WorksEngineering
Adaptive Guidance for 3GPP Ambient IoT Contention-Based Random Access
Abstract
The emerging ambient Internet of Things (A-IoT) envisions connectivity among large populations of low-cost, energy-constrained devices. Supporting such connectivity requires efficient coordination of device access, particularly in reader-triggered operation, where multiple devices may respond to paging over limited shared radio resources. Indeed, simultaneous responses can lead to severe contention, delayed transmissions, and lost reports. This article explores how compact extensions to the current 3GPP Release-19 signaling framework could enable more selective and adaptive device access. The proposed approach allows readers to broadcast compact access guidance, which devices interpret according to their local conditions, while optional feedback supports continuous adaptation. Through a device-reporting case study, we demonstrate how such coordination can improve report delivery and timeliness under congested conditions, while highlighting the trade-offs across different network configurations. We also examine practical signaling considerations and discuss opportunities for more intelligent, adaptive, and resilient protocols.
Introduction
Ambient Internet of Things (A-IoT) enables connectivity for low-complexity, ultra-low-power devices operating with harvested energy and limited energy storage. By minimizing device complexity and maintenance requirements, A-IoT supports large-scale applications such as asset tracking, inventory management, and environmental monitoring. To support such applications, the Third Generation Partnership Project (3GPP) Release-19 framework establishes the architecture and radio-access procedures for A-IoT [1, 2].
A key challenge in scaling A-IoT is coordinating access among large numbers of devices sharing limited radio resources. In the Release-19 reader-triggered procedure [3], a paging message can address individual devices, groups, or the entire device population. Under contention-based random access (CBRA), paged devices select among the configured access occasions (AOs), potentially resulting in collisions and unsuccessful access when many devices respond simultaneously. Managing such contention is particularly challenging when the reader has limited knowledge of pending device reports and access demand varies across paging occasions.
Managing contention in random-access networks has long relied on two complementary approaches: i) adjusting the available access resources and ii) regulating device participation. Classical framed-ALOHA schemes adapt contention resources based on observed outcomes [4], while access barring and backoff mechanisms control simultaneous access attempts, often supported by congestion estimation [5]. These principles also extend to A-IoT, where adaptive multi-channel access and reader-broadcast access probabilities have been investigated to manage contention among large device populations [6, 7]. Despite their different implementations, these approaches primarily regulate access through shared resource configurations or participation probabilities.
Beyond regulating aggregate contention, access decisions can also account for the individual conditions and requirements of devices. For instance, age-dependent access schemes consider information freshness when determining transmission opportunities [8]. More broadly, context-aware protocol design has been recognized as an important enabler of dependable low-power IoT operation, leveraging information freshness, energy availability, task relevance, and communication conditions [9]. Related ideas appear in A-IoT patent disclosures, including mechanisms for suppressing repeated group-paging responses [10], enabling selective responses through paging thresholds [11], and incorporating device energy states and access history into contention control [12].
These developments demonstrate the potential of reader-assisted contention management and selective device responses. However, existing approaches typically focus on particular admission policies, such as probabilistic barring or age-dependent access. A broader protocol framework is still needed to support different device-local admission rules through common signaling, while enabling the reader to refine its guidance using subsequent device feedback. This motivates a more selective and adaptive reader-triggered access while retaining simple A-IoT operation. We address this through four contributions:
We propose a lightweight signaling mechanism that builds on 3GPP Release-19 by extending the paging message to carry access-guidance information (AcGI) and embedding optional device feedback within the device-to-reader (D2R) data payload.
We develop a generic distributed access-admission mechanism, where the reader broadcasts group-level guidance, devices independently determine their eligibility based on local conditions, and the reader adapts subsequent guidance using observed outcomes and optional feedback.
This work is supported by the Research Council of Finland (369116 (6G Flagship), and 362782 (ECO-LITE)); and by the European Commission through the Horizon Europe/JU SNS project AMBIENT-6G (Grant 101192113).
Authors are with the Centre for Wireless Communications (CWC), University of Oulu, Finland (e-mails: {amirhossein.azarbahram, david.ruizguirola, onel.alcarazlopez}@oulu.fi).
