WorksComputer Science
From Confounding to Counterfactuals: A Cross-Layer Causal Model for RAN Scheduling
Abstract
Data-driven control in radio access networks (RANs) selects resource allocations using associations between control actions and key performance indicators (KPIs) learned from network telemetry. Because the shared radio environment influences both the scheduler's decisions and the resulting performance, a bias known as confounding, such associations overstate the effect of an allocation and misdirect resources toward user equipment (UEs) that cannot benefit. Moreover, individual UEs respond heterogeneously to the same decision, so even a corrected average effect does not identify which UEs gain from additional resources. While causal inference has recently been applied to RANs, existing approaches do not model the cross-layer mechanisms that connect an allocation to its outcome. In this paper, a hierarchical structural causal model (H-SCM) spanning the PHY, MAC, and network layers is proposed that separates cell-load confounding from UE-specific response heterogeneity. In particular, a counterfactual scheduling problem is formulated in which physical resource block (PRB) allocations are selected using per-UE counterfactual goodput rather than observational conditional averages. This problem is solved via a pipeline combining constrained causal discovery, structural-equation fitting, and abduction-action-prediction. Simulations demonstrate that naive observational estimation overstates the full-PRB response by 73.3%, whereas adjustment for cell-load confounding reduces the relative error to 9.8%. Held-out interventions are predicted with a coefficient of determination of 0.988. Counterfactual scheduling achieves a mean relative loss in total cell goodput below 0.5% against the ground-truth optimum, compared with 2.0-2.9% for a neural-network baseline.
Introduction
Network control in modern cellular radio access networks (RANs) relies on telemetry and data-driven artificial intelligence (AI) models to support resource-allocation decisions such as physical resource block (PRB) allocation among network slices and scheduling-policy selection [1]. These models learn statistical associations between network states, control actions, and key performance indicators (KPIs), and use these to select the actions that maximize performance objectives. However, such statistical associations can be misleading when the radio environment influences both the scheduler’s decisions and the resulting KPIs. For example, low cell load can simultaneously increase the PRBs available to a user equipment (UE) and improve its link quality through reduced co-channel interference. Consequently, the higher goodput observed at higher PRB allocations may partly reflect lower cell load rather than the effect of allocating more PRBs. This situation, in which a shared environmental variable influences both the control action and the outcome, is known as confounding [2]. A controller that ignores it can misattribute environmental effects to the action being optimized.
While this confounding can be corrected by accounting for the congestion state when estimating the allocation effect, such a correction does not by itself enable UE-specific control. The corrected estimate characterizes only the population-average response, whereas UEs differ in propagation, device, and traffic conditions. The same increase in PRB allocation may therefore yield a negligible gain for a channel-limited UE and a substantial gain for a PRB-limited UE. Per-UE resource allocation instead requires predicting the outcome that a specific UE would attain under an alternative allocation, given the conditions it actually experienced. Such a unit-specific prediction is known as a counterfactual [2]. This work therefore addresses the following question: How can a RAN estimate the causal effects of resource-allocation decisions from observational telemetry, free of confounding by the network state, and predict the counterfactual response of an individual UE to an alternative allocation?
Related Work
Traditional resource-allocation approaches optimize weighted-throughput and proportional-fair objectives [3]. More recently, there has been growing interest in data-driven control for scheduling, resource allocation, and slice management through the RAN Intelligent Controller (RIC), including machine-learning-based xApps and reinforcement-learning methods [1] and [4]. Both classes of formulations optimize network-performance objectives without explicitly distinguishing statistical association from the interventional effect of a scheduling decision.
A principled approach to overcoming these limitations is causal inference, in which structural causal models (SCMs) formalize the distinction between observational, interventional, and counterfactual queries [2], and causal structure can be learned from data [5]. In this framework, confounding biases the estimated effect of an allocation, whereas effect modification means that this effect differs across UEs or network states, and a variable such as congestion can play both roles [6]. In wireless networks, causal reasoning has been advocated as a foundation for AI-native systems, with applications including resilient beamforming, digital-twin modeling, and intent management [7]. CausalTune adjusts for environmental and operational confounders in RAN configuration tuning [8], while BLINC uses a large language model (LLM) to supply domain-knowledge priors for Bayesian structure learning in context-specific configuration [9]. At the counterfactual level, conformal methods have been used to estimate the KPI that would have resulted from deploying a different RAN application [10], and the interaction between a user, an LLM agent, and the network has been modeled as an SCM to evaluate alternative formulations of a user intent [11]. These methods reason at the level of configurations, applications, or intents [8]–[11], without modeling the cross-layer mechanisms that connect a resource-allocation decision to its outcome. Moreover, each addresses either confounding or counterfactual evaluation in isolation, whereas UE-specific allocation requires both.
