Integrated Sensing and Communication in 3GPP: Evolution from 5G-Advanced to 6G

1

Integrated Sensing and Communication in 3GPP: Evolution from 5G-Advanced to 6G Xingqin Lin NVIDIA Email: xingqinl@nvidia.com Abstract—Integrated sensing and communication (ISAC) extends mobile networks from information transfer toward perception of passive objects and environments. By exploiting propagation delay, Doppler, angle, and temporal variation, a mobile network can support detection and tracking while reusing licensed spectrum, infrastructure, and edge computing at network scale and under operator control. Unlike conventional radar, ISAC must additionally address multi-application access, heterogeneous sensing entities, uncertainty, privacy, trust, and integration with communication services. This article reviews the evolution of ISAC in 3GPP from 5G-Advanced to 6G. It explains Release-19 ISAC service requirements and channel models, the Release-20 sensing function architecture and reporting abstractions, and the 5G monostatic baseline for drone detection and tracking. It also examines 6G-native sensing, including passive object sensing, communication assistance, multiple sensing modes, and multi-source data integration. By connecting the evolving 3GPP architecture and radio studies across 5G-Advanced and 6G, this article provides a unified technical perspective on current standardization choices, their implementation tradeoffs, and the open challenges shaping network-grade sensing.

I. INTRODUCTION A mobile communication network has traditionally been designed to transport information between identifiable endpoints. A transmitter encodes information, the wireless channel modifies the transmitted signal, and a receiver estimates enough of the channel to recover the intended data. From a sensing perspective, however, those channel modifications are not merely communication impairments. Propagation delay contains information about path length, Doppler shift contains information about relative motion, spatial phase differences contain information about direction, and temporal variations contain information about changes in objects and environments. Integrated sensing and communication (ISAC) exploits these relationships so that the radio network can also observe the physical world [1]. A base station (BS) may transmit a downlink signal while processing reflected signal to detect an uncrewed aerial vehicle (UAV), estimate its position and velocity, and maintain a track over time. Other potential targets include road vehicles, pedestrians, railway obstacles, automated guided vehicles (AGVs), industrial machinery, and structural features of an environment. The ISAC concept is related to radar but presents a different system problem. A conventional radar generally has a defined sensing mission, a controlled waveform, a known transmitterreceiver configuration, and a processing chain designed for a particular target class. A mobile network must serve multiple applications, cells, and devices. The application requesting sensing may reside outside the radio network. The target may not be a network subscriber and may contain no active radio

device. Measurements may be generated at one node, processed at an edge platform, fused with observations from several sites, and delivered to a third-party application [2]. ISAC therefore requires more than a physical-layer detector. The system must determine who may request sensing, which regions and target categories may be observed, which radio nodes can support the request, what sensing mode should be used, and how radio, computing, and transport resources should be allocated. It must determine whether to transport raw observations, target-related measurements, detected points, or object-level results. The final result must be protected, timestamped, qualified by confidence or uncertainty, and delivered only to authorized consumers. ISAC also differs from conventional cellular positioning. Cellular positioning normally estimates the state of a participating device using measurements associated with that device, such as timing, angle, reference signal, or satellite navigation observations. In passive object sensing, the target may transmit nothing. It is observed because it reflects, blocks, or diffracts a signal transmitted in the network. This enables device-free perception, but it introduces difficult target identification, privacy, and trust problems. The sensing processing must distinguish target reflections from static clutter, moving clutter, transmitter leakage, and unrelated interference. The 3rd generation partnership project (3GPP) initiated ISAC work within 5G-Advanced in Release 19, while Release 20 combines continued 5G-Advanced evolution with the first 6G studies, as illustrated in Figure 1 [3], [4]. In Release 19, 3GPP service and system aspects (SA) working group 1 studied ISAC use cases and established normative service requirements [5], [6]. Building on that, 3GPP SA2 in Release 20 specifies architecture and functional enhancement, end-to-end service operations and procedures to support ISAC [7], and 3GPP core network & terminals (CT) working group 4 develops core network application programming interfaces (APIs) for sensing function services. In parallel, 3GPP SA3 addresses security and privacy across ISAC operations and the sensing data lifecycle [8], 3GPP SA5 studies management and charging aspects of ISAC, and 3GPP SA6 investigates the use of sensing results for vertical applications by studying application-enablement architecture enhancements [9]. On the radio access network (RAN) side, Release-19 work in RAN1 extended the previous 3GPP channel model to account for sensing specific aspects, such as radar cross section (RCS) modeling, sensing target mobility, and spatial consistency [10]. In Release 20, 3GPP RAN1 conducted sensing performance evaluation for BS monostatic sensing for UAV detection and tracking, leveraging the channel model developed in Release 19 [11]. While the sensing functionality relies on existing 5G downlink waveform and reference signals, 3GPP RAN3 studied procedures and

