In this talk we present a study of capacity and traffic models for multiple services in CDMA systems. CDMA transmission provides a better exploitation of the scarce radio bandwidth that greatly enhances cell capacity, making it possible to deploy broadband services so far mainly known in wired systems. In fact, CDMA deployment gradually results in the cellular version of a so-called "multiservice system", where the voice service co-exists with fast data and multimedia services, in the form of internet, file transfer and video applications.

After an introduction to notions of capacity in CDMA, we start by generally studying the Erlang capacity of such systems. We extend the definition of Erlang capacity to comprise multiservice systems and study its behavior upon scaling the rates of transmission. It is shown that a significant increase of Erlang capacity is possible by scaling down these rates, which nonetheless comes at the expense of increased transfer delay and energy consumption. On the opposite side, an increase of Erlang capacity by scaling up these rates is only possible for non-simultaneous transmissions, but -even in the best case scenario- a considerable increase demands rates unrealizable in current systems.

For modeling purposes, we classify multiservice traffic in two categories: real-time (RT) conversational and streaming traffic, with strict QoS requirements, and non-real-time (NRT) traffic, comprising elastic applications.

We then proceed to study a model of elastic data traffic with a homogeneous time-varying service. We consider a CDMA system providing a homogeneous service to different data service users, in spite of possibly different channel conditions. The service is modeled as a generalized processor sharing (GPS) scheme, and xpressions for performance measures such as the sojourn time of flows and their blocking probabilities are derived for different traffic arrival patterns, namely Poisson arrivals and Engset-like arrivals from a finite source population.

Finally, we present a more complicated model of integrated RT calls with guaranteed service and elastic flows with processor sharing. In a Markovian context, we use quasi-birth-death (QBD) process theory to derive performance measures and study the control of shared resources in such systems.