Re: Multi-Hop Bandwidth

In theory after 5 hops the bandwidth stabilizes at 1/5th of the bandwidth of a
single hop link - given that all links are equally good and run at the same
speed. Under laboratory test conditions it is less, the figure I heard several
years ago was 1/7th. But these results are likely to vary from setup to setup
anyway. One reason is the fact that the interference range of the radios is
far greater than the communication range.

However laboratory conditions are still pretty ideal, particularly if we look
at a simple chain of nodes. In a real life mesh cloud (or swarm?) the
interference is coming from many directions. Plus the channel may be busy with
other networks, babyphones, wireless video links and so on.

The results of Scarborough don't show the 1/7th rule because

1) the links are unlikely running at the same bitrate
2) it is a real life mesh network ;-)

In Scarborough the interference in the mesh may be mostly self-inflicted - the
mesh nodes respectively their transmissions interfere with each other. And, as
said above, the interference range is greater than the communciation range.

Re: ETX

ETX values only show the round trip packetloss of OLSR "Hello" broadcast
messages. These are sent at multicast rate which is by default lowest (basic)
rate. ETX values are not really a proper indicator of link speed. A 1Mbit link
without packet loss looks exactly the same for ETX like a 54Mbit link without
packetloss. Besides, together with the beacons the OLSR packets add to a
permanent noise floor. If they are sent at lowest bitrate, they consume a lot
of airtime. Even if the network is idle. Hence it is strongly recommended to
increase the multicast rate. Berlin is using 5.5Mbit broadcast rate. One may
experiment with even higher multicast rates to optimize bandwidth consumption,
routing and throughput. Increasing the beacon interval will also help. If you
don't want to connect wireless clients to the mesh you may get rid of beacons
entirely by using ad-hoc demo mode without beacons. (Can be done with Madwifi)

Increasing the multicast rate will show stronger differences in ETX values for
a perfectly working 1Mbit link compared a 24Mbit link and allows the routing
protocol to make better decisions.

Re: Antennas

Unfortunately we can't do beam forming with the cheap AR2317 chip. And the
ath9k driver for Atheros 802.11n devices is not capable to do this either -
yet.

I understand and appreciate the opinion of David_C - however I'm highly
nervous about the team making the decision for a etched directional antenna.
With our decision to use a built-in antenna only, we have reduced flexibility.
Which is fine for a robust product aiming at low cost. Users can't choose
anymore whether they get the best from a omni or directional antenna. So we
have to make a decision and compromises must be made.

Omnis/dipoles:

With two well working omnis or dipoles or loop antennas inside, we would have
none/slightly directional radiation pattern and no front-to-back attenuation.

Benefits: Users can mount the devices on a pole/broomstick and don't need to
worry where they have to aim to. Simple. We don't have to reduce the TX power
by default to comply with EIRP limits. We can easily accomodate two antennas
in the housing and make best use of receiver diversity.

Disadvantages: We pick up/create interference from/to all sides or from two
preferred directions. Less range.

Directional:

Benefits: More range. We can select where we receive and send best.

Disadvantages: Users have to aim - not just to the next MP, but to one which
is closer to the destination. We have to reduce transmit power to comply with
EIRP limits. We'll probably need more space inside the housing to accomodate
two antennas to make use of receiver diversity.

My 2 cents:
The Ubiquiti Nanostation is IMHO not our role model for this. It is designed
as a long distance wireless bridge or AP that covers a sector from a high
site. And it is very good at that. It is not designed to be used as a mesh
device in a chaotic mesh cloud, that is to be set up by inexperienced users.
Using the same directional pattern and front-to-back ratio means to trust into
the assumption that "it will be possible to connect to a side lobe or within
the main lobe of the directional pattern". However if all nodes aim and
connect to a central point we end up with a star-like topology. I guess nodes
that are more than one hop away from the center will have a hard time to
connect. Plus MP to MP traffic will tend to always hop into the center of our
network first and cause congestion there.

