Congestion control for ATM multiplexers using neural networks: multiple sources/single buffer scenario

Area of Science:

• Computer Science
• Telecommunications Engineering
• Artificial Intelligence

Background:

• Congestion control is critical in Asynchronous Transfer Mode (ATM) networks, particularly at the User Network Interface (UNI).
• Existing methods often adjust coding rates for all traffic sources simultaneously, which can be inefficient.
• Maintaining quality of service (QoS), especially for real-time traffic like voice, is a significant challenge during congestion.

Purpose of the Study:

• To propose a new neural network-based method for congestion control at the ATM network UNI.
• To develop a control strategy that selectively adjusts coding rates for a subset of traffic sources.
• To minimize cell loss rate (CLR) and ensure high-quality information delivery, particularly for voice sources.

Main Methods:

• A neural network model is employed to manage congestion control.
• The proposed method dynamically adjusts the coding rate for only a portion of traffic sources during congestion.
• The controller outputs include the source coding rate and the percentage of sources operating at that rate.

Main Results:

• Simulations involving 150 Adaptive Differential Pulse Code Modulation (ADPCM) voice sources demonstrated the effectiveness of the proposed method.
• The new approach significantly outperformed previous methods in key performance indicators.
• Key performance improvements were observed in cell loss rate (CLR) and overall voice quality.

Conclusions:

• The proposed neural network-based congestion control method offers superior performance compared to traditional approaches.
• Selective adjustment of source coding rates is an effective strategy for mitigating congestion in ATM networks.
• The method successfully balances the need to reduce cell loss with the requirement to maintain high voice quality.