False-CSI Attacks in Power-Domain NOMA for 6G: A Threat Taxonomy and System-Level Impacts

PREPRINT — SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE

1

False-CSI Attacks in Power-Domain NOMA for 6G: A Threat Taxonomy and System-Level Impacts

arXiv:2608.28351v1 [cs.CR] 28 Aug 2026

Samira Jafarli, Aysha Ebrahim, and Suleyman Uludag

Abstract—Power-domain non-orthogonal multiple access (NOMA) remains a widely studied technique for improving spectral efficiency and supporting dense connectivity in beyond-5G and 6G networks. Its main operating mechanisms, however, depend on the integrity of channel-state information (CSI). Power allocation, user ordering, pairing, clustering, and beamforming can all be distorted when the CSI consumed by the base station is deliberately biased rather than merely noisy. This article examines false CSI as an attack surface in power-domain NOMA. We organize the threat space using a compact taxonomy with two primary axes: magnitude, which distinguishes underreporting from overreporting, and ordering effect, which distinguishes order-preserving, boundary, and order-reversing attacks. We then show how coordinated falseCSI behavior, group-changing attacks, direction forgery, pilot spoofing, training-phase injection, and RIS-induced channel manipulation extend this basic taxonomy. Finally, we map each attack family to system-level impacts on power allocation, SIC reliability, scheduler behavior, fairness, throughput, and secrecy. The central message is that false CSI should be treated not only as a channel-estimation problem, but also as a control-input integrity problem for 6G NOMA. Index Terms—6G, non-orthogonal multiple access, physicallayer security, channel-state information, false CSI, CSI integrity, successive interference cancellation, threat taxonomy.

I. I NTRODUCTION Non-orthogonal multiple access (NOMA) remains an actively studied multiple-access technique for beyond-5G and 6G networks because it can improve spectral efficiency and support dense device connectivity in massive machine-type communications and edge applications. Power-domain NOMA, in particular, has dominated the security literature because its three operating mechanisms, superposition coding, successive interference cancellation (SIC), and channel-dependent power allocation, are familiar enough to analyze rigorously yet coupled enough that small perturbations propagate. Recent performance studies confirm that this coupling is also a source of fragility under realistic channel conditions [1]. The fragility traces back to a CSI-integrity assumption. Every operational decision a NOMA base station makes, the decoding order for SIC, the power split between paired users, the choice of pair or cluster on a given resource block, and the This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. S. Jafarli is with Young Talents Lyceum of Baku State University, Azerbaijan. A. Ebrahim is with the Computer Engineering Department at the University of Bahrain. S. Uludag is with the University of Michigan–Flint, Flint, MI, USA (e-mail: uludag@umich.edu).

beam direction in MIMO variants, is computed from channel information obtained through reports, feedback, pilot-based estimation, or hybrid acquisition procedures. The exact CSI path depends on the duplexing mode and system implementation, but the NOMA scheduler ultimately acts on a channel input that is assumed to be trustworthy at allocation time. Prior work has established that the integrity of SIC decoding hinges on the accuracy of the channel ordering derived from this input [2]. When the input is adversarially biased, the entire stack can drift with it [1]: power allocations become locally optimal but globally wrong, decoding margins collapse, and PHY-layer secrecy assumptions may no longer hold. Most of the literature that touches this dependency treats CSI imperfection as a stochastic estimation phenomenon, modeled as zero-mean uncertainty around the true channel. PHYlayer security surveys have often followed this convention [3], [4], while related attack studies appear separately under pilot spoofing, untrusted relays, adversarial scheduling, or malicious reconfigurable intelligent surfaces (RISs) [5], [6]. What is still missing is a compact attack-centered view of the common mechanism behind these cases: the NOMA scheduler acts on channel information whose integrity has been biased before power allocation, pairing, decoding order, or beamforming decisions are made. This article provides that attack-centered view. Rather than offering another NOMA performance survey or a countermeasure survey, it treats false CSI as a threat-modeling problem: scalar misreports, spatial forgery, training-phase bias, and RISinduced channel manipulation are unified by the same NOMA decision chain. Its contributions are three. First, we organize false-CSI attacks along two axes: a magnitude axis, which distinguishes under-reporting from over-reporting of CSI, and an ordering-effect axis, which distinguishes attacks that preserve, collapse, or invert the channel ordering used for SIC. Second, we show how coordinated, group-changing, directionforgery, training-phase, and RIS-induced variants extend the basic taxonomy without changing the underlying dependency on CSI integrity. Third, we map each attack family to systemlevel impacts through a propagation chain that starts at a single corrupted channel input and ends at power misallocation, SIC failure, scheduler distortion, throughput loss, fairness degradation, and secrecy leakage. The remainder of the article is organized as follows: Section II frames NOMA’s CSI-driven attack surface; Section III presents the taxonomy; Section IV maps the system-level impacts; and Section V discusses implications for 6G NOMA security.