We demonstrate the framework and its potential gains through a freshness-based device-monitoring case study, integrating reader-side bandit adaptation and benchmarking against conventional CBRA, device-side backoff, and probability-based guidance. We quantify the effects of traffic load, access resources, and paging-group configurations on report loss and delivery delay, showing the effectiveness of the proposed mechanism.
We identify future research directions toward intelligent and resilient A-IoT access, encompassing advanced reader-side adaptation, digital-twin-assisted protocols, and resilience-aware signaling and access procedures.
Reader-Triggered Access in 3GPP A-IoT
As illustrated in Figure 1, the reader can address a selected group and configure shared AOs for device responses [3]. The addressed group may contain heterogeneous A-IoT devices, such as sensing and tagging devices with different operating conditions, including energy-harvesting availability and, where applicable, stored-energy levels. Under CBRA, each responding device independently selects an AO and transmits its Access Random ID (MSG1). When multiple devices select the same resource, their transmissions may collide, leading to unsuccessful access.

Figure 1. Reader-triggered paging of A-IoT device groups and contention over shared AOs.
The corresponding signaling procedure is illustrated by the black exchanges in Figure 2. Following successful reception of MSG1, the reader can return a Random ID Response (MSG2), allowing the device to proceed with D2R upper-layer data transfer. Subsequent reader-to-device (R2D) data transfer or negative acknowledgement (NACK) feedback may be provided when applicable [3]. These exchanges establish the basic access and data-delivery procedure without requiring the reader to know individual device traffic states beforehand.

Figure 2. 3GPP reader-triggered CBRA procedure and proposed signaling extensions. Dashed boxes identify proposed device-side and reader-side operations, while the lower message structures illustrate potential accommodation of access guidance within paging Fill Bits and optional feedback within the D2R Data SDU. The proposed elements are not standardized 3GPP fields, and field widths are schematic.
The 3GPP paging message includes several fields supporting this procedure. In the CBRA format, these convey the message type, transport block size, security information, access type, transaction identifier, and optional Paging ID. The latter identifies the addressed devices, while its absence allows paging of
the entire population. Additional fields specify the number of AOs, transmission scheduling information, and the response-monitoring configuration. With the 3GPP security parameter present, these fields occupy 170 bits without a Paging ID, or 180 bits plus the variable-length Paging ID when present. The message also includes optional Fill Bits, which support byte alignment and are explicitly identified as a possible location for future extensions.
The D2R Upper Layer Data Transfer medium access control (MAC) protocol data unit (PDU) has a simpler structure. Its 16-bit header identifies the message type, indicates whether additional data remain, and specifies the length of the data service data unit (Data SDU). The Data SDU carries up to 123 bytes of upper-layer information, with segmentation supported when necessary. Optional MAC padding accommodates unused transport block capacity, while reserved header bits have prescribed values. The standardized message structures are summarized in the lower part of Figure 2 in black.
Although these procedures enable communication with large device populations, the reader has limited visibility into their individual conditions. Successful receptions reveal which transmissions were decoded, but do not fully identify devices that remained silent or experienced unsuccessful access, nor the urgency of their undelivered reports. Consequently, the reader cannot reliably infer the complete access demand or the relative importance of pending data from its observations alone. This motivates signaling mechanisms that can account for device-local information without requiring its acquisition before contention.
Proposed Access Guidance and Reader Adaptation
Here, we extend reader-triggered access with compact admission guidance and optional post-access feedback. The central principle is to distribute decision-making between the reader and devices such that the reader provides common access guidance, devices determine their own eligibility, and subsequent observations support adaptation at the reader. The proposed additions to the Release-19 procedure are identified by dashed elements and explicit labels in Figure 2.
Paging Guidance and Local Admission
Building on the 3GPP Release-19 provision for future extensions through the paging Fill Bits, we propose incorporating compact AcGI into R2D paging. This would enable devices to make local admission decisions based on reader-provided guidance and their individual conditions, without requiring the reader to acquire their states beforehand.
