[PLUG-TALK] Satellite

[Keith Lofstrom]

> On 2012-07-27 17:05:52, Keith Lofstrom <keithl at gate.kl-ic.com> wrote:
> > I challenge that.  Wireless on a two dimensional surface with 
> > fixed omnidirectional antennas is indeed limited as you say. 
> > Wireless in three dimensional space with beamforming antennas
> > offers very high bandwidth, with no need for monopolies.  The
> > switches must be off the surface - in the stratosphere or in
> > orbit.  In the 38 GHz band (8mm wavelength), with 5 Gbps/channel,
> > a one square meter phased array antenna can see about 10,000
> > different patches of sky.   A 100 square meter antenna at
> > stratospheric height can see a million different patches of
> > ground, perhaps 100 square kilometers, or 100 meter square
> > patches.  So, we can deliver 100*10K*5Gbps per square kilometer.
> > Portland averages 2000 people per km^2, maybe 50 times that in
> > the high rise core.  So, worst case, we are dividing 5E15 bps
> > by 100K people, or 50 Gpbs each.  If that is not enough
> > bandwidth, we make our stratospheric antennas bigger.

On Sun, Jul 29, 2012 at 05:26:05PM -0700, Bryan Linton wrote:

> I'm curious, the bandwidth of a solution like this would be quite
> impressive, but what would the latency be?  
> 
> The primary complaint of current satellite service is that the RTT
> of a packet is often measured in *seconds* rather than
> milliseconds.  In the past, many hardcore gaming geeks have signed
> up with DSL providers and asked the ISP to disable interleaving on
> the line.

Depends on where your satellite is.  Ping time through a bent pipe
communication system to another spot on the ground is 2 trips up and
down,  four path lengths.  Light in air and vacuum moves at 0.3
meters per nanosecond, 300 kilometers per millisecond. For two 
neighboring endpoints,  4 trips is 1 millisecond per 75 kilometers
distance from the satellite.

An overhead stratospheric local service at 15 km altitude would have
0.2 milliseconds ping time.  The DoD is building huge solar powered
radar blimps now;  megawatts of radar power.  A smaller blimp is
needed for communications, with phased arrays needing a few watts
to produce multiple narrow beams.

Globalstar orbits at 1400km altitude, low MEO.  At 30 degrees elevation,
a single path length is 2800 km (to a terrestrial point next door,
or up to 5000 km away).  The bent pipe ping time is 37 milliseconds.  

GEO comsats orbit at 36,000 km altitudes.  Path lengths are on the 
order of 40,000 kilometers.  That is 530 milliseconds for bent pipe
ping time.  GEO comsat internet often has a terrestrial uplink
adding to the delay, but the main delay is a massive amount of
queueing (keep those transponders busy!), with your upload packets
waiting behind a bunch of other upload packets during high usage. 
I can well imagine multi-second delays, not in the speed of light
path to the satellite, but waiting in RAM at the uplink site. 
So - don't do that!

The server sky constellations (http://server-sky.com) won't be
bent pipe communications, but the actual server endpoints, at 
6400 km altitude, with path lengths of 10,000 kilometers to
latitude 45N.  For endpoint services, only one trip up and down. 
That results in 67 msec ping time to the server, 133 millisecond
if the server is coordinating communications between two users.
That is probably too slow for some gamers, but then, so is 
gaming across the Pacific, and the Shenzhen goldfarmers still
manage to earn their keep.

But most of internet usage is not gaming; a lot of it is streaming
data, interacting with servers, etc.  I am focusing on markets in
the developing world where data communication is practically
impossible, or can take hours (flash drives and bicycles). 

There's a lot more at the server sky website, but the main part of
the plan is to stop building multiton satellites with 1950s aircraft
technology, and instead build cooperating swarms of thousands of 
gram-weight satellites with modern semiconductor technology.  The
cost per kilogram is about the same, but the function-per-kilogram
can increase 100 to 1000 times.  A cooperating swarm acts as a
sparse phased array antenna - each thinsat produces a few milliwatts
of 5 degree beam energy (and multiple simultaneous beams)   The
swarm as a whole (perhaps a kilometer across) constructively adds
up to a few watts of energy, in a 10 microradian-wide beam.  At
10,000 kilometers distance, at 38 GHz, that produces a ground spot
about 100 meters wide at the 3dB points.

The vast majority of the signal will be splattered into sidelobes,
but incoherently and at very low power levels relative to the main
beam.  Spread spectrum correlated receivers can receive signals 
many dB below the noise floor. 

