Index Home About Blog
From: karn@Qualcomm.COM (Phil Karn)
Subject: Re: Digital Cellular Telephony
Organization: Qualcomm, Inc
Date: Wed, 22 Jan 1992 20:51:42 GMT

In article <>, (Bob
Denny) writes:

>> Would someone like to explain how CDMA works?  I understand that it
>> spreads the signal across a huge bandwidth in a way that allows the
>> receiver to select 1 of N signals ... but how?  A CDMA primer, anyone?

> I do not know the details of the specific system being proposed, but I
> am familiar with "spread spectrum" systems, of which the proposed
> cellular "CDMA" system appears to be one. So I'll take a stab at some
> basics in hopes someone else more knowlegable will expand on or
> correct ...

Since I work on the project, I *do* know the details. :-) Many have
already been published; probably the best reference is "On the System
Design Aspects of Code Division Multiple Access (CDMA) Applied To
Digital Cellular and Personal Communications Networks" by Allen
Salmasi and Klein S. Gilhousen, Proceedings of the 41st IEEE Vehicular
Technology Conference, St Louis MO May 19-22, 1991.

Several other papers on CDMA appeared in the May 1991 issue of IEEE
Transactions on Vehicular Technology (Special Issue on Digital
Cellular Technologies). One is an overview of the generic concepts,
another is an analysis of the CDMA system capacity.

All of the info I'm about to give is described in these papers or in
other public forums, so there's nothing proprietary here.

> There are two common forms of CDMA systems: direct-sequence and
> frequency-hopping. I will describe the direct-sequence variation. I do
> not know which is used by Qualcomm. The principles are the same,
> though.

We use direct-sequence PN spreading at 1.2288 megachips/sec. The
spreading bandwidth is 1.23 MHz at 3dB, and that's also our channel
spacing. For a data rate of 9600 bps, this yields a processing gain of
about 21 dB.  There are actually two spreading functions: a "short
code" that is 32768 chips long, and a "long code" 2^42-1 chips long.
Both use the same chip rate.  The signal transmitted by the cell site
includes a "pilot" consisting of a carrier spread only by the short
code; this allows the mobile to rapidly acquire it and use it for
system timing. Actual traffic is spread by both the short and long

> CDMA works by first digitally encoding the audio signal, resulting in
> a stream of bits. Various tricks are used to compress the audio so as
> to minimize the bit rate at the expense of fidelity.

Actually, I dare say our vocoder sounds pretty good.  It runs at a
variable rate (fast when you talk, slow when you stop), and
transmitted power is proportional to this rate.  Since power is
capacity in CDMA, this gives us a nice boost in capacity; whenever a
given user isn't talking (about 60-65% of the time, on average), his
capacity share can be taken by some other user who *is* talking.

[reasonably accurate description of spread spectrum omitted]

> If you aren't confused at this point, great. Now here's why
> spread-spectrum is so great. First, the fact that the signal's energy
> is spread across a very wide range of frequencies makes it very
> resistant to multipath. I won't go into the explanation of why, but
> trust me, a properly engineered CDMA system will be virtually immune
> to flutter and fading, even in the worst urban high-rise areas.

Instead of being a nuisance, multipath can actually be a major asset
in a CDMA system. Our receivers actually consist of three independent
spread spectrum receivers in parallel, and these receiver "fingers"
can be independently assigned to different multipath components which
are then combined before detection. This is an extremely effective

There's a special case of "multipath" in our system called "soft
handoff".  If you're in the region midway between two or more cells on
the same frequency, the system sets up a path through them.  The
additional cells' signals appear to the mobile just as though they
were additional multipath components from the first cell (actually,
this isn't quite true -- the signals are deliberately offset in time
to avoid possible cross-correlation, but the basic idea is that you
combine the signals from multiple cells just as you would combine
multiple signals from a single cell).

