We need to achieve some consensus on the MP01 antenna.  The antenna
design will drive the enclosure design and is hence on the critical path
for getting production Mesh Potatoes out the door.  It's also something
we will be stuck with for a few years, so we need to make a careful
decision.

So the goal of this thread is to come up with an antenna
_specification_, i.e. directional pattern, and gain.  

Proposal 1:  My proposal is a vertically polarised omni-directional
antenna with 0-2dBi gain. The PCB will contain a footprint for an
external connector so that with a few minutes soldering an external
antenna connector can be added and the internal antenna disabled.

Please feel free to present other proposals.

Use Case and Requirements
-------------------------

Please feel free to debate these assumptions, as for me they drive the
antenna specification:

1/ The Use Case I have in my head is MPs sitting on top of poles about
200m apart in a township.  The link is not power limited and they have
line of site but lots of multipath.  Power limited Wifi "long shots"
represent 10% of the possible MP use and can be handled by an external
directional antenna.

2/ I am assuming that by definition a mesh network needs omni antennas.
Otherwise a mesh node can't hear most other nodes and an ad-hoc mesh
can't form.

3/ It was decided at the 2nd Village Telco Workshop to use a printed
antenna as it simplifies the outdoor enclosure and significantly lowers
costs.

Whats Possible and whats not:
-----------------------------

1/ We have an engineer (Jeff W) working on the design who has some
experience with printed antennas, his opinion is that a printed omni
antenna will have a gain in the range of 0-2dBi.  The current external
omni antenna has a gain specification of 3dBi.  This is fed through
several connectors and some coax (estimated loss of 1dB) so about 2dBi
effective.  To get significantly higher gains (e.g. 6dBi omni) means an
external antenna, e.g. like a Picostation.

2/ Redesigning the RF side of the MP is out of scope, so any antenna
will be bolted onto the Atheros reference design via a simple impedance
matching network.  An electronic external/internal antenna switch is out
of scope at present, as is an electronically steerable antenna or extra
radios.

3/ The performance of adding external reflectors is not something Jeff
and I feel is easily repeatable in production.  You are welcome to
experiment, but it's not something we can commit to as a design spec.

Please feel free to discuss over the next few days, we will make a firm
call early next week.

Thanks,

I think it is a good idea to have an external SMA connector for a directional
antenna and I think it should be included for all MPs.

My personal preference will be to have two internal antennas that are software
selectable. The default antenna could be the omni, but with the option to
select a directional when needed.

My 2c
Johann

However, if the 'useful' links are not available at that moment, then of
course the switch should not be activated.

How easy will it be to open the MP01 housing?

If you draw a straight line on a map from A to B and look at a cross
section of the contour, you'll see concave valleys and convex hills,
at one scale or another. The hills don't need to be steep and only
have to be as high as the houses to have major effect on a link.

WiFi signal propagates well across the open air in concave valleys -
there is a lot more Fresnel zone. Conversely, WiFi has a serious
problem with convex hills, even when there's a high site on top of the
hill (because it's convex).

So I find mesh omnis tend to zigzag signal along the sides of the
valleys around the hills, and each node becomes up to 180 degrees
directional anyway.

The MPs probably won't be on top of poles; they'll be strapped to the
side. The pole will shield signal on one side of the MP, so the
installer will be pointing the MP.

Please ensure that the outside of the housing is 8mm away from the PCB
antennae so that the pole/tree/wall will be a good natural ground
plane, or we can stick one on.

Just a quick note.  Unfortunately we had to change our plans for the LCD
panel, at least in this first generation of the Mesh Potato.  The panel
turned out to be a little more expensive than we imagined and also the
added complexity/cost of weatherproofing the LCD panel in the external
enclosure not to mention the work involved in integrating the panel into
the MP circuitry were all factors in the decision.  It's a good idea but
we felt we needed to focus on delivering the core functionality of the
Mesh Potato.

Cheers... Steve

Johann I think two antennas or an electronic switch to select between
internal or external antenna is out of scope with our current resources.
Would be great to see on MP01 V2.0.

David C:

1/ At this stage the housing hasn't been designed, so not sure how easy
it will be to pull apart.  We will no doubt discuss the housing design
on this list so lets keep your 8mm spec and ease of dissasembly in mind.

2/ OK so based on your comments and real world operation of a mesh lets
include internal operation as a Use Case.  How does that affect the
antenna specification we should aim for (e.g. directional pattern, and
gain)?

Thanks,

All the mini-PCI wifi adapters that I make use of supports diversity and they
usually have two UFL connectors that connects to a small antenna switch. One
can usually do a software select between diversity / ant1 or ant2.