2

Figure 1: Evolution of 3GPP ISAC from Release 19 foundations to the dual 5G-Advanced and 6G tracks in Release 20.

signaling between the RAN and core network to support ISAC, investigated RAN architecture for BS monostatic sensing, and is conducting normative work. Looking ahead, 3GPP Release 20 also explores ISAC as part of 6G studies. 3GPP SA2’s study on 6G architecture considers integration of sensing with 3GPP access and other sensing sources [12]. 3GPP RAN study on 6G scenarios & requirements includes sensing key performance indicators (KPIs) and service needs [13]. All six sensing modes (i.e., BS monostatic, user equipment (UE) monostatic, BS-BS bistatic, BS-UE bistatic, UE-BS bistatic, and UE-UE bistatic) are being explored. The 3GPP RAN study on 6G radio tackles physical layer design, higher-layer procedures and protocols, and radio frequency (RF), coexistence, and testability aspects [4]. This article provides a unified technical view of the two concurrent 5G-Advanced and 6G sensing tracks and explains how the 5G sensing baseline informs the native design of sensing in 6G. We first describe the service model, use cases, sensing topologies, and the translation of application requirements into radio and processing requirements. Then we examine the 5G-Advanced sensing function architecture and its extension toward a 6G multi-entity, multi-source sensing framework. Furthermore, we discuss target-aware channel modeling, the 5G sensing processing chain, and the ongoing 6G radio studies. We finally conclude the article and point out future work directions. II. SENSING SERVICES AND REQUIREMENTS The 3GPP sensing service model separates the physical radio observation from the information delivered to a consumer. A sensing signal is a transmission over the 3GPP radio interface that can be used to observe an object or environment. After interaction with the physical scene, a sensing receiver derives sensing data. Depending on where processing is divided, this

data may consist of complex samples, channel estimates, delay or Doppler profiles, angular spectra, or detected target points. Further processing converts the data into a sensing result, such as object presence, position, velocity, trajectory, class, or event indication. Sensing contextual information accompanies the result when additional interpretation is required. It may identify the observation time, coordinate reference, sensing area, target assumptions, achieved accuracy, confidence, or result age. Context is especially important when the result is fused across several nodes or when it is delivered to an application that has no knowledge of the underlying radio configuration. 3GPP also defines a target sensing service area as the geographic region within which requested object or environmental characteristics are to be derived [6]. This area should not be equated automatically with a communication cell, because passive object observability depends on target reflection, geometry, clutter, receiver sensitivity, and radio resource availability. The principal use cases can be grouped into: a) transportation and aerial sensing, b) industrial and logistics sensing, c) humancentric sensing, and d) environmental and public safety sensing, as summarized in Table 1. Transportation and aerial applications require detection and tracking of vehicles, pedestrians, road or railway obstacles, and UAVs. Industrial and logistics applications include AGVs, robots, machinery, assets, and restricted areas. Human-centric applications include presence, occupancy, movement, gestures, and contactless monitoring. Environmental and public safety applications include hazard detection, area monitoring, intrusion detection, and situational awareness. These applications require different output abstractions. A protected airspace service may require a three-dimensional (3D) UAV track with velocity and existence confidence. A factory safety service may require an event indicating that an object entered a restricted region. A gesture recognition service may require a semantic class rather than a

3

Use case family Transportation and aerial Industrial and logistics Human-centric Environmental and public safety Communication assistance