We also use some pretty strong forward error correction (FEC) coding:
rate 1/2 K=9 convolutional coding with Viterbi decoding on the forward
link and rate 1/3 K=9 on the reverse link. This helps us tolerate
errors and lets us operate with a very low average signal-to-noise
ratio (about 7dB). Again, since power is capacity in CDMA, this boosts
our overall capacity.

Another really nice thing about CDMA is that due to the selectivity
provided by the processing gain of spread spectrum, frequencies can be
reused in every cell site.  Current FM cellular systems typically are
able to use only 1/7 of the total number of channels in each cell
because of the need to protect their immediate neighboring cells from

> Secondly, the PN sequence can be generated with an encryption key and
> provide a reasonable level of security. I have no idea if the Qualcomm
> proposal includes security provisions, though.

The PN sequences are generated with linear feedback shift registers.
In article <>, (Steve
Forrette) writes:

>> What it works out to is that while an analog phone transmits at three
>> watts, and the cell-site power requirements are truly horrible, CDMA
>> phones transmit in the milliwatt range, and the entire cell site
>> transmits with only as much power as a couple of the radios in the
>> analog cell.

> Again, not always.  The remote AMPS transmits only at the necessary
> level.  In fact, my handheld is always operating in the "milliwatt
> range."

Changing the transmitter power of a mobile in AMPS requires a "blank
and burst" data trasmission from the cell. This is plainly audible to
the user, so you don't want to do it too often. In CDMA, however,
power control is continuous, fast and completely inaudible to the

In article <>,
(John Nagle) writes:

>> In the long term, digital phones should be cheaper and lighter,
>> however, as VLSI components are employed.

> In the Motorola MicroTAC, the battery is over half the weight.
> Further VLSI integration won't shrink the package that much.

> Actually, the coils and filters probably weigh more than the
> semiconductors already.  Look at the Philips/Signetics chip set, for
> example.  It's down to 12 shrink surface-mount packages.

> Progress will have to come from reducing power consumption.

This is truer than you may realize, John. One of CDMA's biggest
advantages is its very low RF power levels. As Ron Dippold mentioned
earlier, we typically see the mobile RF transmitter operating at the
milliwatt (0dBm) level or lower; -10 dBm is quite typical, and I've
seen it as low as -50 dBm (10 nanowatts) when very close to a cell

All this translates directly into longer talk time and/or a smaller
battery for a CDMA portable phone.


From: (Phil Karn)
Subject: Re: Digital Cellular Telephony
Organization: Qualcomm, Inc
Date: Tue, 28 Jan 1992 03:51:45 GMT

I just found another published reference on the Qualcomm CDMA system.
It's called "Mobile Power Control for CDMA" by Klein S. Gilhousen, and
it appears in the January 1992 issue of Communications magazine.

It describes the mobile's power control system in pretty good detail.
There are two components, open loop and closed loop, with the former
doing most of the work. The mobile simply varies its reverse link
transmitter power in inverse proportion to the total energy received
on the forward link.  The nominal transmit power in dBm is -73dBm^2
minus the received power in dBm. ("Reverse link" == mobile-to-cell,
"forward link" == cell-to-mobile.) Various constants are added to the
nominal -73 dBm^2 figure for each cell to account for differing cell
transmit powers, etc.

Multipath fading tends to be very frequency selective, so in a
narrowband system like TDMA or AMPS it's quite common to be in a
complete null on, say, the reverse link frequency while you're hearing
the forward link just fine. (It's easy to see this frequency-selective
phenomena on on a spectrum analyzer when you ride around town -- you
see little comb notch filters running around the spectrum.) But CDMA
is a broadband system, so it's quite rare for multipath to take out an
entire 1.25 MHz signal. So the forward and reverse links are much more
closely correlated in CDMA than they are in a narrowband system like
TDMA or AMPS. This is what makes open loop power control so highly
effective in CDMA, but not TDMA or AMPS.

