This report investigates the problem of
latency in a wireless network scenario, and
proposes the idea of using Neural Network
models in a Digital Twin in order to predict
the latency in real-time.
Different Digital Twin structures were proposed,
including different amount of Neural
Network models as well as different inputs.
To accompany the Digital Twin, a Physical
Twin with a client-server file transmission
use case was developed, in order to obtain
values of latency. Based on data available
before a transmission, such as the physical
location of the server and file size used, the
Digital Twin was trained to predict.
It was found that the Digital Twin was
capable of making predictions in under
a millisecond by implementing the Neural
Network model as TensorFLow Lite models.
This was significantly faster than Physical
Twin in all scenarios, including when the
observed latency was the lowest. The
latency prediction itself was successful, and
a number of future considerations for more
accurate predictions were proposed. These
considerations include how to accommodate
for temporal characteristics in the observed
latency.