That said: Omnis / horizontal dipoles are problematic from the interference
point of view. However I think we need a "one-size-fits-all" solution. Which is
of course not optimal. But omnis/dipoles are the default in every cardbus
card, SOHO AP, USB WLAN stick for this reason.

So the goal is: Mesh with good data transfer

Need #1 - strong, interference free links

Need #2 - hard to mess up the deployment

...so wanting strong links is pressuring us to take advantage of  
directional antennas and manage power outputs but that is making the  
deployment more risky.

...and wanting to use omnis etc. to make sure that nodes at least get  
some traction in a poor setup is of course making our links less good  
than they could be.

I can understand wanting to be conservative and rather err on the side  
of making deployment easy but I also have very good memories of the  
difference between when we first setup the Scarborough mesh and after  
DaveC had spent a couple of months tweaking router positionings!

Underlying all of this I think there may be an unverbalised (though  
valid!) assumption:

   "Positioning routers & setting power levels is bloody hard"

So - Beelzebub's advocate styling, three questions:

1. Are there things we can do in terms of the hardware, router  
firmware and/or dashboard software to make a high quality physical  
deployment a no-brainer?

2. If there was, would that be a sufficient condition to prefer a  
directional antenna over an omni?

3. If it is _not_ a sufficient condition, would it still make sense to  
spend time doing what we can to make optimizing router positioning &  
power-setting a no-brainer?

As Antoine suggests, patch array may be easier to position than omni,
but I have never worked with diversity - I set to single antenna and
find the spot, sometimes glueing the router in place. Most are indoors
with complex 2.4GHz fields - one chap in the early days had a mark on
his floor to show where the aluminium sliding door should be for good
signal. I hope diversity can sort some of this out.

I'll enquire further and perhaps work with Antoine on building one -
see whether we can figure an adaptive array for the MP2 :)

A dipole. But can I also have an antenna port please?

If I need a directional I'll place a reflector behind my dipole. Voila - its a
directional now ;-) Brainstroming: Hey we could even add support for the
reflector on the back side of the housing... Need a directional? Shove that
aluminium sheet into this slot...

Brainstorm: The wavelength of corrugated iron is 76mm. What's
happening at 3.95GHz?

1/ I see or channel model as nominally good links with occasional
interference mainly from on-mesh sources (other routers).
So I feel the benefit of a directional antenna is interference
reduction, rather than gains in the link budget.  Might even reduce
multipath if the multipath rays are off-axis to the main lobe.  

Lets investigate the idea of having more than one antenna pattern
available, either by an external antenna socket or sliding objects
behind the MP.

2/ One task for the Phase 3 work is to devise an objective measure for
link quality to make alignment and deployment easy.  This would be a
number, possibly presented as a bar graph of LEDs on the back of the
MP01 or plotted on an Afrimesh dashboard.  It would be more than just
signal strength (which doesn't make sense in a mesh anyway), i.e. an
indication of actual voice quality that takes into account many factors
(like the end-end link quality to the gateway).

It will be possible to log this measure over time to gather data on time
varying link quality variations, e.g. by weather, use of microwave ovens
and other services.

If we want to guarantee a certain QoS, we could make the firmware refuse
association of the node to the mesh if it doesn't meet a certain minimum
standard.  This will force the installation guy to find a solid
installation configuration.

3/ Given David C's experience router positioning will give us just as
much variation in signal as a directional antenna, e.g. moving the MP up
30cm on it's mast on a metal roof may vary signal by several dB.

4/ Again given David C's experience we need to come up with a "formula",
or set of guidelines (or possibly MP antenna design) for setting up MPs
to avoid issues like very strong sensitivity to the time-varying
location of furniture, doors, people etc. Given one node may be relaying
calls for many others we can't have a movement in a sliding door
bringing down a whole chunk of the mesh.