But even in CDMA you sometimes have assymetrical paths; e.g., the
noise level at the mobile might be different than at the cell, or
despite the wide signal, there might be a multipath fade that affects
one link more than the other. By itself, open loop power control
brings the reverse link to within about 6 dB of the desired level at
the cell, but this isn't good enough.

Finer control is achieved with a closed-loop mechanism whereby the
cell site measures the signal-to-noise ratio of the mobile and tells
it (via a bit inserted into the forward data link) to adjust power up
or down. It does this at 1.25 ms intervals (800 Hz); each step is
nominally 0.5 dB, so the closed loop mechanism can slew at +/- 400
dB/sec. When the desired power is achieved, the closed loop bit stream
simply consists of alternating 0 and 1 bits. I.e., it's just "delta
modulation". The total closed-loop correction is typically limited to
+/- 24 dB.

The power control bits are sent by simply "puncturing" the forward
data stream, i.e., sending a power control bit in place of the
original data bit.  The user data stream is not signficantly affected,
however, because it has already been interleaved and convolutionally
encoded; the Viterbi decoder at the receiver can easily regenerate the
original data stream without additional errors.

Using delta-modulation for the power control bits has a nice side
benefit: resistance to errors.  The power control bits cannot be error
protected along with the user data because the extra delay would be
intolerable. But due to the inherent robustness of the delta-modulated
control loop, even a high error rate in the power control bits simply
causes the loop to take a little longer to arrive at the correct power
level. In actual field tests, the average difference between actual
and desired SNR at the cell site was essentially zero, with a standard
deviation of about 1.5 dB.

All in all, the CDMA power control scheme is simple, elegant and quite
effective, as shown in our recent field tests. In the beginning, all
of CDMA's critics seemed to go right for the power control issue, but
no longer ...


From: (Ron Dippold)
Subject: Re: Digital Cellular Telephony
Organization: Qualcomm, Inc., San Diego, CA
Date: Thu, 9 Jan 1992 19:26:43 GMT (Anindadeb Vijaykumar Dasgupta) writes:

> I read somewhere that in most large cities cellular carriers are
> switching to digital systems due to saturation of existing cells.

They're moving towards it, as fast as they can.

> I couldn't tell what the advantages of this would be:

> With analog transmission, each equipment would need 4 KHz. while with
> 8 bit PCM, 64 Kbps would be needed, which would surely translate to a
> higher frequency than 8 KHz.  This definitely does not free up any
> bandwidth.  Are these carriers using source coding?  Won't that make
> the cellular phones more expensive/bulky?

Okay, you're assuming that the analog phone is much more efficient in
bandwidth than it actually is.  In actuality, each phone completely
occupies a 30 KHz channel, and each phone is transmitting at thre
watts for marginal quality.  And it still sounds bad.

With digital encoding done correctly (so you can use Viterbi
decoding), you need much less power to get your data through (you're
just looking for on-off instead of a FM sound waveform).  In addition,
if you're using CDMA, CDMA doesn't require that you divide up your
bandwidth into bandwith wasting 30 KHz chunks.  Intead, each phone
uses exactly as much bandwidth as it needs.  More phones just mean
more "noise" (not noise heard by the user, but as in signal to
noise ...)

For example, if you had 50 users that needed computers, then the
analog analagous way to do it would be to buy two mainframes, and let
only one user at a time use each computer, even if all they needed to
do was simple word processing.  The CDMA way would be to buy one
computer that was 20 times more powerful and give each user a
terminal, so they only use as much of the computer power as they need,
and everyone can use it at once.

Even better, we use active power control (something we can do because
it's digital).  The mobile and cell channel elements transmit with
only as much power they need.  If you're close to the cell, you don't
need much power.  As you get farther away, it slowly boosts power.

Finally, with digital we can (and do) use a variable rate vocoder.  In
this way CDMA makes use of the Voice Activity Factor of conversation:
either of the parties involved is not saying something about 60% of
the time.  (Interestingly, even when we force the vocodoer to use a
maximum rate of half its top rate, it still sounds better than my
analog phone).

