Increased user expectations concerning the performance of wireless data networks is putting pressure on GSM service providers to find new tools to efficiently and effectively optimize their networks
There is a lot of talk these days about the wireless data revolution. Both trade journals and the popular press carry articles about applications ranging from wireless e-mail to bidirectional videoconferencing. This exposure is certain to raise user expectations for performance of wireless data networks, which in turn will push operators to improve the coverage, capacity and quality of both voice and data systems.
The general thinking in the industry is that new third generation wireless systems will provide the platforms for the more exotic and bandwidth-intensive data applications. However, mature second generation networks, particularly those using the Global System for Mobile (GSM) air interface, are already providing extensive packet data service in addition to traditional voice telephony.
Most European GSM operators have begun offering General Packet Data Service (GPRS) with data rates that (at least theoretically) rival those provided by dial-up wireline connections. At the same time, voice traffic in most GSM networks continues to grow rapidly.
Even as they allocate a portion of their available spectrum to GPRS, operators are being challenged to increase voice capacity while maintaining service quality. But as GPRS service becomes more generally available, operators are finding that data performance is even more sensitive to network design than is voice quality. To address these growing pressures, GSM service providers need new tools to help them efficiently and effectively optimize their networks.
Measuring Voice and Data Service Quality
Traditionally, wireless operators have used a number of metrics that collectively provide a measurement of network service quality from the user`s perspective. Various operators may define and measure network performance somewhat differently, but overall service quality will generally be determined by a combination of the following four key performance indicators (KPIs).
System coverage
the geographic extent over which the network will reliably provide service. Coverage is a critical issue not only in a regional sense but also in terms of holes caused by localized shadowing or failure of radio signals to penetrate into buildings.
Call blockage
The percentage of attempts to make or receive calls that are blocked due to lack of available voice channels (or, less frequently, lack of available network interconnection trunks). Excessive blockage rates point to inadequate system capacity.
Voice quality
An arbitrary score of how much noise and/or distortion intrudes on the voice conversation. In digital systems like GSM, bit error rate (BER) provides a convenient and objective measure of voice quality.
Dropped call rate
The percentage of calls in progress which terminate before either involved party intentionally ends the call.
These voice network KPIs reflect directly on the three main quality concerns of users:
Can the call be made (or received)?
What will it sound like?
Will the call drop?
By contrast, in wireless packet data networks users have what amounts to continuous online connection so long as they are within service coverage. Therefore, for data users service quality boils down to just two questions:
Is there an online connection?
What`s the effective throughput rate?
The different user perspectives for data service quality suggest that operators need different KPIs to measure data network performance. There are parallels in system coverage, of course, but call blockage, voice quality and dropped call rates are not relevant to packet data service. Instead, focus on effective data throughput suggests these two new KPIs:
Peak and average per-user data throughput.
Average per-channel data throughput.
What Affects Quality of Service?
Wireless networks involve both radio and wireline links as well as switching hardware and software and database operations. For the most part, however, service quality will be determined by the performance of the radio network. Putting aside the issue of regional coverage, there are three critical factors that impact the performance of both voice and data radio networks: RF signal level (in both uplink and downlink); level of radio interference (also in both uplink and downlink); and available radio channel capacity. For each of these factors, the effects are somewhat different for voice and data networks.
To provide any sort of service, the received signal level (RSSI) of both uplink and downlink transmissions must be sufficiently above the level of thermal RF noise to allow for successful recovery of the modulated information stream. In most GSM systems, it is the uplink that provides the limiting factor because of limitations on transmit power of portable handsets.
Within the boundaries of general regional network coverage, holes may exist where localized propagation characteristics cause low RSSI, often when the user is indoors. In voice service, the threshold RSSI required is fairly abrupt, and coverage holes are thus well defined.
GPRS networks can adapt to low RSSI by changing the data coding scheme on the affected channel sort of like downshifting in a car with manual transmission. Increasing the robustness of the coding lowers the required RSSI, but also lowers the data throughput rate.
In urban areas with high wireless usage density, the most significant impact on service quality comes from excessive RF interference, the presence of other, undesired signals while trying to receive a desired signal. The dominant source of interference in urban GSM voice and data networks is transmissions from other mobiles and base stations using the same RF channel at the same time, a principal called channel reuse. (Interference arising from transmissions on adjacent channels are also a factor, but typically have less of an impact.)