Contributions
The main contribution of this paper is a cross-layer causal framework for RAN resource allocation that distinguishes cell-load confounding from UE-specific response heterogeneity. We formulate a counterfactual scheduling problem that allocates PRBs according to each UE’s predicted goodput gain and develop a hierarchical structural causal model (H-SCM) including the physical (PHY), medium access control (MAC), and network layers to predict these individual responses. Simulations demonstrate that accounting for cell load reduces the relative error in estimated goodput under full-PRB allocation from 73.3% to 9.8%. Allocating PRBs based on each UE’s predicted response achieves a mean relative loss in total cell goodput below 0.5% against the ground-truth optimum.
System Model and Scheduling Motivation
Consider a single-cell downlink RAN serving a set of UEs in each scheduling interval. We model the system across the PHY, MAC, and network layers, whose variables and mechanisms, such as link adaptation, hybrid automatic repeat request (HARQ) retransmission, and queueing, jointly determine the application-layer goodput. Cross-layer dependencies are modeled on a per-UE basis, while cell-level quantities such as the congestion state and the available PRB budget are shared among the UEs scheduled in the same interval. Let represent a cell-level congestion state summarizing coupled serving-cell resource pressure and co-channel activity in the surrounding radio environment. As increases, the radio resources available for serving-cell scheduling decrease, and the co-channel interference experienced by the UE grows.
Wireless RAN System Model
For notational simplicity, we omit the UE index in this section. In the PHY layer (L1), let be the nominal received signal-to-noise ratio (SNR), the Doppler spread, and the delay spread. The UE-specific shadow-fading offset is modeled as [12]. The effective signal-to-interference-plus-noise ratio (SINR) can be modeled as
Here, we model the penalties as and , where are penalty coefficients and are reference scales. The parameter scales the interference contribution of the load state, consistent with cellular load-coupling models [13], and is a residual noise term accounting for unmodeled effects on . The block error rate (BLER) is modeled as a function of and . We select the highest-rate MCS satisfying the BLER target from the 3GPP MCS table [14], which specifies the corresponding modulation order and target code rate , which determine the nominal PHY spectral efficiency, , measured in bits/s/Hz.
In the MAC layer (L2), we model the number of PRBs allocated to the UE by a load-dependent scheduling rule,
where is the total number of available PRBs, controls the scheduler’s sensitivity to load, and captures slot-level scheduling variability. Given the allocated PRBs, the MAC-layer service rate is further shaped by retransmission overhead and device-specific radio-frequency (RF) variation. We approximate the overhead of HARQ retransmission procedures [15] using a truncated expected-retransmission surrogate. Let be the maximum number of retransmissions considered in the model, and the number of HARQ retransmissions. Since is a nonnegative integer-valued random variable, its expectation equals the sum of its tail probabilities. Under the simplifying assumption that every transmission attempt fails independently with the same probability , at least retransmissions are required exactly when the first attempts fail, so and
This expression approximates the retransmission overhead without explicitly modeling HARQ combining.
Device-specific RF variation, arising from differences in receiver hardware implementation, such as front-end noise figure and component tolerances across UE models, is represented by . The MAC service rate is , where converts to Mbps and the coefficients scale hardware variation and HARQ overhead. Although hardware impairments physically degrade the received SINR [3], we approximate their net effect through an assumed MAC-layer multiplier, which keeps the device-specific pathway separate from the propagation pathway through .