What it works out to is that while an analog phone transmits at three
watts, and the cell-site power requirements are truly horrible, CDMA
phones transmit in the milliwatt range, and the entire cell site
transmits with only as much power as a couple of the radios in the
analog cell.

Bottom line: We have an officially capacity tested (tests observed by
the major companies in the industry) CDMA system in the field that
gives a capacity improvement of 10 to 30 times (depending on
conditions) over an AMPS system, with better voice quality, better
handoffs, and less dropped calls.

Efficient use of resources is the key (plus a lot of geniuses in the
theory department, and then ignoring those who claimed we couldn't do

One further advantage: The digital medium is a lot more flexible.
When an AMPS phone has to transmit control information, the voice
blanks out.  With CDMA, we can just vocode at half the normal rate and
send the control information in the other half of the frame.  Result:
undetectable loss of voice quality instead of complete loss of voice.

In addition, we can divide the channel between different data sources,
so you could send voice and data (from a modem, perhaps) on the same
channel.  What we can do is limited only by the messages we can think
up to send back and forth.

Another voice quality advantage: because it's a digital vocoder,
rather than sending the analog waveform out, we can more easily do a
lot more filtering on it.  For example, continuous background noises
(such as a car engine or the wind) can be severely reduced.

There are further advantages, but that should be enough.

From: (Ron Dippold)
Subject: Re: Digital Cellular Telephony
Organization: Qualcomm, Inc., San Diego, CA
Date: Tue, 14 Jan 1992 23:48:12 GMT (Steve Forrette) writes:

> In article <> it was written:

>> Even better, we use active power control (something we can do because
>> it's digital).  The mobile and cell channel elements transmit with
>> only as much power they need.  If you're close to the cell, you don't
>> need much power.  As you get farther away, it slowly boosts power.

> Just the way AMPS works today.

Not exactly.  I've worked enough with AMPS to be familiar with the
AMPS power control methods. There are significant differences in the
way it happens in AMPS and in CDMA which significantly affect the
overall power consumption, voice quality, and reaction to areas with
bad coverage.  I wish I could go into this further without giving the
lawyers a heart attack.

>> In addition, we can divide the channel between different data sources,
>> so you could send voice and data (from a modem, perhaps) on the same
>> channel.  What we can do is limited only by the messages we can think
>> up to send back and forth.

> What advantage is this going to be to the individual subscriber?
> Answer: the same advantage that individual subscribers get from ISDN:
> nothing!  The carriers will undoubtedly price the data services at

Yep, that's up to the carriers.  The possibilities are there, however.
If they price it too outrageously, you can always switch the phone to
AMPS and use your trusty CellBlazer.

> A user with a 2400 baud dialup or 9600 baud fax need will have to pay
> whatever rate the carrier sets for these services, as they will not
> operate properly over the voice channel.  And the user has to pay for
> more-expensive equipment to do this, at least initially.

Hey, it's a status symbol!  Only a half-smiley on that.  What we need
is a way to make it obvious from the outside of the car that you've
got one of these ...

From: (Ron Dippold)
Subject: Re: Digital Cellular Telephony
Organization: Qualcomm, Inc., San Diego, CA
Date: Wed, 15 Jan 1992 20:53:33 GMT

meier@Software.Mitel.COM (Rolf Meier) writes:

> In my personal opinion, Qualcomm's claims are designed more to raise
> the price of Qualcomm shares than actual capacity advantages over

Please, we invited all the major carriers and phone companies and some
manufacturers here to observe our formal system tests in November and
went through large-scale capacity testing with actual CDMA mobiles in
cars and vans in the field and multiple CDMA cells.  An actual working
CDMA system.  The results were presented to the CTIA.  If you'd like
to raise doubts with our capacity tests, you're going to have to do
far better than that.

What you'll have to do if you want to dispute CDMA's capacity over
TDMA is let us know the results of all the similar formal capacity
tests with actual mobiles and cells which have no doubt been performed
for TDMA.