A candidate implementation introduces a compact value or index within the A-IoT Paging MAC PDU, potentially referencing a predefined parameter table. For example, a four-bit index could select among 16 predefined access probabilities, while a five-bit index could specify a report-freshness threshold. When multiple guidance policies are supported, an additional indication could identify the applicable rule. The reader and devices must share the corresponding configuration and interpretation. As discussed earlier, the existing paging Fill Bits offer a potential location for such extensions. However, their available capacity is variable, and under the current 3GPP specification, devices treat them as padding rather than interpreting their contents as control information. Using these bits for access guidance would therefore require additional signaling and processing rules. The exact encoding, field allocation, capability indication, and compatibility requirements therefore remain subject to standardization.
Upon receiving a matching paging message, a guidance-aware device evaluates the advertised rule using its local information, such as report age, priority, or access history. Depending on the policy, admission may involve checking whether a local condition satisfies a broadcast threshold or making a probabilistic access decision. An eligible device proceeds with conventional CBRA, while a deferred device remains silent for that paging opportunity. Importantly, deferral does not necessarily imply report deletion. Buffered reports remain subject to application-specific replacement and expiry rules. Guidance therefore regulates participation in contention rather than replacing the underlying access procedure. Devices that do not support the extension would require explicitly defined coexistence and fallback behavior.
Post-Access Feedback and Reader Adaptation
To complement paging guidance, our proposed framework introduces optional reporting of device-local conditions following successful access. Compact indicators may describe earlier report losses, failed access attempts, and the age of a delivered report. Such information helps the reader assess the consequences of its guidance decisions. For example, a successfully reporting device may reveal earlier report expiration or unsuccessful access attempts that the reader could not directly observe.
As illustrated in Figure 2, these indicators could accompany the reading within the existing D2R Data SDU, provided that an appropriate interface exposes them to the reader-side controller. Alternatively, a dedicated MAC extension could provide reader-visible feedback, potentially using otherwise unused padding capacity. However, neither approach is currently defined as a 3GPP access-guidance feedback mechanism. The latter would require explicit encoding and processing rules, while the availability of padding depends on the configured transport block size. In either case, feedback overhead and device energy constraints must be considered.
Using successfully decoded reports and any available feedback, the reader can evaluate the performance of different guidance parameters and refine subsequent broadcasts. Adaptation may operate independently for each paging group, allowing different guidance selections according to their observed traffic and contention conditions. However, feedback is inherently incomplete because devices that remain silent or fail to deliver their reports do not necessarily contribute observations, potentially affecting the accuracy of reader-side decisions.
This framework maintains a clear separation of responsibilities. Devices retain their local states and make admission decisions, while the reader processes feedback, evaluates performance, and updates guidance. Any reward or cost calculations are confined to the reader and need not be communicated to devices. Practical implementations must also define failure handling, how devices confirm delivery and reset counters, and behavior when guidance or feedback is unavailable.
Implementation Options and Trade-offs
The proposed framework supports different levels of signaling and control functionality, allowing implementations to balance adaptability against device complexity and communication overhead. Several configurations can be considered:
Mode-based guidance: A one-bit indication could enable or disable a preconfigured admission rule, while a few additional bits could allow selection among multiple predefined policies. This approach requires minimal paging overhead and no additional device feedback. However, the available admission rules and their parameters must be known beforehand, limiting flexibility under changing network conditions.
Parameterized guidance: The reader could broadcast a policy type (e.g., probabilistic access or freshness-based admission) together with an index selecting a predefined
parameter value, such as an access probability or freshness threshold. This allows the reader to adjust admission decisions dynamically without requiring additional device reports, although shared parameter tables, capability signaling, and sufficient paging capacity remain important implementation considerations.
Feedback-assisted adaptation: Optional D2R feedback could range from simple status indicators to richer representations of report losses, access failures, and information freshness. Such feedback could support more informed reader-side adaptation, but introduces additional payload, processing, and interface requirements. Its effectiveness also depends on the reliability and representativeness of successfully received reports.
These options illustrate how access guidance could evolve from simple policy selection to more adaptive and feedback-driven operation. Their practical adoption would require standardized signaling formats, clear interoperability rules, and careful assessment of the communication and energy costs.
