Voice over DSL (VoDSL) is a technology that enables the transport of data and multiple voice calls over a single copper-pair. VoDSL employs packet voice technology instead of the traditional circuit switched voice. Voice over ATM (VoATM) and Voice over IP (VoIP) are the two main alternatives for carrying voice packets over DSL. ATM is currently the preferred technology, since it offers the advantage of ATM's built-in Quality of Service (QoS) mechanisms. IP, on the other hand, cannot provide QoS guarantees in its traditional form. IP QoS mechanisms have been evolved only in the recent years. VoIP has gained popularity in the core networks. If it could replace VoATM in the access networks, it would open the door for end-to-end IP telephony that would result in major cost savings. In this thesis, we propose a VoIP-based VoDSL architecture that provides QoS guarantees comparable to those offered by ATM in the DSL access network.

Our QoS architecture supports Premium and Regular service categories for voice traffic and the Best-Effort service category for data traffic. Voice and data packets are placed in separate output queues at the bottleneck link. The Weighted Fair Queuing algorithm in used to schedule voice and data packets for transmission over the bottleneck link. Fragmentation of large data packets reduces the waiting time for voice packets in the link. We also propose a new admission control mechanism called Admission Control by Implicit Signaling. This mechanism takes advantage of application layer signaling by mapping it to the IP header. The router can infer the resource requirements for the connection by looking at certain field in the IP header of the application layer signaling packets. This eliminates the need for an explicit signaling protocol.

We evaluate the performance of our QoS architecture by means of a simulation study. Our primary metrics are the end-to-end delay of voice packets across the access network and the bandwidth consumed by a voice call. Our results show that the end-to-end delays of voice packets in our VoIP architecture are comparable to that in the VoATM architecture. ACIS limits the number of voice calls admitted into the premium service class and provides guaranteed service to those calls under all loads. It also provides acceptable service to regular calls under light loads. We also show that PPP is a better choice than ATM as a Layer 2 protocol for our VoIP architecture. PPP offers the advantages of low bandwidth requirement and interleaving of voice packets in between fragments of large data packets during transmission over the bottleneck link. We conclude that our VoIP architecture would be suitable for future VoDSL deployments.