The whole shebang is dependent on very accurate timing and signal
averaging, but I've helped design systems that deliver 10,000
simultaneous pulses with femtosecond accuracy through a lot of
electrical and mechanical noise.  Construction bulldozer blades
are positioned to a fraction of a centimeter with differential
GPS.  We can measure the distance to LAGEOS satellites within
a fraction of a micron.  Welcome to the age of hyperprecision
electronics.  Comsat builders are stuck in the microsecond age; 
your bicycle is more precise.

Speculative technology, sure.  All technology begins that way.

I expect more success than attempting to get a cashstrapped
city to arrange bond loans for the full cost of fiber internet
to hundreds of thousands of endpoints over 377 square kilometers.
Even much richer internet-boom Amsterdam paid less than a third
of the cost of connecting a few percent of its far denser urban
core, then handed the fiber over to an Italian management company.

A more realistic goal might be reenergizing the terabit of dark
fiber already feeding the Convention Center from Seattle, built
for Supercomputer 2009.  Then work outwards through the Lloyd
District, feeding the businesses and hotels that support the
conventions.  Then the high rises in the Pearl (do they already
have CenturyLink FTTH?  Much new buildout does).  Put internet
on the bridges and public buildings, not for websurfing, but
connecting instruments that look for corrosion and concrete
decay, prolonging lifespan and facilitating maintenance.  Then
spread out, a neighborhood at a time, to the rest of the city,
as bake sales and neighborhood activism permit.  If you want
to claim Amsterdam as a precedent, don't expect to reach
farther and faster than they did, with less funding.


> Nevertheless, fiber-optic cabling is a proven, reliable,
> high-bandwidth, low-latency, and low-cost technology.  While I can
> foresee many practical uses for high-bandwidth satellite
> connections, I do not foresee them supplanting fiber-optics any
> time soon.
> 
> Complimenting them, certainly.  Replacing?  Not anytime soon.

You raise many good points, and show a good understanding of
how fiber works in dense and populous first world cities
for entertainment uses.

I'm not anti-fiber, where it makes sense, and where there
can be competition.  Indeed, a few decades ago I helped
organize (over UUCP) what now would be called a flashmob to
"reprogram" the Beaverton City Council so they would allow
Electric Lightwave to run fiber on the poles along Beaverton
Hillsdale Highway.  GTE (later bought by Verizon, then Frontier)
resisted.  But an election was coming up, and a few hundred
letters (many sent by cybercitizens via an "online" letter
printing company!) and a few dozen at the council meetings
let the council know that technovoters valued competition
more than GTE's monopoly.  The GTE reps at the meeting looked
like someone had run over their puppy.  The lone Electric
Lightwave rep was flabbergasted - he offered to give us our
own fiber drop.  At the time, we couldn't afford the endpoint.
But we did like the price erosion for POTS business lines
following the E.L. presence in Washington County.

Nokia got rich by replacing Soviet-era copper landlines with
fiber and cell towers - they paid for eastern Europe's new 
cellular phone systems by selling the copper they dug up.  

Mexico's Carlos Slim Helu is the richest man in the world, 
providing cell communications to the world's poor, bypassing
the POTs network with his own towers and fiber.  A cell
endpoint costs $10 to deploy, while a fiber endpoint costs
$500 and up, depending on how far away the customer is.  More
importantly, a cell endpoint costs $10 to replace and upgrade.
Upgraded fiber endpoints will cost another $500.  

The main problem with fiber in remote places is not the fiber
and endpoint, but powering the endpoint.  Most of the world
has really lousy, unreliable power grids.  Server sky internet
faces a similar problem - but only needs to power one endpoint,
not the switches and repeaters in between.  If the endpoint is
on a cell tower, and internet service starts out as value added
services for cellphone users, then server sky can deliver some
internet to billions of people, and quite a lot where power is
available to villages and shantytowns with real computers.

Closer to home, there are still hundreds of millions of people
in the developed world who live nowhere near towns and cities.
It might cost $10K each to deliver fiber internet to them.  If 
server sky delivers gigabit bandwidth to a $500 electronically
steered ground station, users get much more value and higher
reliability.  The main thing they get is freedom;  the sky
is big, with room for thousands of competing providers, and
switching services takes a fraction of a second.  Reliability
comes from parallelism - no single path system can be as reliable.

