The folks at Novarum think Wi-Fi might actually work city-wide: Ken Biba and Phil Belanger, the founders of Novarum, released a few statistics from their tests of 14 cities in the US that contain large-scale Wi-Fi networks that are substantially deployed or complete. Their results are the strongest endorsement yet that these sorts of Wi-Fi networks can work, but also provide a better sense about the infrastructure cost required to provide full coverage and broadband speeds. (You can read my interview with Belanger from last November here.)

In a set of emails with Biba and Belanger, I asked based on their testing how many nodes per square mile were truly needed to provide service? Biba explained that the intent of the network mattered, but that about 40 nodes per square mile "plus or minus a few depending on tree cover and terrain" was ideal for suburban installations. This is above the 35 nodes per square mile figure EarthLink was putting out last year, and far above numbers provided as recently as a year ago (in the low to mid-20s). A 100-square-mile network that was budgeted for 25 nodes per square mile that really needs 40 would need an extra 1,500 nodes that could cost $3m to $5m additional.

Belanger's sound bite: "40 is the new 20."

St. Cloud, Florida's free Wi-Fi network was the only one of the 14 with 100-percent service availability, which Novarum defines as not just a signal, but the ability to perform useful work over the connection. The No. 2 slot, Anaheim, Calif., scored just 72 percent, all the way down to 44 percent in Galt, Calif., in position 10. That's the area that needs improvement, and it's a direct correlation to infrastructure density, the Novarum founders say.

Toronto Hydro installed as many as 126 nodes over one square mile, which seems insanely high, except for two factors: first, it's a dense "urban canyon" with a huge number of tall buildings; second, performance was remarkable, with 5 Mbps symmetrical in many locations. This speed is highly competitive with wired broadband, and it's the first time that such consistent performance has been demonstrated in a Wi-Fi network. Other networks might be able to achieve those speeds but throttle per-user bandwidth.

The density of population could allow that many nodes to be profitable, too. Novarum didn't test coverage in higher floors yet, but plan to. Toronto scored only 60 percent on service availability in the tested area, which shows that simply putting up nodes doesn't ensure ubiquity.

Also surprising was how cellular outranked Wi-Fi in the cities surveyed. Eight of the top 10 wireless broadband overall scores went to cell networks--and not a single cell provider but divided among Sprint (three cities), Verizon (three cities), and Cingular (two cities). This measure looked at service availability, performance, ease of use (including login, network discovery, etc.), and value (price versus other factors). Cell can't beat Wi-Fi for speed in ideal cases, but it can for overall experience and ubiquity.

Update: David Haskin offers additional detail in his report on this study.