PREPRINT — SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE

II. NOMA IN 6G: W HY CSI I NTEGRITY M ATTERS A. Power-Domain NOMA in Brief Power-domain NOMA allows two or more users to share the same time-frequency resource by separating their signals through transmit power rather than through orthogonal resource blocks. The base station superposes user signals with different power coefficients. The user with the weaker channel is typically assigned more power, while the stronger user receives less power and applies successive interference cancellation (SIC): it first decodes the weak user’s high-power signal, subtracts it, and then decodes its own lower-power signal from the residual. The same principle generalizes to clusters of three or more users, at the cost of additional SIC stages and greater sensitivity to channel-estimation error [7]. This coupling is attractive for beyond-5G and 6G systems because it can improve spectral efficiency and support dense connectivity without requiring new spectrum. It also creates a security-sensitive dependency: the same channel information that determines who is weak or strong also determines the power split, SIC order, pairing or clustering decision, and, in MISO/MIMO-NOMA, the beam direction. B. Why CSI Is the Soft Underbelly Every NOMA allocation begins with CSI. As shown in Fig. 1, the base station uses channel information to choose the SIC decoding order, power coefficients, user pairing or clustering, and beam direction in MISO/MIMO-NOMA. This makes CSI more than a performance variable: it is a control input. This article uses false CSI as an umbrella term for adversarially biased channel information consumed by the NOMA decision logic. The biased input may be a reported channel gain, a channel-quality indicator, a spatial channel direction, or an upstream channel estimate corrupted during training. We do not assume that the base station is physically blind; rather, we assume that the scheduling and allocation logic acts on CSI whose integrity may be biased before the NOMA decision is made. This is the structural difference from orthogonal multiple access. In a TDMA or OFDMA system, a false channel report may mainly affect the reporting user’s own allocation. In NOMA, the same false input also affects the paired user’s decoding role, the shared power split, and the cluster or beam used by nearby users. The next subsection captures this dependency through the SIC margin that false-CSI attacks manipulate. C. The SIC Decoding Margin For a two-user NOMA pair, let the near user have the stronger channel and the far user have the weaker channel. The far user is typically assigned more transmit power, while the near user applies SIC by decoding and subtracting the far user’s signal before decoding its own. The exact SIC feasibility condition depends on the receiver, coding and modulation choices, channel model, and residual-interference assumptions.

2

For the taxonomy below, it is sufficient to use a simplified received power-gap condition as an illustrative SIC margin [2]:
𝑃far − 𝑃near |ℎnear | 2 ≥ 𝜁, (1) where 𝑃far and 𝑃near are the allocated powers, |ℎnear | 2 is the near user’s channel power gain, and 𝜁 denotes an effective SIC-margin threshold. False CSI does not violate (1) directly. Instead, it causes the base station to choose the wrong near–far roles or the wrong power split before the inequality is evaluated. Three attack effects follow: the false CSI may preserve the ordering but skew the power allocation, push two users to a boundary where SIC becomes fragile, or reverse the ordering so that the SIC-margin check is satisfied for the wrong assignment. These three effects define the ordering-effect axis of the taxonomy below. III. T HREAT TAXONOMY OF FALSE -CSI ATTACKS A. The Two Axes A useful taxonomy of false-CSI attacks must be compact enough for a magazine article but precise enough to distinguish attack mechanics, affected NOMA decisions, and system-level consequences. We use two axes, summarized in Fig. 2. The first is magnitude: the adversary can understate or overstate the relevant channel value. The second is ordering effect: the false CSI can preserve the reported channel ordering, push users to a boundary where ordering becomes ambiguous, or reverse the ordering used for SIC. Crossing these axes yields six base cases. Coordinated, group-changing, spatial, and trainingphase attacks then extend these cases without changing the central dependency: NOMA decisions are computed from corrupted CSI. The two axes are intentionally operational. Magnitude is the lever the adversary controls; ordering effect is the NOMA decision consequence. In the order-preserving case, the base station keeps the same near–far roles but computes the wrong power split. In the boundary case, the reported gains become too close for a stable SIC margin. In the order-reversing case, the base station assigns near–far roles to the wrong users and therefore satisfies the SIC condition for the wrong power assignment. The remaining rows of Fig. 2 clarify scope. A groupchanging attack crosses a pairing or clustering threshold and changes which users share a resource block. A coordinated attack distributes bias across multiple users, making the aggregate allocation error larger than any one false report. Direction forgery and training-phase injection sit outside the scalar magnitude axis, but they corrupt the same control input: the CSI used by the base station before NOMA allocation, beamforming, or SIC decisions are made. B. Underreporting: Strong-to-Weaker False CSI A central scalar false-CSI attack is a strong-channel user claiming a weaker channel than it has. NOMA assigns more power to weaker users, so an attacker that successfully appears weaker captures a larger share of the transmit power. Three

