Abstract
Heterogeneous network (HetNet) is a key enabler to largely boost network coverage and capacity in the forthcoming fifth-generation (5G) and beyond. To support the explosively growing mobile data volumes, wireless communications with millimeter-wave (mm-wave) radios have attracted massive attention, which is widely considered as a promising candidate in 5G HetNets. In this article, we give an overview on the end-to-end latency of HetNets with mm-wave communications. In general, it is rather challenging for formulating and optimizing the delay problem with buffers in mm-wave communications, since conventional graph-based network optimization techniques are not applicable when queues are considered. Toward this end, we develop an adaptive low-latency strategy, which uses cooperative networking to reduce the end-to-end latency. Then, we evaluate the performance of the introduced strategy. Results reveal the importance of proper cooperative networking in reducing the end-to-end latency. In addition, we have identified several challenges in future research for low-latency mm-wave HetNets.
Introduction
As the industry moves toward 5G and beyond, achieving "ultra-low latency"—specifically end-to-end (E2E) delays between 1 and 5 milliseconds—is a primary goal. Heterogeneous Networks (HetNets), which integrate large macro cells (MeNBs) with numerous small cells (SeNBs), combined with millimeter-wave (mm-wave) communications, offer the massive bandwidth necessary to meet these demands.
However, a significant hurdle in these networks is the buffer problem. In high-traffic scenarios, data packets are not just transmitted; they are queued at various nodes. Traditional graph-based optimization methods, such as shortest-path algorithms, fail to account for these queuing delays. E2E latency in buffer-aided mm-wave HetNets depends not only on link capacity but also on the sequence of packet arrivals and the dynamic state of the queues.
Adaptive Low-Latency Strategy
To address these challenges, we propose an adaptive low-latency strategy rooted in cooperative networking. The system architecture is bifurcated into two functional planes:
- Control Plane: The MeNB serves as the central controller, gathering channel state information (CSI) from all nodes to determine optimal routing paths.
- Data Plane: This plane utilizes a fork-join system where traffic is split at the source, routed through various paths (including multiple SeNBs), and re-synchronized at the User Equipment (UE).
The strategy optimizes traffic allocation by calculating the minimum E2E latency across a range of traffic allocation coefficients ($\alpha$ and $\beta$). These coefficients define the fraction of data sent over specific links, such as MeNB-to-SeNB1 versus SeNB1-to-SeNB2. The optimization specifically accounts for First-In-First-Out (FIFO) queuing dynamics, ensuring that the "arrival of the last bit" is balanced across all parallel paths.
Performance Evaluation and Key Insights
The strategy was evaluated through numerical simulations representative of 5G environments, with link capacities ranging from 0.8 to 12 Gbps and a standard 2 Mb downlink file size. Key findings include:
- Sub-millisecond Latency: The system achieved a minimum E2E latency of 0.396 ms in the primary test scenario.
- The Necessity of Queue Awareness: In buffer-aided networks, the "fastest" link is not always optimal if it is congested. A slower, less-congested link can often yield lower overall latency.
- Benefits of Cooperation: Utilizing mm-wave backhaul between small cells consistently reduced latency compared to non-cooperative routing. Heatmap analysis of the allocation coefficients demonstrated that splitting traffic across multiple paths is superior to relying on a single "best" path.
- Coordinator Choice: Simulation results showed that switching which small cell acted as the primary coordinator for traffic merging could impact E2E latency by nearly 10%.
Future Challenges
While the results are promising, several challenges remain for real-world implementation:
- Scalability: Computational complexity increases significantly as the number of small cells grows.
- Mobility and Doppler Effects: High-frequency mm-wave links are sensitive to the Doppler effect in mobile scenarios, complicating accurate CSI acquisition.
- Multi-user Interference: Managing scheduling and interference in dense environments with many simultaneous users adds another layer of complexity.
- Signaling Overhead: The control plane's signaling requirements must be carefully managed to avoid negating latency gains.
Summary updated