Representative targets or scenario

Required sensing result

Dominant requirements

UAVs, vehicles, pedestrians, railway obstacles AGVs, robots, machinery, restricted areas Presence, gesture, movement, contactless monitoring Hazard, intrusion, area monitoring Blockage, geometry, environmental state

Detection, position, velocity, trajectory Position, activity, safety event Occupancy, motion, activity class Event, map, environmental state Beam or mobility assistance information

Coverage, missed detection, false alarms, accuracy refresh rate Accuracy, latency, availability, continuity Fine motion sensitivity, latency, privacy Wide-area coverage, reliability, long-term stability Freshness, confidence, low processing latency

Table 1: Representative 3GPP ISAC use cases and sensing requirements.

globally referenced position. An environmental monitoring service may require an aggregate spatial map updated over minutes rather than a low-latency target track [5], [6]. The geometry is determined by the placement of the sensing transmitter and receiver [6]. In monostatic sensing, the functions are co-located at one RAN node or UE. The measured delay corresponds to the outward and return propagation between that node and the target. A common clock simplifies timing, but a monostatic receiver must suppress strong leakage from its own transmitter. In bistatic sensing, transmitter and receiver are at different locations. A measurement then depends on the combined transmitter-to-target and target-to-receiver path. Bistatic operation provides spatial diversity and may reduce local self-interference, but it requires accurate node positions, timing alignment, configuration coordination, and a method for associating the transmission with the received reflection. Multistatic sensing uses several transmitters, receivers, or both. It can improve coverage, reduce geometric ambiguity, mitigate blockage, and provide more observations for target tracking. The cost is increased synchronization, transport, calibration, association, and fusion complexity. Different nodes may observe different scattering centers on the same target, and their observations may share common clutter or model errors. Sensing performance contains several coupled dimensions. Detection performance includes probability of detection, missed detection probability, and false alarm probability. These values depend on the detector threshold. A low threshold improves sensitivity to weak targets but produces more false alarms and more candidate points for downstream processing. Estimation performance includes range, position, angle, velocity, trajectory, target size, orientation, and classification. The coordinate system and observation timestamp are as important as the numerical estimate. A target position expressed in a local antenna coordinate system cannot be fused with another node until the frames have been aligned. For fast targets, a result delivered tens of milliseconds after observation may already require prediction to the current time. Resolution differs from single target accuracy. Range resolution describes whether two targets at nearby path lengths can be separated, whereas accuracy describes the error in estimating one target. As a reference, approximately 300 MHz of effective monostatic sensing bandwidth provides about 0.5 m of conventional range resolution. Velocity resolution improves with coherent observation time. A longer observation can separate closely

spaced Doppler components but increases latency and sensitivity to target acceleration, oscillator drift, and phase noise. Angular resolution improves with physical aperture and calibrated spatial sampling. Large arrays provide narrow beams and high potential precision, but array phase, antenna position, orientation, and beam-dependent RF responses must be calibrated. Service-level KPIs include maximum latency, refresh period, availability, confidence, and area coverage, among others. Endto-end latency includes radio observation, local processing, transport, fusion, result generation, and exposure. Refresh period determines how often a new state or event is delivered. A fast UAV track may require updates several times per second, whereas a slowly changing environmental map may be updated every minute. Translating from application intent to a feasible radio operation is a central function of the sensing architecture. A mature service must support feasibility assessment and quality degradation. The network should reject an unsupported request, negotiate a lower service level, or report the quality actually achieved. These service-to-radio tradeoffs motivate the common sensing framework being studied for 6G. The Release-20 6G studies broaden the sensing framework in two important ways. First, sensing is treated as a native 6G capability rather than as an add-on to 5G. Second, sensing supports both detection and tracking of passive objects (covering at least UAVs, humans, vehicles, and AGVs) and communication assistance [13]. For the latter, a sensing result may be consumed internally by a communication function rather than exposed to a vertical application. Environmental and object information may assist beam selection, blockage prediction, mobility preparation, or interference management [1]. III. SENSING ARCHITECTURE A. Sensing Function and Service Lifecycle in 5G-Advanced The Release-20 5G-Advanced architecture introduces a logical sensing function (SenF) into the 5G system, as illustrated in Figure 2 [7]. The architecture builds on existing network exposure and service-based principles, while enhancing the core network and next-generation RAN (NGRAN) to support sensing operations. The SenF is responsible for functions such as sensing service authorization, sensing entity discovery and selection, provisioning of sensing parameters, collection and transport of sensing data, generation