PREPRINT — SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE

3

Beam leakage Four CSI-driven decisions User 1

Pairing distortion Decoding-role swap

SIC decoding order

False CSI

SIC-margin collapse Power coefficients

Selfish power capture

Reported CSI

CSI-based Decision

Normal Outcome Scheduler manipulation

User 2 Base Station

User pairing/clustering

Secrecy-rate degradation

Beam direction (MISO/MIMO-NOMA)

Selective overhearing Corrupted channel est. Throughput loss

User 3

DoS variant Fig. 1. False-CSI attack surface in power-domain NOMA. User 1 sends biased channel information, while Users 2 and 3 are legitimate. The base station’s CSI-driven decisions can produce normal operation or one of several adversarial effects, which are organized in Section III and mapped to system-level impacts in Section IV.

subcases follow from how aggressively the attacker shrinks its reported gain (Fig. 3). In the order-preserving case (Fig. 3, case 1A), the attacker reports a value that is lower than its true gain but still above its victim’s. The base station’s ordering of users is unchanged, and so is the SIC decoding sequence; only the size of the power gap is distorted. This is the stealthiest variant because nothing in the BS’s view of the system looks abnormal: the attacker remains the strong user, the victim remains the weak user, and the SIC decoding margin of Section II-C continues to be satisfied. Only the rate split has shifted in the attacker’s favor. In the boundary case (Fig. 3, case 1B), the attacker reports a gain at or near its victim’s level. The two users are now indistinguishable to the ordering rule, and the power separation is insufficient to support reliable SIC. The decoding margin collapses, residual interference accumulates, and outage probability rises sharply for both users in the pair. In the order-reversing case (Fig. 3, case 1C), the attacker reports a gain below its victim’s. The BS now treats the attacker as the weak user and grants it the larger power coefficient. Because the attacker’s true channel is strong, it decodes its own signal easily despite being assigned the role of weak user, while the genuine weak user is forced into a decoding position its channel cannot support. These cases also admit group-changing variants. If the false report crosses a pairing or clustering threshold, the scheduler may move the attacker into a different NOMA pair or cluster, creating resource-block churn beyond the original two-user interaction. Coordinated versions are discussed in

Section III-E. C. Overreporting: Weak-to-Stronger False CSI Overreporting is the symmetric case: a user claims a better channel than it actually has. The mechanics mirror Section III-B, but the incentives differ. A weak user that overreports gets a smaller power share under NOMA, which directly hurts its own rate. The attack therefore rarely makes sense as pure rate-grabbing; instead, overreporting is most effective when the goal is to manipulate ranking-driven decisions such as user selection, proportional-fair scheduling priority, beam pointing, or pairing assignment. The same three ordering effects recur (Fig. 4). In the order-preserving case (Fig. 4, case 2A), the weak user’s report is inflated but still below the strong user’s true gain. The pair ordering is preserved, but a scheduler that combines NOMA with proportional fairness or quality-ofservice ranking may treat the attacker as more deserving of resources than its actual channel justifies, capturing service that would have gone to other users in the cell. In the boundary case (Fig. 4, case 2B), the inflated report ties the strong user. As with boundary underreporting, the SIC decoding margin collapses, but the responsibility is now reversed: it is the attacker’s report that pulls the gap closed. In the order-reversing case (Fig. 4, case 2C), the weak user reports a gain above the strong user’s. The BS now treats the weak user as the strong one and assigns it the smaller power coefficient, while the genuine strong user is assigned the larger share and is expected to decode without applying

