[Paper Review] Sum-Rate Maximizing Cell Association via Dual-Connectivity

Minho Kim, Sang Yeob Jung+1 more|arXiv (Cornell University)|May 12, 2015

Advanced MIMO Systems Optimization4 references3 citations

TL;DR

This paper proposes a suboptimal dual-connectivity (DC) profile allocation algorithm for heterogeneous cellular networks (HCNs) that maximizes sum rate by intelligently assigning 1A or 3C DC profiles to users based on channel conditions and interference. The algorithm reduces the feasible solution space using profile characteristics and achieves 97% of the optimal sum rate with up to 10^6 times lower complexity than brute-force search, especially effective for networks with up to 20 UEs.

ABSTRACT

This paper proposes a dual-connectivity (DC) profile allocation algorithm, in which a central macro base station (MBS) is underlaid with randomly scattered small base stations (SBSs), operating on different carrier frequencies. We introduce two dual-connectivity profiles and the differences among them. We utilize the characteristics of dual-connectivity profiles and their applying scenarios to reduce feasible combination set to consider. Algorithm analysis and numerical results verify that our proposed algorithm achieve the optimal algorithm's performance within 5 \% gap with quite low complexity up to $10^6$ times.

Motivation & Objective

• To address the challenge of maximizing network capacity in heterogeneous cellular networks (HCNs) with dual-connectivity (DC) while minimizing computational complexity.
• To overcome the NP-hard nature of optimal DC profile assignment by reducing the feasible solution space through profile-specific constraints.
• To design a suboptimal algorithm that closely approaches the optimal sum rate performance with significantly reduced complexity.
• To evaluate the trade-off between sum rate performance and computational cost in DC-enabled HCNs under realistic deployment scenarios.

Proposed method

• The algorithm models a two-tier HCN with one macro base station (MBS) and multiple small base stations (SBSs) operating on different carrier frequencies.
• It defines two DC profiles: 1A (UE receives user plane from SBS only) and 3C (UE receives user plane from both MBS and SBS simultaneously), leveraging their distinct interference and capacity characteristics.
• The algorithm uses a greedy, iterative search strategy that evaluates combinations of DC profiles per UE, prioritizing those with higher potential sum rate gains.
• It reduces the search space by exploiting the fact that 1A and 3C profiles have different impact patterns on interference and spectral efficiency, avoiding full enumeration.
• The method computes sum rate using SINR expressions based on path loss and noise power spectral density, with rate calculation performed via Shannon capacity formula.
• Complexity is bounded by limiting the number of profile combinations considered per iteration, avoiding the exponential $K \cdot 3^K$ cost of full search.

Experimental results

Research questions

• RQ1 How can DC profile allocation be optimized to maximize sum rate in a heterogeneous cellular network with minimal computational complexity?
• RQ2 What are the performance and complexity trade-offs between 1A and 3C DC profiles in terms of spectral efficiency and interference management?
• RQ3 Can a suboptimal algorithm achieve near-optimal sum rate performance while reducing the solution space to a tractable size?
• RQ4 How does the proposed algorithm scale with increasing numbers of UEs compared to brute-force optimal search?
• RQ5 What is the impact of UE distribution and SBS density on the effectiveness of profile assignment strategies?

Key findings

• The proposed algorithm achieves 97% of the optimal sum rate capacity, with a performance gap of less than 5% compared to the full-search optimal solution.
• The algorithm’s computational complexity is reduced by a factor of more than $10^6$ compared to the brute-force optimal algorithm, especially for networks with up to 20 UEs.
• The 3C-only and 1A-only schemes show no performance gain with increasing UE density, indicating limited adaptability to traffic variation.
• The 'Stronger' algorithm (assigning to the closest BS) achieves only about 70% of the proposed algorithm’s capacity, highlighting the benefit of intelligent profile selection.
• The proposed algorithm maintains high performance even as UE count increases, with rate calculation operations staying below $10^4$ for up to 20 UEs.
• The complexity gap between the proposed algorithm and full search grows exponentially, exceeding $10^6$ times for more than 16 UEs, confirming its scalability advantage.