betaFeedback

WorksEngineering

Ghost Target Elimination in 5G Reference Signals ISAC via Multi-static Multilateration

Abstract

Integrated Sensing and Communication (ISAC) systems can reuse 5G New Radio (NR) reference signals to enable joint communication and environmental sensing. Among these signals, the Positioning Reference Signal (PRS) is particularly suitable for range and Doppler estimation. However, its comb structure introduces periodic ambiguities that generate ghost targets. Existing solutions resolve these ambiguities by combining PRS with additional signals such as the DeModulation Reference Signal (DMRS), which increases sensing overhead. This paper proposes a PRS-only sensing framework that resolves ambiguities using multi-static observations. Instead of removing ghosts locally at each receiver, the proposed method exploits spatial diversity and performs ambiguity resolution at a fusion anchor. The anchor jointly estimates target locations and ambiguity indices via multilateration across multiple sensing nodes. This approach eliminates the need for additional reference signals while maintaining low receiver complexity. Simulation results show that the proposed method reliably resolves ghost targets and achieves superior detection performance compared to benchmark methods across various scenarios, while significantly enhancing resource efficiency by enabling sensing using PRS-only transmissions.

Introduction

Future 6G wireless networks are expected to extend the role of communication systems beyond data delivery by enabling environmental sensing capabilities within the same infrastructure. This vision is embodied in the Integrated Sensing and Communication (ISAC) paradigm, which supports applications such as object tracking, autonomous driving, environmental monitoring, and smart city services [1, 2]. In this context, the reference signals already defined in 5G New Radio (NR) have attracted considerable attention for sensing purposes. Their well-designed structure, favorable correlation properties, and robustness to noise make them appealing for radar-like sensing tasks, while allowing seamless integration with communication functionalities [3, 4].

Several studies have investigated the use of 5G reference signals for sensing [4]. For instance, [3] compares the sensing performance of Positioning Reference Signals (PRS) with other pilots such as DeModulation Reference Signals (DMRS), Channel State Information Reference Signals (CSI-RS), and Synchronization Signals (SS). Among these, PRS, introduced in 3GPP Release 16, has emerged as a strong candidate due to its flexible configuration, high processing gain, and large time–frequency occupancy, which enable accurate time-of-arrival estimation and extended sensing range [3]. Despite these advantages, PRS-based sensing suffers from an inherent limitation caused by its comb-like frequency-domain allocation. The regular spacing between active subcarriers leads to periodic ambiguities in the range domain, resulting in multiple ghost targets [5]. These ambiguities significantly degrade detection reliability, particularly in cluttered or multi-target environments, and represent a major obstacle to PRS-only sensing.

Various approaches have been proposed to address this ambiguity problem. A prominent line of work resolves PRS ambiguity by jointly exploiting multiple reference signals [6]. In particular, 3GPP-compliant solutions combining PRS with DMRS have been proposed to eliminate range or Doppler ambiguities [5, 7]. Other methods attempt to mitigate ambiguities through irregular resource allocations, such as the irregular time-frequency patterns proposed in [8]. Despite their effectiveness, these approaches present several limitations. Leveraging multiple reference signals increases system complexity and sensing overhead, as additional resources must be allocated for sensing tasks. Irregular resource allocation strategies also deviate from standard signal structures and can affect correlation properties, potentially increasing sidelobe levels [9]. More broadly, allocating additional resources to sensing directly impacts the communication capacity, making resource-efficient sensing strategies challenging [10, 11, 12].

In this paper, we propose a PRS-only sensing framework that resolves range ambiguities without requiring additional reference signals or non-standard resource allocations. The key idea is to exploit spatial diversity through a multi-static network of sensing nodes and to perform ambiguity resolution centrally rather than locally. Instead of attempting to eliminate ghost targets independently at each receiver, the proposed approach jointly processes multi-static observations at a fusion anchor, which estimates both the true target locations and the associated ambiguity indices via multilateration. By leveraging geometric consistency across spatially separated sensing nodes, the method effectively suppresses ghost targets while maintaining low receiver complexity. Simulation results demonstrate that the proposed framework reliably resolves PRS-induced ambiguities and significantly outperforms benchmark methods in terms of detection accuracy across a wide range of scenarios. Importantly, this is achieved without increasing sensing overhead, thereby enabling efficient and fully standards-compliant PRS-based ISAC operation.

The rest of the paper is organized as follows: § 2 introduces the system model, § 3 presents our proposal, § 4 provides the numerical results, and § 5 concludes the paper.

Multi-Static ISAC System Model

We consider the multi-static scenario illustrated in Figure 1, where a 5G gNB acts as the transmitter while QQ spatially distributed receivers observe the reflections from the sensing scene with q∈{1,…,Q}q \in\{1,\ldots,Q\} representing the receiver index. The receivers operate in a passive sensing mode and estimate bistatic ranges using the echo received by the transmitted signal. The sensing region contains KK point-like targets present in the line-of-sight of both the transmitter and the receivers.