The ratio of the desired RF carrier signal power to the linear sum of interference power is called C/I. As in the case of marginal RSSI, the required C/I to support voice service in GSM networks is fairly well defined while the adaptive coding in GPRS causes data throughput rates to vary over a substantial range of C/I.
Channel reuse is essential in large GSM networks because the RF spectrum available to each operator is limited. The density of channel reuse–how closely located are sectors using the same RF channel–largely defines the capacity of the network. Increasing reuse density in a given area increases the number of channels and capacity available there but also increases the likelihood that a radio link will suffer from excessive interference from one or more nearby co-channel sectors.
Capacity obviously affects network service quality, but the ramifications are quite different in voice and data service. Inadequate capacity causes excessive call blockage in voice networks while excess capacity provides little benefit. In GPRS networks, localized capacity will directly weigh on per-user throughput rates. When available channels are overly loaded a given user`s data will not stop like a blocked voice call, but rather will slow to an infuriating trickle.
Optimizing Service Quality
To maximize its voice service KPIs, a GSM operator will aim to provide adequate uplink and downlink RSSI wherever a user might reasonably expect service, adequate localized capacity to meet peak demand everywhere within the network, and minimal probability of excessive levels of interference (areas of low C/I). In GPRS networks, optimization objectives are not so simply stated because there might be other definitions for adequate RSSI, capacity or C/I.
Instead, data service KPIs will be maximized when coverage, capacity and interference levels are balanced in each area so as to maximize per-user and total throughput, a task made particularly complex in GPRS networks by the multiple levels of coding schemes used. However, GSM voice and GPRS networks do share common categories of design factors, as follows:
Frequency Planning. The allocation of channels to each sector defines co-channel relationships and thus areas and levels of co-channel interference. Optimal frequency plans allow for maximized channel reuse density while maintaining adequate C/I, whether by increasing density of base station deployments or more desirably by increasing the average number of channels assigned per sector.
Powerful automatic frequency planning (AFP) tools are available to generate optimized plans, but they must take into account the different characteristics of voice and data service. For example, the frequency hopping used in GSM voice networks to improve apparent quality must be factored in frequency planning. In GPRS networks the possibility of separate allocations of uplink, and downlink channels and the dynamic time slot allocation feature, must be considered.
Network Physical Configuration. This includes all aspects of network infrastructure deployment–locations of base stations, heights of towers, sector azimuth orientations, antenna selection and tilting, and so forth. An optimal configuration will provide adequate coverage while minimizing excessive sector-to-sector RF footprint overlap that increases interference.
Unlike frequency planning, configuration optimization for mature networks needs to be tempered with the practical realities of cost and time involved in making wholesale changes. Software tools for optimizing physical configurations are currently being developed, but until they are available operators can use AFP tools to determine the performance impacts positive or negative of any proposed configuration changes.
Operational Parameter Settings. Operators of GSM systems must determine a variety of parameter settings that define operational characteristics. These include such factors as handoff thresholds, power control levels, handoff neighbor lists and so forth. GPRS networks have their own list of operational parameters.
Optimization of these parameter settings is crucial to maximizing network performance, and generally will depend upon physical configuration design. Therefore, an effective configuration optimization tool should also include parameter optimization per cell, while taking into consideration network-wide influence.
Two elements are critical to the performance of all wireless network optimization tools. First, they must use a very powerful optimization algorithm such as a genetic algorithm capable of finding global optima in extraordinarily complex systems. For example, consider the AFP task for a network of 1000 sectors with four channels to be assigned to each from an available set of 50 channels. The number of different possible frequency plans would be (50*49*48*47) 1000, an unimaginably large number!
The second critical element required for these tools is their ability to use extensive detailed data to accurately model the network in terms of RF propagation characteristics and geographical distribution of usage. Sources for this data include drive testing, operational performance data collected by the network itself, and possibly computer simulations of RF propagation (although the latter is often quite inaccurate in urban regions). To be most effective, an optimization tool must be capable of ingesting, processing and correlating a wide range of network data from a variety of sources.
The task of wireless network optimization is highly complex and specialized, with somewhat different approaches required for GSM voice and GPRS systems. It is also a task with enormous potential rewards, as each incremental improvement in system performance can translate to huge cost savings and increased revenues for the operator. For assistance in this effort system engineers can use mature and highly effective AFP tools and, in the near future, new and exciting tools for configuration and parameter optimization.
Roni Abiri is Vice President-Wireless Products for Schema, Herzelia, Israel. He can be reached at roni.abiri@schema.com. The company`s U.S. offices are located in Rochelle Park, NJ.
To View Article 
.gif)