4

Figure 2: 5G-Advanced Sensing Function architecture.

of final results, and result exposure [7]. The SenF is logical and need not be implemented as one centralized software instance. Control oriented functions may be centralized while high-rate processing is placed at an edge site near the RAN. Such a split may reduce transport and latency, but it requires coordination across sensing control function and sensing processing function. A representative service begins when an application function submits a sensing service request, either directly through an authorized exposure mechanism or through the network exposure function (NEF). The request may identify the application, external target sensing area, requested sensing service type, desired result type, target characteristics, start time, duration, reporting mode, and service quality requirements. The exposure layer authenticates the application and applies operator policy. It may translate an externally defined geographic region into a network internal representation and discover an appropriate SenF. The SenF then applies sensing specific authorization [7], [8]. Authorization must be more granular than a binary permission to use the sensing service. A public safety application may be permitted to receive UAV tracks over a defined perimeter but not detailed human motion information in adjacent areas. A factory application may receive an event indicating that a restricted zone is occupied but not raw radio measurements from which activities elsewhere could be inferred. After authorization, the SenF selects one or more sensing entities. In the 5G-Advanced scope, the sensing entities are next-generation node B’s (gNBs). Selection can consider geographic coverage, carrier frequency, available bandwidth, antenna configuration, receiver capability, processing capacity, and expected sensing quality. The SenF derives radio-facing sensing parameters from the application request. These can include the target sensing service area, sensing QoS, reporting behavior, and target information. The gNB maps these serviceoriented parameters onto implementation specific resources,

such as beams, downlink signals, antenna panels, receiver chains, coherent observation periods, and processing levels. The gNB then performs the radio operation. It may schedule dedicated resources or reuse suitable communication transmissions. It receives the target dependent response, suppresses leakage and clutter, extracts measurements, and reports the configured information. The report may contain detected reflection points or an object-level result, depending on the selected processing split. The SenF receives the report and produces the application-facing result. It may combine observations across time or gNBs, transform coordinates, associate detections with an existing target, perform tracking, and attach confidence and contextual information. The result is then delivered through the exposure framework as a one-time response, a periodic report, a subscription notification, or an event indication. An ongoing service can be updated. An application may change the sensing area, duration, refresh period, or required quality. The SenF may modify the selected gNBs or radio configuration. Termination may be initiated by the application, authorization revocation, resource shortage, completion of the requested period, or sensing failure. The relationship between application requests and radio sensing operations is potentially many-to-many. Several authorized applications may consume different results derived from one observation, avoiding duplicate use of radio resources. Conversely, one request may require several gNBs or repeated sensing occasions. The SenF must therefore maintain a service context that links authorization, requested quality, selected entities, operation identifiers, reporting state, and processing state. B. Processing Abstraction and Architectural Extension in 6G Processing placement determines how much sensing information leaves the RAN and where target interpretation occurs. Four abstraction levels studied in 5G-Advanced are particularly useful for 6G sensing architecture analysis: sample

5

Abstraction

Representative information

Transport load

Parameter profile level

Raw observations indexed by time, frequency, antenna, and beam Delay, Doppler, angle, or power profiles

Detected point level

Range, velocity, angle, power, timestamp, uncertainty

Moderate

Object level

Position, velocity, class, track ID, confidence, lifecycle state

Low

Sample level

Very high High

Processing and fusion implications Maximum centralized flexibility; substantial calibration and privacy burden Preserves multi-target structure; requires common binning and normalization Supports centralized association and fusion; local detection may discard weak components Direct application use; greatest dependence on local tracking and classification

Table 2: Sensing information abstractions and their transport, processing, and fusion tradeoffs.

