Abstract
This paper details the use cases and architectural design principles of the SPEED-5G project. Focused on Dynamic Channel Selection, Load Balancing, and Carrier Aggregation, the proposed architecture leverages Software-Defined Networking (SDN) and Network Function Virtualization (NFV). We explore various functional virtualization splits and define Key Performance Indicators (KPIs) aligned with 5G-PPP standards to support the massive capacity and reliability requirements of next-generation wireless networks.
Introduction
The transition to 5G necessitates a paradigm shift in how wireless resources are managed. The SPEED-5G project introduces enhanced Dynamic Spectrum Access (eDSA), a framework designed to manage heterogeneous wireless resources across both licensed and unlicensed bands. By decoupling high-level control from hardware-specific execution, this virtualized architecture aims to meet the 1000x capacity increase demanded by 5G.
The Spectrum Crunch and 5G Requirements
Current 4G/LTE architectures are often too rigid, with wireless resources statically allocated and poorly utilized. This leads to unbalanced loads and capacity bottlenecks. 5G must support a diverse range of requirements, from massive IoT (low power, high density) to ultra-reliable low-latency communications (URLLC). The challenge lies in dynamically managing these resources across dense small-cell deployments while maintaining manageable infrastructure costs (CapEx/OpEx).
Proposed Methods: The eDSA Framework
The eDSA framework operates on three primary degrees of freedom: network densification, rationalized traffic allocation, and improved load balancing. Key technical approaches include:
MAC Layer Decomposition
A key technical innovation is the split of the Medium Access Control (MAC) layer into:
- High MAC: A Radio Access Technology (RAT)-independent layer handling traffic steering and load balancing.
- Low MAC: A RAT-specific layer responsible for scheduling and framing.
Functional Virtualization Splits
The architecture analyzes six potential "split points" in the 3GPP protocol stack, from the Radio Resource Control (RRC) down to the Physical (PHY) layer. The goal is to determine which functions should remain at the remote antenna and which should be centralized in the cloud, balancing fronthaul constraints like latency and bandwidth.
SDN/NFV Orchestration
Utilizing OpenStack, the architecture dynamically instantiates virtualized Radio Resource Management (RRM) and MAC entities. This allows the network to "slice" resources for specific use cases, such as Dynamic Channel Selection and Carrier Aggregation, on demand.
D2D Offloading
Device-to-Device (D2D) communication is treated as a separate tier, utilizing unlicensed spectrum (such as LTE-U/LAA) to offload traffic from primary cellular bands.
Evaluation and Key Insights
The architecture is evaluated through simulation environments modeling user data rates and hardware testbeds measuring end-to-end (E2E) latency.
The MAC Split "Sweet Spot"
The MAC layer is identified as the most effective split point. It enables centralized scheduling and improved coordination without the extreme fronthaul bandwidth requirements of a full PHY layer split.
Backhaul and Reliability
The paper emphasizes that Point-to-Multi-Point (PtMP) wireless backhaul must achieve ~1ms latency and >2Gbps throughput. By implementing a 1:1 redundancy mode for PtMP backhaul hubs, system availability can be increased from "five nines" to "ten nines," reducing annual downtime to approximately 1.5 milliseconds.
Context-Aware RRM
The architecture incorporates a "KPI Collector" and "Spectrum Manager," allowing the RRM to make proactive decisions based on real-time interference and traffic types, such as prioritizing high-bandwidth video streams over background IoT data.
Performance Metrics and Results
Concrete target KPIs for the SPEED-5G architecture include:
- Broadband Wireless: 300 Mbps downlink and <5s U-plane latency.
- Massive IoT: Support for 200,000 devices per $km^2$ with multi-year battery life.
- Ultra-Reliable: 99.999% availability and <1ms E2E latency.
- Fronthaul Trade-offs: A PHY split (Split 6) requires 5Gbps bandwidth and 250µs latency, while an RRC split (Split 1) requires only 200Mbps and <6ms latency.
Conclusion and Future Work
While the SPEED-5G architecture provides a robust blueprint for 5G resource management, challenges remain. Future work will address fronthaul error propagation in copper-based technologies and the impact of centralization on synchronous HARQ processes. Subsequent project phases will focus on developing machine-learning-based traffic steering and Filter Bank Multi-Carrier (FBMC) MAC protocols.
Summary updated