Technical Preparation

So how do you build a functional cellular basestation from a USRP on short notice and with limited funds? Here's one answer.

Radio Gear

We borrowed a 30-ft Army surplus AS-2236 mast from a consulting client. These run $200-$400. We also bought a 8'-long 9 dBi marine cellular antenna, new, for $100, to put on top of it.

We scored a dB Products DB4090 cavity duplexer on eBay for $60.

We used our existing USRP, RFX900 daughter cards and an Apple MacBook Pro for the digital radio, GSM stack and Asterisk server. (Those are Nokia 3310 and 5110 handsets on top of the USRP.) That USRP equipment cost about $1,500. Running at full load (7 active lines) the multi-threaded basestation consumed about 50% of the Intel dual-core 2.33 GHz MacBook.

Pesky Power Amps

The power amp turned out to be a problem. First, we bought a 50 W Vocom 900 MHz PA off of eBay for about $60.

The problem was that the USRP didn't generate a hot enough signal to drive it. When we looked inside this Vocom amp, we found a perfectly good Motorola UHF module, wrapped with a bunch of other stuff we didn't want or need.

So now we were one week from a site test without a serviceable power amp. We had three choices:

  1. Gut the Vocom amp, leaving just the Motorola module and a few caps.
  2. Borrow the Down East Microwave 3318PA offered by a friend, a 18W/33cm power amp based on discrete transistors.
  3. Order our own Down East 3340PA and hack it as needed to get the full use of the Mitsubishi RA20H8994M module inside it. That's about $200 with shipping if you buy it in kit form.
We decided to pursue options 2 and 3 in parallel since 2 was low-effort and both were low-risk.

A couple of days later, we picked up the 3318PA and hooked it up to the USRP. We put 23 dBm in, tweaked a few trim pots and got 33 dBm out. Our STA allows 39 dBm ERP. So 33 dBm at the PA, plus 9 dBi antenna gain, minus 3 dB of cable and insertion losses and bam! We are right on the money. We'll bring the 3340PA kit just in case we need it, but time is tight so we're moving on to the next hurdle.

The Receive Chain

Our calculations showed that our receive chain needed 16 dB of external gain to be thermal noise limited (instead of gain limited). But experience with the USRP also showed that there was a lot of crosstalk noise from the transmit section into the receive section, so extra gain would be required to overcome that. [Note: We later discovered that this crosstalk could be greatly reduced through proper gain settings in the RFX cards.] Fortunately, a surplus house in Atlanta was unloading a whole pile of Angle Linear GaAs FET preamps for $50 each on eBay. We bought 3. The only hitch is that the preamps had TNC connectors and everything else in our system was SMA or N.

Duplexer Drama

On the Friday before Burning Man, we finally got some TNC adapters we'd been waiting for and we could fire up the full transmit and receive chains together for a bench test. To our horror, we realized that our duplexer had insufficient isolation and the transmitter was bleeding into the receiver to the tune of about -30 dBm. Since the thermal noise floor for GSM is about -113 dBm and the narrowest part of the USRP receive chain only had about 40 dB of dynamic range, this was a serious problem, especially on a Friday afternoon just before we were supposed to hit the road. We needed an extra 40-50 dB of isolation and we needed it in a weekend.

Our "technical team" conferred by e-mail and arrived at a four candidate solutions:

  1. Re-wire the radio path so that we can use the two duplexer cavities in series on the receive side. At 65 dB per cavity with 20 dB of combiner isolation this would give about 150 dB of transmitter suppression, which would be more than adequate. (Bill Alexander suggested this.) This was low risk and maybe an hour of effort.
  2. Build a notch filter with stubbed lines and Smith-chart voodoo. This would be about four hours of effort and might give enough isolation, but it will be hard to tune.
  3. Build an interdigital bandpass filter on the receive side. Given the time available, this filter will have to be built entirely of components from local retail stores. This would take about six hours, but it was the most solid technical solution.
  4. Hack the GSM stack so that we only operated on timeslots 0-3, effectively making the system half-duplex. Frankly, this would suck and might not even work with a lot of phones since it would result in a discontinuous beacon.
We decided to pursue options 1 and 3 in parallel. Harvind started rewiring the duplexer and David took off to the local hardware store.

Harvind rewired and retuned the cavities for option 1. He got great transmitter suppression, but the USRP was outputting so much receive-band noise on the transmit side that the overall system performance wasn't any better. The good news, though, the cavities had been mistuned the first time around and we picked up an extra 20 dB of isolation by fixing that. Harvind also made some adjustments to the internal gain stages in the USRP that bought us about 10 dB more useful dynamic range. The system was serviceable again and only needed about 10 dB more isolation for our range to be downlink-limited.

Meanwhile, David built an 840 MHz 2-element filter with a hacksaw and hand drill.

It was ugly, leaky and poorly-matched, but thanks to all those GaAs-FET preamps we had gain to spare and it gave us a another 40 dB of isolation. With this filter in place, our range should be thermal-noise limited, although we were still concerned about crosstalk between the PCBs inside the USRP case. [Note: We later discovered that this crosstalk could be greatly reduced through proper gain settings in the RFX cards.]

Electrical Power Gear

We needed a power source that could produce an average of 80 W all week. We looked at solar, but then we read about how windy the Playa is most of the time, so we ordered a Southwestern Windpower Air X turbine for about $500 and picked up an RV battery at Sam's for $80. Here they are getting bench-tested.

(Yes, that's an R&S CMD-57 test set and some classic Tektronics mainframes in the corner. The Tek stuff's going to travel, but the R&S box will probably stay home.)

We built up a power distribution board that included fusing, safety disconnects and a shut-brake for the wind turbine, along with regulated 6-volt and 5-volt buses so that we could the run USRP and network equipment without wasting power through inverters and wall-warts. This took about $400 in parts, with most of that going to regulation for the low-voltage buses. The clever part was the shunt brake, which shorts the turbine generator when it is disconnected from the rest of the system. Normally the shunt brake and disconnect are built together in high-current double-pole switch. What we did instead was get a matched pair of pull-out disconnects and then hide one of the pull-out handles. The probably saved us $200 in electrical hardware.

Network Equipment and SIP Support

A campmate volunteered an outdoor WiFi bridge for backhaul. We bought a PAP2 adapter, ethernet hub and a wired PSTN terminal (a 'telephone') just in case the basestation got flakey and we needed a reliable line. Together, that was about $80. We also set up an account with link2voip and leased ten E.164 DIDs ('phone numbers') in the NANP 650 and 707 area codes.

Another key component of the site would be the Multitech GSM/GPRS modem, which will be used for interference surveys, power monitoring, and to read the SIMs of test subscribers for provisioning.

At this point, we're as ready as we're going to be. Let's pack.