Disco Antenna

Michael Shuler mike at bwsys.net
Tue Dec 10 14:59:14 EST 2002

     Well it looks like the general consensus is that the Prism
implementation of antenna diversity isn't very useful.  So I am
proposing a different solution and want to see what other people think
of it.  I've had this idea for quite some time but haven't know quite
how to implement it, maybe someone else here could help?

     As a WISP covering many different cities/terrains we have seen all
kinds of weird interference issues.  Most of our problems DO NOT come
from other radio interference but rather completely unexplainable
interference.  Usually the unexplainable interference seems to come from
specific directions but for varying amounts of time.  The level of
interference seems to vary quite a bit too.  Other times it just seems
to go completely away for a while and nobody knows why, sometimes for
days at a time.  In every one of these situations we found that our
signal levels never dropped but only the noise level changed.  When the
noise floor would increase enough the connection would obviously break
or be impaired.  In a big city you would expect this type of behavior as
more and more WiFi devices get sold every day, but this is happening in
small towns (under 1000 population) too and it just doesn't add up.
I've noticed a general increase in all of my noise floors since about
the time XM Radio went live and from what I understand their frequency
is right next to the ISM band...but anyway where I was actually going
with this is that to improve performance (get a better SNR) we would
replace omni's with 3 120 degree sector panels and 3 radios (channels
1,6,11).  Not only would the signal levels get higher because of the
higher gains of the antennas but the noise floors would drop.  Obviously
since a single sector panel is only looking at 120 degrees of potential
interference its noise floor should improve...and it always did.  Taking
this concept even further we have a 500ft tower that we rent space on.
We ran CAT5e & 110 AC (Power over Ethernet wouldn't make it, but despite
the 100m limit for Ethernet data it runs without any line errors...go
figure) to the 400ft and 500ft levels.  At the 500ft level we put 3 60
degree vertically polarized sector panels and 3 radios (channels
1,6,11).  At 400ft we put 3 60 horizontally polarized sector panels and
3 radios (channels 1,6,11 again).  We also put an 8db omni at 400ft to
test the performance differences between sector panels and the omni.
The test proved exactly what we expected.  The noise floor on the omni
was considerably higher then each sector panel.  We were able to get
distances in excess of 50 miles (to a 250 water tower using a 24db grid
antenna) that actually would hold a pretty decent link.  The water tower
worked because it had pretty good Frenel Zone clearance but nothing else
lower would work that far out, pretty cool anyway.  The 3 panels that
covered the larger city (150,000 population) that was about 4 miles from
the tower massively out-performed the omni (obviously the signal levels
were higher due to the higher gain of the antennas but I'm mainly
referring to noise levels).  With the exception of thick trees and hills
we are able to hit about everything as long as we can get reasonable
line of site to the customer.

     Finally, here is where I'm going with all this :)  

     Since sectorizing to 60 degrees caused and even greater improvement
(many db reduction in noise floor) vs. 120 degrees and the added bonus
of increase antenna gain, then why not cut it down even further into
hundreds of sectors not only horizontally but vertically too?
Essentially creating an antenna that looks like a "Disco" mirror ball.  

Problem 1 - Only 3 non-overlapping frequencies and 2 available
polarizations....good luck finding an ISM circular sector panel.  Pretty
much maxes you out at 6 panels in *near* proximity of each other and
even then they still interfere just due to the RF leakage coming off the
PCMCIA cards alone.

Problem 2 - The 6 could be increased to 12 if diversity worked
right...looks like no luck there :)

Problem 3 - If the gain of your AP's antenna became too high in a multi
point situation would exceed the max EIRP allowed by the FCC

Looking for ideas from anyone out there but here are my ideas so far:

The Antenna (120 degree of coverage per panel):  
     The "Disco" antenna would consist of 16x16 evenly spaced antenna
elements in a 120x16 degree arc (the idea is to cover a total of 120
degrees since most towers have 3 faces to them).  I really don't know
much about antenna design so it may have to be flat instead of an arc.
I'm guessing that it will be about 1ft high and about 3ft wide.  I
picked 256 elements because my Soekris boards I use for our existing APs
have a general purpose 8 bit I/O port...although using a serial
interface would be better if anyone knows how to make a decoder.  Based
off the value of the I/O ports it would create a connection between the
radio cards and one of the elements using a simple set of logic gates.
I'm sure some all-in-one chip exists for this type of thing but I'm not
a hardware guy, any suggestions would help.

The Radios:
     3 Soekris boards with 3 radios each (2 PCMCIA and 1 Mini-PCI).
SENAO seems to make the best radio card for the buck...23db for
~$85/each.  They would be linked together via the second Ethernet ports
on a private subnet for driver communications.  The primary Ethernets
would be for actual data traffic.  Each radio would have an antenna lead
attached to its spot on the input to the gates.  ONLY ONE radio is for
transmitting, the remaining 8 all passively listen.  More about this in
the next section.

How it works:
     It's easiest to conceptualize if you think of the antenna as just a
big grid if smaller flat panel antennas that have a beam width of 7.5x1
degrees.  We will label the rows 1-16 and the columns A-P, you sunk my
battle ship :)  The 9 radios will attach to nine different antenna
elements simultaneously with only the center element actively
transmitting.  The surrounding 8 radios will be listening to determine
if their antenna element is better suited to transmit and receive on.
Here is an example of listening on C3:

P = Passively Monitor
* = Transmit/Receive
  = Unused at the moment

  A B C D E F G H I J K L M N O P
2   P P P
3   P * P
4   P P P

Only C3 is actually doing any real work.  The remaining 8 are only
watching signal levels.  I'm guessing that the passive radios should be
checked about 10 times per second or so but I think we are going to have
to just fiddle with it until a "magic number" is found.  When one of the
passive radios has a better receive SNR the whole set of radios will
move in that direction.  For example lets say that D4's SNR level is a
db higher, the driver would change the I/O pins to make the gates adjust
to the following:

P = Passivly Monitor
* = Transmit/Receive
  = Unused at the moment

  A B C D E F G H I J K L M N O P
3     P P P
4     P * P
5     P P P

When the radio is not busy transmitting or receiving it can go into an
antenna scan mode where it is essentially sending out its usual beacons
or whatever else it has to do while being quickly moved throughout the
different elements of the antenna.  When one of the 9 radios picks up a
signal from one of the associated clients it will immediately return to
the last know position for that client and then adjust accordingly.
Whenever it has something to transmit to a client it will also return to
the last know position for that client.

Summary (finally):
	If all goes well this antenna should create a 7.5x1 degree beam
towards a client, which should keep the noise level low for everyone
else in the area and should have a much higher gain like 30db+radio
power.  On the receive side the noise level should be decreased
dramatically and since the panel is 3ftx1ft (guessing) it should have a
pretty good receive gain too, maybe 25-30db gain.  With all that it
sounds to me like this would solve everyone's interference issues and
would provide enough receive gain to get though some extra trees or
walls :)

Before I start to build this thing I wanted to get some feedback from
some other people who might have already tried this.


More information about the HostAP mailing list