From: (Ron Dippold)
Subject: Re: Digital Cellular Telephony
Organization: Qualcomm, Inc., San Diego, CA
Date: Mon, 20 Jan 1992 01:59:59 GMT

mole-end! writes:

> Would someone like to explain how CDMA works?  I understand that it
> spreads the signal across a huge bandwidth in a way that allows the
> receiver to select 1 of N signals ... but how?  A CDMA primer, anyone?

Okay, let me explain as much as I can without (a) giving away anything
proprietary, and (b) getting too technical.

First, think of a standard system with frequency multiplexing or time
multiplexing.  You know where to find the information by looking at a
specific time or frequency.  This is easy to implement, but if
whatever you have allocated that frequency or time slot to doesn't use
all of it, you have wasted capacity.  And because you usually need a
"guard" to make sure that slots don't bleed over into each other, you
have built in inefficiency.

(Note: CDMA has been implemented before in bulky and expensive
military systems.  However, anything here about CDMA will refer to
Qualcomm's consumer CDMA technology).

CDMA (Code Division Multiple Access) is frequency multiplexed in the
sense that one CDMA band takes up a 1.25 MHz frequency band, so that
if you need more CDMA channels you allocate another 1.25 Mhz band.
However, within that band, it's a free for all.  All transmissions are
spread and stuffed all across the band, seemingly beyond recovery.

Each transmission is spread across the band by encoding it with a
pseudorandom sequence of ones and zeros that can be calculated from
the absolute system time.  Different "channels" are obtained by using
an offset from this sequence.

Now here's the beautiful part.  The recieved signal is demodulated
using the same sequence.  If you demodulate with one sequence of ones
and zeros, _only_ those signals that have been encoded with that
sequence will fall out.  All other signals look just like background
noise.  So if you have seven cells near each other, they can all share
the same frequency band.  They just use a different offset.  When the
mobile powers up it attempts to find a cell at each of the possible
offsets until it finds a transmitting cell.  You can get channels
within a cell by offsetting another small "distance" in the sequence.

So the limit to your capacity is a "soft" one.  Each new transmitter
adds more "background noise" for other transmissions and makes them
more error prone.  You have reached your capacity when you have
reached your limit of acceptable transmission errors.  The cell has
all the quality information and can determine when not to allow any
more mobiles on the system.  In emergency situations you can always
allow a new mobile on, however, and allow it to transmit with higher
power to ensure it gets through.  This makes other calls worse, but in
an emergency situation this should be worth it.

We use some other techniques to channelize and to improve error
correction, but the basic technique is fairly simple: spread the
signal with a unique sequence so that only by demodulating with the
same sequence produces a coherent signal.

I realize I'm being vague.  However, patents have been filed, and the
CTIA will be reviewing our CAI (Common Air Interface) spec, which
describes everything in detail (it's the CDMA equivalent of IS-54).
As soon as they get through with it and it is released, everything
will be available in _excruciating_ detail (it makes excellent reading
to put you to sleep).

From: (Ron Dippold)
Subject: Re: Digital Cellular Telephony
Organization: Qualcomm, Inc., San Diego, CA
Date: Wed, 22 Jan 1992 00:08:18 GMT (Don Newcomb) writes:

> OK. So they are probably using a Linear Maximal Pseudorandom Sequence
> to modulate and demodualte the signal. But, doesn't this imply some
> rather heavy duty signal processing? Is a matched filter adequate to

We did the prototype system with a whole bunch of DSPs, and the code
is enough to give you nightmares.  The current system uses a custom
CDMA ASIC (Application Specific Integrated Circuit) to do the
modulation and demodulation.  This allows us to do whatever is
necessary to get the best possible signal even with fading, multipath,
and a whole host of other nasty things that happen when you're driving
at 65 MPH through a canyon.  A simple filter just would not have been

From: (Ron Dippold)
Subject: Re: CDMA Impact on Cellular Software
Organization: Qualcomm, Inc., San Diego, CA
Date: Fri, 31 Jan 1992 22:13:35 GMT (Mitsutaka Ito) writes:

>   I would like to know CDMA technology impacts on base-station
> software and mobile unit software.  I will appreciate any information
> on this topics.  Thank you in advance.