In the network layer (L3), traffic burstiness is characterized by , the squared coefficient of variation of the UE’s inter-arrival times , where and are the mean and variance of . This is a standard variability measure in queueing theory [16]. Poisson arrivals have , while indicates greater inter-arrival variability. Using the heavy-traffic approximation for mean waiting time [16], assuming no service-time variability conditional on the modeled MAC rate, and applying Little’s law, we model the mean queue backlog as
where , is the rate of the queued traffic component that loads the UE’s buffer and sets its backlog through , caps utilization, captures residual queueing variation, and . Further, we model the packet-loss probability, which captures residual decoding errors and queue buildup, as
In (5), the first factor is the probability that decoding succeeds within attempts under the same independent-attempt assumption as (3), and the second factor models the probability of avoiding queue-induced drops as decreasing smoothly once the backlog exceeds a threshold . The parameter controls the rate of this decrease. The two loss mechanisms are treated as independent. Finally, the application-layer goodput is , the goodput attainable by a saturated (full-buffer) flow, i.e., the service capacity net of residual decoding and buffer-overflow losses.
Conventional Scheduling
Using the per-UE model, we now consider scheduling across the UEs , where and denote the PRB allocation and resulting goodput of UE , respectively. The network state available to the scheduler when selecting the PRB allocation is . A conventional scheduler assigns each UE a scheduling weight , which encodes its priority under the operator’s fairness policy, and allocates PRBs from a total budget accordingly. A data-driven weighted sum-rate scheduler can then be formulated as [3]
In (6), the conditional expectation is estimated from historical telemetry. Since includes cell load , conditioning on this state blocks the modeled confounding path. However, the tight coupling between load and allocation limits observational support for some candidate allocations. The conditional mean also averages over unobserved UE-specific conditions. We propose an H-SCM to infer these conditions from factual telemetry and predict each UE’s response to alternative allocations.
Proposed H-SCM Framework for Causal Resource Allocation
We model the cross-layer cause-effect relations as an H-SCM, defined as [2], where each endogenous variable is determined by the structural equation , where represents its causal parents and is an exogenous noise term. The model is hierarchical because its endogenous variables, comprising the PHY, MAC, and network quantities from the effective SINR through the allocation and service rate to the goodput, are partitioned by the three protocol layers, , with no edges directed from higher to lower layers. The inputs to these mechanisms divide into context variables, which remain fixed within a scheduling epoch, and mechanism-level noise, which is redrawn in each realization. Let collect the context variables, namely the congestion state , the radio-link variables , and the unit-specific factors . The set contains the noise terms , and the remaining mechanisms are deterministic given their parents.
The structural equations induce the directed acyclic graph (DAG) in Figure 1. Acyclicity follows from the layer partition together with the feedforward processing order within each layer. The resulting observational distribution over context and endogenous-variable values , including PRB allocation and the resulting cross-layer KPIs, factorizes as , where each conditional is induced by the mechanism with marginalized out. For deterministic mechanisms it is a point mass.

Figure 1. Causal DAG spanning the PHY, MAC, and network layers. The DAG represents the simulated setting; operational RAN datasets may require additional nodes and dependencies.
In the H-SCM, cell load is a common cause of the PRB allocation and effective SINR , creating the backdoor path . Accounting for blocks this path when estimating the population effect of PRB allocation. Herein, we define the do-operator that severs but leaves active, since forcing the scheduler does not alter the UE’s interference environment. Cell load can also modify the goodput gain from additional PRBs. In our DAG, the UE-specific factors , and modify this gain without confounding the allocation effect and are preserved during unit-level counterfactual inference.
Three Rungs of Causal Inference
The H-SCM supports observational, interventional, and counterfactual queries about PRB allocation and goodput [2]. At the observational rung, the mean goodput among UEs observed with allocation is
which, unlike the conditional mean in (6), does not condition on and is therefore confounded under the assumed DAG. At the interventional rung, the backdoor structure established previously identifies the population response to setting by adjustment for as
Computing from observational data via (8) additionally requires the overlap condition over the relevant load distribution [2]. Because the scheduler couples tightly to , extreme allocations are rarely observed under unfavorable loads, so the controlled interventions in our evaluation provide direct coverage of these allocations.
At the counterfactual rung, the predicted goodput of UE under allocation is , obtained by fixing the allocation to through and evaluating the fitted mechanisms with the UE-specific state , inferred from factual telemetry and held fixed.
