Optimal Hierarchical Radio Resource Management for HetNets With Flexible Backhaul

Zhang Fan (Michael), Naeimeh Omidvar, An Liu, Vincent K. N. Lau, Fan Zhang, Danny H. K. Tsang, Mohammad Reza PakravanIEEE Trans. Wireless Communications 2018 · DOI

TL;DR

This paper tackles the complex challenge of efficiently managing radio resources in future mobile networks (HetNets) that use "flexible backhaul," where not all base stations are directly connected to the main network. They propose a novel two-timescale hierarchical approach that optimizes resource allocation across different network layers, achieving near-optimal performance with reduced complexity and signaling overhead.

What Problem Does This Paper Address?

Future mobile networks, known as Heterogeneous Networks (HetNets), are envisioned to have a dense deployment of macro and pico base stations (BSs). Traditionally, all BSs require a high-capacity backhaul connection to the core network. However, this leads to significant infrastructure costs. The concept of "flexible backhaul" proposes connecting only a subset of pico BSs to the backhaul, with others relaying data through neighboring BSs. This drastically reduces costs but introduces significant complexity in radio resource management (RRM). Specifically, coordinating flow control, routing, interference mitigation, and link scheduling across different network layers and timescales becomes a formidable challenge. Existing approaches often fail to jointly optimize these aspects or are heuristic, lacking guarantees of optimality.

Methods

The authors propose a two-timescale hierarchical RRM framework to address the complexity. This framework separates RRM control into:

  • Long-term control variables: Determined at a slower timescale (superframe), these include flow control (how much data each flow can carry) and routing control (how data travels across the network, allowing multi-path routing). They also include Discontinuous Transmission (DTX) control, which manages when BSs transmit to mitigate interference, and is determined probabilistically over superframes.
  • Short-term control variables: Determined at a faster timescale (subframe), this primarily involves dynamic link scheduling, where each BS decides which links to activate on which subbands to maximize user throughput, exploiting instantaneous channel conditions.

The core of their solution involves formulating the problem as a two-timescale stochastic optimization problem. They then employ a primal decomposition method to break it down into two subproblems: a network layer subproblem (P1) and a physical layer subproblem (P2). P1 is solved using convex optimization techniques. P2, which is non-convex, is tackled by identifying a hidden convexity and deriving a global optimality condition. Based on this condition, they propose an iterative algorithm that alternates between updating the short-term link scheduling and long-term controls (DTX, routing, flow control) to converge to the global optimum.

Datasets

Not discussed in the paper. The evaluation is based on simulations using a defined HetNet topology and standard channel models.

Key Insights and Ideas

  • Flexible backhaul is a cost-saver but RRM complexity driver: The core trade-off is clear: significant infrastructure cost savings come at the price of a much more intricate RRM problem.
  • Two-timescale RRM is crucial: Separating controls into fast (link scheduling) and slow (routing, DTX) timescales is key to balancing performance (exploiting instantaneous CSI) with scalability and robustness to backhaul latency.
  • Joint optimization across layers is essential: Optimizing flow control, routing, interference, and link scheduling simultaneously, rather than in isolation, yields significant performance gains.
  • Hidden convexity unlocks optimality: The non-convex RRM problem can be transformed into a convex equivalent, allowing for the derivation of a global optimality condition and the design of a convergent algorithm.
  • Probabilistic DTX control: Instead of deterministic DTX patterns, using time-sharing probabilities for DTX patterns provides flexibility and robustness.

Key Results

  • The proposed two-timescale hierarchical RRM scheme significantly outperforms various baselines (fixed routing, fixed DTX, fast-timescale, slow-timescale) in terms of network utility.
  • The algorithm converges to the global optimal solution, a feat not guaranteed by other methods like alternating optimization.
  • The proposed scheme demonstrates robustness to backhaul signalling latency due to its hierarchical, two-timescale design.
  • Computational complexity scales polynomially with key system parameters, making it scalable.
  • Signalling overhead is comparable to simpler slow-timescale designs and significantly lower than fast-timescale approaches, despite achieving superior performance.
  • The paper quantifies the trade-off between cost saving and performance loss by reducing backhaul connections, showing substantial cost savings are achievable with manageable performance degradation.

Limitations and Future Work

  • Off-line DTX pattern selection: The paper assumes the set of feasible DTX patterns is determined off-line. Determining this set optimally is a separate problem.
  • Power control extension: While the framework can be extended to include power control, it's not explicitly optimized in this work.
  • Future work: The authors suggest that the proposed model can be adapted to a broader range of scenarios beyond flexible backhaul, including conventional HetNets, relay networks, D2D networks, and multi-hop cellular networks. The two-timescale design and solution techniques are highlighted as being highly scalable and practical.

Summary updated

BibTeX
@article{michael2018optimal,
  author = {ZHANG Fan (Michael) and Naeimeh Omidvar and An Liu and Vincent K. N. Lau and Fan Zhang and Danny H. K. Tsang and Mohammad Reza Pakravan},
  title = {{Optimal Hierarchical Radio Resource Management for HetNets With Flexible Backhaul}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2018},
  doi = {10.1109/twc.2018.2809745},
}
Figure