Performance Analysis of ISAC Slicing with Sensing-Assisted Communication and Communication-Assisted Sensing

© 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. DOI: 10.1109/XXXXXX.XXXX.XXXXXXXX.

Performance Analysis of ISAC Slicing with Sensing-Assisted Communication and Communication-Assisted Sensing Daniel “Dante” Campos and Jacek Kibiłda†

I. I NTRODUCTION One of the major innovations behind future 6G networks is Integrated Sensing and Communication (ISAC), which is based on the idea of co-designing wireless systems to enable simultaneous provisioning of communication and sensing services. While much of the innovation in terms of integration and performance analysis is currently occurring at the physical and signal-processing layers [1], the system-level view remains somewhat limited, positing the existence of two types of slices: Communication as a Service (CaaS), and Sensing as a Service (SaaS) [2]–[4]. Viewed as distinct slices with distinct objectives and performance requirements, sensing and communication will inevitably compete for resources [5], [6]. In this article, we make the case that, unlike conventional slicing, ISAC slicing introduces interdependencies between SaaS and CaaS slices. Communication performance may benefit from sensing via Sensing-Assisted Communication (SAC), leveraging sensing information to enhance control decisions, such as beamforming based on position estimates of served User Equipments (UEs) [2], [7]. Simultaneously, sensing accuracy and coverage may benefit from Communication-Assisted Sensing (CAS), which includes providing communication links for sensors in bi- and multistatic sensing modes to share data [8], [9]. These interdependencies are likely to become particularly poignant in scenarios with high target densities. Sensing

Sensor

ISAC Base Station

Antenna Boresight

400 Both 200 SaaS 0 CaaS

−200

ISAC Slice SLAs Satisfied

Abstract—Integrated Sensing and Communication (ISAC) will allow 6G networks to provide both sensing and communication services through shared infrastructure and spectrum resources. However, network slicing that views these services as distinct slices with distinct objectives and performance requirements will inevitably miss the emergent value of integrating them. By considering Sensing-Assisted Communication (SAC) and Communication-Assisted Sensing (CAS), which create interdependencies between services, we can observe trade-offs in resources dimensioned to support each slice and, ultimately, the ISAC slicing performance. As such, this paper proposes a performance analysis of an ISAC slicing framework that accounts for resource allocation and sensor density, as well as SAC and CAS, dimensioned to achieve slice-specific Service-Level Agreement (SLA) satisfaction. Numerical simulations, across different bandwidth allocations, sensor deployments, and demand scenarios, show that ISAC slicing performance is fundamentally constrained by sensing resolution, shaped by demand, and scaled by sensor deployment. These findings provide insights for the design of future ISAC slicing mechanisms. Index Terms—Integrated Sensing and Communication, Network Slicing, Sensing-Assisted Communication, Bistatic Sensing.

Y Position (m)

arXiv:2610.03879v1 [cs.NI] 2 Oct 2026

† Commonwealth Cyber Initiative, Virginia Tech, USA, emails: {dcamposfsa, jkibilda}@vt.edu

None −400 −400

−200

0

200

400

X Position (m)

Fig. 1: Sample SLA satisfaction coverage for one SaaS slice and one CaaS slice in a single-cell ISAC slicing scenario with three deployed sensors performing bistatic sensing. in such scenarios requires additional bandwidth to maintain sufficient sensing resolution, ultimately limiting the number of distinguishable targets [1], [10]. Beyond resource allocation, sensor density itself becomes a critical design variable for ISAC networks supporting bi- and multistatic sensing [9], [11]. By impacting sensing distances, sensor density directly affects sensing accuracy and coverage [6], [9], [11]. Consequently, dimensioning resources to satisfy SaaS and CaaS Service Level Agreements (SLAs) presents a challenging question that requires accounting for both demand and network-specific factors. In this article, we evaluate these factors together in an ISAC slicing scenario and characterize their impact on sensing, communication, and, ultimately, network-level slicing performance. Fig. 1 illustrates a snapshot of network-level slice SLA satisfaction across a cell provisioning ISAC. Recent work has investigated resource allocation, cooperative sensing, and ISAC slicing, but these aspects are studied independently and evaluated with different performance metrics. The works [5], [6] optimize beamforming, sensor-target assignment, and sensing coverage in bi- and multistatic cooperative sensing deployments, evaluating performance through Signal-to-Noise Ratio (SNR) under fixed sensing topologies. Sensor deployment has also been investigated as a sensingdesign problem. In [11], the authors optimize sensor placement