Case Study: Freshness-Guided aiot cbra
To illustrate the proposed framework, we consider an indoor environmental monitoring scenario aligned with the 3GPP A-IoT sensor data collection use case [13]. A reader collects measurements, e.g., temperature, humidity, and equipment-status, from A-IoT devices deployed in an indoor facility. Each device generates reports according to an independent Poisson process, modeling asynchronous sensor updates triggered by environmental changes or monitoring events. This also allows for evaluating different traffic loads without assuming fixed reporting intervals. The reader pages one group per simulation slot in round-robin order, with each slot encompassing paging, contention, and data transfer. Devices attempting access select a configured AO uniformly at random. This slot-based collision model abstracts from physical-layer and energy-consumption details. For this reporting-oriented evaluation, only paged devices with valid buffered reports are modeled as access candidates, isolating contention among devices seeking to deliver sensor measurements.
Each device has a single-report buffer that retains only the newest undelivered reading, with three states:
Idle: No report is buffered and a new arrival makes the device active.
Active: A report is buffered but has not yet been attempted. An unsuccessful attempt makes it pending.
Pending: A previously attempted report awaits another opportunity while unsuccessful retries leave it pending.
Successful delivery clears the buffer, while a new arrival replaces any buffered report. If several readings arrive in a slot, only the newest is retained. Reports also expire after their application-defined lifetime. Thus, losses arise from replacement or expiry.
Freshness-Based Admission and Compared Policies
We implement AcGI as a threshold on report freshness. Each device uses its report age, the group count, and the report lifetime to assign the reading to an age band spanning one group-cycle length. For each paged group, the reader broadcasts an index identifying the oldest band eligible for access. The lowest index admits only the freshest band, whereas the highest admits every unexpired report. Devices apply this rule through a simple local comparison, without using loss counters or previous access failures to determine eligibility.
For example, a 30-slot report lifetime and five paging groups yield six age bands. Index two admits reports younger than 15 slots. The bands describe elapsed group cycle lengths, not how many paging messages a report has actually encountered. Our evaluation assumes that the reader and devices share the lifetime, group configuration, slot-age convention, and index mapping, while heterogeneous lifetimes would require additional configuration.
We compare four access policies:
Unguided CBRA: Following the 3GPP Release-19 reader-triggered CBRA procedure [3], paged devices randomly select an available AO and attempt access without additional admission control. Only devices holding valid buffered reports are considered, excluding responses without data to avoid unnecessarily degrading the baseline and ensure a fair comparison.
Device-side backoff: Each device initially attempts access with probability one, which is halved after each failed attempt for the same report, down to approximately 0.1%. A new report resets the probability to one, and no guidance is broadcast.
Probability-only guidance: The reader selects among 16 geometrically spaced access probabilities from 0.01 to one. Devices holding valid reports combine the broadcast admission probability with their local exponential backoff decisions.
Freshness-threshold guidance: Devices first check whether their report-age band satisfies the broadcast threshold. Eligible devices then apply local exponential backoff before attempting access.
The two guided policies use the same reader-side adaptation rule but differ in their candidate sets and device-local admission decisions.
Device Reports and Reader-Side Adaptation
For the feedback process in Figure 3, each device tracks reports replaced or expired and failed access attempts since its last delivered report, together with the age of its current reading. The first two counters accumulate across reports, while the age refers to the reading eventually delivered. In the collision model, an attempted access without a successful response increments the failure counter. After successful delivery, the sensor reading and these indicators are assumed to be available to the reader-side controller through the upper-layer interface discussed in Section III.

Figure 3. Freshness-guided admission and reader-side adaptation. Dashed elements denote the proposed guidance and feedback mechanisms. The compact additional message structures are illustrated at the bottom.
The reader uses an -greedy multi-armed bandit [14], maintained independently for each paging group. Each arm corresponds to a predefined guidance setting (i.e., an access probability or freshness threshold) and maintains an estimate of its expected reward.
For each successfully delivered report, the reader assigns one unit of reward, subtracting 0.25 for each previously lost report, 0.10 for each earlier failed access attempt, and 0.50 times the delivered report age normalized by its lifetime. The loss and failure counts are obtained from the device’s post-access feedback. It sums the contributions and divides by the number of configured AOs, while rounds without deliveries receive zero reward. Each delivered report’s reward contribution is floored at zero, preventing negative contributions from making a round with deliveries score below an empty round. The reward is computed from the feedback exactly as received, i.e., from the saturated four-bit loss and failure counters and the report age.
