Ramsey/All I think DR. Viterbi addressed some of the service provider concerns last year. Sorry about the length but it's good reading. W.
Andrew Viterbi of QUALCOMM
With all the alphabet soup in the titles of the multitude of wireless
conferences these days, we lose sight of what the initials stand for. Here UPC
stands for Universal Personal Communication.But perhaps more importantly, it can
also stand for universal service to all the world's population, more than half
of which has never had access to telephony.
Universal is a term which has two meanings: first it can stand for "Anywhere,
Anytime" availability and access, which only mobile communications can offer.
For these, mobility is hardly meaningful; the issue is how possibly can they be
connected to the developed world within the first decade of the new century? And
the answer can only be through wireless fixed local loops, another theme of this
conference.
Now let me turn to the word Personal, usurping its true meaning in the
conference's title to describe rather some personal experiences and express some
personal views. I've been extremely fortunate to have participated in the
four-decade evolution of digital wireless and spread spectrum from its origins
in military and space applications through its current commercial diffusion. The
abridged titles of my three books gives a brief summary of my technical
activities: "Coherent Communications"' "Digital Communications and Coding"; and
the most recent "CDMA and Spread Spectrum." And this last one relies heavily on
all the previous subject matter.
It's also the topic which continues to fully occupy my interests and activities.
But, more importantly, CDMA is the technology which in my opinion will best
fulfill the promise of Universal Personal Communication. And let me tell you
why. The primary issue in wireless multiple access communication is how to cope
with interference. This can be approached by separating users in frequency, time
and space, all of which limit the number of users supportable by each base
station; and even this does not completely avoid the self-interference through
multipath, resulting in fading, or the interference from remote base stations
which reuse the same frequency band; the smaller the reuse factor, the greater
the user capacity, but the more severe the interference.
Or one can tackle the problem head-on, accept that every user or every base
station will interfere with every other user or base station and share the
spectrum among all user transmitters in one direction and among all base station
transmitters in the other. To achieve this, one must create benign interferers,
and employ signal processing techniques to extract individual user signals from
their benignly interfering neighbors. The first task is fulfilled by spreading
the spectrum such that every user appears as wideband noise to every other user.
The second task is fulfilled by a collection of tightly related and mutually
supportive techniques. Foremost among these is power control. For the challenge
to true spread spectrum (meaning that generated by direct sequence spreading
rather than frequency hopping) has always been the "near-far" problem. In the
earlier military scenarios, directional antennas were the only remedy. But for
civilian systems, tight and accurate power control of each individual user's
transmitter, so that it transmits just enough power to ensure reliable
reception, solves the near-far problem; beyond this it serves the dual purposes
of minimizing interference to other users and maximizing the battery life of
portable terminals. Other important processing measures to enhance spread
spectrum signal reception include "Rake " receivers to turn detrimental
multipath into cooperative signal energy and "Soft Handoff" among base stations
or among different antenna sectors of the same base station. These two features
can be implemented by the same receiver technique since two or more base
stations transmitting the same signal appear to the "Rake " receiver as
time-delayed versions of the same signal, which is equivalent to the case of
multipath. Another means for reducing interference, by better than half, is
through use of a variable rate voice coder, which reduces its transmission rate
and hence transmitted power, during periods of voice inactivity or reduced
activity. Finally, error-correcting coding permits the receiver to operate at a
lower signal-to-interference ratio, and antenna switching or adaptive techniques
provide spatial interference reduction.
All these tightly connected and integrated signal processing techniques are
either made possible or at least facilitated by spread spectrum signaling. They
provide a capacity, in terms of users per MHz per base station, which is several
times that of other multiple access techniques, along with extended coverage for
each base station or enhanced portable battery life, and better voice clarity.
This message has only recently been widely accepted and understood. QUALCOMM
began this development in 1989, held numerous demonstrations, large-scale field
tests, forums and debates over the next four years, in a climate that was
generally less than friendly. A number of political and public relations
roadblocks were erected by the competition to delay or derail the process. Only
a few service providers and fewer manufacturers agreed with us initially. But by
1993, with their help we had gained enough support to establish the TIA
standard, IS-95. Thereafter scores of manufacturers, for both handsets and
infrastructure joined the CDMA consortium, now known as "cdmaOne". Hong Kong was
the first region to be offered commercial CDMA and South Korea became the first
country where CDMA digital users exceeded the number of analog users. The vast
majority of North American operators have selected CDMA for digital telephony,
with continent-wide coverage and serving the majority of the population. Even
though service initiation has begun in earnest in the U.S. only this year, there
are already well over one million CDMA subscribers with service in over 100
cities of North America, two million subscribers in South Korea and five million
worldwide. There is now little risk in predicting that CDMA will ultimately
dominate the North American market and, even with the considerable head start of
the European digital standard GSM, that CDMA will gain wide acceptance in Asia
and South America as well.
