Searching for Quality Metrics in an Asynchronous Universe

Delay and delay variability and packet loss are important quality metrics, particularly
if we need to deliver consistent end-to-end application performance. The change in
handset hardware and software has increased application bandwidth—the need to
simultaneously encode multiple per-user traffic streams, any one of which can be
highly variable in terms of data rate and might have particular quality of service
requirements. This chapter demonstrates how offered traffic is becoming increasingly
asynchronous—bandwidth is becoming burstier—and how this exercises network
hardware.
In earlier chapters we described how multiple OVSF codes created large dynamic
range variability (peak-to-average ratios) that can put our RF PAs into compression.
This is a symptom of bursty bandwidth. On the receive side of a handset or Node B
receiver, front ends and ADCs can be put into compression by bursty bandwidth (and
can go nonlinear and produce spurious products in just the same way as an RF PA on
the transmit path). As we move into the network, similar symptoms can be seen.
Highly asynchronous bursty bandwidth can easily overload routers and cause buffer
overflow. Buffer overflow causes packet loss. Packet loss in a TCP protocol-based
packet stream triggers “send again” requests, which increase delay and delay variability
and decrease network bandwidth efficiency.
We need to consider in detail the impact of this increasingly asynchronous traffic on
network architectures and network hardware. In practice, we will see that neither traditional
circuit-switched-based architectures nor present IP network architectures are
particularly well suited to handling highly asynchronous traffic. We end up needing a
halfway house—a circuit-switched core with ATM cell switching in the access network,
both optimized to carry IP-addressed packet traffic.
In the first chapter of this Part, we study the RF parts of the network and how the RF
subsystems need to be provisioned to accommodate bursty bandwidth. We will find
that adding a radio physical layer to a network implicitly increases delay and delay
variability. It is therefore particularly important to integrate radio layer and network
layer performance in order to deliver a consistent end-to-end user experience.