to improve sensing accuracy, while [9] surveys deployment, selection, and coordination mechanisms for cooperative ISAC. Still, these works do not address slicing and resource allocation. On the other hand, studies on these two topics, such as [3], [12], [13], optimize bandwidth, power, or deployment decisions under sensing and communication requirements, but, importantly, treat sensing and communication as independent. Recent works have separately demonstrated the benefits of SAC for communication through sensing-assisted beamforming [7] and highlighted the communication overhead of CAS in cooperative sensing [14]. Nevertheless, these interdependencies are not analyzed within the context of ISAC slicing. To the best of our knowledge, this work is the first to evaluate, in a bistatic ISAC slicing scenario, how SAC, CAS, and sensing deployment impact on sensing, communication, and networklevel slicing performance. Our contributions can be summarized as follows: • We propose an ISAC slicing framework that jointly models the impact of resource allocation, sensor deployment, SAC, and CAS, linking network configuration to sensing, communication, and slicing performance. • We develop a publicly available simulation tool and evaluate trends and trade-offs between SaaS and CaaS with SAC and CAS across different sensor deployment densities, bandwidth allocations, and demand scenarios. • We find that ISAC slicing performance is: (i) constrained by resolution, which impacts both slices and defines the SAC trade-off; (ii) shaped by demand, especially from CaaS slices; and finally (iii) scaled by sensor deployment, significantly improving SaaS slice performance. II. N ETWORK M ODEL This section describes the key network and channel modeling parameters and assumptions, with a focus on the SNR, which is the basis for describing the link budget and associated performance of both sensing and communication services. A. Network We consider a single Urban Microcell (UMi) in a vehicular scenario where an ISAC Base Station (BS), denoted by M , provides communication and sensing services to SaaS and CaaS slices. The cell is divided into a set of K non-overlapping sectors. Without loss of generality, we assume that each sector is equipped with a Uniform Linear Array (ULA) of AM antenna elements. Within the cell range R, there is a set of vehicles V, of which U ⊆ V are active UEs attached to the BS and served by the CaaS slice; and T ⊆ V are known targets tracked by the SaaS slice. Since Orthogonal FrequencyDivision Multiplexing (OFDM) continues to be the waveform of choice for the upcoming releases of 6G [4], we assume that the BS implements OFDM-based ISAC, as in, e.g., [1], [15], leveraging time and frequency resources. Without loss of generality, we assume a frame of T time slots, and an operating bandwidth of B Hz around a center frequency fc . To support the sensing service, a set of N sensors is deployed at fixed and known locations within the cell range, each equipped with

a ULA of AN antenna elements, collecting sensing data, and transmitting it in-band to the BS to support both slices. B. Sensing Link Budget We assume a bistatic sensing scenario in which the ISAC BS acts as the sensing transmitter, while a sensor n ∈ N acts as the sensing receiver. As such, we assume that the two paths of the cascaded sensing channel, Dv , from the BS to v ∈ V, and Dv,n , from v to n ∈ N , are both Line of Sight (LOS), following [5], [10]. Thereby, we obtain the bandwidthnormalized sensing SNR, as follows γS (v, n) =

PM GP (v)G1 (M, v)G1 (n, v)λ2 σrcs (v, n) , (4π)3 (Dv Dv,n )α N0 ϱv ϱv,n

(1)

where PM is the per-beam transmit power uniformly allocated between downlink beams, GP (v) is the sensing processing gain (defined as below), G1 (M, v) is the directional gain of the transmitter BS antenna, G1 (n, v) is the directional gain of the receiver sensor antenna, λ is the wavelength of fc , σrcs (v, n) is the Radar Cross-Section (RCS) (calculated as in [16, Table 7.9.2.1-4]), N0 is the noise power spectral density, and ϱv , ϱv,n ∼ log N (0, σF2 ) are the large-scale fading on the two paths with variance of σF2 . For simplicity, we consider G1 (M, v) and G1 (n, v) as the maximum directional gain (calculated as in [16, Table 7.3-1]). We generalize the path-loss exponent α ≥ 2 to model scenarios with additional attenuation. Finally, the processing gain GP (v) is obtained for OFDM-based estimation following [1]: GP (v) = AN NB min(lv τ, τCPI ),

(2)

BS where NB = ∆ is the number of sensing subcarriers with BS f the sensing bandwidth and ∆f the subcarrier spacing, lv is the number of beam time slots allocated to v, τ is the number of OFDM symbols in a slot, and τCPI is the number of OFDM symbols in the Coherent Processing Interval (CPI).