Notably, failed attempts proxy access effort, not directly measured energy consumption. Moreover, the reward is computed only at the reader and is never transmitted to devices. After each paging round, only the selected arm’s reward estimate is updated, including with a zero reward if nothing is delivered. At the group’s next opportunity, the reader selects the highest-estimated candidate, except with an exploration probability under which it selects a candidate at random, while this probability decays from 0.1 towards 0.01 as the group accumulates observations.
The signaling requirements are illustrated in Figure 3. Specifically, 16 probability candidates need four information bits, while freshness indices need three bits with five groups or five bits with one group for a 30-slot lifetime. A common one-byte AcGI design could allocate one bit to enable/disable the preconfigured policy, five to the index, and two as spare bits. Feedback can use four bits each for report losses and failed attempts, and five for report age: 13 bits packed into two bytes, with three spare bits. Loss and failure counts exceeding 15 are saturated. These are illustrative payload budgets rather than standardized 3GPP field allocations. As stated earlier, allocating additional bits to the policy indication could enable dynamic switching among different admission strategies.
Numerical Evaluation
The default simulation configuration is motivated by the 3GPP indoor environmental monitoring scenario, which specifies a device density of 1.5 devices/m and an end-to-end latency requirement of 20–30 s [13]. We consider 380 devices, corresponding to a nominal area of approximately 253 m, without explicitly modeling spatial deployment. Assuming a one-second paging round, the report lifetime is set to 30 slots, aligning the application-level deadline with the upper end of the specified latency range. The reader configures 16 AOs per round and a single paging group. The average report-generation rate is 0.02 per device per slot and is varied to assess the impact of traffic load, as 3GPP does not prescribe a reporting interval for this scenario.
Both guided schemes operate on top of device-side backoff. Neither uses offline calibration, and each starts from its most restrictive candidate index, with all reward estimates initialized to zero. Learning runs online, including during warm-up, and guidance adaptation is performed independently per group.
Results are averaged over 50 independent realizations, paired across the evaluated schemes, with 300 warm-up and 1,000 measurement slots per realization. Performance is assessed through the report-loss ratio, defined as the fraction of reports generated during the measurement period that are replaced or expire before successful delivery, and the mean delivery delay of successfully delivered reports, measured in slots from report generation. These metrics capture the trade-off between timely access and report preservation, since restrictive admission may reduce delay while increasing losses. Delay is omitted when the loss ratio reaches 95%. In addition, the average number of transmissions per delivered report can serve as a proxy for the effort/energy per successful report.
Figure 4 reveals three regimes as the report rate grows. Unguided CBRA collapses as soon as traffic exceeds what the AOs can carry. At light and moderate load, all adaptive schemes deliver most reports, and near the capacity transition, the two guided schemes are comparable. Under overload, threshold guidance loses the fewest reports, whereas device backoff degrades the most. Threshold guidance also keeps the delay of delivered reports low and nearly constant across all loads and needs the fewest transmissions per delivered report, while the transmissions required by device backoff escalate as collisions accumulate.
Figure 4. (a) Report-loss ratio (left) and (b) delay of delivered reports (right) versus the report-generation rate per device. Numbers next to open markers give the average number of transmissions per delivered report and shaded areas show 95% confidence intervals.
Figure 5 shows the same regimes as a function of resources. When AOs are scarce, CBRA loses almost all reports and device backoff degrades the most among the adaptive schemes, whereas threshold guidance preserves the most reports with the shortest delay and the fewest transmissions per delivered report. Near the capacity transition, the guided schemes are comparable, and device backoff loses slightly fewer reports. With sufficient AOs, all schemes perform well.
Figure 5. (a) Report-loss ratio (left) and (b) delay of delivered reports (right) versus the number of AOs per round. Numbers next to open markers give the average number of transmissions per delivered report and shaded areas show 95% confidence intervals.