data, parameter profiles, detected points, and object-level results [11]. • Sample level reporting transports raw observations indexed by time, frequency, antenna, and beam. It provides maximum flexibility for centralized processing, retrospective analysis, and algorithm updates. But its transport requirement is extremely high. • Parameter profile reporting carries delay, Doppler, angle, or reflection power profiles. It retains multi-target structure with lower volume than raw samples, but the interface must define bin spacing, normalization, dynamic range, timestamping, and calibration. • Detected point reporting carries target point related components such as range, velocity, azimuth, elevation, reflection power, and timestamp. This level is compact enough for network transport while retaining information needed for centralized clustering, association, fusion, and tracking. The limitation is that the gNB has already applied clutter suppression and a detector threshold. Weak components discarded locally may not be recovered through later fusion. • Object level reporting carries a target track or semantic result, including position, velocity, class, track identifier, existence confidence, uncertainty, and lifecycle state. It minimizes transport and allows the gNB to exploit local calibration and clutter history, while leaving less room for centralized fusion. Table 2 summarizes the four levels of sensing information abstractions and their transport, processing, and fusion tradeoffs. A practical network may adapt the reporting level. Routine surveillance could use object level reports. A cooperative fusion operation could trigger detected point reporting. Selected profiles or samples might be collected temporarily for calibration, fault analysis, or model validation. Such adaptation requires capability exchange, policy control, and continuity rules when the abstraction changes. The Release-20 6G architecture study extends 5G sensing to a broader set of sensing entities and modes [12]. RAN nodes and UEs may serve as sensing entities, leading to various sensing modes as illustrated in Figure 3. The architecture studies discovery and reselection of sensing entities, sensing mode selection, configuration and policy provisioning, authorization and revocation, data collection, and result exposure. Current solution discussions include evolution of the 5G sensing function architecture to support UE involvement and selection among RAN- and UE-involved sensing modes.

UE involvement introduces architectural concerns absent from a gNB-only model. A UE can move during an operation, change serving nodes, enter a power saving state, or lose radio coverage. It may require capability exchange, user consent, subscription policy, and control over how its observations are used. A UE can act as a sensing entity, a sensing service consumer, or potentially both. Sensing mode selection becomes a system decision. The appropriate mode may depend on target geometry, target area, UE capabilities, synchronization, radio load, and privacy policy. A RAN node may be the transmitter and receiver, two RAN nodes may form a bistatic pair, a UE may transmit while a RAN node receives, or a UE may receive a RAN transmission. The architecture must identify which function proposes candidate modes, which entity makes the final radio decision, and how mode changes are coordinated. Multiple sensing data sources broaden the problem further. The 6G SenF may receive 3GPP radio measurements, UEgenerated observations, RAN-generated object tracks, or information originating from other sensing technologies [12]. In order to fuse multiple sources such as a camera detection, lidar point cloud, and radio reflection point, their timestamps and coordinate frames need to be aligned. Multi-source fusion also creates trust and privacy questions. A sensing result should identify enough provenance for the network to assess whether sources are independent, correlated, authenticated, and authorized. A non-3GPP source may have different ownership, retention, consent, or integrity guarantees. Communication assistance creates another type of consumer. Instead of exposing a result through an external API, the SenF or RAN may deliver it to a beam management, mobility, scheduling, or interference control function. Such internal consumers need low latency and a quality representation suited to automated control. The 6G architecture should therefore preserve the useful abstractions of the 5G SenF while extending them for entity mobility, mode selection, multi-source data, and communication assistance. The challenge is to maintain one coherent service and data model across increasingly heterogeneous sensing operations. IV. RADIO SENSING The credibility of cellular sensing depends on two tightly coupled elements: a channel model that preserves the relationship among transmitter, target, receiver, and environment, and a receiver that converts known radio transmissions into target related measurements. A sophisticated detector evaluated with an isolated point target and no realistic

6

Figure 3: Monostatic and bistatic sensing modes involving BSs and UEs; multi-static sensing combines multiple such links.

