MayFirst2012NYCCrateDeployment

From The Commons
Jump to: navigation, search
Crate open in Union Square
Opened crate in Union Sq

As a technical experiment and outreach project, some New York City FNF members used existing FreedomTower equipment to deploy a free and open mobile "internet-in-a-suitcase" WiFi access point at march/demonstration sites in mid-Manhattan on May 1st, 2012. This was something of a last-minute deployment using the equipment on hand, with backhaul over a regular 4G service provider without any tunneling or encryption, and does not represent the ultimate topology or (fixed) infrastructure of a Free Network, but was a lot of fun and a great way to get hands on experience in an extremely crowded radio environment.

We experimented with a rolling "basestation crate" (which contained a commercial 4G modem, pfSense router, and seperate 2.4GHz and 5GHz WiFi radios all powered by battery) and a smaller "repeater unit" (dual-band commercial router powered by a battery).

More photos online here.

Equipment

Pre-deployment crate photo
Crate and equipment pre-deployment. Repeater is on the right in a ziplock bag.
Mid-deployment crate photo
Better photo of crate internals.

Basestation Crate

  • Rolling Pelican Suitcase/Crate
  • 1x 5GHz Ubiquity NanoStation ("N5")
  • 1x 2.4GHz Ubiquity NanoStation ("N2")
  • Intel Atom x86 router (running pfSense)
  • Ethernet switch
  • 12v sealed lead-acid battery (absorbed glass mat)
  • 120v AC inverter and power strip

The basestation/crate was pretty heavy with the battery; care was taken to make sure the battery wouldn't rattle too much, everything was secured with adhesive-backed velcro to the inside of the case. Lifting the case up and down stairs we a bit of a pain, but it was easy to roll, and only suffered a bit of wear being rolled more than a mile over endless curbs, potholes, and cobblestones.

Electrical Block Diagram
Crate Electronics Block Diagram

Total crate electrical draw coming out of the inverter (as measured with a Kill-a-Watt device) was about 25 watts idle. We chose to use a lead-acid battery because they are simple and (relatively) easily available. Unfortunately they are very large and heavy (for the desired capacity), somewhat dangerous (high current), and more expensive for the sealed "absorbed glass mat" (AGM) style required for safety. Regular car batteries have sloshing liquid acid and should not be left at an angle, can leak or spray acid, are damaged by deep discharges, and are generally unsafe for mobile deployment (though they do have large capacity and are cheap if found used; might be appropriate for towers neading DIY large capacity battery backup). AGM batteries cost around $100 for the capacity needed and can be found at marine supply stores or in the marine/recreation section of car supply stores. We got a smallish AGM battery new from AutoZone (was able to return it a week later) which had a 19 amp-hour capacity. Large batteries usually don't list an amp-hour capacity, they list "Reserve Capacity" (RC); in our case we wanted at least an RC of 40 (messy units, see wikipedia).

No special efforts were taken to reduce power consumption via software settings ("sleep modes").

Network Block Diagram
Crate Network Block Diagram

Above is the network structure block diagram. The nanostations operate as "dumb" radios, all DHCP is handled by the central router.

The 2.4GHz Nanostation was configured to only use 802.11bg (not 802.11n), on the wild guess that high throughput speeds would be an impossibility and that sticking to the older protocols supported by all devices would reduce complexity.

Otherwise the router and NanoStations were configured the same way as a regular FreedomTower, see documentation elsewhere.

NanoStation Beam Shape
NanoStation Beam Shape, from datasheet

The above diagram (from NanoStation documentation) shows the expected RF signal strength. The radios were mounted to the front of the case facing forward vertically, such such that the bubble of WiFi coverage was directed forwards as if the case was a large speaker. The "squashed cone" of coverage was about 50 degrees "wide" and 30 degrees "tall" according to the above diagram.

Obviously the overall functionality of the basestation could be replicated with a tiny cheap off-the-shelf battery powered WiFi hotspot and a 4G USB dongle. Our equipment and software was much higher performance, but if we wanted to scale out this kind of deployment the package size could be reduced significantly.

Repeater

The motivation for the repeater was to test extended range using the (relatively clean and reliable) 5GHz band as backhaul and making shorter-hop 2.4GHz connections to users with 2.4GHz-only devices.

Equipment:

  • Netgear WNDR2700v2 ("N600") WiFi gateway/router running OpenWRT
  • Small sealed lead-acid battery (12v)

The battery was connected directly (with a fuse inline) to the router's barrel jack power connector by canibalizing the wall-wart cabling; this router takes 12v DC input, so not conversion was necessary. The router pulled 6-11watts of AC power (measured with Kill-a-Watt). No special software efforts were taken to reduce power consumption.

The router was configured to use the 5GHz radio as uplink (WAN) and to provide general connectivity via the 2.4GHz radio (LAN), with mostly default settings. This means the router ran an internal DHCP server (dnsmasq?) and handled all aspects of client configuration locally, instead of forwarding request packets on to the primary router in the basestation. A seperate SSID ("The Free Network Repeater") was used.

Narrative

Router traffic chart
Router Traffic Chart (UTC timezone)

Above is a chart of 4G network uplink bandwidth throughout the day. Measured by the pfSense router, this is traffic to the internet from the basecrate, including that from the repeater, but not including "local" traffic, eg serving up the chart itself. Time is UTC, so "16:00" is noon EDT (rough start of deployment), grey lines are 10minutes. Maroon in download, grey is upload.