C. Communication Link Budget Downlink – On both downlink and uplink communication channels, we consider the path-loss following 3GPP’s Urban Microcell Street Canyon LOS model [16]. Given n ∈ N , the sensor assigned to track u ∈ U , we write the bandwidthnormalized downlink communication SNR as PM GB (u, n)GU γU (u, n) = , (3) L(Du )N0 ϱu where L(Du ) denotes path-loss at distance Du (calculated as in [16, Table 7.4.1-1]), GU is the receiver UE singleantenna isotropic gain, and ϱu ∼ log N (0, σF2 ) is the largescale fading. GB (u, n), the sensing-based beamforming gain, is defined following [17]:
( 2σB (u,n) 2 3 − 10 φA G (M, u)10 if σB (u, n) ≤ Θu 1 GB (u, n) = G2 otherwise, (4) where φA = π − 2 arccos( 1.391λ πAM d ) is the broadside Half-Power Beamwidth (HPBW) of the ISAC BS with d the antenna ele3π ment spacing [18, Table 6.2], G2 = (AM sin2 ( 2A ))−1 is the M

q G1 (M,u) ) is side lobe beamforming gain, Θu = φ2A 10 3 log10 ( G2 the mainlobe width of the beam serving u, and σB (u, n) is the beam misalignment error (as defined in Section III). Uplink – In the uplink, which the sensor uses to report the sensing measurements to the network, a sensor n ∈ N positioned at a distance Dn of the BS experiences the bandwidthnormalized SNR γN (n) =

PN G1 (n, M)G1 (M, n) , L(Dn )N0 ϱn

(5)

where PN is the sensor’s transmit power, G1 (n, M) is the directional gain of the transmitter sensor antenna, G1 (M, n) is the directional gain of the receiver BS antenna, and ϱn ∼ log N (0, σF2 ) is the large-scale fading. Since the BS and sensor positions are fixed, we assume maximum directional gain. III. ISAC S LICING F RAMEWORK In this section, we describe our ISAC slicing framework, comprising bandwidth and beam time slot allocation for SaaS and CaaS slices, target-sensor assignment, SAC and CAS interdependencies, and performance metrics. A. ISAC Slicing SaaS and CaaS – We consider the SaaS slice to support sensing-based applications, such as environment monitoring and ISAC-aided Vehicle-to-Everything (V2X), by providing a localization service that tracks detected vehicles [2]. To evaluate its performance, we choose the position estimation error metric. We assume that a single position estimate per frame is sufficient to accurately track vehicle motion. Also, we consider the impact of sensing resolution at the network level, limiting the number of resolvable vehicles. Accordingly, we describe the SaaS SLA as a max. position estimation error QS . The CaaS slice, on the other hand, provides a V2X Guaranteed Bit Rate (GBR) communication service [19, Table 5.7.4-1]. We then define its SLA as a minimum throughput per frame QC . SAC and CAS – The sensing functions of the ISAC BS can be leveraged not only to satisfy the SaaS SLA, but also to improve communication performance and CaaS SLA satisfaction through SAC. We model SAC as the opportunistic use of sensing bandwidth for a sensor n ∈ N to track a UE u ∈ U and reduce its beam misalignment σB (u, n) = (u,n) arctan( σPD ), for σP (u, n) the position estimation error. u However, UE and target position estimations can be used efficiently for SAC and SaaS only if they are continuously transmitted from sensors to the ISAC BS, i.e., if the CAS demand is satisfied. Further in this section, we model the position estimation error and the impact of unsatisfied CAS demand on tracking success. B. Resources Allocation Bandwidth Allocation – Assuming Frequency Division Duplexing (FDD) operation, the bandwidth B is divided between downlink BDL , to sense and serve vehicles, and uplink BUL , for sensors to report measurements. We further split downlink bandwidth between BC , communication bandwidth associated