clutter can produce misleading results, while a detailed propagation model provides little value if the receiver assumes perfect calibration or negligible transmitter leakage. The ISAC channel model extensions to TR 38.901 and the 5G ISAC study in TR 38.765 provide the initial 3GPP framework for connecting target-aware propagation with radio processing [10], [11]. The Release-20 6G studies broaden this framework to additional targets, sensing modes, and RF conditions. A. ISAC Channel Modeling A communication channel model describes the composite propagation between a transmitter and an intended receiver. For sensing, the model must retain enough physical structure to identify which components are associated with the target. A useful model therefore separates the target channel, the background channel, and local transmitter-to-receiver coupling. The target channel includes propagation to the target, the target’s scattering response, and propagation to the sensing receiver. The background channel represents buildings, terrain, vegetation, machinery, vehicles, and other non-target objects. Local coupling represents direct transmitter leakage and reflections from the sensing site. TR 38.901 provides the target and background propagation framework, whereas local transmitter leakage and coupling require an additional RF or implementation impairment model. Separating target and background components is essential for evaluating detection, as illustrated in Figure 4. The receiver must distinguish the target using observable delay, Doppler, angle, reflection strength, polarization, and temporal behavior. Missed detections and false alarms depend strongly on the structure and dynamics of the background. The sensing topology determines the geometry that the model must preserve. In monostatic sensing, transmitter and receiver are colocated, so delay represents a round trip and Doppler is associated with radial motion relative to one node. In bistatic sensing, delay depends on the combined transmitter-to-target and target-to-receiver paths, while Doppler depends on motion relative to both legs. Multistatic sensing adds spatial diversity, but observations across links must remain consistent with one physical target state. Independently generating each link can create unrealistic diversity and exaggerate cooperative sensing gains. 3GPP TR 38.901 represents a physical target using one or more target scattering points [10]. A scattering point is an

abstraction for one or several nearby physical scattering centers with a common position, motion, RCS behavior, and polarization response. A single point may be sufficient for a compact target at long range. Extended targets require several points when they occupy multiple delay or angular resolution cells. Vehicles, humans, AGVs, and UAVs therefore require different levels of target detail. Besides, RCS should not be treated as one fixed value for a target class. It varies with frequency, material, shape, polarization, target orientation, and illumination and observation angles. A UAV can move through deep reflection minima even when its range changes only slightly. This aspect dependence is important for tracking because a valid target may disappear temporarily from one view and reappear as its orientation changes. The model should preserve consistency between target motion, orientation, and reflection strength over time. Extended and articulated targets may also produce microDoppler. Detailed micro-Doppler is particularly relevant to classification studies and may require target models beyond the baseline detection-oriented representation. Rotating UAV propellers, vehicle wheels, human limbs, and industrial machinery create Doppler components around the bulk target motion. The required detail depends on the study objective. Basic detection may need only a dominant target component and realistic power variation. Tracking needs temporally consistent reflection changes. Classification requires richer representation of moving target parts. The background channel requires equal care. Static clutter includes buildings, terrain, parked vehicles, and fixed infrastructure; dynamic clutter includes traffic, people, birds, and rotating equipment. These components should evolve consistently across time, beams, and sensing links. Such consistency is necessary for evaluating receivers that learn or subtract a long-term background [14]. B. Radio Processing For the gNB-based monostatic baseline considered in 5G ISAC, the essential processing sequence includes five steps [11]. Step 1: transmitted-signal reconstruction: The first requirement is accurate reconstruction of what was radiated. A gNB knows the downlink reference sequences and transmitted data symbols, but the sensing processor must also account for resource allocation, physical channel mapping, precoding, and beamforming. This is particularly important when data-bearing

7

Figure 4: An illustration of 3GPP ISAC channel model comprising target channel and background channel.