Counterfactual Scheduling Problem
For each UE , we select an allocation to maximize the weighted sum of counterfactual goodput under a shared PRB budget. The counterfactual scheduling problem is then
A conditional causal response can already support UE-specific scheduling through the observed state , but averages over unobserved conditions. The objective in (9) additionally uses each UE’s residual state inferred from factual telemetry to predict its response to alternative allocations, assuming that state persists.
Proposed Solution via a Causal Inference Pipeline
We solve (9) in three stages to predict each UE’s goodput under alternative allocations while preserving its inferred conditions. Accordingly, we recover the causal graph, fit the structural mechanisms, and infer each UE’s residual state from factual telemetry. The fitted model then predicts the counterfactual goodput used in the scheduling objective.
1) Causal Discovery: In practice, the causal graph is not given and must be recovered from data, a task known as causal discovery [5]. The DAG in Figure 1 serves as the simulator ground truth against which discovery is evaluated, and the recovered graph supplies the parent sets used for structural fitting and the causal order used for counterfactual prediction. We use the Peter–Clark (PC) algorithm [5] to learn an initial causal skeleton from observable telemetry through conditional-independence testing. We restrict edge directions using protocol-layer and within-layer processing orders [14], [15] and orient the remaining edges with the Direct Linear Non-Gaussian Acyclic Model (DirectLiNGAM) [17]. DirectLiNGAM exploits non-Gaussianity to estimate a causal ordering under a linear acyclic model. Since our mechanisms are nonlinear, we use this ordering as a heuristic, followed by interventional refinement. Regression on pooled observational and interventional data proposes additional edges, which are admitted in significance order while preserving acyclicity.
Candidate edges are then refined using simulator interventions that hold selected mediators fixed. For an edge , the mediator set contains variables between and in protocol order that respond to interventions on and are either ancestors of in the candidate graph or children of connected by previously retained edges. These variables are held fixed jointly, yielding the contrast
| Parameter | Value |
| OFDM grid | 76 subcarriers, 15 kHz spacing, 14 symbols |
| Coding and MCS set | LDPC with 512 information bits; 12 MCSs |
| BLER target / HARQ limit | 0.1 / 3 retransmissions |
| PRB and load parameters | , , dB |
| Queue loading rate / utilization cap | 5.5 Mbps / |
| Queue-loss parameters | , |
Table 1. Key simulation parameters
The two arms share identical exogenous draws. A candidate edge is retained if an effect remains significant after mediator blocking, but unblocked indirect paths can prevent interpreting it as a direct edge. Mediator selection uses the candidate graph and protocol order, without consulting the ground-truth edges.
2) Structural Equation Fitting: For each endogenous variable , its structural mechanism is estimated from its causal parents as
We fit each structural mechanism with a gradient boosting machine (GBM) using regression-tree base learners [18], which capture nonlinear dependencies and interactions among its tabular parent variables. Telemetry fits use observable parents in the recovered graph. Cross-fitting [19] computes each UE’s residual using a model trained without that UE.
3) Counterfactual Inference: Counterfactual predictions follow three steps [2]. First, abduction estimates each UE’s residual state from factual telemetry using the assumed mechanism-specific noise rules. Its counterfactual interpretation assumes correctly specified parent sets and noise forms, accurate mechanism fits, and exogenous conditions that remain fixed under the intervention. The inferred residuals approximate these conditions but also contain model and estimation error. Second, action replaces the PRB-allocation mechanism with the assignment . Finally, prediction evaluates the fitted GBM mechanisms in topological order, using updated parent values and applying each UE’s abduced residual according to the mechanism-specific rule.
Simulation Results and Analysis
We evaluate the proposed framework in a single-cell downlink RAN simulation using Sionna 1.2 for the physical layer [20]. We use the 3GPP clustered delay line model C (CDL-C) channel at 2.6 GHz, with a fixed delay spread of 100 ns and a UE speed of 3 m/s [12]. The simulated MCS-specific BLER curves determine the highest-rate MCS satisfying the BLER target at each UE’s effective SINR. The resulting MCS and BLER values feed the MAC and network-layer equations in § 2 to compute service rate, queue backlog, packet loss, and goodput. Table I lists the key simulation parameters.