If the MP supports diversity then it would be nice to at least have one of
them come out on a SMA connector. If it's not the case then I would rather go
for a wide angle directional than for an omni.

Regards
Johann

When positioning a router in a home, there is usually an obvious place
to site the router - as high up as possible, away from metal/walls and
with a good view in the general direction of the nearest gateway.
Usually the router ends up on the window sill, or still better,
hanging from the curtain rail at the top of the window, but sometimes
it ends up deeper inside the house with a view through the window.

Once the general site is chosen, I look for signal quality. If links
are marginal or sparse, I then move the router a few centimetres to
one side or the other. After a few moves, I usually find good signal -
this is what I assign to the diffraction pattern and point source
antenna. With difficult sitings, I fix the router in the exact spot
that I have found, so that the owner doesn't move it and lose signal.
This is where I think a patch array will avoid problems.

I do not think that a directional antenna detracts from the ability of
the nodes to form an ad-hoc mesh network.

I am not sure how it all comes together in the software, but I am
guessing Antenna 1 is used most of the time.  I think the idea is that
Antenna 2 is used on rx if multipath wipes out the signal from Antenna
1, but I am not sure if this is a hardware or driver level feature.
Actually the data sheet block diagram suggests the rx signals from
Antenna 1 & 2 are just summed, rather than being switched.  

So it would be possible to use an internal PCB antenna for Antenna 1,
then route Antenna 2 to an external antenna connector, but this external
antenna connector would then just be for rx.

Cheers,

David

Have you (or anyone else on the list) found patch antennas mounted
indoor to be less sensitive to positioning?

Cheers,

Johann

running mesh nodes from a indoor location is problematic, but works if the
mesh is dense enough. I usually recommend to place them on the roof on a small
pole. There is no point finding a sweet spot indoors. Lift your arm, move the
TV, open the window or welcome a guest and the sweet spot is moving or just
disappearing.

Re: Internal antenna

Lets not forget we need two internal antennas. Since the RX only port is more
sensitive than the TX/RX port it deserves a well-working antenna as well.

Internal delta loops / cubical quads could also be an option.

Delta loops or quads have relatively wide bandwidth. So they are less
sensitive to capacitive influence from the environment that may alter their
resonance frequency and shift it away from the frequency range that we use.
Quads are popular DIY antennas in the Freifunk community (mostly Bi-Quads with
reflectors)

Gain of a single loop element is 0.67dBd / 2.8dBi (delta loop), respectively
1dBd / 3.15dBi (quad). Feed impedance is 80-120 Ohms (depending on capacitive
influence/antenna height), symmetrical input. Hence they can be fed without
Balun. I guess that the impedance matches more or less with the RF ports of
the AR2317 chip.

The directional pattern is bidirectional with 80 degrees angle in the E-plane  
and large opening angle in the H-plane.

The shape of the delta loop antenna is a triangle with ~1/3 lambda length per
side multiplied by the shortening factor of the pcb material. Respectively 1/4  
lambda for the quad.

A optional loop reflector element could be mounted on the backside of the MP
housing. The reflector element would have the same shape but needs to be a few
percent larger. Placed about 0.19 lambda behind the internal loop the gain
should be around 7-8dBi, with a front-to-back ratio of 12-15dBi.

1/ The tiny ceramic antennas you get in USB Wifi devices have 1 - 1.5dBi
gain.  In the ball park with a printed omni antenna but would save
significant PCB area so you would get a smaller potato.

I suspect these antennas perform worse than their specs suggest
otherwise why wouldn't all routers use them instead of the typical
rubber antennas?

2/ What would happen if the external antenna is connected in parallel
with an internal antenna, i.e. no switching at all? I am moving into
speculation land here, so pls correct the following if I have missed
something fundamental:

I think this would split the power in two (a 3dB loss). However at the
receiver you get the sum of the external plus internal antenna signals.
For example if you had a 12dBi directional antenna connected it's
effective gain would be 9dBi, but you would also have the ability to
receive signals from the omni.

On the tx side signal would go out both antennas.

Another bonus on the rx side is that one antenna would passively
re-radiate half the energy it receives out of the other antenna, which
may (or may not) be a good thing for a mesh.  I am not sure how to
calculate path loss in this case (i.e. is the total loss two paths or
one?) - it could be the re-radiated signal is too weak to be useful.

Disadvantage would be strong interfering signals would be picked up by
the omni, overcoming some of the benefit of a directional antenna.

Perhaps to make this work some impedance match is required to make all 3
ports 50 ohms, I am not sure how reflections at microwave frequencies
would affect performance.

Anyway, just a crazy idea, and with my project manager hat on I would
have to say out of scope for this version of the MP.

Cheers,

really interesting ideas. I do agree with your conclusions.