PREPRINT — SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE

Ordering effect −→

Attack type

Magnitude

Underreporting (strong → weaker)

4

Order-preserving

Boundary

Order-reversing

(stealthy)

(SIC-margin collapse)

(label swap)

Selfish power capture

SIC margin collapses; high outage probability

Decoding-role swap; weak user forced to SIC

§ III.B · Fig. 3 (1B)

§ III.B · Fig. 3 (1C)

§ III.B · Fig. 3 (1A)

Overreporting

Scheduler / ranking manipulation

SIC margin collapses; tied ordering

Decoding-role swap; attacker fails own SIC

(weak → stronger)

§ III.C · Fig. 4 (2A)

§ III.C · Fig. 4 (2B)

§ III.C · Fig. 4 (2C)

Crosscutting

Group-changing

Coordinated

(overlays)

Crosses clustering boundary; forces re-pairing

Colluding users mask individual bias

Orthogonal

Direction forgery

(points of injection)

Beam leakage; RISinduced bias (§ III.D)

Applies to any cell Overlays, not separate families

Training-phase injection

Layer onto any cell

Pilot spoofing / false CSI (§ III.D)

Combine with magnitude attacks

Fig. 2. Threat taxonomy of false-CSI attacks in power-domain NOMA. The magnitude axis captures whether the relevant CSI value is understated or overstated; the ordering-effect axis captures whether the reported ordering is preserved, collapsed to a boundary, or reversed. The six base cells describe scalar CSI attacks, while group-changing and coordinated behavior act as overlays. Direction forgery and training-phase injection are separate points of CSI corruption that can combine with the base cases.

Actual |ℎ| 2

Reported

Actual |ℎ| 2

Reported 2C (order-reversing)

Strong (attacker)

1T (truthful)

Strong (victim)

2B (boundary)

1A (order-preserving) 2A (order-preserving)

Weak (victim)

1B (boundary)

Weak (attacker)

2T (truthful)

1C (order-reversing)

Fig. 3. Underreporting by a strong user. Lowering the reported channel gain can preserve the ordering, collapse the users to a boundary, or reverse the near–far roles used for SIC.

SIC. The configuration is doubly wrong: the attacker, with a weak true channel, cannot reliably perform the SIC subtraction it has been assigned, and the genuine strong user receives no decoding instructions suited to its actual channel. A strong user can also overreport, exaggerating an already favorable channel; this rarely changes scalar ordering but may still bias ranking-driven scheduling. Spatial manipulation of beam direction is treated separately in Section III-D. D. Spatial and Training-Phase False CSI The six cells in Fig. 2 describe scalar false CSI: the attacker biases a reported or estimated channel value upward or downward. 6G NOMA also depends on non-scalar and

Fig. 4. Overreporting by a weak user. Raising the reported channel gain can manipulate ranking decisions, collapse the SIC boundary, or reverse the near– far roles assigned by the base station.

pre-report channel information. We include those attacks here not as new taxonomy cells, but as additional points of CSI corruption that feed the same NOMA decision pipeline. Spatial false CSI corrupts the direction of the channel rather than its magnitude. In MISO and MIMO-NOMA, the base station uses channel direction to form beams and spatial nulls. A malicious user may report a direction correlated with a victim’s channel so that the resulting beam leaks energy toward the attacker. A compromised or malicious RIS can produce a similar effect by changing the apparent spatial channel observed at the base station, even without user-side collusion [5], [6]. Training-phase false CSI corrupts the channel estimate before the NOMA scheduler acts. Pilot contamination, pilot