Using data generated by combined PHY, MAC, and network-layer simulation, we evaluate causal discovery on the 16 observable telemetry variables, connected by 21 of the complete DAG’s 27 edges. The remaining six nodes are the latent factors and noise terms , which are unavailable in the assumed telemetry. Their outgoing edges are therefore excluded from discovery and its evaluation, while their effects remain present in the generated observations. Each discovery run uses 500 observational and 2940 interventional samples, with an additional 2000 paired UE realizations per mediator-testing contrast described in Section III-C1. Across 50 independent realizations, the proposed pipeline recovers 19 of the 21 observable edges without false positives, achieving precision 1.000, recall 0.905, and F1 score 0.950. The interventional regimes and the mediator-blocking contrasts set internal variables that only a simulator can control, which is a limitation of the discovery stage, whereas fitting for the counterfactual (CF) queries and held-out validation uses observational telemetry only.
After recovering the causal graph, we fit the SCM according to (11) using 500 observational samples. Each GBM uses 200 trees, a maximum depth of 4, and a learning rate of 0.05. Prediction accuracy is measured by the coefficient of determination (), the fraction of ground-truth outcome variance explained by the predictions. For counterfactual prediction, residuals are additive for most mechanisms, multiplicative for the MAC service rate, and rescaled only in the traffic-dependent component for queue backlog. For CF-2 (hardware normalization), the MAC residual multiplier is set to one, representing nominal hardware under the assumed decomposition. On 5000 held-out UEs, the per-UE counterfactual predictions achieve values of 0.979, 0.988, and 0.994 against the simulator’s per-UE outcomes under , , and hardware normalization, respectively.
Figure 2 compares goodput predictions from naive ordinary least squares (OLS) regression of on PRB allocation alone, confounding-adjusted OLS regression of on both and cell load , and the H-SCM, against ground-truth goodput from simulator interventions. In panels (a) and (b), naive OLS regression predicts a mean goodput of 49.20 Mbps at , exceeding the simulator’s ground-truth interventional mean of 28.39 Mbps by 73.3%. This naive estimate omits , unlike the conventional scheduler in (6). Including cell load in the regression and averaging over the observed load distribution reduces the predicted mean goodput to 31.8 Mbps and the relative error to 9.8%. The recovered and reference graphs yield identical adjusted estimates because the two edges missed during causal discovery do not change the required adjustment variable . In panel (b), averaging cross-fitted H-SCM counterfactual predictions over the 500 observational UEs gives 27.64 Mbps, a 2.7% relative error against the simulator mean over 5000 held-out UEs. Panels (c) and (d) show the goodput distributions and responses across cell-load ranges. The mean interventional goodput increases from 2.42 to 14.12 and 28.39 Mbps as the allocation increases from 10 to 50 and 100 PRBs. However, fixing the allocation leaves the effect of load on link quality active, so performance still differs across load conditions.

Figure 2. Confounding analysis. (a) Goodput versus PRB allocation. (b) Full-PRB goodput estimates compared with interventional ground truth. (c) Goodput distributions. (d) Mean goodput by cell-load range.
In Figure 3, we show the results of three CF queries. CF-1 and CF-3 evaluate the prediction that constitutes the objective of (9) at the fixed allocation , whereas CF-2 evaluates hardware normalization as a model-based simulator diagnostic. CF-1, shown in panels (a) and (d), asks what goodput the UEs with moderate factual goodput, Mbps, would have attained had they received the full PRB allocation . This criterion selects 179 UEs, whose mean goodput increases from 19.53 to 29.04 Mbps, an improvement of 48.7%. Under full-PRB allocation, the mean increase in goodput relative to factual values is 5.65 Mbps for UEs in the most channel-limited residual range and 11.35 Mbps for those in the most favorable range. CF-2 asks what goodput the UEs underperforming their peers would have attained had their MAC residual multiplier been set to one. We select UEs whose MAC throughput is below 95% of the median among peers with the same MCS and PRB decile. Panel (b) shows the sorted per-UE goodput changes. Panel (e) shows the mean predicted goodput change across scaled MAC-residual ranges. Mean predicted goodput increases from 14.15 to 15.80 Mbps (11.7%), with gains concentrated among UEs with a negative scaled MAC residual, i.e., a residual multiplier below one; those with a positive residual lose goodput. CF-3, shown in panels (c) and (f), asks what goodput the UEs selected by the joint criteria of nominal SNR dB and factual goodput Mbps would have attained under . These criteria select 113 UEs, whose mean goodput increases from 3.55 to 5.52 Mbps, an improvement of 55.6%. However, UEs in the channel-limited residual range gain only 0.11 Mbps on average, compared with 6.88 Mbps for UEs in the favorable, PRB-limited range. These differences show that the mean gain alone cannot identify which UEs benefit from additional PRBs.

