A very clear explaination of how cell systems work

[David Farber <dave () farber net>]

Begin forwarded message:

From: Phil Karn <karn () ka9q net>
Date: February 10, 2010 8:29:30 PM EST
To: Bob Frankston <bob2-39 () bobf frankston com>
Cc: dave () farber net, "'ip'" <ip () v2 listbox com>, karl () cavebear com, Phil Karn <karn () ka9q net>
Subject: Re: [IP] Re: Broadband truth in advertising, redux

On 2/9/2010 1:29 PM, Bob Frankston wrote:
> If this is true then we have another example of how the cellular
> architecture creates scarcity by forcing us back into the circuit model
> thus losing the benefits of best efforts and sharing. Sort of like
> having to dial up again for each web page.
>
> I'd like to hear from those with technical expertise about how the
> various protocols are actually implemented. It would make a lot more
> sense to allocate the resources on demand and cede them for idle periods
> and reacquire them. This is very different from running PPP over a
> dedicated circuit.  One can still hold the IP address though they
> latency might vary.

Yes, CDMA cellular has connection-oriented traffic channels. But the
data rate is variable to take advantage of a variable rate codec that
adapts to speech activity.

There are four data rates in the voice-type CDMA systems: full, half,
quarter and eighth. When you stop talking, the channel idles at 1/8
rate. With voice, this carries "comfort noise" so it doesn't sound like
the other party got cut off. For data, why not shut it down entirely? To
keep the closed-loop transmitter power control system going.

To adapt to rapid changes in path loss and interference level, the base
station constantly computes the received Eb/No (digital signal-to-noise
ratio) and tells the mobile to increase or decrease its transmitter
power with a single non-error-protected bit sent at 800 Hz. (It's
actually "punctured" out of the FEC-encoded data stream.)

A '1' tells the mobile to increase power by 1 dB, a '0' tells it to
decrease power by 1 dB. There's no error protection because that would
introduce delay. It's also unnecessary. If an error occurs and the
mobile adjusts power in the wrong direction, the base station will
simply tell the mobile again to adjust it in the right direction. So as
long as the error rate is less than 50%, the loop will converge.

1xEVDO, being designed specifically for data, works a little
differently. The transmitter power and symbol rate remain constant and
the modulation and FEC (turbo) coding are changed over a wide range to
vary the data rate according to feedback from the receiver. There's also
an adaptive ARQ scheme whereby the transmitter sends additional parity
bits for a given frame until the receiver acknowledges a successful decode.

WiFi is similar in that it also keeps the transmit power constant and
varies the data rate according to channel conditions.

For constant signal and noise power, halving the data rate increases the
Eb/No (per bit signal-to-noise ratio) by 3 dB. Increasing the data rate
by 10x drops the Eb/No by 10 dB, and so on. For both WiFi and 1xEVDO the
relationship between SNR and data rate is not exactly linear because the
faster modes require higher Eb/No to fit into a constant channel bandwidth.

Both the voice (IS-95, CDMA-2000) and data (1xEVDO) flavors of CDMA send
IP over traffic channels with PPP. Nothing says we can't drop the
traffic channel entirely after an inactivity timeout, and in fact we do.
It just takes a little more work to get it back when new traffic arrives.

To establish (or re-establish) a traffic channel the mobile transmits a
short request message on a shared random-access access channel. Without
active power control the mobile can only estimate the proper transmitter
power level from the base station received power. The estimate is Ptx =
73 dBm^2 - Prx: the sum of the received and transmitted powers, in
decibels relative to 1 milliwatt, should be 73 dBm squared.

To minimize interference to other mobiles in case the estimate is high,
the mobile sends its first message on the access channel at a power
below the open-loop estimate, backing off and increasing power after
each unsuccessful attempt.

So the inactivity timeout is obviously a compromise between dwelling too
long on a traffic channel and repeating the access procedure to get
another traffic channel when new traffic arrives. But CDMA can still
handle connectionless IP traffic just fine, and because of the variable
rate CDMA traffic channel it does so much more efficiently than a
connection-oriented system like GSM with constant rate traffic channels.

I note that every mobile radio channel has a very similar set of
problems, especially rapid and unpredictable changes in signal-to-noise
ratio, and because you can't predict them or even your future traffic
demands you have to use heuristics like inactivity timeouts.






-------------------------------------------
Archives: https://www.listbox.com/member/archive/247/=now
RSS Feed: https://www.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com