Nevertheless, the negative propaganda and public relations campaigns continue.
This in spite of Forbes magazine recognizing its past misunderstanding of the
technology ("Wrong Call" October 6), and even the Wall Street Journal now
desisting from general and personal attacks and admitting instead on September
20 that "South Korea's bet on cellular phone technology may pay off." The
battleground seems to have now shifted to the trade press with an article in
Telephony magazine ("Blind Faith" September 8) rehashing for the nth time
previously discredited negative claims about CDMA capabilities, complexity and
cost. In fact, these claims are belied by the dozens of cellular and PCS
operators and their millions of customers which use the technology daily. But in
addition to the commercial success of CDMA, and the ever expanding universe of
manufacturers and service providers joining the "cdmaOne" organization, we now
have the grudging admission of the two dominant European (GSM and TDMA) wireless
manufacturers, and the dominant Japanese telecommunications operator, that CDMA
is a valid technology. In fact, they have gone as far as placing their
considerable power and prestige behind a 3rd Generation proposal to the
International Telecommunications Union, which uses a 4.1MHz spreading sequence
and which they have labeled Wideband CDMA (W-CDMA). So at this point, we should
simply declare "Victory" and retire. But not so fast, for this too is a clever
PR ploy. It lets our erstwhile detractors and future competitors label IS-95 as
the Narrowband CDMA standard. Now "wide" is a relative term. No one can argue
that 4.1MHz is wider than the 1.23 MHz of the IS-95 spreading sequence by more
than a factor of 3, but then 1.23 MHz is more than 4.5 times the 271K symbol
rate of GSM and more than 25 times the 48.6K symbol rate of the North American
TDMA standard IS-136. But, so what? Is wider necessarily better? Perhaps and
perhaps not. What matters is what one gains or loses by wider-spreading.
To begin with, we need to distinguish between voice telephony and high speed
data. It's easy to say that data at 2 Mbps is better served with a 4 MHz
modulation bandwidth than by a 1 MHz bandwidth, but, for digital telephony, very
high quality vocoders require only a maximum data rate of 8 Kbps to 13Kbps, and
with improved algorithms, over time this may yet be reduced to 4Kbps or less.
Spreading to 1.23 MHz provides a healthy processing advantage over interference
as well as multipath resolution of better than 1msec. Wider spreading permits
resolution of more closely spaced multipath components. But is this necessarily
good? Aside from the complexity required to acquire and track the additional
components, which may be minimized by advancing technology, there's the more
serious issue of channel measurement inaccuracy, which is inversely proportional
to the energy per component and hence increases as the number of multipath
components increases. Thus wider bandwidths and lower data rates (requiring less
transmitted energy) are counter-productive to measurement accuracy and for too
wide spreading they can thus cause overall degradation.
But enough of past and present standards battles, politically and economically
motivated. What about the future? Especially, what about data? Here, contrary to
common wisdom, I'll make another controversial distinction: "All bits are not
created equal." Voice and data can coexist, but to the detriment of both. Let me
hold off on data and on explaining the difference. Let's first explore
improvements in voice telephony. Only a knave or a fool would argue that what we
have today is the end-all of CDMA perfection. I have argued that IS-95 has
incorporated many critical and excellent features, which in fact, are being
imitated by the newly proposed standards just mentioned. Some of these proposals
involve just parameter changes, designed to counter the competition, without
noticing that it would also impact operators which have implemented CDMA,
forcing them to scrap their current systems and start over, hardly a palatable
alternative given their large capital investments. Other proposed changes may
also be ill conceived but for less obvious reasons; time prevents me from
elaborating.
