Ideally, there’s a system to simplify and automate the task of radio network management, but is such a thing even remotely feasible given the added complexities of data networks?
One of the hallmarks of the earliest days in the wireless industry was the extent to which the realities of network design frustrated the attempts of carriers to impose order on the process. Engineers approached the problems of providing more-or-less ubiquitous coverage over large metropolitan areas armed with notions of nice, neat hexagonal grids and orderly frequency reuse patterns. Of course, the real world was, and is, far messier than the simplistic models suggested by these early engineering tools. Irregular terrain, wildly fluctuating RF propagation characteristics of urban “clutter” and unpredictable usage distribution all contributed to make the process of wireless network design vastly more complex than anybody realized.
Successful network engineers have learned that management of wireless networks, particularly in the areas of design and accommodation of traffic growth, is far more a stochastic process than a deterministic one. What that means is that the process does not lend itself particularly well to the application of standardized, “rules based” solutions. At this point, it should be noted that the “network management” I am talking about here refers to the radio part of a cellular network. That is, primarily, the location, antenna configuration and provisioning of base stations. Other parts of the network – backhaul facilities, for example – are far more deterministic.
The stochastic nature of cellular radio networks is a fact of life, but that doesn’t make it any less frustrating. More importantly, network RF design always has been highly laborintensive and therefore costly. To be sure, a number of developments in basic cellular technology and planning tools have served to somewhat simplify the process. For example, CDMA air interface technologies reduce the complexity of interference management. Automated frequency planning software tools do the same for narrowband systems. But the overall process for designing and optimizing a wireless network largely remains one of applying human expertise, based primarily on trial-and-error experience.
DATA TRAFFIC PROJECTIONS
For obvious reasons, the challenges in engineering of cellular voice networks were never greater than during the period of intensive growth, mainly during the 1990s. Unfortunately, the costs and complexities of managing anticipated growth in wireless data networks will almost certainly be even greater. For one thing, wireless data growth isn’t just in the number of users: Compared to voice service, the traffic demand per data user is far more difficult to predict. What’s worse, as Internet applications come and go, usage patterns on wireless data networks likely will change significantly, placing a premium on rapid engineering response.
Faced with these realities, it’s hardly a surprise that high on wireless data network operators’ wish lists is a “silver netbullet” in the form of a system that would simplify, and ideally automate, the task of radio network management. But is such a thing even remotely feasible, particularly given the added complexities associated with data networks? Surprisingly, it may indeed be possible to apply a healthy dose of automation to wireless data network management if we build the enabling features into the networks themselves.
To see how these integral features might work, let’s first separate out the network design element. In designing a “greenfield” wireless network, we take into consideration predictions of geographic traffic distribution and localized characteristics of RF propagation. If we have accurate traffic and propagation data, there are actually some good tools available that will assist in coming up with a reasonably optimum network design. Getting accurate data is therefore the key, but until a network is built and operational, it cannot participate in providing such data. So, integrated management features won’t help in initial network design.
Fortunately, the reactive part of wireless network management is generally more important than the initial design. Once the network is up and running, its engineers will need to make modifications, first to fix deficiencies in that design and then to deal with ongoing growth and changes in traffic characteristics. That’s where the biggest headaches lie and where integrated management features offer the greatest promise. To see how such features might work, it is helpful to consider the process by which network management reacts to current and evolving service issues.
IDENTIFYING QOS ISSUES
The first step in this process is identification of radio issues that require management attention. These might include current problems such as poor service quality in a particular location, or anticipated problems such as growth in traffic in a particular cell or sector that will, based on current trends, outstrip channel capacity in the foreseeable future. Monitoring peak usage trends against capacity is fairly straightforward, and most networks already have features that can alert the operator to predicted shortfalls.
Identification of location-specific service quality problems is far more of a challenge. Fortunately, most current types of user equipment (UE) have positioning capabilities that could be used to location-stamp a report of poor service quality. Such a report could be triggered automatically upon various predetermined conditions that are assumed to indicate less than acceptable quality of service (QoS). Alternatively, UEs could be programmed to periodically measure and report QoS by uploading and downloading predetermined (and very brief) test files. Of course, these methods would not directly indicate locations with no service whatsoever (that is, coverage “holes”), but over time such “holes” could be identified, or at least suggested, by a lack of test reports.
Once QoS problems are identified, an automated management system would need to diagnose the cause or causes. For example, poor QoS in the form of low downlink speeds in a particular location could be caused by excessive channel loading or by poor signalto-noise ratio due to low receive signal level or high levels of interference. Between the UE and the serving base station, there is a substantial array of data just waiting to be captured and processed in order to identify the specific cause or causes of a given incidence of poor QoS, and there is really no reason this process cannot be automated.
The final step in the management process is to determine optimal corrective measures for existing and anticipated network problems. This is by far the part that would be most difficult to automate because it requires a comprehensive understanding of radio characteristics, primarily those related to propagation in and around the problem location. For example, excessive interference on the downlink in a particular area might be addressed by adjusting antenna downtilt on the interfering base station sector. But that change could very easily introduce a whole slew of QoS problems elsewhere. To reliably engineer a “fix” for a known problem, the automated system would have to know this in order to avoid making things worse. But the really good news is that with modern optimization software algorithms, an automation system that does have accurate propagation data can rapidly consider far more complex combinations of configuration changes than even the most experienced team of human engineers.
In other words, the real key to developing a silver bullet for automating wireless data network management is finding a way to obtain accurate and comprehensive data on RF propagation characteristics. Automating that task would require – surprise! – yet another silver bullet. In the meantime, we can at least develop and take advantage of integral network features to help with problem identification and diagnosis.
Drucker is president of Drucker Associates. He may be contacted at email@example.com.