Multi-Static ISAC system model

Figure 1. Multi-Static ISAC system model.

OFDM Radar Signal Model

The gNB transmits an OFDM waveform composed of NN symbols in time and MM subcarriers in frequency. The transmitted continuous-time signal can be written as [5]:

s(t)=∑n=0N−1rect⁡(t−nT0T0)∑m=0M−1d(m,n)ej2πmΔf(t−nT0),(1)s(t)=\sum_{n=0}^{N-1}\operatorname{rect}\left(\frac{t-nT_{0}}{T_{0}}\right)\sum_{m=0}^{M-1}d(m,n)e^{j2\pi m\Delta f(t-nT_{0})}, \tag*{(1)}

where Δf\Delta f denotes the SubCarrier Spacing (SCS), rect⁡(⋅)\operatorname{rect}(\cdot) denotes the rectangular pulse, T0=TCP+TsT_{0}=T_{\mathrm{CP}}+T_{s} is the OFDM symbol duration with Ts=1/ΔfT_{s}=1/\Delta f and TCPT_{\mathrm{CP}} the cyclic prefix length, and d(m,n)d(m,n) represents the symbol transmitted on subcarrier mm of OFDM symbol nn. The signal received at receiver qq can be modeled as [13]:

rq(t)=∑k=1Kβk,qs(t−τk,q)ej2πfd,k,qt+uq(t),(2)r_q(t)=\sum_{k=1}^{K}\beta_{k,q}s(t-\tau_{k,q})e^{j2\pi f_{d,k,q}t}+u_q(t), \tag*{(2)}

where βk,q\beta_{k,q} denotes the reflection coefficient, τk,q\tau_{k,q} is the bistatic propagation delay associated with the bistatic path, fd,k,qf_{d,k,q} is the Doppler shift, and uq(t)u_q(t) represents additive white Gaussian noise with zero mean and variance 2σ22\sigma^{2}. After sampling and applying the Fast Fourier Transform (FFT) at the receiver, the received sample on subcarrier mm and symbol nn is written as:

yq(m,n)=∑k=1Kβk,qej2πnT0fd,k,qe−j2πmΔfτk,qv(m,n)+ISIq+zq(m,n),(3)y_q(m,n)=\sum_{k=1}^{K}\beta_{k,q}e^{j2\pi nT_{0}f_{d,k,q}}e^{-j2\pi m\Delta f\tau_{k,q}}v(m,n)+\mathrm{ISI}_q+z_q(m,n), \tag*{(3)}

where zq(m,n)z_q(m,n) denotes the complex AWGN component with zero mean and variance 2σ22\sigma^{2}. As conventional OFDM radar processing, uncompensated delays or Doppler shifts may introduce Inter-Symbol Interference (ISI) under challenging sensing conditions.

Bistatic Range Estimation

Each receiver estimates the bistatic ranges of potential targets using a periodogram-based approach due to its low computational complexity. The received samples yqy_q are first divided element-wise by the transmitted symbols dd to obtain a normalized matrix gqg_q. Range estimation is performed by applying an MM-point Inverse FFT (IFFT) along the frequency dimension, yielding sn,q(l)s_{n,q}(l) for each OFDM symbol. Averaging across the NN symbols improves the signal-to-noise ratio, producing sˉq(l)≜1N∑n=0N−1sn(l)\bar{s}_q(l) \triangleq\frac{1}{N}\sum_{n=0}^{N-1}s_n(l). The peak index l^k,q\hat{l}_{k,q} corresponds to the estimated bistatic range B^k,q=l^k,qΔfM\hat{B}_{k,q}=\frac{\hat{l}_{k,q}}{\Delta f M}. Similarly, Doppler estimation can be obtained after delay compensation by applying an FFT along the time dimension [13]. However, in this work, the primary measurements used for localization are the bistatic ranges estimated at each receiver. These measurements are then forwarded to the fusion anchor, where spatial processing is used to resolve ambiguities and estimate the target positions. For brevity, detailed derivations are provided in [5].

A PRS-Only Ambiguity Elimination Proposal

This section addresses the ghost target problem inherent to PRS-based sensing and introduces a PRS-only ambiguity elimination framework suitable for multi-static ISAC deployments. We first briefly review a benchmark solution that combines PRS and DMRS to suppress ambiguities. We then present the proposed multilateration-based approach, which exploits spatial diversity to eliminate ghost targets without relying on additional reference signals.