Rather let me comment on improvements to IS-95 which are easily implemented and
are backward compatible to existing infrastructure. The first is to abandon the
reverse link Walsh-function modulation, which lends itself to non-coherent
reception at the base station, replacing it by a simpler biphase modulation,
aided by an auxiliary continuous pilot on the quadrature channel, which lends
itself to coherent demodulation at the base station. This, in fact, makes the
reverse channel quite similar to the forward channel modulation. Why then did we
not use this fairly commonplace approach for the reverse link in the first
place? Conservatism is the reason. We feared that in a rapidly varying multipath
channel, phase could not be tracked accurately unless an inordinate percentage
of the power was assigned to the unmodulated pilot; hence we opted for
noncoherent reception. This proved to be an overly cautious decision. Second, if
tight power control is a necessary feature, tighter power control is even
better. This has been found to be particularly true for the forward link from
base station to user, where power control was correctly deemed to be less
critical and hence made very infrequent, but this turned out to limit
performance. Also, means for reducing latency of the power control loops in both
directions, will considerably improve performance. Another improvement certain
to enhance capacity and coverage involves adaptive antenna array techniques. A
combination of spatial and signal processing, this will primarily benefit the
reverse link and will be rendered more feasible by the coherent demodulation
just mentioned. I'm convinced that spatial signal processing will be far more
effective at interference reduction, as well as simpler, than purely temporal
signal processing techniques such as interference cancellation. And finally, as
noted previously, vocoder algorithms will continue to improve, yielding lower
peak and average bit rates.
And last what you've all being waiting for - what about high speed data?
Everyone wants to connect to the Internet and why not by wireless means,
avoiding the nuisance of unfriendly line connectors for our laptops in each new
location. Here common wisdom dictates higher bit rates. ISDN is insufficient;
384 Kbps is better; so why not 2 Mbps? Faster is better. But is speed really the
main issue? For wireline connections, we buy a 56 Kbps modem and find that our
line will only support 24 Kbps. So what is the consequence? Latency: the delay
time we have to sit at our terminal waiting for the downloading from the Net to
be completed. But if we're used to varying rates, and hence latencies, from our
wireline modems, why shouldn't we expect this in wireless. If the base station
is in a closet down the hall in our office, we should expect a 2 Mbps
downloading rate, but if it's five miles away from our vehicle which is being
driven (by someone else) on an expressway, might we not be quite satisfied with
ISDN speeds or lower? This is what differentiates data from voice. We can
tolerate and even expect latencies of many seconds in receiving data, but voice
conversations are intolerant of delays any greater than 100 milliseconds. So for
voice telephony, we must allocate an inordinate amount of our common resources,
usually power, to the weakest users, thereby limiting the total throughput for
the collection of users served by one base station. With variable latency, we
have much more flexibility. To begin with, latency need not be inversely
proportional to data rate, because in a packet-switched network, we can assign
more packets to lower rate users. We may thus impose a "fairness criterion" that
the most disadvantaged user's latency be no greater than N times that of the
highest rate user. For voice, we must make N = 1. For data, studies with real
traffic and measured channel quality statistics show that we can more than
double overall throughput in the forward direction by letting N = 8.
The flexibility of packet transmission allows other improvements. One is the use
of longer interleaving depths, intolerable by voice because of the inherent
delay. In this context, I would be forgetting my heritage, if I did not mention
"turbo codes", a mixture of simple short convolutional codes, long interleavers
and better soft decision decoding, which permit data rates to approach within
60% to 80% of the Shannon coding limit (an amazing feat), thus increasing
current throughputs by more than 60%. Last, we should note that data
transmission is often asymmetric. We may wish to download megabytes from the
Web, but need only send kilobits worth of requests. For the forward link all
these data packet network features and improvements taken together may lead to a
forward link throughput enhancement of at least a factor of 4. This is in sharp
contrast with circuit-switched voice and serves to demonstrate why "all bits are
not created equal."
All this argues for the desirability of a separation of voice and data, which
assumes, of course, that customer data requirements approach those for voice.
For the near-term, data can coexist with circuit-switched voice, employing the
IS-99 and IS-657 data standards. The quite attainable goal then should be for a
smooth transition to packet-switched higher throughput data service protocols
with an air interface which is compatible with current voice and data services.
Let me conclude then by summarizing my various messages. Spread spectrum CDMA is
here to stay. The religious wars are over. The heretics have been confounded and
a few may even have been converted. Some conversions have been under duress and
some may not yet have fully shed their heresies, but it's only a matter of time.
Various improvements can and should be implemented, but Third Generation Systems
better be backward compatible if they're to find a willing market among
operators. Data transmission differs fundamentally from voice through relaxed
latency requirements. With Internet-type packet protocols, we can exploit these
differences to increase throughputs several-fold.
Data and voice users can coexist on the same carrier but to the detriment of the
capacity of both. As long as voice requirements dominate wireless demand, this
is a tolerable condition. As connectivity to the Internet or any other massive
data resource becomes the dominant application, great improvements in spectrum
efficiency will mandate the separation of voice and data on separate frequency
carriers with different network protocols. The technologies for this transition
are already available. Market forces and operator choices will determine how
rapidly they are deployed. |