resources are reused. They can increase effective bandwidth and energy, but communication scheduling creates gaps and may change the precoder or beam across time and frequency. Those changes alter target illumination and must not be interpreted as target motion. Step 2: calibration and self-interference suppression: Calibration removes deterministic radio and array effects. Frequency-dependent transmit and receive responses can distort the delay profile, timing and frequency offsets can appear as range and velocity bias, and antenna phase errors can bias angle estimates. Communication-grade calibration may not be sufficient because errors with little effect on throughput can produce significant sensing bias. Self-interference suppression is the defining challenge of monostatic gNB sensing. The target echo may be many orders of magnitude weaker than local transmitter leakage. Antenna separation, directional isolation, shielding, filtering, and analog cancellation are needed to prevent receiver saturation. Digital cancellation can then remove residual leakage using the reconstructed waveform and an estimate of the coupling channel. That estimate may need to capture frequency selectivity, phase noise, IQ imbalance, and transmitter nonlinearities. Since saturation cannot be corrected digitally, realistic RF isolation and cancellation assumptions are essential in performance evaluation. Step 3: delay-Doppler-angle processing: After calibration and leakage suppression, the receiver estimates delay, Doppler, and angle. Frequency-domain phase variation provides delay information, temporal phase evolution provides Doppler, and spatial variation across antennas provides direction. The resulting delay-Doppler-angle response is an intermediate radio representation rather than an object list. Strong cells may correspond to the target, environmental clutter, sidelobes, or residual leakage. The actual 5G resource pattern matters. Sparse subcarriers create delay sidelobes and ambiguities; gaps between sensing occasions degrade Doppler processing; and beam or precoder changes couple spatial and temporal

observations. Windowing can suppress sidelobes from strong components, but it broadens the main response and makes nearby targets harder to separate. Step 4: clutter suppression and detection: Clutter suppression is then applied to the delay-Doppler-angle response. Static background subtraction can reduce persistent environmental components, while temporal or Doppler filtering emphasizes moving targets. A practical receiver may combine a moving-target branch with a slower scene-change branch that compares the current observation with a long-term environmental map. Detection identifies statistically significant components in the clutter-suppressed response. Adaptive thresholds, including constant-false-alarm-rate techniques, are useful because the background varies across range, direction, and environment. The threshold controls the balance between weak target sensitivity and false alarms. Step 5: target-parameter estimation: The radio processor finally estimates parameters for each detected component. A point-level measurement may contain delay or range, Doppler or radial velocity, azimuth, elevation, reflection strength, timestamp, uncertainty, and a quality indicator. Extended targets may produce several points, and preserving them is often preferable to forcing the physical layer receiver to declare one object. Clustering, tracking, classification, and multi-node fusion can then operate above the radio measurement layer. This point-level boundary is useful for standardization. It allows 3GPP to define measurement semantics, coordinate and time references, quantization, capability exchange, and reporting procedures without prescribing radio processing details [11]. While point-level measurements provide one natural standardized radio boundary, object-level reporting remains possible when the selected processing split places clustering and tracking at the sensing entity. The 6G radio study can reconsider each stage. More regular sensing resources may improve delay-Doppler ambiguity and coherent integration. Bistatic modes can reduce local self-

8 interference but require tighter synchronization and assistance information. Cooperative sensing can improve geometry but requires consistent measurements across nodes. 3GPP RAN4 evaluation remains critical because dynamic range, leakage, phase noise, nonlinear distortion, coexistence, and testability determine whether a standardized ISAC feature is deployable [15]. V. CONCLUSIONS AND FUTURE OUTLOOK 3GPP ISAC is evolving from an exploratory radio capability into an end-to-end network function. Release 19 established the essential foundations, including sensing use cases, service requirements, and channel models. Release 20 advances two complementary tracks. The 5G-Advanced track develops a sensing framework based primarily on gNB monostatic sensing for UAV detection and tracking. In parallel, the 6G studies treat sensing as a native capability, covering detection and tracking of objects beyond UAVs, communication assistance, multiple sensing modes, and heterogeneous sensing data sources. The 6G work can reuse 5G-Advanced developments while reconsidering limitations inherited from a communicationcentric air interface. It can introduce more suitable signals, resource structures, beam procedures, feedback mechanisms, and architectural functions where the benefits justify their cost. The central 6G challenge is to provide network-grade perception that is timely, secure, privacy-preserving, and useful to both applications and communication functions. We conclude by pointing out some fruitful avenues that require further work. Sensing-communication co-design: 6G sensing must progress beyond opportunistic reuse of communication signals. Sensing-aware scheduling, beam management, and admission control are needed to provide predictable sensing quality without excessive communication overhead. Communication assistance results must also be delivered within the timescale of beam, mobility, and scheduling decisions. Cooperative and multi-modal sensing: Multi-node and multi-modal sensing can improve coverage, accuracy, and resilience by combining radio, camera, lidar, or other observations. These gains require aligned timestamps, coordinate frames, calibration states, and provenance, together with robust cross-node and cross-modal association. Security and privacy: Passive targets may include individuals with no subscription or signaling relationship with the network. Authorization should therefore constrain the sensing area, target category, result detail, purpose, duration, and retention policy. Edge processing and minimum necessary result exposure can reduce the risks associated with transporting detailed radio observations. Reproducible evaluation: Fair comparison requires shared target and clutter models, RF-impairment assumptions, calibration procedures, and ground truth methods. These elements should be complemented by open-source ISAC evaluation tools that implement reference waveforms, channel and target models, receivers, baseline algorithms, and reference outputs. They should support simulation, measured datasets, and hardware-in-the-loop validation, enabling repeatable results without prescribing a single implementation.

