One of the trickier challenges in the telecommunications business right now is building out wireless network systems without going broke. Competition between carriers is fierce, marketing budgets are booming, and consumer demand is heated. Soon a deluge of data traffic will pour onto the airwaves, providing important new sources of revenue—at least in theory. But there are big risks: Serving floods of new subscribers requires major investments in new hardware—around $1,000 per subscriber is an industry rule of thumb—such as towers, antennas, radio transceivers, and back-haul links. Build out the network too far ahead of traffic growth or install gear in the wrong locations, and a carrier risks diminished returns on investment. Build it out too late, and too many customers will suffer poor service and switch to competing carriers. Indeed, wireless companies today are coping with a customer churn rate of 2.5% per month.

Predicting the performance and service quality of any communications network is difficult, but make it a radio network and the complexity of the problem grows exponentially. Radio waves reflect off buildings, don`t always like to go around corners, and quickly fade with distance. The pattern of coverage from a cellular base station varies with every change in the height, vertical and horizontal angles, and shape of its antennas. Turn up a transceiver`s power too far or point its antenna too high and it may disrupt signals in other network cells. Too little power, on the other hand, may cause whole swaths of the surrounding landscape to suffer poor or no service.

By the numbers

It`s this precarious balancing act, involving thousands of interdependent variables, that Schema is out to solve with an impressive piece of software employing genetic, or evolutionary, algorithms. Its Falcom software finds the mathematically optimal solution to network configuration and expansion problems, weighing all the technical, marketing, and financial factors involved and showing the path to maximum ROI. By indicating exactly where a provider ought best to split an existing cell into three new ones, for example, or back off on the power for a certain radio frequency, the Schema software helps service providers make the best use of their two scarcest resources: radio spectrum and cash.

Schema is hardly the only company out to improve the planning and performance of wireless networks. With carriers worldwide spending some $50 billion a year on wireless infrastructure, there`s an estimated $2 billion market for related tools and services. Among others, a French outfit called Forsk, LCC International, and Salient 3 Communications offer network design and frequency planning tools. ScoreBoard sends trucks to prowl cities and suburban streets and identify areas of poor service quality. ScoreBoard presents its analyses and suggestions for improvements almost immediately, via the Web, so carriers can tackle problems quickly. ScoreBoard is probably Schema`s most direct competitor, but as yet it does not optimize entire networks in one fell swoop.

Shuffle and deal

In fact, Schema tells us that its software could take good advantage of the field data that ScoreBoard collects. For now, though, the Falcom program makes do with mountains of call data collected from switches throughout a wireless net. The software mines this data and shuffles it together with a detailed description of the network`s many elements: the locations and positions of antennas, the radio channels used at each location, transmitter settings, and the topography of the local terrain. Finally, customers can enter the various goals of their organization. Marketing may wish to add 10,000 new subscribers to the network over the next six months, while finance prefers to hold down capital and operational expenditures.

Loaded with all this data, Falcom is ready to start crunching, but what the program does exactly is something that Schema`s executives decline to discuss in much detail. Over the years, we`ve seen our share of such black-box technologies, some more worthy of hiding than others, and we appreciate the intrigue and allure that secrecy can give. In Schema`s case, however, we believe the secrecy is necessary, since the basics of genetic algorithms are now well understood and what`s most valuable is the knowledge about how to engineer them for specific industrial problems.

Optimal excitement

Like personalization, optimization is one of those amorphous terms that seem to get applied to more techniques and products with each passing day. From what we can tell, however, the field is truly enjoying growing attention right now from academics, users, and investors. One major optimization success story has been i2 Technologies, which sells its advanced planning and scheduling (APS) product to manufacturers. Advanced optimization techniques are helping in supply-chain execution, airline scheduling, and the production of chemicals.

Any company entering the field can build on a large body of knowledge; mathematically based optimization got its start several centuries ago. It enjoyed particularly rapid improvement during World War II when the Allies, particularly, worked out new techniques for calculating how to deploy scarce resources—B-17 bombers, for instance—for maximum effect, taking into consideration many interdependent constraints and objectives. Thus was born the field of operations research, relying heavily on a technique called linear programming.

Linear programming continues to be popular, and it gains power with every improvement in computing oomph. But the search for new methods able to attack problems that can`t be described in terms of linear equations led to the first work on genetic algorithms in the 1960s. First identified by researchers in artificial intelligence, the idea was to build programs that mimicked the processes of cross-fertilization and Darwinian natural selection to gradually zero in on difficult-to-find solutions.

As in the natural world, genetic algorithms start with several random sets of parameters (analogous to chromosomes containing genes) and combine them according to predefined rules to create new sets. The process repeats again and again, but at each stage, or generation, the new set of parameters is tested to see how well it solves the problem. Every so often, a mutation may be thrown in (a randomization of one value, for instance). Over time—minutes to as much as a day of computing time, depending on the situation—this process homes in on the best possible set of parameters.

Darwin`s children

In the past few years, we`ve seen genetic algorithms successfully commercialized for designing everything from optical networks to aircraft wings and factory schedules. In the early 1990s, a team at General Electric applied the technology to designing jet engines. The GE team left in 1994, with a license to the technology, to form Optimum Technologies, since renamed Engineous Software. Backed by venture money, the firm is applying its design optimization techniques to nuclear plants, aircraft wings, and heart pacemaker antennas.

The list goes on. Farm machinery maker John Deere is using genetic algorithms to optimize factory production schedules. Through its acquisition of Optimax, i2 has added genetic programming to its supply-chain management toolbox. Natural Selection is working in the defense, biotech, and Internet sectors.

Up in the air

Schema itself has evolved since Yuval Davidor founded the company in 1994, after working for the Israeli military and the Weizmann Institute. Originally, he applied his knowledge of evolutionary programming to the problem of scheduling the movement of shipping containers. The company switched in 1997 to helping wireless network operators, winning contracts with Bell Atlantic Mobile, BellSouth, Cellcom Greenbay, and Pelephone, an Israeli cellular operator. Its Falcom software was designed first to handle analog and TDMA/GSM digital networks, but Schema recently began work with Watchmark, a network management venture owned by Lucent Technologies, on optimizing CDMA networks. The company provides its software under license—typically $100,000 and up, and lasting two to five years—and also charges customers for services.

Human factors

We don`t imagine that Schema will ever be more than a niche player in the wireless wars, a provider of specialty software and services interesting to a relatively small number of wireless companies. But its niche could grow into a quite lucrative one, especially as the expected boom in all-new wireless data networks takes place.

Potentially more interesting, over the long term, will be Schema`s role as a pioneer in an exciting field of technology with uses in many fields. As Mr. Davidor sees it, evolutionary programming contains the seeds of a computational theory of human innovation. Brainstorming, for instance, involves a cross-fertilization and exchange of important ideas. Useful ideas survive to the next round of conversation and useless ones are left behind. Not so long from now, he maintains, scientists and engineers, especially, could find themselves relying regularly on computers to help with analyzing new ideas and identifying new avenues of thought. Skeptical as we usually are about the prospects for artificial intelligence, we think the idea of machine-enhanced innovation bears, well, thinking about.

To View Article