with the CaaS slice, and BS , sensing bandwidth associated with the SaaS slice. The sensing bandwidth is used solely for sensing and is entirely allocated to each simultaneous beam. With frequency reuse, each UE u ∈ U is allocated a downlink communication bandwidth bu ≤ BC , while each sensor is allocated an uplink communication bandwidth bn ≤ BUL . Beam Allocation – Each sector k ∈ K can simultaneously sense/serve AM vehicles and AM sensors in multi-beam operation by distributing a budget of AM T downlink and AM T uplink beam time slots. Assuming at most one beam per device, each vehicle v ∈ V and sensor n ∈ N are allocated lv ≤ T and ln ≤ T slots, respectively. Remark 1: To manage communication interference with frequency reuse, as in [3], we assume resources are allocated orthogonally. On the downlink, if a UE u ∈ U has a set of overlapping P UEs Uu ⊆ U that might be served simultaneously (lu + i∈Uu li > T ), we consider they are assigned nonP overlapping channels (bu + i∈Uu bi ≤ BC ). Likewise, on the uplink,Pif a sensor n ∈ N has an overlapping set Nn ⊆ N P with ln + j∈Nn lj > T , we assume bn + j∈Nn bj ≤ BUL . C. Service Performance Communication Performance – The uplink data rate ψN (n) of a sensor n ∈ N and downlink data rate ψU (u, n) of a UE u ∈ U tracked by n are obtained as

ln bn log2 (1 + γN (n)), (6) T lu ψU (u, n) = bu log2 (1 + γU (u, n)). (7) T Sensing Accuracy – We obtain the accuracy of estimating the position of a vehicle v ∈ V tracked by a sensor n ∈ N as q σP (v, n) = σR (v, n)2 + (σA (v, n)Dv,n )2 , (8) ψN (n) =

where σR (v, n) and σA (v, n) denote the errors in range and angle estimation relative to n, respectively. For both, we assume efficient, unbiased estimators, and approximate their average estimation error as the square root of the Cramér-Rao Lower Bound (CRLB) [10], which can be written as c p σR (v, n) = , (9) 4πBS 2γS (v, n) √ λ 6 p , σA (v, n) = 2πd max(ϵ, | cos(θv,n )|) γS (v, n)AN (A2N − 1) (10) where c is the speed of light, θv,n is the relative angle between v and the boresight of the ULA of sensor n. Since the term cos(θv,n ) in σA (v, n) comes from CRLB’s far field assumption and does not apply as θv,n → 90º, we lower-bound it by ϵ. Sensing Capacity – Tracking is only feasible if sensors meet their CAS demand, and the sensing resolution is fine enough to distinguish each vehicle. Given I(v, n) ∈ {0, 1}, the vehiclesensor assignment, where each vehicle is assigned one sensor, we write the rate of successfully tracked vehicles as    min(|V|, Y ) X ψN (n) X ρ= min , I(v, n) , (11) |V|2 Z n∈N

v∈V

TABLE I: Simulation Parameters

2

where Y = πR ∆c approximates the maximum number of resolvable vehicles for OFDM-based sensing [1] and ∆c = λ c 2BS R AN d , is the resolution cell area [10] for R assuming the worst-case cross-range resolution. D. Slice Performance Satisfaction We evaluate the impact of resource allocation and target assignment using slice- and network-level aggregated metrics. On the SaaS slice, we use the mean position estimation error ρ XX σT = (1 − ρ)R + I(t, n)σP (t, n), (12) |T | t∈T n∈N

where R is the error experienced by unsuccessfully tracked targets. On the CaaS slice, we calculate the mean throughput (1 − ρ) X ρ X X ψU = ψU2 (u) + I(u, n)ψU (u, n), |U| |U| u∈U u∈U n∈N (13) where ψU2 (u) is the throughput achieved by u ∈ U if it is not successfully tracked, thus experiencing the side lobe beamforming gain G2 . Finally, we evaluate the ISAC slicing satisfaction at the network level by adapting the SLA Satisfaction Rate (SSR) metric from [12]: ρWS X  Q= 1 P I(t,n)σP (t,n)≤QS |T | n∈N t∈T X  ρWC 1 P + I(u,n)ψU (u,n)≥QC |U| n∈N u∈U X  (1 − ρ)WC + 1 , (14) ψU2 (u)≥QC |U| u∈U

