New training pattern selection method for ATM call admission neural control

Estrella, Antonio Díaz; Jurado, A.; Hernández, Francisco

doi:10.1049/el:19940371

Cited by 9 publications

(2 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As every window corresponds to a range of the input-space, patterns are classified according to their inputs and stored in tlie corresponding window. There are many ways to split the pattern memory and the input-space [9]. In this case an n-dimensional niesli (where n is tlie nunibcr of traffic types) is used according to the number of .…”

Section: Neural Cac Based On the Network Statementioning

confidence: 99%

“…Tlicse patterns are stored in a pattern memory, and a random pattern is taken periodically from that memory to train tlie ANN. If patterns from all input-space zones in which tlie CAC works are to be kept, thus avoiding information loss from previously visited zones when the pattern memory becomes full, tlie pattern incinory mrist be split into blocks called 'windows' [9]. As every window corresponds to a range of the input-space, patterns are classified according to their inputs and stored in tlie corresponding window.…”

Section: Neural Cac Based On the Network Statementioning

confidence: 99%

See 1 more Smart Citation

Neural call admission control through virtual links estimates

Herrero

Díaz-Estrella²,

Sandoval³

IEEE ATM '97 Workshop Proceedings (Cat. No.97TH8316)

Self Cite

View full text Add to dashboard Cite

A Neural Call Admission Control (NCAC) decides whether a new connection can be accepted, according to a neural network decision. Such a decision is based on traffic information (e.g cell loss rate (CLR) ) measured in real time., at tlie ATM node, which ensures continual training of an artificial neural network (ANN) during NCAC performance, allowing the CAC to adapt to possible changes in traffic conditions. This paper addresses the problem of providing accurate CLR information to pcrforin optimal ANN training , which results in efficient NCAC performance. This paper proposes a novel estimation method based on measuring tlie CLR at virtual links with different and slower output rates. This information can be related to the real CLR, by incans of an ANN, thus solving tlie accuracy and estimation-duration problems of real-link estimates. Prior inforination which allows tlie ANN to interpolate the real CLR is also rcquircd to establish the relationship between the virtual and tlie real CLRs. This information lias been named Zero Loss Bandwidth patterns. INTRODUCTIONCall Admission Control (CAC) is a key element for traffic control i n ATM networks CAC must decide which combinations of conneclions of the different traffic types existing In tlie network can be supported without any loss in the quality of service (QoS) required by each type The only QoS parameter considered in this case is the cell loss rate associated uitli the trafic aggregate accessing to tlie ATM node Many attempts have been made to compute such a loss rate analytically, [1][2], but have always come up against the same problem. namely that tlie algorithms developed have been so eoinplex as to render their rcal-time application impossible, in inany cases for computational reasons; and secondly, that inany approxiinations are accomplished with the traffic and queue models used to develop these algorithms The advent of new traMic services means that tliese formulae must be continuously altered, and new forinulae developed Other adinission control methods, based on artificial intelligence systeins such as artificial neural networks (ANNs), have emerged as an alternative to analytical methods Thcsc neural admission controls have

show abstract

Section: Neural Cac Based On the Network Statementioning

confidence: 99%

Section: Neural Cac Based On the Network Statementioning

confidence: 99%