Delivering Quality of Service

From the user’s perspective, the critical measurement is application performance. From
the network operator’s perspective, the criteria required is to be able to demonstrate and
prove that pre-agreed application performance metrics have been delivered, which
requires some form of proof-of-performance reporting. Delivering a particular quality of
service, therefore, requires us to measure and manage access, network, and application
latency, and access, network, and application jitter (application delay variability).
Many parameters interact with quality of service and are dependent on our ability
to measure and manage radio and network bandwidth quality. These include the
following:

Error protection choices. We must decide whether we put error protection at bit
level, packet level, or protocol level. A send-again request implemented at protocol
level will be more bandwidth-efficient provided it is not used very often. If
it is continuously triggered by a discontinuity in the radio or network path, it
will absorb radio and network bandwidth and introduce lots of delay and delay
variability.

Circuit switching choices. We need to decide whether we circuit-switch, packetswitch,
or cell-switch our offered traffic. This depends on whether the traffic is
synchronous or asynchronous. If the traffic is asynchronous, how asynchronous is
it, which means how bursty is our bandwidth (the frame-to-frame bit rate excursion)?

We must decide how much of our traffic needs to be isochronous (packets
arrive in the same order they were sent), how much can be non-isochronous. We
must know or be able to predict whether our radio and network bandwidth needs
to support symmetric or asymmetric traffic. If asymmetric, how asymmetric?
Bandwidth delivery choices. We need to decide whether we deliver transparent
or nontransparent bandwidth. A nontransparent channel can give us constant
bit error rates but at the cost of variable delay.

Impact on the user’s experience. We need to be aware of the consequences of all
these decisions on the quality and consistency of the end-to-end user experience.
Source coding issues. We must be aware of the quality issues of high-level
source coding, including the impact of error rates and error distribution (over
the radio layer), and packet delay, delay variability, and packet loss on highly
compressed source-coded rich media channel streams.

Error rate considerations. We must consider that wireless voice networks have
traditionally been planned on the basis of delivering 1 in 103 bit error rates,
whereas wireline networks typically deliver a consistent 1 in 1010 bit error rate.
Multimedia requirements. We need to consider the very particular requirements
of multimedia, including the need to multiplex multiple per-user channel
streams at the application layer and multiple per-user channel streams at the
physical layer and the relative requirement to synchronize these multiple
streams with each other (or reclock the multiple streams with each other when
they arrive in the receiver).