"Impacts" is a bit fuzzy.  Basically, the software is far more
complex, if you mean as far as that goes.  FM is simple enough that
you can easily program the phone on a simple microprocessor in
assembly language.  Speaking as one of the CDMA programmers, I think
you could write a CDMA system in assembly, but I sure wouldn't want to
do it.  "Living hell" comes to mind.  We wrote everything in C (and
I'm sure we'll get some snide remarks about C being structured
assembly ...)

Comparing CDMA software to FM software is pretty much like comparing
the control software for a USR Dual Standard modem to a 300 bps modem.
You get far better performance in exchange for more complexity.

From: (Ron Dippold)
Subject: CTIA Board Asks TIA For CDMA Standard
Organization: Qualcomm, Inc., San Diego, CA
Date: Mon, 15 Jun 1992 20:32:58 GMT

 From a CTIA press release, June 10.  Any typos are due to my


Washington, D.C. -- The CTIA Board of Directors today asked the
Telecommunications Industry Association (TIA) to immediately begin
development of a technical standard for a Code Division Multiple
Access (CDMA) digital technique.

The Board stipulated that this request not slow the TIA's continuing
work on enhancements to Time Division Multiple Access (TDMA), another
digital technique that already has been standardized by the cellular

The Board further resolved to launch a separate association effort,
unrelated to either TDMA or CDMA, to determine what standards will be
needed to support the enhanced wireless services that cellular will
provide in the future.

[ other agenda actions ]

The Board's vote on the request to TIA follows a six-month study
conducted by the CTIA Technology Committee's wideband/spread spectrum
subcommittee.  During this study, three companies presented
information on their CDMA techniques.

"CTIA's job is to foster new technologies, but not to endorse any
single company's approach to a technological concept," said Thomas E.
Wheeler, the president and CEO of the Cellular Telecommunications
Industry Association.  "The specifications for a CDMA standard will
now be developed by TIA and CTIA stands ready to assist."

The TIA, which represents equipment manufactureres, has organized a
subcommiuttee to write a new standard in response to a CTIA Board
request of last January.  That subcommittee, known as the TIA 45.5
panel, already has begun setting the stage for development of a CDMA

[ description of digital cellular at a very basic level ]

There are different techniques for digital transmission, however.  In
January 1989, the cellular industry selected TDMA as the first digital
technique to go through the TIA standards-writing process.  Now, the
board has agreed that CDMA should be the second digital technique to
be standardized.

For additional information:  Contact Norman Black or Lynne Mallonee at

                    ---------  End of Release --------

While this may not sound like much on the surface, it's incredibly
important once you read between the lines.  We've never doubted the
CDMA technology, or our ability to pull it off, but the politics has
always been a big red "?".  It's 1989, and the industry has
standardized on TDMA, and tens of millions of dollars have been spent
on it.  Suddenly a small company comes out of nowhere with another
method, which they claim is far superior -- you can see the precarious

Things finally came to a head this past November, when we did our
system capacity testing with most of the industry invited, and showed
everyone that CDMA did as well as we claimed under actual field
conditions, and better, and let them see for themselves.  We then had
the most extensively tested system, at least as far as publically
available figures, and I don't believe that has changed.  Nor have we
been resting, and we've converted quite a number of our erstwhile

And now this, which is something akin to the bastard child (we've been
called far worse, I assure you) being legitimized.  The other child
isn't too pleased, as his share of the inheritance isn't guaranteed
anymore.  The politics continue, but we've gained one.

Actually, it's either interesting or frustrating, depending how you
look at, that so much has depended purely on the political savvy of
our president and CEO and other high-level people.  "Build a better
mosetrap and the world will beat a path to your door," hah!

Index Home About Blog