Figure 6 shows how the guidance adapts to the paging configuration, with separate estimates per group. More groups mean fewer devices per paging occasion. With four or more paging groups, contention decreases sufficiently for unguided CBRA to recover, narrowing the performance gap between the compared access schemes. However, excessive grouping also extends the interval between paging opportunities, increasing the likelihood of report replacement or expiry in single-report buffers. This leads to report loss eventually increasing even as contention becomes less severe. Probability guidance deteriorates as the number of paging groups increases, potentially reflecting the reduced number of learning opportunities per group and its larger candidate set. Freshness guidance requires fewer candidate thresholds under these configurations and retains the shortest delay. However, this advantage narrows
Figure 6. (a) Report-loss ratio (left) and (b) delay of delivered reports (right) versus the number of paging groups. Transmissions per delivered report are not annotated, as they are similar for all adaptive schemes (1.4–3.1) across the curves.
as grouping reduces contention and all schemes need similar numbers of transmissions per delivered report.
Overall, freshness-threshold guidance performs best under severe contention, achieving the lowest report loss together with the shortest delay and the fewest transmissions per delivered report, while its outcomes vary far less across realizations than those of probability guidance.
Future Research Directions
Beyond the immediate benefits demonstrated in this work, the proposed framework opens the door to a broader evolution of AIoT access protocols, encompassing several promising directions for future research and standardization.
Signaling and Feedback Design
Further research and development should investigate the encoding and placement of AcGI within existing paging and control procedures, including optional MAC extensions and device-capability indication. Particular attention is needed to ensure coexistence between guidance-aware and legacy devices, quantify additional signaling overhead, and define how optional access-related feedback can be made available to the reader.
Beyond signaling design, feedback reliability is equally important, since silent and unsuccessfully contending devices cannot directly report their conditions. Future designs should investigate compact feedback representations, counter saturation, delayed reporting, and mechanisms for confirming successful delivery. Guidance policies must also remain effective despite missing, quantized, or outdated observations.
Context-Aware Admission and Intelligent Adaptation
In addition to the freshness-based strategy, device-local admission could also incorporate service priority, energy availability, transmission history, application-specific deadlines, or combinations of these factors. Class-based admission, probabilistic filtering, and multi-criteria eligibility rules offer alternative implementations of the same signaling framework. Their design should also address fairness and starvation, ensuring that repeatedly deferred devices are not systematically excluded from access.
Complementing these device-local decisions, more advanced reader-side adaptation could improve guidance selection under changing contention conditions. Contextual bandits, reinforcement learning, and traffic-prediction-assisted policies represent promising approaches. Building on emerging developments in Artificial Intelligence (AI)-assisted decision-making under uncertainty [15], generative and agentic AI could further support joint coordination of access guidance, paging-group selection, and contention resources. However, their benefits must be assessed against simpler controllers in terms of performance, feedback requirements, computational complexity, and convergence.
Digital Twins, Validation, and Resilience
Network digital twins could reproduce traffic, paging, contention, and device behavior to evaluate guidance policies and protocol extensions before deployment. Beyond calibrating admission thresholds and resource configurations, they could help identify which additional signaling fields and feedback mechanisms justify their overhead. During operation, digital twins could also be updated using observed statistics to anticipate changing conditions and refine guidance decisions.
Establishing the practical value of these approaches also requires evaluation under realistic operating conditions, including correlated and event-triggered reporting, heterogeneous traffic requirements, intermittent energy harvesting, imperfect decoding, capture effects, and multi-reader deployments. Protocol-level and hardware-oriented validation would help quantify practical gains and implementation costs, while addressing modeling accuracy, incomplete observability, and digital-twin mismatch.
An important extension is resilience to unexpected disruptions, including traffic surges, device unavailability, and changing service priorities. Such conditions may require joint adaptation of access guidance, paging configurations, and contention resources, potentially beyond conventional freshness-based admission.
Conclusion
This article presented a candidate AcGI extension for reader-triggered 3GPP A-IoT CBRA, enabling devices to make local admission decisions based on reader-provided guidance while retaining the underlying access procedure. Through a freshness-based sensor-monitoring case study, we demonstrated how compact guidance and optional device feedback can support adaptive contention management. The numerical results showed improvements in report loss and delivery delay, particularly under congested conditions, while highlighting the influence of traffic load, access resources, and paging-group configuration. More broadly, these findings motivate a shift toward context-aware A-IoT access, where reader-side coordination and device-local decisions jointly regulate contention within the existing radio-access framework.