where WS and WC are normalized weights defined by the network operator to prioritize slices. The first term is the weighted SaaS SSR, while the second and third are the weighted CaaS SSR for tracked and untracked UEs, respectively. IV. S IMULATED E NVIRONMENT In this section, we detail the parameters and allocation strategies implemented in our ISAC slicing simulation1 . Table I summarizes simulation parameters, which we adopt from the mmWave OFDM-based ISAC work [1] and the 3GPP UMi model [16]. A. Network Configuration We consider a single cell with the ISAC BS at the origin. Sensors, targets, and UEs are independently distributed within the cell area according to a homogeneous Poisson Point Process of arbitrary densities, while sensor orientations are uniform at random. The ISAC BS is equipped with 3 sectorized ULAs transmitting at 35 dBm, following [16], which we uniformly distribute across 16 downlink beams as 23 dBm/beam. We consider a single isotropic antenna UE, and a ULA sensor. Sensors report back their collected data once per OFDM frame, in a message we assume is similar in size to a Collection Perception Message [14]. We define two 1 github.com/DanteCampos/2026-ISAC-DJ

Parameter

Symbol

Value

Cell Radius Number of Sectors Carrier Frequency Wavelength Effective System Bandwidth Subcarrier Spacing Slots per frame OFDM Symbols per Slot OFDM Symbols in the CPI ULA Antenna Elements ULA Antenna Spacing UE Single Isotropic Antenna Gain Sensing Path Loss Exponent BS Transmit Power per Beam Sensor Transmit Power Noise Power Spectral Density Large-Scale Fading Variance Sensing Data Generation per Vehicle Cosine Lower Bound Correction Slice Weights SaaS Max Pos. Est. Error Req. CaaS Min Throughput Req.

R |K| fc λ B ∆f T τ τCPI AM , AN d GU α PM PN N0 2 σF Z ϵ WC , WS QS QC

400 m 3 [16] 28 GHz [1] c/fc =10.71 mm [1] 60 MHz [1] 120 kHz [1] 80 [20] 14 [20] 512 [1] 16 [1] λ/2=5.35 mm [1] 0 dBi [17] 2.4 23 dBm [16] 23 dBm -174+9 dBm/Hz [16] 4 dB [16] 29.72 kbps [14] cos(75º) 0.5 0.1 m [8] 24 Mbps [21]

slices with equal weights: one SaaS slice provides sensing to Urban Environment Reconstruction [8], while one CaaS slice provides communication to V2X applications [21]. B. Resource Allocation For each slice, we design resource allocation to guarantee orthogonality in the time or frequency domains. To isolate the impact of resource allocation, we consider upper-bound sensor assignments with minimum position estimation error via exhaustive best-selection. We leave the study of practical assignment algorithms for future work. On the downlink, we allocate beam time slots to each vehicle v ∈ V in a Round-Robin manner, limiting lv ≤ τCPI /τ if v ∈ / U to allocate the remaining slots to UEs. Since sensing and communication operate on separate bandwidths, we allocate the entire communication bandwidth to UEs: bu ← BC ∀u ∈ U. We then identify the set Uu for each UE u ∈ U, containing u and all UEs whose Pbeams it overlaps with. If Uu overlaps in the time domain ( j∈Uu lj > T ), we enforce frequency-domain orthogonality by updating: bi ← BC min(bi , |U ) ∀i ∈ Uu . u| In the uplink, we do the opposite. Due to the maximum directional gains between the fixed sensors and the ISAC BS, and to the large subcarrier spacing, we are able to satisfy the CAS demand of all sensors by reusing 1 Physical Resource Block (PRB). Therefore, we fix this allocation to dedicate the remaining PRBs to downlink communication and sensing, while enforcing orthogonality across beam time slots. Sensors start with ln ← T ∀n ∈ N , then we check thePset of sensors Nn with n and the beam it overlaps with. If j∈Nn lj > T , we update lj ← min(lj , |NTn | ) ∀j ∈ Nn . V. E VALUATION

In this section, we evaluate the performance of ISAC slicing across different downlink sensing-communication bandwidth