PREPRINT — SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE

spoofing, or training-phase injection can bias the estimate used for ordering, power allocation, pairing, clustering, or beamforming [8]. Reactive training jamming is the denialof-service version: instead of biasing the estimate toward a favorable allocation, it prevents the base station from obtaining a usable estimate. Thus, spatial and training-phase attacks do not expand the taxonomy horizontally. They specify where false CSI enters the system, while the magnitude and ordering-effect axes specify how the corrupted input changes NOMA decisions. E. Coordinated and Group-Changing Attacks Every attack family above admits a coordinated extension in which two or more malicious users combine their reports. Coordination changes the attack surface in three ways. First, it makes the boundary and order-reversing cases of Sections III-B and III-C far easier to engineer: a single user reporting near a victim’s gain may be flagged as a statistical outlier, but two users reporting consistent near-victim values look like a plausible pair. Second, coordination opens attacks that no single user can mount alone, such as 𝑁-attacker spread underreporting in which a small bias is distributed across many users so that no individual report appears anomalous, while the aggregate effect on power allocation is large. Third, in direction forgery, coordinated multi-attacker schemes can reconstruct enough of the victim’s spatial signature to enable selective eavesdropping that single-attacker schemes cannot. The same logic applies to the training phase: cooperative pilot spoofing distributes the false-pilot burden across several attackers, defeating detectors that rely on energy or spatial signatures of a single source. Two features of 6G amplify the coordinated threat. Ultradense deployments raise the prior probability that any given pair of users includes at least one compromised endpoint [9], and history-aware scheduling widens the window over which a coordinated bias can act. Coordinated false-CSI behavior therefore deserves explicit treatment in any 6G threat model, not relegation to a footnote on the single-attacker case. IV. S YSTEM -L EVEL I MPACT The taxonomy of Section III describes what an attacker does; this section describes what breaks. The impacts propagate through a coupled chain (Fig. 5), and their severity depends jointly on the magnitude axis of the attack and on which downstream stage absorbs the corrupted input. Table I maps the four attack families onto their manipulated CSI input, primary system-level effect, and the point at which the attack becomes operationally visible. Power allocation distortion and pairing errors. The base station’s power-allocation routine is a constrained optimizer that takes CSI values as input and returns quasi-optimal allocations. When the inputs are biased, the optimizer still produces a locally optimal solution with respect to the CSI it received, which is what makes the distortion difficult to distinguish from a legitimately unfavorable channel realization. The misallocation manifests as both an efficiency loss, aggregate spectral efficiency falls below the truthful baseline, and an

5

unfairness, since power redirected to attackers is taken from genuinely weak users whose channel information is accurate. The same corrupted input feeds the pairing and clustering stage: false CSI near a clustering threshold can move an attacker into a different NOMA pair or cluster, leading to scheduler churn that ripples across the cell. Coordinated falseCSI behavior amplifies this effect by making the perturbation persistent rather than transient [9], and dense 6G deployments compound it further because each biased CSI input can affect more potential pair partners than in 5G. SIC failure and error propagation. Once power allocation is misaligned with the true channels, the SIC decoding margin of (1) is checked against the wrong quantities. In the boundary case from Section III-B, the BS satisfies the inequality on paper while the genuine post-channel power gap is at or below the sensitivity floor. The strong user’s first-stage decoding then fails to cleanly subtract the weak user’s signal, residual interference leaks into the second decoding stage, and outage probability rises sharply for both users.At the link layer this surfaces as elevated block-error rates, retransmission storms, and adaptive-modulation fallback to lower-rate codes; effects that can resemble ordinary channel-quality degradation rather than an adversarially biased CSI input. The order-reversing case is more severe: the user the BS expects to perform SIC is the one with the weak true channel and cannot reliably decode the high-power signal it has been told to subtract. Errors propagate not just within the pair but to the ranking inputs used by the scheduler in the next slot, since hybridARQ feedback and channel-quality reports both depend on the broken decoding chain. Throughput, fairness, and quality of service. The aggregate effect on cell throughput depends on how persistently false CSI biases scheduling and power allocation. Orderpreserving variants are especially damaging at scale because they have low operational visibility: a small steady bias across many users can shift a fairness-weighted rate distribution toward attackers while still resembling natural channel variation. Selfish-user equilibria may emerge in which truthful reporting becomes individually disadvantageous, and quality-of-service guarantees for genuine weak users degrade because resources nominally reserved for them have been redirected. Over time, this hollowing-out of QoS becomes a form of system instability: the scheduler may report that service targets are being met while genuine weak users experience persistent under-service. The 6G targets of 106 devices/km2 and ultra-reliable lowlatency communications make this regime particularly fragile because fairness must hold across very large user populations on tight timescales. Confidentiality and secrecy leakage. The direction-forgery and beam-leakage subfamilies of Section III-D act on a different layer: rather than distorting the rate split, they bend the spatial pattern of the transmission so that signals intended for legitimate users illuminate the attacker [5], [6]. The result is a degradation of secrecy rate that classical eavesdropping models do not predict, because the eavesdropper here is also an authenticated user whose feedback the system relies on. Coordinated direction forgery sharpens the leakage further by allowing several attackers to triangulate a victim’s spatial

PREPRINT — SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE

6

TABLE I FALSE -CSI ATTACK FAMILIES IN POWER - DOMAIN NOMA, ORGANIZED BY MANIPULATED CSI INPUT, PRIMARY SYSTEM EFFECT, LAYER OF IMPACT, AND ATTACK VISIBILITY.