Biographies
Amirhossein Azarbahram [S] is a postdoctoral researcher at the Centre for Wireless Communications (CWC), Oulu, focused on IoT, ISAC, and AI-native wireless systems.
David E. Ruiz-Guïrola [S’19, M’25] is a postdoctoral researcher focused on sustainable IoT, machine learning, and traffic prediction at the University of Oulu, Finland.
Onel López [S’17, M’20, SM’24] is an Associate Professor of wireless communications engineering, focused on sustainable IoT connectivity, at 6G Flagship, University of Oulu, Finland.
References
- [1]X. Shen et al., “State-of-the-art overview of 3GPP ambient IoT radio access aspects,” IEEE Communications Standards Magazine, vol. 9, no. 4, pp. 215–223, 2025.
- [2]3GPP, “Architecture support for ambient power-enabled internet of things; stage 2,” 3GPP, Technical Specification TS 23.369, 2026, release 19. Accessed: October 23, 2026. [Online]. Available: https://portal.3gpp.org/desk topmodules/Specifications/SpecificationDetails.aspx?specificationId=4397
- [3]3GPP, “NR; ambient IoT medium access control (MAC) protocol,” 3GPP, Technical Specification TS 38.391, 2026, release 19. Accessed: October 23, 2026. [Online]. Available: https://portal.3gpp.org/desk topmodules/Specifications/SpecificationDetails.aspx?specificationId=4391
- [4]F. Schoute, “Dynamic frame length ALOHA,” IEEE Transactions on Communications, vol. 31, no. 4, pp. 565–568, 1983.DOI
- [5]S. Song et al., “Two-level estimation enabled online congestion control for massive IoT networks,” IEEE Communications Letters, vol. 29, no. 8, pp. 1968–1972, 2025.DOI
- [6]S. Song et al., “Dynamic multi-channel random access procedure for 6G-enabled ambient IoT,” IEEE Journal on Selected Areas in Communications, vol. 44, pp. 2606–2621, 2026.DOI
- [7]D. E. Ruiz-Gu irola et al., “Adaptive contention-based random access for uplink reporting in 3GPP ambient IoT networks,” 2026, arXiv:2605.04966. Accessed: October 23, 2026. [Online]. Available: https://arxiv.org/abs/2605.04966
- [8]H. Chen, Y. Gu, and S. C. Liew, “Age-of-information dependent random access for massive IoT networks,” in 2020 IEEE Conference on Computer Communications Workshops (INFOCOM Workshops), 2020, pp. 930–935.arxiv.org/abs/2001.04780
- [9]D. E. Ruiz-Gu irola et al., “Context-awareness for dependable low-power IoT,” IEEE Communications Magazine, vol. 64, no. 8, pp. 86–92, 2026.
- [10]G. Jo and H. Lee, “Method and apparatus of preventing data transmissions for ambient IoT in wireless communication system,” International Patent Application WO 2025/198416 A1, Sep. 2025, published September 25, 2025. Accessed: October 23, 2026. [Online]. Available: https://patents.google.com/patent/WO2025198416A1/en
- [11]Z. Wu et al., “Paging techniques for ambient IoT,” International Patent Application WO 2025/212854 A1, Oct. 2025, published October 9, 2025. Accessed: October 23, 2026. [Online]. Available: https://patents.google.com/patent/WO2025212854A1/en
- [12]Z. Wu et al., “Ambient Internet of Things device random access procedure,” International Patent Application WO 2025/231815 A1, Nov. 2025, published November 13, 2025. Accessed: October 23, 2026. [Online]. Available: https://patents.google.com/patent/WO2025231815A1/en
- [13]3GPP, “Service Requirements for Ambient Power-Enabled IoT,” 3GPP, Technical Specification TS 22.369, 2024, release 19. [Online]. Available: https://portal.3gpp.org/desk topmodules/Specifications/SpecificationDetails.aspx?specificationId=4323
- [14]R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, 2nd ed. The MIT Press, 2018.
- [15]A. Azarbahram et al., “Toward proactive RF charging scheduling: Generative AI for decision support,” 2026, arXiv:2606.10600. Accessed: October 23, 2026. [Online]. Available: https://arxiv.org/abs/2606.10600


