UE-BS Initial Connection Optimization

• UE-BS Initial Connection Optimization is a process that rapidly and reliably connects user equipment to base stations using advanced beamforming and load balancing techniques.
• It integrates multi-dimensional criteria including latency, energy efficiency, control signaling, and fairness to optimize performance in dense mmWave environments.
• Practical schemes, such as hierarchical search, compressive IA, and collaborative beam management, demonstrate significant gains like up to 6 dB SNR improvement and 100× latency reduction.

UE–BS initial connection optimization encompasses the methods and algorithms enabling user equipment (UE) to rapidly, reliably, and efficiently associate with a base station (BS) upon entering a cellular network. In contemporary mmWave and dense-cellular environments, the process is shaped by multi-dimensional optimization criteria, including coverage, latency, energy efficiency, control signaling, and fairness in resource allocation. State-of-the-art research develops both mathematical models and practical schemes addressing beam management, cell and beam selection, multi-BS cooperation, adaptive codebooks, and joint radio/physical layer integration.

1. Mathematical Formulations and System Models

Canonical optimization models for UE–BS initial connection focus on bi-objective (throughput and load-balance) frameworks under practical constraints. For a set of UEs $U=\{1,\dots,M\}$ and BSs $B=\{1,\dots,N\}$, UE associations are described by binary indicators $x_{u,b}\in\{0,1\}$, subject to:

• Single connection: $\sum_{b=1}^N x_{u,b} = 1\,\,\,\,\,\forall u$
• BS capacity: $\sum_{u=1}^M x_{u,b} \leq L_b^{\max}$
• Connection quality: $\mathrm{SINR}_{u,b} \geq \gamma_u$ where $$\mathrm{SINR}_{u,b} = \frac{P_b g_{u,b}}{\sum_{b'\neq b} P_{b'}g_{u,b'}+\sigma^2}$$

Objectives include maximizing system throughput $$F_1(x)=\sum_{u=1}^{M}\sum_{b=1}^{N} x_{u,b} R_{u,b}$$ with $R_{u,b}=\log_2(1+\mathrm{SINR}_{u,b})$, and minimizing cell load imbalance $F_2(x) = \max_{b\in B} L_b$ (Ren et al., 20 Jan 2026).

Modern mmWave systems also require integrating a detailed beam management model:

• BSs sweep a predefined codebook of $N$ or $n$ beams, each UE measures received power or synchronization on each.
• Connection–seeking may be augmented by multi-BS cooperative strategies, reinforcement learning-based codebook adaptation, or robust data-driven beam design (Aroua et al., 29 Oct 2025, BarghiZanjani et al., 3 Jun 2025, Thoota et al., 2023).

2. Classical and Hybrid Initial Access Algorithms

Initial access (IA) methods span exhaustive (sequential) and iterative (hierarchical) search procedures for BS and UE beam alignment:

• Exhaustive Search: All pairs of transmit and receive beams are scanned, yielding $$T_{\mathrm{scan}}=B_{\mathrm{tx}}B_{\mathrm{rx}}T_{\mathrm{sig}}$$, with high reliability at the cost of increased access latency and overhead (Barati et al., 2015, Giordani et al., 2016).
• Hierarchical/Iterative Search: Coarse beams identify rough direction, followed by fine–resolution scanning in promising sectors, reducing total delay by up to 70% in certain configurations, at the expense of possible SNR loss at cell edges (Barati et al., 2015, Giordani et al., 2016).
• Compressive IA: Pseudorandom quasi-omni beamformers at the BS/UE, with energy detection and maximum-likelihood refinement, enable joint cell discovery, synchronization, and fine beam training in a single protocol step. This approach can reduce discovery latency by two orders of magnitude, with reliability and SNR comparable to hierarchical methods (Yan et al., 2019).

These procedures are governed by inherent trade-offs among detection probability ($P_d$), misdetection probability ($P_{md}$), and overhead, parameterized by SNR regimes, array geometry, and scanning and reporting policy.

3. Advanced Beam and Codebook Optimization

Recent research advances initial connection via robust and adaptive beam/codebook design:

• Robust Environment-Specific Beamforming: Pre-computation of codebooks tailored to the statistical and spatial properties of the environment using stored channel state information (CSI). When IA is triggered, a neighborhood of likely channel vectors is identified, and a max–min–sum beamforming optimization is solved:

$$\max_{\{\mathbf f_\ell\}} \min_{i=1...K} \frac{1}{\|\mathbf h_i\|^2}\sum_{\ell=1}^L|\mathbf h_i^H\mathbf f_\ell|^2,\,\,\, \|\mathbf f_\ell\|^2\leq P$$

This approach minimizes outage probability in realistic indoor/outdoor layouts and consistently achieves up to 7 dB higher worst-case beamforming gain than MRT or eigen-based codebooks (Thoota et al., 2023).

• Reinforcement Learning for Adaptive Codebooks: Framing the codebook design problem as a POMDP, RL agents adaptively select or combine expert-designed beams based on real-time per-beam UE connectivity feedback. The actor–critic policy operates in the expert beam pool, complying with hardware constraints, and yields an average improvement in user-discovery coverage of 10.8% versus static codebooks, with negligible latency and computational cost increase (Aroua et al., 29 Oct 2025).
• Collaborative Filtering for Beam Recommendation: Historic UE–beam RSS data is factorized (e.g., via truncated SVD), learning a latent representation space for UEs and beams. Upon UE arrival, a cold–start set is probed, the UE's latent embedding estimated, and beam recommendations are made via weighted collaborative filtering against nearest historical UEs. This achieves over 94% of the oracle (full-sweep) RSS with only ≈20 probes in a 60-beam codebook, outperforming smart-search heuristics, and can be scaled for multi-BS coordination (Yammine et al., 2022).
• Bayesian Optimization for Beamset Selection: The BS builds a Gaussian process surrogate of RSRP over its beam codebook, and greedily selects the set of beams which balance exploration of uncertain, potentially high-power beams and the cost of UE feedback. Guarantees arise from submodular maximization, and in 3GPP-compliant settings, BO yields >90% beam selection accuracy with <20% overhead, adapting automatically to UE mobility (Maggi et al., 2023).

4. Multi-BS Cooperation and Joint Transmission

Coordinated multi-BS SSB beam design leverages repetition-based joint transmission (JT) and phase-tuple orthogonality to enhance initial access coverage:

• Repetition-Based JT Procedure: $B$ BSs agree on a Hadamard-like set of $K=B$ phase tuples. For each joint beam/phase-configuration, $K$ back-to-back SSBs are transmitted, and the UE coherently combines receptions. This yields, for grid-point $x$,

$$\gamma_{JT}(x) = \frac{B}{N_0}\sum_{b=1}^B | \sqrt{P_b}\,h_b^H(x)f_b |^2$$

Optimal beam/phase-sets are selected via a greedy coverage maximization. Simulations report up to 6 dB median SNR gain and cell-center coverage boost from 66% to 94% under LoS and resource constraints (BarghiZanjani et al., 3 Jun 2025).

• Resource Trade-Offs: Matching the SSB beam budget of independent transmission, the JT approach preserves or improves coverage with identical signaling and overhead, especially when cooperation is restricted to up to $B=4$ strongest BSs—additional BSs offer diminishing returns. Location-based pruning further reduces repetition overhead with minimal SNR loss.

5. Energy and Load-Aware Initial UE–BS Association

In ultra-dense networks (UDNs), initial connection is interwoven with load balancing and energy-efficiency objectives. The process is formalized as follows:

• Three-Step Algorithm (Ren et al., 20 Jan 2026):
  1. Feasibility Screening: For each UE–BS pair, ensure $\mathrm{SINR}_{u,b}\geq \gamma_u$.
  2. Redundant Link Removal: For UEs with feasible links to multiple BSs, select the BS yielding highest instantaneous SINR.
  3. Overload Redistribution: If any BS exceeds its connection capacity, retain only highest-SINR UEs and redistribute the remainder to alternate feasible BSs, respecting capacity and SINR requirements.

• Computational Complexity: Dominated by $\mathcal O(MN\log N)$ due to candidate list sorting and iterative overload resolution.

• Systemic Effects: This sequential refinement secures one-to-one UE–BS matching, precludes overloaded "hotspots" or idle "cold cells," and lays the groundwork for BS sleeping and energy conservation without service degradation.

6. Specialized Physical Layer Integration and Emerging Schemes

• Intelligent Reflecting Surface (IRS) Enhancement: Mounting a hybrid IRS (HIRS) on the UE and alternating between sensing and reflecting modalities enables simultaneous AoA and AoD estimation for both ends of the link, dramatically shrinking initial access latency. Beam alignment is accomplished with multi-slot ML estimation, achieving sub-degree accuracy at low SNR with $O(N_{\mathrm{slot}})$ protocol overhead (Muhr et al., 2023).
• System-Level Recommendations:
  • Employ hierarchical search for very large arrays.
  • Utilize digital receivers with low-bit ADCs to collapse scan space and latency.
  • Use compressive IA protocols for single-stage synchronization and beam training under channel sparsity (Barati et al., 2015, Yan et al., 2019).
  • Integrate precomputed, environment-specific codebooks for IA–beam coverage (Thoota et al., 2023).

7. Performance Results and Design Guidelines

Methodology	Key Metric	Reported Gain/Feature
JT SSB Beam Design (BarghiZanjani et al., 3 Jun 2025)	Median SNR Gain (dB)	Up to 6 dB
RL Codebook Optimization (Aroua et al., 29 Oct 2025)	Coverage Gain (%)	10.8% (vs. expert codebook)
Compressive IA (Yan et al., 2019)	Latency Reduction	100× (0.2 ms vs. 20 ms)
Collaborative Filtering (Yammine et al., 2022)	# Probes for >94% RSS	≈20 (60-beam case)
Bayesian Optimization (Maggi et al., 2023)	Overhead for >90% accuracy	<20% of codebook
Robust Data-Driven Beamforming (Thoota et al., 2023)	Worst-case gain vs. MRT	+7 dB
Energy-Load Balanced UE–BS Match (Ren et al., 20 Jan 2026)	Max BS Overload prevented	Admits highest-SINR UEs, balances load

Optimized UE–BS association, beam management, and codebook/beam selection schemes offer proven multi-dB SNR and substantial access-latency and coverage gains, while also supporting energy-efficient and scalable network operation. These approaches have been validated under 3GPP-compliant scenarios, with robust performance under practical constraints such as SSB/CSI-RS budgets, UE distribution heterogeneity, and synchronization errors.

---

In conclusion, effective UE–BS initial connection optimization synthesizes multi-objective combinatorial assignment, advanced beamforming protocol design, data-driven or learning-based adaptation, and physical layer integration, collectively enabling next-generation wireless networks to meet stringent performance, efficiency, and coverage demands (BarghiZanjani et al., 3 Jun 2025, Aroua et al., 29 Oct 2025, Barati et al., 2015, Yammine et al., 2022, Ren et al., 20 Jan 2026, Yan et al., 2019, Thoota et al., 2023, Maggi et al., 2023, Muhr et al., 2023, Giordani et al., 2016).