Attack family

Manipulated CSI

Primary system effect

Layer of impact

Attack visibility

Underreporting (§ III.B)

Reported channel gain |ℎ| 2

Selfish power capture; SICmargin collapse; decoding-role swap

Scheduler, PHY, link

Low (1A) to high (1C)

Overreporting (§ III.C)

Reported channel gain |ℎ| 2

Scheduler/ranking manipulation; pairing distortion; possible role swap

Scheduler, PHY

Medium

Spatial (§ III.D)

Beamforming, air interface

Low / silent

PHY, scheduler

Medium high

false

CSI

CSI direction; RISinduced apparent channel

Beam leakage; rate degradation; overhearing

Training-phase CSI (§ III.D)

false

Pilot signal sounding-phase estimate

Corrupted channel estimate; throughput loss; denial-ofservice variant

False CSI (§ III)

or

secrecyselective

to

wrong inputs ⇒ locally optimal, globally wrong

Eq. (1) satisfied for the wrong assignment

block-error rate and HARQ storms

selective overhearing of targeted users

Power allocation distortion

SIC margin violation

Throughput & fairness loss

Secrecy leakage (if MIMO/RIS)

scheduler symptom: low

PHY-layer symptom: medium

link-layer symptom: high

silent secrecy leakage

hybrid-ARQ feedback corrupts next-slot ranking

Fig. 5. Impact-propagation chain for a false-CSI attack. A single corrupted channel input cascades through power allocation, SIC decoding, link-layer throughput and fairness, and, in MIMO/RIS-assisted NOMA, secrecy. The lower labels indicate where the attack first appears as an operational symptom: scheduler bias may be subtle, SIC failure is more visible, and secrecy leakage can remain silent. Hybrid-ARQ feedback closes a loop back to next-slot scheduling, making the perturbation persistent.

signature, enabling selective overhearing of the specific users targeted by the scheduler. V. I MPLICATIONS FOR 6G NOMA S ECURITY The attack families above point to a broader lesson for 6G NOMA: false CSI is not merely a channel-estimation nuisance. It is a control-input integrity problem. Because the same CSI values determine ordering, power allocation, pairing, clustering, beamforming, and SIC behavior, even a small adversarial bias can propagate across multiple layers of the system. Five implications follow. False CSI should be modeled as adversarial input. Most NOMA studies treat imperfect CSI as random estimation uncertainty. That model is useful for performance analysis, but it does not capture strategic behavior by selfish or malicious users. In a false-CSI attack, the distortion is not zero-mean noise; it is chosen to alter a specific decision made by the base station. A strong user may underreport to capture power, a weak user may overreport to manipulate scheduling, and a coordinated group may distribute small biases across many reports so that no single report appears suspicious. Attackaware NOMA analysis therefore needs an explicit adversarialCSI profile, not only a statistical error model.

Dense deployment amplifies small reporting biases. The 6G vision of ultra-dense connectivity changes the scale of the problem. A biased CSI input in a sparse system affects a limited number of pairings; in a dense NOMA cell, the same false-CSI value can alter which users are clustered, how resource blocks are shared, and which users inherit difficult SIC responsibilities. The danger is therefore not only the size of one false report, but the number of downstream decisions that depend on it. At the scale of massive machine-type communications, many low-amplitude false reports may be more damaging than a single obvious outlier. Hybrid attacks blur the boundary between PHY-layer and control logic. The taxonomy shows that false CSI can enter through several points: reported channel gain, spatial channel direction, uplink training, pilot spoofing, or RISinduced manipulation. These vectors are often studied separately, but in a 6G NOMA system they affect the same decision chain. A pilot-spoofing attack can bias the channel estimate before allocation; a direction-forgery attack can bend the beam pattern after user selection; and a malicious or compromised RIS can change the apparent channel seen by the base station [5], [6]. Treating these as isolated attack classes risks missing the combined effect of corrupted CSI