We started the day with fully charged batteries at Bryant Park, where a series of stationary events were taking place and a crowd had gathered. The crate was powered up at 11:40am EDT right next to a row of television vans (with their own radio equipment). The crate was on the sidewalk at street level at 41st St and 6th Ave; the park itself is elevated about 2 meters, with a broad stairwell leading up.

By 12:05 we had tested and found "good signal" for the main SSID around the entire park lawn. Some users had trouble connecting directly to the basestation using iPhones; other phones had no trouble connecting. The bandwidth chart shows very low usage for a period here, there may have been a misconfiguration?

At 13:20 (17:20 in chart) we turned on the repeater; it's battry voltage was 12.44v. The repeater's 2.4GHz radio was set to channel #1 to avoid interference with the basestation radio. There was another strong station on channel 1 (SSID: "warroom"), and we may have been clobbering that. Around this time we moved the crate up into the park itself and stood it on folding chair to get it off the ground by a foot or two; this greatly improved signal strength. The bandwidth charts show mostly intermitant download.

Around 14:23 (18:23 in chart) we began marching/rolling down 5th Ave to Union Sq, which took about 40 minutes. We saw pretty heavy upload traffic during this period. Bryan rolled the case on the opposite side of the street from the march and monitored uplink connectivity while Pablo took the repeater in the crowd and told marchers that connectivity was available.

By 15:00 we had arrived at Union Sq and took a break at the north end of the plaza for some food. We opened the crate and powered down for a few minutes at 15:15 (19:15 in the chart, note "gap"). Crate battery voltage was 12.00v open circuit (no load).

After about 40 minutes we made our way south in the square. The repeater battery died at about 16:00 (with 11.90v), so it ran about 2.5 hours.

At around 16:30 we made a few signal strength measurements across about 15 meters of very dense crowd (in front a concert stage just north of 14th St and Broadway) and got poor signal and throughput. At this point the batteries died in the smartphone being used for testings, so the signal quality measurement could be due to low power. Chart shows we had some strong down/up usage during this period. We had the crate up off the ground at about chest height on a stone wall, but not above heads or with clear lines of sight to the crowd, which may have diminished signal propagation. After about 40 minutes the crate was moved south of the crowd to a new location before the start of a large organized march, and the batteries probably died shortly after that.

The pfSense logs indicate no traffic after 17:30, and we have notes of only "questionable connection quality" at 17:00 (checked with a borrowed phone). It was discovered that the batteries were dead and inverter was in alarm at 17:55. This means the batteries held out for almost 6 hours, which is pretty good.

The final basestation battery voltage was 11.10v, which is very low ("deep cycle", would cause damage a regular non-AGM lead-acid battery).

Conclusions

Healthy use of the network while "on the move" around 16:30 was surprising. Providing mobile/roaming connectivity isn't a short-term priority for the FNF, but might be possible in certain situations.

We did little to promote the existance or goals of the network to users, because we thought a captive portal would be frustrating and not in the spirit of open access. Using a domain or short URL as the SSID would have allowed curious users to look us up.

It is (obviously) important to get the radios up above head height for this sort of deployment, thus the usual "tower" structure. It was equally important to keep the mobile directional antennas at least relatively fixed with respect to users; rotating the crate even a small angle could cut off many connected devices in a crowd, which is a very frustrating user experience.

It seems like the OpenWRT router was more accessible to user's mobile devices compared to the 2.4GHz NanoStation; this may have been due to accidental misconfiguration or confusion, more lab testing to confirm is called for.

Overall, in a very crowded chaotic environment, we provided about 250MB of up and down data transfer, averaging to around 100kbps up and down throughput. It is encouraging that users made relatively symmetric use of the network!

Discussion

Some builders mailing list notes from the time of deployment:

 RE: Lead-Acid batteries -- IME w/ AGM batteries, you MUST pop for a microcontroller operated 
 battery charger with the appropriate AGM curve if you want your AGM battery to live much longer
 than a Chinese NiMH cell.  I made the mistake of using inexpensive float chargers on mine, and now
 suffer from greatly reduced capacity as a result.  I suggest using gel-cell packs instead.  They're less expensive, 
 far more tolerant of the most readily available / inexpensive battery chargers that people who don't know lots about 
 battery chemistry will tend to buy. and have all the leak-proof capabilities of AGM.  I strongly recommend using CTEK 
 chargers on AGM batteries.  BTW, if you do use a car battery, don't count on it being able to last long if it's put 
 back in a car.  Cranking batteries die quickly when you deep cycle them.
 RE: Small inverters -- If it doesn't have "PURE SINE" plastered all over it, it's not.  If it does, it might be but 
 I'd still check.  Instead of running wall warts (that can be difficult and too costly to replace when you destroy 
 them) from a cheap inverter, how about using equally inexpensive "universal" DC-DC converters made to plug into a 
 standard auto cigar socket (and sometimes EmPower airline seat jacks), with switchable voltage outputs?


There were after-the-fact mailing list discussions about this deployment on the dicuss and builders mailing lists in June 2012.