150T,50U/km2 300T,50U/km2

Network SSR (%)

splits, sensor deployments, and demand scenarios, defined by target and UE densities. Without loss of generality, we consider T ∩ U = ∅. Since targets have a lower impact on beam slot allocation due to the τCPI bound, we consider target densities higher than UE densities. For short, we refer to the bandwidth split as the Normalized Sensing Bandwidth (NSB) S , with BDL = BS + BC , and abbreviate targets, UEs, = BBDL and sensors as T, U, and S, respectively. Results are averaged over 1000 independent realizations for statistical significance. A. ISAC Slicing Satisfaction

150T,150U/km2 300T,150U/km2

100

100

80

80

60

60

40

40

20

20

better

0

Peak SSR NSB at Peak SSR

better

0 0

25

50

75

100

0

25

50

75

100

Fig. 2 displays the network SSR as a function of NSB under different sensor deployments and demand scenarios. We observe the SaaS-CaaS trade-off: at 100% NSB, no CaaS SLA is satisfied, whereas at 0% NSB, no SaaS SLA is satisfied. The NSB at the highest network SSR is therefore a compromise between the two slices. The best-performing bandwidth split is primarily determined by demand, particularly CaaS demand, since we do not constrain UE beam time slot allocation. Consequently, increasing demand shifts the best-performing NSB towards a larger share of the bandwidth for communication. Increasing the sensor density has little effect on the trade-off itself, scaling the achievable SSR without significantly altering the curve’s shape. This behavior is illustrated in Fig. 2a for the deployment with 5 sensors/km2 . The low-demand scenario (150T, 50U/km2 ) achieves a peak network SSR of 79% at 27% NSB and remains within 2 percent points of it over the NSB interval [10%, 54%]. In contrast, the high-demand scenario (300T, 150U/km2 ) reaches a lower peak SSR of 47% at 15% NSB, with a narrower interval of [7%, 20%]. This result indicates that ISAC slicing becomes increasingly sensitive to bandwidth allocation as demand increases. Furthermore, CaaS demand has a stronger impact than SaaS demand. Increasing only the UE density (150T, 150U/km2 ) reduces the peak SSR by 18 points, whereas increasing only the target density (300T, 50U/km2 ) reduces it by 4 points. Increasing the sensor density to 15 sensors/km2 , shown in Fig. 2b, provides an almost uniform increase of 12 points across all demand scenarios. Overall, bandwidth allocations that satisfy the SaaSCaaS trade-off are constrained by CaaS demand, whereas achievable network performance scales with sensor density. To better understand these effects, we next analyze the CaaS and SaaS slices individually.

In the low-demand scenario (150T, 50U/km2 ), the mean throughput peaks at 152 Mbps with 5% NSB. In the highdemand scenario (300T, 150U/km2 ), this peak decreases to 29 Mbps and requires a higher sensing allocation of 7% NSB to resolve all vehicles. Consistent with the network SSR results, CaaS demand dominates the achievable performance. Increasing only the UE density (150T, 150U/km2 ) reduces throughput by approximately 73 % relative to the low-demand scenario. In contrast, increasing only the target density (300T, 50U/km2 ) reduces throughput by approximately 39 %.

B. CaaS Performance and SAC Trade-Off

C. SaaS Performance

Fig. 3 evaluates the CaaS performance. We observe the SAC trade-off: at 100% NSB, no bandwidth remains for communication, whereas at 0% NSB, throughput is degraded by beam misalignment. Peak performance occurs at the lowest NSB with enough resolution to successfully track all vehicles. Beyond this point, the SNR gain from improved beamforming is outweighed by the loss of communication bandwidth. Since beamforming requires substantially lower positioning accuracy than the SaaS slice, communication performance is not limited by sensing accuracy, but by resolution and demand. As a result, CaaS performance is largely insensitive to sensor density.

Fig. 4 depicts the SaaS performance. We observe that sensing performance is first limited by resolution, then by sensor density. Once all vehicles are resolved, increasing the NSB yields only marginal improvements because the additional receiver noise offsets the processing gain. In contrast, sensor density has a substantial impact on sensing performance. Denser deployments reduce the target-sensor distance, improving the link budget and decreasing the cross-range component of the position error, as shown in Eq. (8). Finally, SaaS is less sensitive to demand than CaaS, since targets require fewer beam slots to achieve the CPI.

Norm. Sensing Band. (%) (a) 5 sensors/km2

Norm. Sensing Band. (%) (b) 15 sensors/km2

Fig. 2: ISAC Slicing performance at network-level.