PREPRINT — SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE

on scheduling, power allocation, SIC, and secrecy. AI-driven scheduling increases the value of manipulation. Future 6G systems are expected to rely increasingly on learning-based or history-aware resource allocation. This creates an additional incentive for stealthy false-CSI behavior. An attacker does not need to cause an immediate outage to gain an advantage; it can bias the scheduler’s view of channel quality, reliability, or service demand over time. Adversarial behavior at higher 6G layers can therefore compound falseCSI attacks at the NOMA layer [9]. The most dangerous cases may be those that remain below short-term anomaly thresholds while gradually changing long-term scheduling decisions. Standardized adversarial-CSI evaluation is missing. The final implication is methodological. There is no common benchmark for comparing false-CSI attacks in NOMA: different papers use different channel models, attacker objectives, user densities, pairing rules, and CSI-error assumptions. This makes it difficult to compare attack severity across underreporting, overreporting, direction forgery, pilot spoofing, and RIS-induced manipulation. A useful benchmark would specify the attacker’s knowledge, the manipulated CSI field, the intended NOMA decision error, and the observable systemlevel consequences. Such a benchmark would also connect naturally to emerging 6G trust and zero-trust frameworks. Recent surveys of trust evaluation and 6G security [10], [11] and reference zero-trust architectures [12]–[14] offer a foundation, but the benchmark’s first purpose should be clear attack modeling rather than defense comparison. VI. C ONCLUSION False CSI creates a distinctive attack surface in powerdomain NOMA because the same channel input shapes user ordering, power allocation, pairing and clustering, beamforming, SIC reliability, fairness, throughput, and secrecy. This article treated false CSI as an adversarial control-input problem rather than as ordinary channel estimation error. We organized the resulting attacks along two axes: magnitude, which distinguishes underreporting from overreporting, and ordering effect, which distinguishes attacks that preserve, collapse, or reverse the channel ordering used for SIC. The central lesson is that a false channel input may appear local, but its consequences are system-wide. A small bias can alter the power split, weaken the SIC margin, distort scheduler behavior, shift resources away from truthful users, or create silent secrecy leakage in MIMO/RIS-assisted settings. For 6Gscale NOMA, CSI integrity should therefore be treated as part of the attack model itself: before resource allocation, trust management, or evaluation methodology can be made robust, the field needs a clear understanding of how false CSI enters the system and how its effects propagate. R EFERENCES [1] V. Özduran, M. Mohammadi, N. Nomikos, I. S. Ansari, and P. Trakadas, “On the performance of uplink power-domain noma with imperfect csi and sic in 6g networks,” Journal of Communications and Networks, vol. 26, no. 4, pp. 445–460, 2024. [2] Z. Ding, R. Schober, and H. V. Poor, “Unveiling the importance of SIC in NOMA systems—Part I: state of the art and recent findings,” IEEE Communications Letters, vol. 24, no. 11, pp. 2373–2377, 2020.

