![]() |
|||
![]()
|
![]() |
![]() |
![]() |
5.2. WIRELESS OR SATELLITE COMMUNICATIONA cellular network is a telecommunications network with some distinct features. It consists of a mobile switch center and a number of radio base stations, each responsible for covering a geographical service area called a cell. A mobile unit (portable or handheld) communicates with the base station via a voice channel. All base stations are linked by some transmission facilities to the switch center, which coordinates the operation for the entire system and serves as a connection point to the public switched telephone network. For satellite communication, each satellite can be treated as a big cell in this context. The cellular industry has been experiencing tremendous growth in recent years and needs AI technology to assist the operation and management of cellular networks. AutoCell is a distributed client/server expert system operated by Singapore Telecom (Low, 1995). AutoCell periodically acquires cellular network status and traffic data. Based on these, it generates traffic forecasts for all cells, and performs automatic frequency reassignments to make more channels available at cells that are congested due to unexpected high demands or faulty channels. AutoCell also provides a performance reporting facility. AutoCell is implemented based on a multi-agent architecture where each agent is assigned a specific function, and agents communicate via the exchange of messages. It uses a heuristic search approach that combines hill climbing and branch-and-bound pruning to perform dynamic channel assignment. The revenue generated by improved traffic capacity due to AutoCell is estimated at over 1 million Singapore dollars in the first year (1994) alone. PERFEX is a performance analysis and tuning expert system for cellular networks developed at GTE (Tan, 1996). Like AutoCell, it collects and displays the network status and traffic data. Based on performance and configuration data, PERFEX uses a neural network to discover generic performance problems in the network, and then uses rules to generate expert advice on how to fine-tune the system parameters to improve performance before resorting to adding or reassigning channels. These parameters include handoff thresholds, reassignment of handoff neighbors, configuration errors, and dynamic power control parameters. PERFEX provides a set of cellular tools to examine the network in finer detail and tightly integrates its different information presentation forms such as the map, tools, graphs, reports, and templates. PERFEX is in daily use at most GTE mobile switch centers. InCharge is a system developed by SMARTS for real-time isolation and handling of network system problems (Kliger, 1996). InCharge employs a coding approach that reduces the event correlation time significantly by replacing a causal graph with simple codebooks. In a codebook, each problem is associated with a collapsed set of symptoms. At runtime, the actual symptoms are compared with the ones in the codebook by calculating the Hamming distance. The stored problem with the smallest Hamming distance is selected as the actual problem. InCharge has been adopted by Motorola Satellite Communications, which is using InCharge's codebook event correlation for their IRIDIUM project. The IRIDIUM project is a worldwide satellite-based communications system that will receive thousands of problems or symptoms at very high rate in real-time. CHAMP is a churn analysis, modeling, and prediction system developed for GTE Mobilnet (Masand, 1996). Churn, a term for customer disconnecting the cellular service, is a very serious problem for the cellular industry, with churn rates ranging between 20 and 30% a year in most markets. CHAMP tries to identify those customers most likely to churn, which can be contacted for proactive churn prevention. CHAMP analyzes billing data using neural networks, decision trees, case-based reasoning, or a combination of the three methods. Its training and test data are preprocessed by eliminating irrelevant data fields, pruning unrelated subscribers, merging data from different months, sampling data, and selecting best fields using decision trees. The CHAMP system is able to identify a large part of predictable churn, with prediction rates (lift) usually 5 to 6 times better than random.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |