Incremental Port Selection (IPS)

• Incremental Port Selection (IPS) is a set of low-complexity algorithms that iteratively select optimal network or antenna ports using partial observations and surrogate models.
• IPS methods reduce probing, measurement, and computational costs compared to exhaustive search, achieving near-optimal performance in diverse applications.
• Key applications include fluid antenna systems, IPv4 service discovery, and multi-port FAMA, where IPS improves metrics like SNR, SINR, and service recall.

Incremental Port Selection (IPS) is a class of low-complexity algorithms that enable high-performance selection of network or antenna ports through iterative, data-efficient decision-making. IPS frameworks are driven by the fundamental need to identify optimal port configurations or service exposures under resource constraints, offering near-exhaustive performance at vastly reduced measurement or computational overhead. IPS methodologies have achieved notable prominence in fluid antenna array selection for wireless communication systems, rapid large-scale service discovery in IPv4, and multi-port selection in fluid antenna multiple access (FAMA) for massive connectivity.

1. Mathematical Problem Formulation and General Principles

IPS seeks to maximize a system-specific metric (e.g., received signal-to-noise ratio (SNR), signal-to-interference-plus-noise ratio (SINR), or service discovery recall) by incrementally selecting (or predicting) the best port(s) based on partial observations and learned correlations.

• Fluid Antenna Systems: For $K$ available ports, the exhaustive optimum is $i^* = \arg\max_{1 \leq i \leq K} \gamma_i$, demanding $O(K)$ direct SNR measurements. IPS targets $i \in S_{obs} \cup S_{est}$, with $S_{est}$ populated via fast surrogates based on spatial or statistical channel dependencies (Chai et al., 2022).
• IPv4 Service Discovery: Given $65\,535$ TCP ports per IP, full scanning is prohibitively costly. IPS (as instantiated in GPS) leverages small “seed” samples to incrementally predict which ports are likely to be open using multi-feature co-occurrence models. Candidate probes are selected adaptively by maximizing empirical conditional probabilities (Izhikevich et al., 2023).
• Multi-Port FAMA: For users with $N$ ports and $N_{\rm RF}$ RF chains, IPS incrementally selects $$\kappa = \{k_1, \dots, k_{N_{\rm RF}}\} \subset \{1, \dots, N\}$$, greedily optimizing instantaneous SINR at each step. Selection proceeds one port at a time to achieve near-optimal performance at polynomial cost (Hong et al., 22 Nov 2025).

IPS relies on exploiting correlations—spatial fading in antennas, inter-port/service co-occurrence in networking, or channel structure in multi-user environments—and designs surrogate models to fill in unmeasured or unprobed ports. Surrogate modeling may be analytical, table-driven, or use machine learning.

2. Core IPS Algorithms and System Instantiations

IPS is instantiated in several domains, each leveraging unique fast-selection algorithms:

Fluid Antenna Systems (Chai et al., 2022):

• Analytical Approximation (AA): Uses a parameterized spatially correlated Rayleigh model to infer unmeasured port SNRs via Rician-type conditional pdfs.
• LSTM-Based Prediction: Interprets the spatial SNR sequence as a time series, using a trained many-to-one LSTM to predict missing ports from a small, evenly spaced input set.
• Smart Predict-and-Optimize (SPO): Minimizes the decision error directly, not the estimation error, with a convex surrogate loss ($\mathcal{L}_{\rm SPO^+}$) and linear predictors.
• Hybrid Cascades: AA–SPO–LSTM hybrids empirically yield the lowest outage probability.

IPv4 Service Discovery (Izhikevich et al., 2023):

• Seed+Co-Occurrence Table Model: GPS acquires a “seed” of sampled IPs/ports, then builds highly parallel conditional-probability tables over three feature classes (transport, application, network). Subsequent probes are incrementally selected based on descending predicted probability.
• Two-Phase Expansion: Phase 1 (“priors”) focuses on guaranteeing discovery of the first service per host using exhaustive scans in small subnets, Phase 2 predicts candidate (ip, port) pairs using extracted co-occurrence features.
• **All computation is SQL-based, feature extractions are reproducible, and per-port predictions are entirely empirical; no neural networks are trained.

Multi-Port FAMA (Hong et al., 22 Nov 2025):

• Greedy Port Addition: At iteration $i$, candidate port $k \in \Omega$ is scored via the digital combining SINR ($\gamma_{i,k}$). The port maximizing $\gamma_{i,k}$ is assimilated into the chosen set, and matrix-inverse helpers are updated recursively. IPS concludes at $N_{\rm RF}$ ports.
• Matrix-Inverse Lemma: Successive updating avoids full recomputation, reducing complexity to $O(N N_{\rm RF} \min(N_{\rm RF}^2, U^2))$ (with $U$ the number of users).
• **Pseudocode and analytical complexity bounds are explicit in the source (Hong et al., 22 Nov 2025).

System Domain	IPS Algorithm(s)	Primary Surrogate Model(s)
Fluid Antenna	AA, LSTM, SPO, Hybrids	Rician pdf, LSTM, Linear SPO
IPv4 Discovery	GPS (tabular increments)	Conditional probability tables
Multi-Port FAMA	Greedy recursion	SINR-based greedy selection

3. Computational Complexity and Scaling Properties

IPS achieves substantial reductions in probe, measurement, or computational cost compared to exhaustive enumeration or combinatorial methods:

• Fluid Antenna Systems: Each IPS method scales as $O(K)$ or $O(K n)$ per frame, where $n$ is the number of observed ports ($n \ll K$). Hybrid AA+SPO+LSTM methods outperform best-fixed-antenna selection schemes with similar aperture (Chai et al., 2022).
• IPv4 Service Discovery: GPS achieves 92.5% service recall over 65K ports with 131$\times$ less bandwidth and 204$\times$ more precision compared to exhaustive scans. Parallel table-construction and probe scheduling yields four orders of magnitude faster discovery (Izhikevich et al., 2023).
• Multi-Port FAMA: Greedy IPS is polynomial (contrasted to combinatorial scaling in exhaustive port selection), supporting multiplexing of 40-80 users at feasible computational budgets for RF-chain-limited FAS (Hong et al., 22 Nov 2025).

4. Performance Analysis, Metrics, and Bounds

IPS methods are evaluated on domain-relevant metrics, demonstrating near-optimal selection probabilities and coverage.

• Fluid Antenna Systems: Outage probability $$P_{\rm out} = \Pr[\max_{i \in S} \gamma_i < \gamma_{\rm th}]$$ is suppressed by two orders of magnitude when only 10% of ports are observed, and significant improvements persist even with single port observations. AA+SPO+LSTM yields $P_{\rm out} \sim 10^{-3} - 10^{-2}$ for 10% probing at typical SNRs (Chai et al., 2022).
• IPv4 Service Discovery: Service recall (${\rm Recall}_{\rm all}$, ${\rm Recall}_{\rm norm}$), precision, and bandwidth savings are reported. GPS achieves 204$\times$ gain in precision at high coverage and consumes fewer probes than current XGBoost classifiers. Full empirical trade-off curves quantify incremental gains per probe budget (Izhikevich et al., 2023).
• Multi-Port FAMA: Symbol error rates (SER), block error rates (BLER), and multiplexing gain are bounded by theoretical upper/lower limits and match exhaustive search for $N_{\rm RF} \ge 3$. Analytical upper and lower bounds for average SEP are derived and validated by simulation (Hong et al., 22 Nov 2025).

5. Algorithmic Extensions and Generalization

IPS frameworks exhibit extensibility to diverse environments and channel models.

• Fluid Antenna: The IPS protocol generalizes to Rician fading channels, multi-antenna (MIMO) receivers (per-RF chain selection), and multi-user scheduling in correlated environments. Parameter tuning ($n/K \sim 5-15\%$) balances probing cost and outage probability (Chai et al., 2022).
• GPS/IPv4: Incremental selection strategies can be transferred to IPv6 scans, industrial subnets, or captive cloud arrangements, contingent on multi-feature availability. Limitations arise with sparse responder populations and randomized port mappings (Izhikevich et al., 2023).
• Multi-Port FAMA: The port selection paradigm adapts to wider classes of access structures, with complexity scaling maintained and performance advantages persisting across wideband channel models (3GPP CDL-C/E) (Hong et al., 22 Nov 2025).

6. Practical Considerations, Limitations, and Comparative Analysis

IPS practical deployment hinges on system latency, offline training, and domain-specific constraints.

• Fluid Antenna: Port switching has sub-ms latency. Online computation (few hundred MACs) is tractable on DSP/AI accelerators. SPO and LSTM require channel trace-driven offline training, with retraining needed as channel statistics vary (Chai et al., 2022).
• GPS/IPv4: Table computation is highly parallelizable; actual packet probing constrained by network bandwidth. A minimum of 2 responsive IPs per port is needed to derive accurate conditional probabilities. Feature expansion entails manageable increases in table size (Izhikevich et al., 2023).
• Multi-Port FAMA: EPS is infeasible beyond $N_{\rm RF}=2$ for large FAS ($N \geq 150$). IPS delivers near-optimal multiplexing at practical computational cost, outperforming decremental approaches unless $N_{\rm RF} / N$ is high (Hong et al., 22 Nov 2025).

Method	Complexity Order	Notable Trade-offs
Exhaustive/EPS	Combinatorial	Unmanageable for large N
IPS	Polynomial	Near-optimal, scalable
DPS	Polynomial	Higher cost when $N_{\rm RF} \ll N$

7. Significance and Research Impact

IPS marks a pivotal advance in multi-port selection for interference-limited wireless communications, large-scale service discovery in networks, and adaptive sensing applications. By leveraging local observations and domain-optimized correlations, IPS achieves performance close to optimal, yet at feasible probe, measurement, or computational budgets. Its adoption across fluid antenna systems, Internet-wide scanning, and massive connectivity multi-access underscores the generality and operational gains associated with incremental selection frameworks. IPS thus enables scalable, efficient solutions to previously intractable port selection problems in contemporary and future communication environments (Chai et al., 2022, Izhikevich et al., 2023, Hong et al., 22 Nov 2025).