Figure 3. Counterfactual goodput gains: (a)–(c) sorted UE gains, where is the number of UEs selected by each query, and (d)–(f) mean gains by inferred residual range. Dashed lines show the mean gain. Bar labels give UE counts.
Figure 4 evaluates (9) for four UEs () sharing the load-dependent PRB budget derived from (2), with at least 10 PRBs per UE. Training data comprise two scheduling epochs from 500 four-UE cells in the nominal environment, first under equal allocation and then a random feasible split. H-SCM uses 4000 unpaired rows. For scheduling, we fit the MAC mechanism to using its remaining observable parents and reconstruct its prediction by multiplying by the candidate PRB allocation. An MLP with two hidden layers of 64 ReLU units each uses 2000 pairs to predict second-epoch goodput from full first-epoch telemetry and candidate allocation. Two reference policies are the same H-SCM without abduction, setting residuals to their population defaults, and the conventional scheduler of (6), fitted as a GBM of goodput on allocation and context using the same rows. All policies are tested without refitting across four environments over 50 realizations. Relative to nominal conditions, high load raises mean from 0.50 to 0.71 and traffic-burstiness log-variance from 0.12 to 0.30. Poor channel lowers the maximum SNR from 28 to 16 dB, raises minimum Doppler/delay spread from 5 Hz/30 ns to 100 Hz/165 ns, and increases shadowing standard deviation from 7 to 9 dB. The diverse-UE environment increases hardware-factor standard deviation from 1.5 to 3, shadowing standard deviation from 7 to 10 dB, and traffic-burstiness log-variance from 0.12 to 0.40. Panel (a) reports mean relative loss in realized total cell goodput against the ground-truth optimum. Loss is 0.17–0.35% for H-SCM, 0.8–1.8% without abduction, 2.0–2.9% for MLP, 8–10% for goodput-proportional allocation, and 19–42% for (6). Abduction reduces loss in every environment with identical mechanisms. Panel (b) confirms that H-SCM closely approaches ground-truth optimal goodput.

Figure 4. Scheduling performance across four environments: (a) loss relative to the ground-truth optimum and (b) realized cell goodput. Error bars show approximate 95% confidence intervals over 50 realizations.
Conclusion
In this paper, we have developed an H-SCM for cross-layer causal reasoning in wireless RANs. We have combined causal discovery, structural-equation fitting, and counterfactual inference to evaluate resource-allocation decisions. Simulation results have shown that accounting for cell-load confounding improves estimates of the effect of PRB allocation on goodput. Unit-level analysis has further shown that the same allocation can produce substantially different gains across UEs. Together, these results show that confounding adjustment and counterfactual inference address different needs in causality-aware RAN control. Future work will consider interventions that can be actuated on an operational RAN. We will also validate the recovered structure on Open RAN (O-RAN) traces.
References
- [1]M. Polese, L. Bonati, S. D’Oro, S. Basagni, and T. Melodia, “ColO-RAN: Developing machine learning-based xApps for open RAN closed-loop control on programmable experimental platforms,” IEEE Transactions on Mobile Computing, vol. 22, no. 10, pp. 5787–5800, 2023.arxiv.org/abs/2112.09559
- [2]J. Pearl, Causality: Models, Reasoning, and Inference, 2nd ed. Cambridge, U.K.: Cambridge University Press, 2009.
- [3]E. Björnson and E. Jorswieck, “Optimal resource allocation in coordinated multi-cell systems,” Foundations and Trends in Communications and Information Theory, vol. 9, no. 2–3, pp. 113–381, 2013.
