probably top out at about 56k, and typically function at half or a quarter that rate.
Promised for sometime in 2002 to compete with HDR, EDGE now also promises an enhanced version that will go
beyond the originally specified 384 kbps all the way to some vaporous megabits. But the strain shows. Every
theoretical enhancement to a TDMA system—including smart antennas, dynamic channel control, even
significantly upgraded power control—is being conjured on its behalf. But every proposed tweak is equally available
to CDMA, and confers no competitive advantage. Edge will remain a bandwidth wasting technology that may well
not ever be built (although AT&T may cobble the camel together and call it an Angel). |