CaaS: Mean Thr. (Mbps)

150T,50U/km² 300T,50U/km²

150T,150U/km² 300T,150U/km²

15S/km2 5S/km2

Peak Thr.

150 better

100 50 0 0

20

40

60

80

100

Normalized Sensing Bandwidth (%)

Fig. 3: CaaS performance and SAC trade-off.

SaaS: Mean Pos. Error (m)

150T,50U/km2 300T,150U/km2

15S/km2 5S/km2

All Vehicles Resolved

(No. 2554414) grants, and from the Commonwealth Cyber Initiative (CCI) in Virginia, US, an investment in advancing cyber R&D, innovation, and workforce development. R EFERENCES

102 better 101 100 10−1 0

20

40

60

80

100

Normalized Sensing Bandwidth (%)

Fig. 4: SaaS performance of tracking SaaS targets. These effects are reflected in the position estimation error. With sparse deployments (5S/km2 ), the mean error at 100% NSB reduces only slightly from 0.81 m in the highdemand scenario (300T, 150U/km2 ) to 0.62 m in the lowdemand scenario (150T, 50U/km2 ). In contrast, increasing the sensor density to 15S/km2 significantly reduces the error, 0.11 m and 0.08 m in the high- and low-demand scenarios, respectively. Therefore, unlike the CaaS slice, SaaS performance tolerates demand increases and is primarily determined by sensor density once sensing bandwidth is sufficient to resolve all vehicles. Combined with the communication performance results, this explains the behavior of the network SSR. Both slices are fundamentally constrained by resolution, while the bestperforming bandwidth split is dominated by CaaS demand. Consequently, as demand increases, the peak network performance tends to be achieved by the NSB with the smallest sensing bandwidth share that still resolves all vehicles. VI. C ONCLUSIONS In this work, we proposed a model for evaluating the sensing, communication, and slicing performance of an ISAC cell with bistatic sensing via a network of sensors. Our results reveal a hierarchy of design decisions for ISAC slicing networks. First, sufficient sensing bandwidth is allocated to resolve all vehicles, as sensing resolution is the primary bottleneck for both SaaS and CaaS slices. Second, the sensor deployment is dimensioned to meet the SaaS SLA requirements, as this is the main factor determining sensing accuracy. Finally, the sensingcommunication bandwidth split that defines the SaaS-CaaS trade-off is considered in relation to the demand scenario, with a primary focus on how CaaS SLA satisfaction degrades. Future work will focus on designing resource-allocation algorithms to dynamically resolve the observed trade-offs and ensure SLA satisfaction for both SaaS and CaaS slices. ACKNOWLEDGMENTS This work received support from the National Science Foundation under IUCRC WISPER (No. 2412872), and Miageru