- [4]F. Lotfi, O. Semiari, and F. Afghah, “Evolutionary deep reinforcement learning for dynamic slice management in O-RAN,” in Proc. of the 2022 IEEE Globecom Workshops (GC Wkshps), Rio de Janeiro, Brazil, Dec. 2022, pp. 227–232.arxiv.org/abs/2208.14394
- [5]J. Peters, D. Janzing, and B. Schölkopf, Elements of Causal Inference: Foundations and Learning Algorithms. Cambridge, MA, USA: MIT Press, 2017.
- [6]T. J. VanderWeele, “On the distinction between interaction and effect modification,” Epidemiology, vol. 20, no. 6, pp. 863–871, 2009.DOI
- [7]C. K. Thomas, C. Chacour, W. Saad, M. Debbah, and C. S. Hong, “Causal reasoning: Charting a revolutionary course for next-generation AI-native wireless networks,” IEEE Vehicular Technology Magazine, vol. 19, no. 1, pp. 16–31, 2024.arxiv.org/abs/2309.13223
- [8]L. Xue, B. Zhang, Y. Ma, K. Marina, H. Yan, Y. Zhou, C. Y. Ip, S. Dhandapani, and J. Klosowski, “CausalTune: Causal learning based automated cellular RAN configuration tuning framework,” in Proc. of the ACM SIGCOMM 2026 Conference, Denver, CO, USA, Aug. 2026, pp. 242–262.DOI
- [9]R. Prasad, M. Polese, and T. Melodia, “BLINC: Context-specific causal learning for automated RAN configuration,” arXiv preprint arXiv:2604.27084, 2026.arxiv.org/abs/2604.27084
- [10]Q. Hou, S. Park, M. Zecchin, Y. Cai, G. Yu, and O. Simeone, “What if we had used a different app? Reliable counterfactual KPI analysis in wireless systems,” IEEE Transactions on Cognitive Communications and Networking, vol. 11, no. 5, pp. 3529–3543, 2025.arxiv.org/abs/2410.00150
- [11]A. Farzaneh, S. D’Oro, and O. Simeone, “Should I have expressed a different intent? Counterfactual generation for LLM-based autonomous control,” arXiv preprint arXiv:2601.20090, 2026.arxiv.org/abs/2601.20090
- [12]3GPP, “Study on channel model for frequencies from 0.5 to 100 GHz,” 3rd Generation Partnership Project, Tech. Rep. TR 38.901 V17.0.0, Mar. 2022.
- [13]I. Siomina and D. Yuan, “Analysis of cell load coupling for LTE network planning and optimization,” IEEE Transactions on Wireless Communications, vol. 11, no. 6, pp. 2287–2297, 2012.arxiv.org/abs/1201.4116
- [14]3GPP, “NR; physical layer procedures for data,” 3rd Generation Partnership Project, Tech. Spec. TS 38.214 V19.4.0, Jun. 2026.
- [15]3GPP, “NR; medium access control (MAC) protocol specification,” 3rd Generation Partnership Project, Tech. Spec. TS 38.321 V19.3.0, Jun. 2026.
- [16]W. Whitt, “The queueing network analyzer,” Bell System Technical Journal, vol. 62, no. 9, pp. 2779–2815, Nov. 1983.DOI
- [17]S. Shimizu, T. Inazumi, Y. Sogawa, A. Hyvärinen, Y. Kawahara, T. Washio, P. O. Hoyer, and K. Bollen, “DirectLiNGAM: A direct method for learning a linear non-Gaussian structural equation model,” Journal of Machine Learning Research, vol. 12, pp. 1225–1248, 2011.
- [18]J. H. Friedman, “Greedy function approximation: A gradient boosting machine,” The Annals of Statistics, vol. 29, no. 5, pp. 1189–1232, 2001.DOI
- [19]V. Chernozhukov, D. Chetverikov, M. Demirer, E. Duplo, C. Hansen, W. Newey, and J. Robins, “Double/debiased machine learning for treatment and structural parameters,” The Econometrics Journal, vol. 21, no. 1, pp. C1–C68, 2018.
- [20]J. Hoydis, S. Cammerer, F. Ait Aoudia, A. Vem, N. Binder, G. Marcus, and A. Keller, “Sionna: An open-source library for next-generation physical layer research,” arXiv preprint arXiv:2203.11854, 2022.arxiv.org/abs/2203.11854