7

[3] S. Pakravan, J.-Y. Chouinard, X. Li, M. Zeng, W. Hao, Q.-V. Pham, and O. A. Dobre, “Physical layer security for noma systems: Requirements, issues, and recommendations,” IEEE Internet of Things Journal, vol. 10, no. 24, pp. 21 721–21 737, 2023. [4] M. M. Saeed, R. A. Saeed, M. K. Hasan, E. S. Ali, T. Mazha, T. Shahzad, S. Khan, and H. Hamam, “A comprehensive survey on 6g-security: Physical connection and service layers,” Discover Internet of Things, vol. 5, no. 1, p. 28, 2025. [5] W. Khalid, T. Van Chien, W. U. Khan, Z. Kaleem, Y. B. Zikria, T. Kim, and H. Yu, “Malicious reconfigurable intelligent surfaces: Security threats in 6g networks,” IEEE Internet of Things Journal, vol. 12, no. 15, pp. 30 063–30 085, 2025. [6] H. Alakoca, M. Namdar, S. Aldirmaz-Colak, M. Basaran, A. Basgumus, L. Durak-Ata, and H. Yanikomeroglu, “Metasurface manipulation attacks: Potential security threats of ris-aided 6g communications,” IEEE Communications Magazine, vol. 61, no. 1, pp. 24–30, 2022. [7] S. A. H. Mohsan, Y. Li, A. V. Shvetsov, J. Varela-Aldás, S. M. Mostafa, and A. Elfikky, “A survey of deep learning based noma: State of the art, key aspects, open challenges and future trends,” Sensors, vol. 23, no. 6, p. 2946, 2023. [8] D. Nashat and S. Khairy, “Statistical-based detection of pilot contamination attack for noma in 5g networks,” Scientific Reports, vol. 15, no. 1, p. 3726, 2025. [9] B. D. Son, N. T. Hoa, T. Van Chien, W. Khalid, M. A. Ferrag, W. Choi, and M. Debbah, “Adversarial attacks and defenses in 6g network-assisted iot systems,” IEEE IoT Journal, vol. 11, no. 11, pp. 19 168–19 187, 2024. [10] E. Saeedi Taleghani, R. I. Maldonado Valencia, A. L. Sandoval Orozco, and L. J. Garcı́a Villalba, “Trust evaluation techniques for 6g networks: a comprehensive survey with fuzzy algorithm approach,” Electronics, vol. 13, no. 15, p. 3013, 2024. [11] P. Scalise, M. Boeding, M. Hempel, H. Sharif, J. Delloiacovo, and J. Reed, “A systematic survey on 5g and 6g security considerations, challenges, trends, and research areas,” Future Internet, vol. 16, no. 3, p. 67, 2024. [12] S. iTrust6G Project, “iTrust6G: Zero-trust security for 6G networks,” Smart Networks and Services Joint Undertaking (SNS JU), Tech. Rep., 2024, submission to FNWF 2024. [13] B. Trinh-Nguyen, S. Berri, S. G. Teo, T. Truong-Huu, and A. Chorti, “A framework for global trust and reputation management in 6g networks,” in Machine Learning for Networking: 7th International Conference, MLN 2024. Springer, 2025, p. 55. [14] Y. Wang, X. Kang, T. Li, H. Wang, C.-K. Chu, and Z. Lei, “Six-trust for 6g: Toward a secure and trustworthy future network,” IEEE Access, vol. 11, pp. 107 657–107 668, 2023. Samira Jafarli Samira Jafarli studies at the Young Talents Lyceum of Baku State University, Azerbaijan. Her research interests include next-generation wireless communications, 6G networks, artificial intelligence, cybersecurity, and intelligent infrastructure systems. She received a Gold Medal at the GENIUS Olympiad 2025 and a Second Place award in the National Selection Tour for the International Science and Engineering Fair (ISEF) for her engineering and scientific research projects. Her current work focuses on the security and reliability of future communication networks and the application of emerging technologies to real-world challenges. Aysha Ebrahim (Senior Member, IEEE) received her B.Sc. degree in Computer Engineering with first class honors in 2009 from the University of Bahrain. She completed her M.Sc. degree, with distinction, in Electronic Engineering from the University of York in 2011. In 2016, she received her Ph.D. degree in Electrical and Electronic engineering from the University of Manchester. Aysha is currently working in the University of Bahrain as an assistant professor in the department of computer engineering. Aysha is a Fellow of the UK Higher Education Academy (HEA), a member of the Institute of Electrical and Electronic Engineers (IEEE), a board member in IEEE ComSoc Bahrain chapter. In 2019, she received the best paper award in the prestigious IEEE Wireless Communication and Networking Conference (IEEE WCNC 2019). Her research focuses on 5G and beyond wireless networks, MAC layer design for wireless communication systems, green wireless networking, interference and radio resource management. Suleyman Uludag Suleyman Uludag is the David M. French Professor of Computer Science in the College of Innovation and Technology at the University of Michigan–Flint, where he also serves as Accreditation Director and as Special Assistant to the Vice Provost for Academic Strategy and Effectiveness. He is an ABET Computing Accreditation Commission (CAC) Commissioner and Team Chair. His research interests include cybersecurity, AI/ML applied to security problems, wireless and smart-grid security, and trust management, with a focus on the intersection of security and emerging communication systems.

Related documents

• Improving the Selection Rule of Correlation Attacks for Remote Power Analysis#179408
• Power-law distributions in empirical data#6196
• Violation detection in power operation sites based on multi-scale detection and few-shot learning.#330768
• Data Agents Under Attack: Vulnerabilities in LLM-Driven Analytical Systems#267584
• Out-of-Distribution Detection in Wireless Multimodal Foundation Models for 6G ISAC#394352