Dili Village Telco Part 4

Several days have passed with little progress. Setting up a reliable first link in our mesh has proven very difficult. I have been reporting my progress back to the Village Telco Google Group and VT team members – they have been very helpful with support and ideas. It makes me feel like I am not alone when debugging these frustrating problems.

We have a Nanostation 2 at Fongtil HQ, and the first node at the Blind Society about 300m away. The link quality changes from fair to completely dead. When fair we have 10% packet loss on short ping packets, and 50% loss on 1400 byte ping packets. When the link dies it’s 100% packet loss on all packets. The Batman score moves between 0 and 180. Even at a score of 180 the ping performance on long packets is poor. The link might be fair for a few hours, then completely dead for a few hours.

Fongtil HQ is on the edge of the Dili city centre, Lemi wants to use the Blind Society site to spread the mesh to several nearby NGOs. The terrain here is flat, with 20m trees, two story buildings, and terrestrial microwave towers in the approximate path we are aiming for. We have a 20m tower at Fongtil and a 15m pole at the Blind Society. Both ends report signal strengths in the -65 to -70dBm range, so it’s not a link budget issue.

We have tried many tests, each of which take a few hours to organise and carry out. Possible problems include line of site, long power lines, interference from Wifi and other services, and dud Supernode hardware.

For example; to test another Wifi channel we need to change the channel at both ends, which requires ssh access to the Blind Society node. This means driving around to the Blind Society with my laptop and another MP, booting them, some ifconfig, ssh, reboot on the new channel, run batmand -d1 -c, a few ping tests, find this channel is also poor, change channel back, reboot, test again, ahhhh the sun is too bright I can’t read the laptop, and it’s 45 degrees C and 100% humidity inside this car and the sweat is stinging my eyes and dropping into my keyboard!

One test yesterday helped point the finger at interference:

We set up a NS2 at the rear of the Fongtil building, at about 10m
height. Good links to the MPs inside the building. With a mobile MP (mounted on a 2m stick, powered by a small 12V battery) we could make calls 200m down the road.

We raised the NS2 to a height of 18m and the links to the MPs inside of the building dropped out. However the signal strengths to the MPs inside the building were still OK at about -70dBm. It’s like the NS2 suddenly went deaf with the extra height.

So my theory is the extra height caused the NS2 to pick up some very strong Wifi or non-wifi microwave signals that jam the signals coming from the MP’s inside the building. However we need the height to get good LOS to our 1st node at the Blind Society. Catch 22.

My current inclination is to get some 5 GHz gear to get around and 2.4GHz problems. I can’t obtain this here in Timor so it means shipping it a few weeks after I have left Timor. We might also have better results with Nanostations at either end of this first link, sort of blast through the interference with higher signals and a little more directivity. There is also the possibility of an intermediate node if we can get access to a site.

There was one nice innovation on Thursday: a way for anyone to test a mesh link with a few minutes training. I was tired of carrying my laptop and an extra mesh node around for testing. So we put a MP01 on a 2m broomstick and connected a battery and a phone. I added some dialplan:
exten => 4005,1,Dial(SIP/4001@10.130.1.5)

4001 is the echo extension we have in each MP’s dialplan. So this test makes a call to MP01 10.130.1.5 (which is sitting inside the Fongtil building) and runs the Echo application. The net results is that you here your own voice on the mobile MP. The quality of your echoed voice is a clear indication of link quality. It works really well.

In 5 minutes I trained 4 Timorese guys how to test the link this way. We walked down the road 200m and I could clearly demonstrate the need for line of site to the Supernode, for example by jumping on a fence to get more height. How long would it take to train them in all the Linux command line ifconfig/ssh/ping/batmand magic to test a link? This is the sort of innovation we really need for the Village Telco.

This test also showed us we could get good signal over 200m as long as we had line of site and the Supernode was fairly low down (and perhaps not subject to crippling interference). We could have gone further but couldn’t get a line of site path at that height.

Links

Dili Village Telco Project Wiki
Dili Village Telco Part 1
Dili Village Telco Part 2
Dili Village Telco Part 3

4 thoughts on “Dili Village Telco Part 4”

  1. Omni’s hear everything… which is both feature and flaw. Doesn’t surprise me you had trouble higher up.

    I deployed my first two nanostation M5 nodes today. I plan to do the next 2 nodes tomorrow or friday.

    Deployment didn’t go too well, I had to write a script to get the darn things to beacon, and after that couldn’t convince babel to route a whole /24 subnet properly (I’m hoping I’m just being stupid).

    adhoc mode only works in 802.11a, not n, mode, on these radios.

    But after I convinced them to beacon right and got the antennas aligned reasonably, I got 18Mbit out of them in a 280m mountain shot, using iperf. The next shot is 580m. After that, I add a true mesh node or two…

    It’s nice to have these narrow profile antennas that are in the M5s. No wind is going to move them around much.

    I don’t have good tools on the things for signal strength or leds. I do have the second POE port working in software.

    I can get a good openwrt build for the m5s your way. (actually, if you get around to having ipv6, it’s
    http://opkg.sjds.teklibre.org/snapshots/trunk/ar71xx/ at least for now)

    I however am using babel, not batman, haven’t bothered to build asterisk, or a gui, or much else. It’s nice to have two ethernet ports to hang other stuff off of, if needed….

    I’m told the beacon problem I hit has been fixed but I have no good full update-in-place facility – I have to manually install new software AND the wifi drivers after polling

    I may defer installing the next two until AFTER I figure out these three problems.

    My stuff is WAY more raw than yours, but I ran smack dab into the 2.4ghz interference problem ages ago and it never really struck me how common this must be elsewhere in the third world until you started writing about it.

  2. Hello Mike,

    Thanks for your support and the update on your project. Your experience with interference in Nica matching mine in Dili is an interesting data point. It hadn’t ocurred to me, either.

    Yes it’s a catch 22 with the omnis, we need them for the mesh paradigm but they remove any protection from interference.

    Your firmware issues are familiar to me. I feel we still need some decent tools to help with and dramatically simplify installation, alignment. and monitoring.

    “wlanconfig ath0 list” dumps signal strengths for us, but might be different for your driver.

    Best of luck with your 5GHz mesh.

    Cheers,

    David

  3. Hi David:

    I ditched the madwifi code in favor of the mac80211 stack back around linux 2.6.28, and am currently up to kernel 2.26.32.10.

    The equivalent tool, using that, is:

    root@lat2:~# iw dev wlan0 scan dump
    BSS 4e:c2:59:9a:8d:ec (on wlan0) — joined
    TSF: 80099021064 usec (0d, 22:14:59)
    freq: 5180
    beacon interval: 100
    capability: IBSS (0x0002)
    signal: -69.00 dBm
    last seen: 0 ms ago
    SSID: teklibre
    Supported rates: 6.0* 9.0* 12.0 18.0 24.0 36.0 48.0 54.0

    I’d always planned on a hybrid 5.x/2.4ghz mesh, but given how rapidly 802.11n in this spectrum is coming along, and the huge number of channels available, I am thinking of dropping most of the 2.4 support from the design, using that for conventional APs only.

    Last night we had a major lightning and rainstorm and performance hovered around 12Mbit, with only once suffering a few seconds of packet loss, dropping performance to 4.9Mbit….

    (this is only a 280m shot, but I’m also running at 17dbm power, rather than at max – 27 – but was by no means the worst rain Nica can dish out.)

    I have 580m, 1.2km, 6km, and 12km tests in the pipeline.

    Best of continued luck with your project. It looks like a lot of work, but also fun and fulfilling.

Comments are closed.