Dynamically Resource Allocation in Beyond 5G (B5G) Network RAN Slicing Using Deep Deterministic Policy Gradient

Table 2

RL-based resource allocation.

Ref.	Algorithm	Focus	Optimization objective	Use case/vertical app	Training	Development
[17]	DQN	RAN	Improve resource consumption and slice isolation	Continuous bit rate and lowest bit rate	Centralized	Simulation
[18]	-learning		Maximization of resource use while assembling the elements of successful communication	Haptic	Centralized	Simulation
[19]	-learning, SARSA, and Monte Carlo	RAN	Assurance of efficient resource use while fulfilling the demands for low latency	Internet of Things	Centralized	Simulation
[20, 21]	DDQN and duelling DQN	RAN	Maximize long-term profits while offering the services that different multitenant customers require	Manufacturing, automotive, and utilities	Centralized	Emulation (TensorFlow)
[22]	DQN	RAN	Maximize the utilisation of radio resources while preserving QoS	eMBB, mIoT, and uRLLC	Centralized	Simulation
[23]	DQN	E2E (RAN, TN, CN, edge)	SFC traffic variations should be accommodated when VNF placement is optimised	eMBB	Centralized	Emulation (OpenAI gym)
[24, 25]	LSTM	RAN	It is necessary to maximize spectrum efficiency and the SLA satisfaction ratio	VoLTE, eMBB, and uRLLC	Centralized	Simulation
[8]	A3C	RAN	Making the most of resources while preserving slice separation	Undeclared	Distributed	Emulation (TensorFlow)