08700494.pdf (442.26 kB)
Download fileReconfigurable and traffic-aware MAC design for virtualized wireless networks via reinforcement learning
journal contribution
posted on 2019-05-24, 13:10 authored by Atoosa Dalili Shoaei, Mahsa DerakhshaniMahsa Derakhshani, Tho Le-NgocIn this paper, we present a reconfigurable MAC
scheme where the partition between contention-free and
contention-based regimes in each frame is adaptive to the
network status leveraging reinforcement learning. In particular,
to support a virtualized wireless network consisting of multiple
slices, each having heterogeneous and unsaturated devices, the
proposed scheme aims to configure the partition for maximizing
network throughput while maintaining the slice reservations.
Applying complementary geometric programming (CGP) and
monomial approximations, an iterative algorithm is developed
to find the optimal solution. For a large number of devices, a
scalable algorithm with lower computational complexity is also
proposed. The partitioning algorithm requires the knowledge of
the device traffic statistics. In the absence of such knowledge, we
develop a learning algorithm employing Thompson sampling to
acquire packet arrival probabilities of devices. Furthermore, we
model the problem as a thresholding multi-armed bandit (TMAB)
and propose a threshold-based reconfigurable MAC algorithm,
which is proved to achieve the optimal regret bound.
History
School
- Mechanical, Electrical and Manufacturing Engineering