Abstract:
One of the major challenges facing the realization of cognitive
radios (CRs) in future mobile and wireless communications
is the issue of high energy consumption. Since future network infrastructure
will host real-time services requiring immediate satisfaction,
the issue of high energy consumption will hinder the full
realization of CRs. This means that to offer the required quality
of service (QoS) in an energy-efficient manner, resource management
strategies need to allow for effective trade-offs between QoS
provisioning and energy saving. To address this issue, this paper
focuses on single base station (BS) management, where resource
consumption efficiency is obtained by solving a dynamic resource
allocation (RA) problem using bipartite matching. A deep learning
(DL) predictive control scheme is used to predict the traffic load
for better energy saving using a stacked auto-encoder (SAE). Considered
here was a base station (BS) processor with both processor
sharing (PS) and first-come-first-served (FCFS) sharing disciplines
under quite general assumptions about the arrival and service processes.
The workload arrivals are defined by a Markovian arrival
process while the service is general. The possible impatience of customers
is taken into account in terms of the required delays. In
this way, the BS processor is treated as a hybrid switching system
that chooses a better packet scheduling scheme between mean
slowdown (MS) FCFS and MS PS. The simulation results presented
in this paper indicate that the proposed predictive control scheme achieves better energy saving as the traffic load increases, and that
the processing of workload using MS PS achieves substantially superior
energy saving compared to MS FCFS.
