Wireless Controlled from the Cloud

Rolling out next-generation wireless networks can be painstakingly slow and patchy at the best of times, as the U.S. deployment of 3G has shown. But IBM researchers in China reckon that shifting the signal-processing requirements from base stations into the cloud will make it cheaper and easier to upgrade networks. Ultimately, the approach could lead to wireless networks that can provide better coverage by rapidly adapting to user demand.

A new architecture called the Wireless Network Cloud (WNC) marks a step away from using dedicated hardware in the radio base stations that serve wireless networks like GSM and 3G cell phone networks, says Ling Shao, senior manager of System Software and Appliances at IBM’s China Research Lab, in Beijing. With WNC, the radio antennas are physically decoupled from the base stations, with the latter existing virtually, within general-purpose data centers at distant locations.

All of the signal processing–the modulation and encoding of the signals to and from the physical antennas–is carried out using software radio technology, adds Yonghua Lin, manager of IBM’s Next Generation Network Appliance. With multicore and multithreading techniques, it is possible to use general-purpose data centers to carry out the signal processing entirely in software, she says.

This allows the network to be managed in a more centralized way, with the raw signals being relayed to and from multiple antennas, which IBM calls “remote radio heads,” via optical fibers from as far away as 40 kilometers.

The main attraction to network operators is cost, says Lin. Traditional base stations currently account for about 40 percent of a network’s total cost, she says. And yet because of the proprietary design of the hardware used, whenever a network is upgraded, almost all of this equipment has to be replaced. WNC upgrades can be implemented relatively cheaply by installing new software.

The centralized nature of this approach also means operators can manage their networks more efficiently. In areas where cell traffic may vary dramatically depending upon the time of day–business districts where daytime traffic is heavy but evening traffic is light, or residential areas where the opposite is true–WNC should allow the network operator to allocate resources when they’re needed, says Lin.

Existing base stations tend to be linked directly to the network gateway, but not to other base stations. Pooling the software radio resources within a data center makes the network much more adaptive to user demand, says Lin. “We can dynamically reallocate resources across different base station cells,” she says.

“IBM’s concept is not totally new but rather a combination of familiar themes,” such as software-defined radio, network equipment virtualization, and networks as software,” says Dipankar Raychaudhuri director of the Wireless Information Network Laboratory (WINLAB) at Rutgers University. But the work is part of an important trend, Raychaudhuri says, with research moving steadily toward early technology implementations, he says. “For example, there is a U.K. company called picoChip that offers a variety of base stations in software using the same parallel computing platform,” he says.

“I think there is a good, strong argument to use software radio in base stations,” says David Grace head of the communications research group at the University of York, U.K., and chair of the World Universities Network Initiative on Cognitive Communications. Grace says the approach is capable of handling the signal processing required of base stations, but he is less convinced about the need to relocate this software and absorb it within the cloud. “It’s the old argument of whether to centralize or distribute,” he says.

Centralizing software in the cloud would allow for more efficient use of processing resources and would make future upgrades considerably faster and cheaper, Grace says. “The disadvantage is, you have to introduce the additional communication between devices to allow that to happen,” which can cause problems. Even with analog-to-digital converters at the antennas, the raw streams of data they would create would be of very high frequencies. Processing this is notoriously difficult because it requires even higher frequency sampling, he says.

Even so, IBM appears to have managed it. “We already have a prototype for WiMax,” says Lin. This has demonstrated that a general-purpose data server can handle the software radio requirements, she says. The team is now developing a prototype 4G mobile phone network and is talking to mobile operators in China.

The potential for WNC is huge, says Shao. Beyond reducing infrastructure costs for network providers, it will improve their ability to manage mobile virtual network operators–phone companies that own no infrastructure and instead lease it off larger companies. And although WNC would be most suitable for urban areas with very high density, it could also help in rural areas where network nodes can be hard to access and have unreliable power supplies, he says.