[1] Z. Xiao, R. Liu, M. Li, Q. Liu, and A. L. Swindlehurst, “A Novel Joint Angle-Range-Velocity Estimation Method for MIMO-OFDM ISAC Systems,” IEEE Transactions on Signal Processing, vol. 72, pp. 3805–3818, 2024. [2] F. Liu, Y. Cui, C. Masouros, J. Xu, T. X. Han, Y. C. Eldar, and S. Buzzi, “Integrated Sensing and Communications: Toward DualFunctional Wireless Networks for 6G and Beyond,” IEEE Journal on Selected Areas in Communications, vol. 40, no. 6, pp. 1728–1767, 2022. [3] F. Dong, F. Liu, Y. Cui, W. Wang, K. Han, and Z. Wang, “Sensing as a Service in 6G Perceptive Networks: A Unified Framework for ISAC Resource Allocation,” IEEE Transactions on Wireless Communications, vol. 22, no. 5, pp. 3522–3536, 2022. [4] M. Shafi, E. G. Larsson, X. Lin, D. Panaitopol, S. Parkvall, F. RonteixJacquet, and A. Toskala, “Industrial Viewpoints on RAN Technologies for 6G,” arXiv preprint arXiv:2508.08225, 2025. [5] R. Li, Q. Zhang, D. Ma, K. Yu, and Y. Huang, “Joint Target Assignment and Resource Allocation for Multi-Base Station Cooperative ISAC in AAV Detection,” IEEE Transactions on Vehicular Technology, vol. 74, no. 5, pp. 7700–7714, 2025. [6] R. Li, Z. Xiao, and Y. Zeng, “Toward Seamless Sensing Coverage for Cellular Multi-Static Integrated Sensing and Communication,” IEEE Transactions on Wireless Communications, vol. 23, no. 6, pp. 5363– 5376, 2023. [7] S. Zhou, H. Yang, L. Xiang, and K. Yang, “Temporal-Assisted Beamforming and Trajectory Prediction in Sensing-Enabled UAV Communications,” IEEE Transactions on Communications, vol. 73, no. 7, pp. 5408–5419, 2024. [8] A. Bayesteh, J. He, Y. Chen, P. Zhu, J. Ma, A. W. Shaban, Z. Yu et al., “Integrated Sensing and Communication (ISAC) —- From Concept to Practice,” Tech. Rep., 2022. [9] I. Chih-Lin, Z. Han, R. Xi, T. Wang, X. Zhang, T. Jiang, S. Wang et al., “Cooperative ISAC: An End-to-End Perspective,” IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation, 2025. [10] M. Richards, Fundamentals of Radar Signal Processing, Third Edition, 3rd ed. Columbus, OH: McGraw-Hill Education, Apr. 2022. [11] G. Fatima, P. Stoica, A. Aubry, A. De Maio, and P. Babu, “Optimal Placement of the Receivers for Multistatic Target Localization,” IEEE Transactions on Radar Systems, vol. 2, pp. 391–403, 2024. [12] Y. Guan, Q. Song, T. Chen, W. Qi, L. Guo, and A. Jamalipour, “SlicingAware Aerial Networks for Integrated Sensing and Communication: 3D Placement and Adaptive Allocation of Resources,” IEEE Vehicular Technology Magazine, vol. 19, no. 2, pp. 79–88, 2024. [13] M. A. Hossain, A. Xiang, A. Kiani, T. Saboorian, J. Kaippallimalil, and N. Ansari, “AI-Assisted E2E Network Slicing for Integrated Sensing and Communication in 6G Networks,” IEEE Internet of Things Journal, vol. 11, no. 6, pp. 10 627–10 634, 2023. [14] A. Figueiredo, J. Amaral, P. Rito, M. Luı́s, and S. Sargento, “Improving Object Selection for Collective Perception Messages Under Congestion,” Ad Hoc Networks, vol. 185, p. 104175, 2026. [15] Z. Wei, Y. Wang, L. Ma, S. Yang, Z. Feng, C. Pan, Q. Zhang et al., “5G PRS-Based Sensing: A Sensing Reference Signal Approach for Joint Sensing and Communication System,” IEEE Transactions on Vehicular Technology, vol. 72, no. 3, pp. 3250–3263, 2022. [16] 3GPP, “5G; Study on Channel Model for Frequencies from 0.5 to 100 GHz (3GPP TR 38.901 v19.3.0),” 3GPP, Tech. Rep., 2026. [17] M. S. Zia, D. M. Blough, and M. A. Weitnauer, “Effects of SNRDependent Beam Alignment Errors on Millimeter-Wave Cellular Networks,” IEEE Transactions on Vehicular Technology, vol. 71, no. 5, pp. 5216–5230, 2022. [18] C. A. Balanis, Antenna Theory: Analysis and Design. John Wiley & Sons, 2016. [19] 3GPP, “5G; System architecture for the 5G System (5GS) (3GPP TS 23.501 v19.7.0),” 3GPP, Tech. Rep., 2026. [20] ——, “5G; NR; Physical Channels and Modulation (3GPP TS 38.211 v19.3.0),” 3GPP, Tech. Rep., 2026. [21] X. Wang, S. Mao, and M. X. Gong, “An Overview of 3GPP Cellular Vehicle-to-Everything Standards,” GetMobile: Mobile Computing and Communications, vol. 21, no. 3, pp. 19–25, 2017.

Related documents

• Performance Analysis of ISAC Slicing with Sensing-Assisted Communication and Communication-Assisted Sensing#1199153
• SESO-ISAC: Service-Aware End-to-End Sensing Orchestration for 6G ISAC#667962
• Integrating Wi-Fi into 3GPP 5G Network Slicing: An Experimental Prototype Study#924775
• Integrating Wi-Fi into 3GPP 5G Network Slicing: An Experimental Prototype Study#920541
• AFDM-Enabled ISAC in Dynamic Environments: Fundamentals, Technologies and Opportunities#971537