baseline-comparison.md

Publication Study

In this document, we compare different publications and their suitability for the problem we want to solve. Most publications propose a solution that may be effective but requires modifications to become coherent with the project aims and constraints.

Sub-Band Assignment and Power Control for IoT Cellular Networks via Deep Learning

Paper: Sub-Band Assignment and Power Control for IoT Cellular Networks via Deep Learning

Aspect	Details
Tasks	Subband Allocation, Power Control
System	IoT Cellular Networks (MIMO Systems)
Objective	Maximize the achievable sum rate of IoT users with low complexity
Method	Two-Stage Optimization Method with Deep Learning Models SubBand Allocation stage with CNN architecture Power Allocation stage with FNN architecture
Validity	✖ Not Valid The assumption is that users can't share subbands. In our case, K < N, so subband sharing is necessary.
Required Modifications	Adapt the subband allocation stage of the approach for our system.

Deep-Learning-Based Resource Allocation for Transmit Power Minimization in Uplink NOMA IoT Cellular Networks

Paper: Deep-Learning-Based Resource Allocation for Transmit Power Minimization in Uplink NOMA IoT Cellular Networks

Aspect	Details
Tasks	Subband Allocation, Power Control
System	IoT Cellular Networks with K users and N subbands (swapped notation)
Objective	Balance between power minimization and rate constraint satisfaction
Method	Two-Step Optimization Method with Deep Learning Models SubBand Allocation stage based on GA search (genetic algorithm) Power Allocation stage with DNN architecture with unsupervised learning
Validity	✅ Valid The assumptions of the method can be extended to the current constraints. The subband allocation step uses a non-optimal method (we cann't assure its optimality) and it's computationaly expensive
Required Modifications	Adapt the subband allocation stage of the approach for our system.

An Energy-Efficient Downlink Resource Allocation In Cellular IoT H-CRANs

Paper: An Energy-Efficient Downlink Resource Allocation In Cellular IoT H-CRANs

Aspect	Details
Tasks	Subband Allocation, Power Control
System	NOMA-Based Vehicular Communication Networks
Objective	Maximize the sum rate of vehicular users while ensuring fairness and low latency
Method	Deep Reinforcement Learning (Deep-Q Learning) Learning to dynamically allocate resources and power
Validity	❗ Partial Valid Different Type of Networks with lower density NOMA systems Allows dynamic allocation to deal with rapid changes, although it may respond uncorrectly to small variations
Required Modifications	The latency-aware RRM is a good approach that can adapted to factory settings. We could consider a model-based reinforcement learning approach to work with better scenarios. Requires changes to model hyper-dense and mobile in-X subnetworks settings

An Energy-Efficient Downlink Resource Allocation In Cellular IoT H-CRANs

Paper: An Energy-Efficient Downlink Resource Allocation In Cellular IoT H-CRANs

Aspect	Details
Tasks	Resource Allocation; Network Slicing
System	NOMA-Based Vehicular Communication Networks
Objective	Maximize the sum rate of vehicular users while ensuring fairness and low latency
Method	Deep Reinforcement Learning (Deep-Q Learning) Learning to dynamically allocate resources and power
Validity	❗ Partial Valid Different Type of Networks with lower density NOMA systems Allows dynamic allocation to deal with rapid changes, although it may respond uncorrectly to small variations
Required Modifications	The latency-aware RRM is a good approach that can adapted to factory settings. We could consider a model-based reinforcement learning approach to work with better scenarios. Requires changes to model hyper-dense and mobile in-X subnetworks settings

User Subgrouping and Power Control for Multicast Massive MIMO Over Spatially Correlated Channels

Paper: User Subgrouping and Power Control for Multicast Massive MIMO Over Spatially Correlated Channels

Aspect	Details
Tasks	Resource Allocation, Network Slicing
System	MIMO Network with muticast users
Objective	Optimize resource allocation for network slicing to enhance system performance and user experience
Method	Deep Reinforcement Learning (DRL) Utilizes a DRL-based approach to dynamically allocate resources across network slices
Validity	❗ Partially Valid Focuses on network slicing in RAN with Massive MIMO, which differs from in-X subnetwork scenarios Employs DRL for resource allocation, which is relevant but may need adaptation for hyper-dense and mobile in-X subnetworks
Required Modifications	The DRL-based resource allocation approach can be adapted to handle the unique challenges of in-X subnetworks, such as high density and mobility. Incorporating considerations for rapid interference variations and dynamic sub-band allocation may be necessary.