IP: Trade-offs vs wishful thinking (compression and UWB)

[David Farber <dave () farber net>]

> Date: Wed, 09 Jan 2002 20:08:24 -0800
> To: dave () farber net
> From: Lars Poulsen <lpoulsen () ucwireless com>
> Subject: Trade-offs vs wishful thinking (compression and UWB)
>
> Lay people and the press generally fail to understand that marvellous
> technical advances are not miracles: They are born out of trade-offs.
> In the case of data compression, the tradeoffs involve
> - file size (compression ratio)
> - processing time (CPU power)
> - size of storage buffer (dictionary)
> - latency
>
> Data compression is not magic: It is an analysis of data to remove
> redundancy, and is valid within a defined application space where
> the redundancy is understood. Indeed, if you start with data such as
> - bitmaps of pictures (TIFF or BMP)
> - english text (or any other language)
> - computer programs
> - HTML pages
> ... it is very easy to remove enough redundancy to reduce the file size
> by a factor anywhere between 4 and 10. In the case of pictures, it has
> been standard for the last 10 years to always store and move these
> files in compressed form. The rest of the above tend to be small files,
> compared to the aggregate of what we are moving around today.
>
> The large data streams that fill up our data pipes these days are
> already compressed: GIFs, MP3s, video streams. Any further gains
> must come from recognizing repetitions of the same files. Analyzing
> network usage, one can derive models that indicate that compression
> factors in the 10-100 range are possible with Gigabyte sized
> storage buffers; the true bottleneck is memory bandwidth between
> the comparators and the storage buffer. However, the operational
> behavior of such a compression engine would be very similar to
> that of a simple web cache (like squid) which is much simpler to
> build.
>
> Two severe limitations on a black box compression appliance,
> (which can serve an arbitrary bit stream, not just HTTP file access)
> are that  it works only for a point-to-point link and it requires an
> error-free link. If data may have been lost, the dictionaries must be
> reset, and the compression factor goes back to 1:1 and then climbs
> slowly back until the dictionary has been rebuilt. For a T3 link
> (43 Mbps) and a Gigabyte dictionary, it would take several minutes
> to recover from such an event.
>
> I apply similar skepticism to all the great projections of unlimited
> bandwidth in unlicensed wireless. This is also a game of trade-offs:
> - power and sensitivity
> - distance
> - jam-resistance
> - speed
> When you push on one of these dimensions, it comes out somewhere
> else. We hope for co-existence between outdoor (fixed) broadband
> internet access, indoor 802.11, HomeRF etc. In principle, this can
> be achieved with spread spectrum of either the FH or DS variety,
> but only by sacrificing some amount to link speed for forward error
> correction at one level or another of encoding. The parameters can
> be tweaked in any direction, but everyone always wants to hear that
> you can get the good numbers in all dimensions simultaneously.
> Sorry, that just won't happen.
>
> We have come a long way in 10 years of DSP development, and
> there may be room for more stretch, but I am still waiting to see
> plausible confirmation of the practical implementation of UWB pulse
> modulation. As far as I can see, the pure pulse modulation shifts
> the receiver bottleneck to the problem of maintaining time synchro-
> nization to an unprecedented degree. IIRC, the Time Domain web
> site talks about synchronizing to picosecond accuracy.
> (But then again, I'm not a DSP or RF engineer, so I don't really
> understand either type of modulation.)
>
> I do believe, is that there will be progress when we focus
> concentrated effort on a new approach (such as shifting from
> multi-threshold level detection to time synchronization); I'm less
> sure that it will come in exactly the predicted way.
> / Lars Poulsen                  E-mail:    <lpoulsen () UCwireless com>
>   Senior Software Engineer      Telephone: +1-805-964-5848 ext 279
>   UC Wireless, 323 David Love Place, Santa Barbara, CA 93117, USA

For archives see:
http://www.interesting-people.org/archives/interesting-people/