Cheers,

However looking at you question another way - there are serious capacity
questions with Voip over mesh Wifi, which get much worse with many
nodes.  When I looked at Kyleisha my first thought was about reducing
the "cell" size, and winding back antenna gain, power etc, and linking
adjacent "cells" with Ethernet!

Cheers,

On 29 Oct 2009, at 17:23 , elektra wrote:

With tech & communities it often seems that the responsibility tends  
to magically find its way to those folk with a latent talent for it so  
while I see the direction of the solution being "anyone can do it" I  
also tend to expect that in practice by far the majority of nodes will  
be deployed by the MP-wallah-fundi-whiz.

Maybe another way to reframe the question:

   How many steps are there in the physical deployment task that  
random user w/ latent aptitude needs to have internalized before s/he  
becomes a MP-wallah-fundi-whiz?

That would give us a nice metric to work against and, if we should  
ever reach the magical number '3', then probably even random user  
could do the setup.

:-D

I was asked to add an opinion on this from academic literature - so
here goes:

        At an intuitive level, directional antennas will give you
bettter
        spatial reuse, routing paths with less hops and increased
network
        capacity (up to 90% extra capacity according to literture [1])

        But they carry the disadvantages of the deafness problem -
where
        collisions occur because sensing only occurs by nodes inside
beam of
        antenna , less flexibility (sometimes antennas need to be
reorientated
        if a new node is added)

        But smart antennas will solve most of the disadvantages and
will create
        the ultimate solution for mesh.

        Good papers on the topic are:

        1. Opportunities and Challenges for Mesh Networks Using
Directional
        Antennas:
         http://www.cs.ucdavis.edu/~prasant/WIMESH/p13.pdf

        2. DMesh: incorporating practical directional antennas in
multichannel
        wireless mesh networks
        https://engineering.purdue.edu/ECE/Research/TR/2005pdfs/TR-ECE-05-16.pdf
        Claims throughput improvements of up to 231% but used a new
distributed
        algorithm to perform routing and directional channel
assignment in the
        DMesh architecture. Improvement simply because more
simultaneous
        transmissions can happen in the network.

        When there are long distances between nodes directional
antennas are
        clearly all thats possible but when the density of nodes is
high enough
        is there still a case for directional antennas? From the above
        literature it seems to still be the case - but it comes at a
cost. Well
        to me it becomes a question of maintaining a threshold of node
neighbors
        and this can be achieved by decreasing power or using
directional
        antennas and a combination of these. But clearly it is always
an
        advantage to have more nodes able to communicate
simultaneously without
        interfering and a combination of spatial and frequency reuse
achieve
        this.

[1] is an overview of the possibilities and problems of mesh nodes using
electronically steerable smart antennas.  [2] reports on a practical
network using multiple radios at each node connected to an omni and a
directional (but non-steerable) antennas.

Unfortunately neither technique is possible with our current Mesh
Potato, due to the constraints of the current hardware.  

So I figure the best MP01 antenna design is a built-in omni.  An option
for an external antenna connector can be placed on the PCB for the rare
cases where a directional antenna is needed.

Cheers,

David

As posted earlier I tend to favor a horizontal dipole. Not strictly omni-
directional - two preferred directions and no front-to-back attenuation.

Perhaps we could add support to mount a flat metal reflector on the back side of
the housing (2-4 screws). So the user or MP-wallah can convert the internal
antennas into directional antennas (Dipole(s) in front of a reflector).

If the reflector is sufficiently sized the front-to-back ratio is high (>20 dB)
and the gain can be in the range of 7 dBi. It may be worth experimenting with
it.

As an interloper, I'll just toss in my agreement here.

I think a vertical dipole with an optional reflector is a good  
solution here.

The difficulty, though, may be designing the antenna around the PCB.  
By putting that slab of grounded metal in the near field, you are  
likely end up with a directional antenna whether you want it or not.

-- David

On Nov 3, 2009, at 7:36 AM, Corinna Elektra Aichele wrote: