Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!

Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


3. TELECOMMUNICATIONS DOMAINS AND POTENTIAL TASKS

The telecommunications domains are very diverse, ranging from wireline telephone service, wireless communication, and satellite communication, to Internet service. Each domain has its own requirements. However, they all share the following characteristics, which makes them quite different from other application domains. These characteristics present challenges and opportunities for expert system applications:

  1. Large amount of data -- A typical network switch generates alarms on the order of a million per week and can handle millions of calls per day. Large databases are needed to store various types of data.
  2. Real time -- Many network operations must be carried out online in real-time. Delayed reactions can result in loss of revenue and customer confidence.
  3. Mission critical -- Hardware equipment and software systems often have redundancy built-in and expert systems must be equally reliable. Sometimes, misdiagnosis of a key transmission trunk can have a catastrophic result.
  4. Dealing with legacy systems -- Some switch technologies have existed for decades and it is expensive to replace them right now. Communicating and integrating with them is essential for many expert systems.
  5. Embedded AI -- Expert systems are often just a portion of a large system that uses AI techniques to accomplish special subtasks. The success of an expert system also depends on the rest of the system.
  6. Different user skills -- Some operation support personnel are highly trained and others are not computer literate. Knowledge acquisition must be specialized accordingly.
  7. Adaptive -- Switch version, usage pattern, telecommunications technology, and policy change frequently. Any expert systems should be adaptive to changes.

With these characteristics in mind, the following are some potential AI application tasks in each of four management layers:

  1. Element management layer -- Alarm filtering, monitoring and correlation, and admission control.
  2. Network management layer -- Alarm correlation, fault isolation and diagnosis; repair and maintenance; performance monitoring and tuning; traffic control and routing; configuration management and dynamic channel allocation; and workflow management that coordinates task assignments.
  3. Service management layer -- Service order, customer helpdesk, fraud detection, churning management, Internet service offering, billing automation, international market customization, and language translation and speech recognition.
  4. Business management layer -- Growth planning, facility design, resource scheduling, finance and contract management, and workforce training.

Some tasks are across the layers. They include data classification and interpretation, cooperation and negotiation among different systems, and data mining and trending analysis. In the next section, we discuss which AI techniques can be used for these tasks in telecommunications domains.

4. AI TECHNIQUES

Expert systems use AI techniques to emulate the behavior of experts and automate the operation of systems. Determining a right technique for a specific application is crucial for any AI-based product. There are many AI techniques that have been and can be applied to telecommunications domains. The following AI techniques have been reported extensively. The detail of each technique can be found in many AI textbooks.

4.1. RULE-BASED SYSTEMS

Rule-based systems emphasize declarative knowledge rather than search in the problem space. Knowledge is organized in IF-THEN-style rules. Rules are fired in sequence by an inference engine built in a rule-based shell. This kind of forward/backward chaining of fired rules captures a problem-solving strategy that is highly data driven. One quality of rule-based systems is that individual rules can be added, deleted, or changed independently. On the other hand, rule acquisition is a challenging aspect in the construction of rule-based systems.

Rule-based systems have been widely used in telecommunication domains where a "rule-of-thumb" or data-driven method is suitable. Their tasks include alarm correlation, fault isolation and diagnosis, repair and maintenance, traffic control and routing, service order, customer helpdesk, and resource allocation.

4.2. SEARCH

Search is a generic problem-solving technique. A typical problem space consists of a number of decisions and each decision has a range of possible choice. A search technique allows one to systematically examine each choice at each decision until one finds a satisfactory solution. To speed up the search, one can use problem-specific heuristics to reduce the number of decisions or choices and prune the problem space. Another strategy is to modify the representation of the problem space to simplify the search. The search technique can also be used in different forms, e.g., by genetic algorithm and simulated annealing.

The search technique has been employed for a class of constraint satisfaction problems encountered in dynamic channel allocation, reconfiguring capacity, growth planning, and resource scheduling.

4.3. NEURAL NETWORKS

Typical neural networks consist of several layers of nodes and links that connect the nodes between adjacent layers. They must be trained first before they can classify a pattern. Using an architecture loosely analogous to how neurons are organized in a brain, a neural network learns patterns by strengthening and weakening the weights of link connections between nodes when given a sufficient number of training patterns. Neural networks are suitable for knowledge-poor and data-rich domains. They are adaptive and noise tolerant. Training a large-size neural network is time consuming. This limits the scope of neural network applications.

Neural networks have been applied to the task of classification and interpretation, including admission control, performance tuning, data mining, traffic control and routing, dynamic channel allocation, fraud detection, network reconfiguration, and speech recognition.


Previous Table of Contents Next

footer nav
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.