REFERENCES [1]

F. Liu et al., “Integrated sensing and communications: Toward dualfunctional wireless networks for 6G and beyond,” IEEE J. Sel. Areas Commun., vol. 40, no. 6, pp. 1728–1767, Jun. 2022. [2] J. A. Zhang et al., “Enabling joint communication and radar sensing in mobile networks—A survey,” IEEE Commun. Surveys Tuts., vol. 24, no. 1, pp. 306–345, Firstquarter 2022. [3] X. Lin, “The Bridge Toward 6G: 5G-Advanced Evolution in 3GPP Release 19,” IEEE Communications Standards Magazine, vol. 9, no. 1, pp. 28-35, March 2025 [4] X. Lin, “A Tale of Two Mobile Generations: 5G-Advanced and 6G in 3GPP Release 20,” IEEE Communications Standards Magazine, early access. [5] 3GPP, “Study on Integrated Sensing and Communication,” 3GPP TR 22.837, V19.4.0, Jun. 2024. [6] 3GPP, “Integrated Sensing and Communication,” 3GPP TS 22.137, V19.1.0, Apr. 2024. [7] 3GPP, “Integrated Sensing and Communication; Stage 2,” 3GPP TS 23.137, V0.3.0, Jun. 2026. [8] 3GPP, “Study on Security and Privacy Aspects of Integrated Sensing and Communication,” 3GPP TR 33.777, V0.6.0, Jun. 2026. [9] 3GPP, “Study on Use of Sensing Results for Vertical Applications,” 3GPP TR 23.700-15, V20.1.0, Jul. 2026. [10] 3GPP, “Study on Channel Model for Frequencies from 0.5 to 100 GHz,” 3GPP TR 38.901, V19.4.0, Jun. 2026. [11] 3GPP, “Study on Integrated Sensing and Communication (ISAC) for NR,” 3GPP TR 38.765, V20.0.0, Jun. 2026. [12] 3GPP, “Study on Architecture for 6G System; Stage 2,” 3GPP TR 23.80101, V0.8.0, Jul. 2026. [13] 3GPP, “Study on 6G Scenarios and Requirements,” 3GPP TR 38.914, V20.0.0, Jul. 2026. [14] W. Hong et al., “Integrated Sensing and Communication (ISAC) Channel Model Toward 3GPP 6G Standardization: Modeling, Validation, and Application,” IEEE Journal on Selected Areas in Communications, vol. 44, pp. 3459-3472, 2026. [15] A. Ghosh et al., “A Unified Future: Integrated Sensing and Communication (ISAC) in 6G,” IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation, vol. 1, no. 1, pp. 365-374, Sept. 2025.

Related documents

• Evaluation of the effects of 3GPP-specific beamforming and channel estimation on the 3D EIRP profile of a 5G gNB#134514
• Challenges and Opportunities in the Transition from 5G to 6G Networks#1000113
• Challenges and Opportunities in the Transition from 5G to 6G Networks#919325
• Effets environnementaux de la 5G (partie 3)#350846
• Change Point Detection in the Phase of a Signal : Application to Remote Sensing by GNSS-Reflectometry#213214