So - Amsterdam in 500 unit, 10 story apartment buildings?  By
all means, run fiber, and more than one along redundant paths.  
GNA & Citynet.nl is the /second/ fiber provider for many of
these homes and businesses!  Waiting for the $1 billion dollar
metro area fiber buildout?   Waiting to replace all that fiber
because a corrupt contractor used polyvinyl chowmein fiber armor
instead of the 50 year plastic specified?  Sites that can't
afford downtime can afford to rent gigabit satellite internet.

It is insane to compete head-to-head with established services
(about 4% chance of business survival) but a fifty-fifty bet
if your customers have no place else to go.  While server sky
is risky, it is not as risky as starting a new ISP in a crowded
market.  Amsterdam had a dozen public ISPs a decade ago.  They
have three now, plus a school & institute net.  Open fiber is
a great idea, but if there is only one ISP still in business
at the other end, what's the point?


> Want a 25-250 Mbps connection with 5-25 msec latency, rock-solid
> reliability, and the ability to frag your opponents before their
> packets even have time to get half-way to your router?  Get fiber.

On the continental scale, fiber is slower than MEO satellite.
Glass takes twisty routes through a lot of switching centers,
and has a prop delay about 1.5x vacuum.

I suspect the really dangerous opponents of the next few
decades will be black hats playing "frag your national
infrastructure" using the anonymizing "features" of the
switched internet.  With single-hop steered beam internet,
you have a better idea of who you are talking to.  One of
my worst nightmares is a server sky network becoming the
world's only surviving meta-network, as ground fiber is
saturated with DOS packets flying between p0wned switches. 
Guess who gets blamed, and executed slowly and painfully?

 
> Any real geek would choose the lower latency and higher
> reliability.    

Real geeks have better things to do than play games.  They
build businesses in India that do your (former) job better 
and cheaper.  With high bandwidth satellite internet, Jorge
can stay in Guatemala City with his family and use your
robots to mow your lawn.  That will allow you to afford
servants on a food stamp budget.

There are certainly serious jobs that require really low
latency - realtime trading in the stock market is one,
controlling high speed machinery is another.  But most
tasks can be done with 100 millisecond latency.  Again,
rich urban areas will use fiber intranets, no question. 
But over thousands of kilometers, satellite is cheaper
and faster.  Rich (as opposed to poor) urban core areas
are less than 10% of the world's population.  Guess where
the growth markets are and aren't?  


> As a side note, I have read on many occasions that fiber-optic
> cabling is actually *cheaper* to run in rural locations than urban
> or suburban areas.

Per kilometer, yes.  You can staple it to poles, rather than
trenching or vaulting it.  30x cheaper per distance for a 
single fiber.  But the distances are indeed kilometers, not
meters to the pedestal where your fiber joins your neighbor's
fiber at the splitter, joining hundreds of other fibers in
bundles back to the switch.  Provisioning dense first world
cities is cheap.  Rural/temporary/mobile/third-world is hard.

We passed a milestone in 2010.  More than 50% of the world's
population now lives in cities.  By 2040, this is expected
to rise to 80%.  Cities are ecologically cleaner and more 
energy efficient per capita, so this is a Good Thing.  High
rise urban cores are easy to provision with fiber and many
other needs.  But cities depend on resources drawn from
thousands of kilometers away, and highly interconnected
global trade.  If the machines extracting and transporting
the resources can be connected by satellite internet, then
even more people can move to cities, with the planes and
ships and trucks monitored by, and in some cases operated
by, data centers and operators halfway around the world. 

Operators at Nellis AFB in Nevada fly armed drones in
Afghanistan.  They might not win a thumb-twitch battle
with an online gamer, but they have the tools and skill
and adequate reaction time to vaporize one.  How fast
can you run from a 100 kg HE smart bomb?

I expect this nasty military technology will "civilianize". 
Rather than games, your kids may tour Paris or a Costa Rican
rain forest - over the net, by robot, during a coffee break. 
In my imagined future, people will travel the world, indeed
the inner solar system, using robots and high bandwidth
mobile internet.  The body stays home (perhaps in a haptic
suit in a resistance machine), the mind wanders the universe. 
Will fixed infrastructure be flexible enough for all that?

There is a lot more to the Real World than games and 
entertainment.  I want to build tools for the people who
keep the world ticking, not the people mindlessly clicking.

Keith

-- 
Keith Lofstrom          keithl at keithl.com         Voice (503)-520-1993
KLIC --- Keith Lofstrom Integrated Circuits --- "Your Ideas in Silicon"
Design Contracting in Bipolar and CMOS - Analog, Digital, and Scan ICs