Ghost Targets and How to Use DMRS to Remove Them

When used for passive target sensing, the comb-structured allocation of the PRS in 5G NR introduces empty subcarriers that cause periodic ambiguities (replicas of the real targets) in the range and Doppler domains. The work of [5] proposes removing these ghosts by jointly exploiting the PRS and the DMRS embedded in the Physical Downlink Shared Channel (PDSCH), which is repurposed for sensing, as presented in Figure 2. Since PRS and DMRS exhibit different frequency-domain structures, their corresponding ambiguity patterns differ. By appropriately combining their sensing outputs, ghost targets can be suppressed while preserving the true target response. Different fusion strategies can be applied such as for range ambiguities [5]:

PRS–PDSCH resource allocation: 2 slots, 2 PRBs, comb size 12

Figure 2. PRS–PDSCH resource allocation: 2 slots, 2 PRBs, comb size 12 [5].

  • For smaller comb sizes (2, 4), the range profiles from the two signals are combined using a Hadamard product since the ghost in one profile typically corresponds to a low value in the other.

  • For larger comb sizes (6, 12), the method sums the time-domain representations of PRS and DMRS, thereby disturbing the periodicity of the ghost pattern and reducing the magnitude of ghost peaks below detection thresholds.

This solution maintains range resolution and maximum detection range and the two cases (PRS-only and PRS+DMRS) are illustrated in Figure 4 and will be discussed in § 4. Despite its effectiveness, the joint PRS–DMRS approach increases sensing overhead and receiver complexity, as it requires allocating additional reference signals and performing multiple sensing chains. These limitations motivate the PRS-only solution proposed next.

Figure 4. Range estimation with PRS only (top) and PRS + DMRS (bottom) for a two-target scenario (Kcomb=4K_{\mathrm{comb}}=4, fc=28f_c=28GHz and SCS = 120kHz).

Ghost Elimination via Multilateration

We propose a PRS-only ambiguity elimination strategy that leverages spatial diversity in a multi-static sensing configuration. Instead of attempting to remove ghost targets locally at each receiver, the proposed method resolves ambiguities centrally at a fusion anchor by jointly processing multi-static bistatic range measurements.

Due to the comb-based structure of PRS, ghost targets appear periodically in the bistatic range domain (e.g., Figure 4 top plot). For a given comb size KcombK_{\mathrm{comb}}, the spacing between successive ghost replicas is given by [8]:

D=cKcombΔf,(4)D = \frac{c}{K_{\mathrm{comb}}\Delta f}, \tag*{(4)}

where cc denotes the speed of light and Δf\Delta f is the subcarrier spacing. As a result, each receiver observes multiple candidate bistatic ranges corresponding to the true target and its ghost replicas.

In practical ISAC systems, the ultimate objective is not the bistatic range itself but the estimation of the target position. This motivates a shift from local ambiguity resolution to position-domain processing. In the proposed framework, bistatic range estimates from multiple spatially distributed receivers are forwarded to a fusion anchor, which performs multilateration to estimate the target location [14]. While ghost targets act as ambiguities at individual receivers, they do not exhibit geometric consistency across multiple sensing nodes. By extending the multilateration formulation to account for ambiguity indices associated with each bistatic measurement, the fusion anchor can jointly estimate the true target position and discard geometrically inconsistent ghost hypotheses. This enables reliable ghost suppression without modifying the PRS waveform or allocating additional reference signals.

The proposed approach is particularly well suited for scenarios in which new targets appear sequentially, allowing ambiguity resolution to be performed at each update. In the following subsection, we extend this principle to more general scenarios using a successive estimation and matching strategy.

Multi-static Multilateration with Ambiguity

We consider a multi-static sensing configuration composed of one transmitter and QQ spatially distributed receivers observing a common target. Each receiver estimates a bistatic range based on PRS processing. Due to the comb-based structure of PRS, these bistatic range estimates are affected by periodic ambiguities, such that multiple candidate ranges are observed for a single physical target.

Let BqB_q denote the bistatic range between the transmitter, the target, and receiver qq. At a given receiver qq, and due to the presence of ghost targets, BqB_q can be expressed as

Bq=bq+nq⋅D,(5)B_q = b_q + n_q \cdot D, \tag*{(5)}

where nq∈Nn_q \in\mathbb{N} is the ambiguity index associated with receiver qq, DD is the ambiguity period defined in (4) and bqb_q is the minimum peak value (whether real target or ghost target). Each receiver therefore provides a set of candidate bistatic ranges corresponding to different values of nqn_q, which can be interpreted as a family of ellipses in the spatial domain. For receiver qq, the foci of these ellipses are given by the transmitter and the receiver locations, while the bistatic range BqB_q determines the length of the corresponding major axis. Rather than resolving these ambiguities independently at each receiver, the proposed approach exploits spatial diversity by performing joint processing at a fusion anchor. The key observation is that while ghost targets are locally indistinguishable from true targets in the range domain, they do not satisfy the geometric consistency required across multiple sensing nodes. This motivates an ambiguity-aware multilateration formulation, whose objective is to jointly estimate the target position defined by p∈R2\mathbf{p} \in\mathbb{R}^{2} (or R3\mathbb{R}^{3}), and the ambiguity indices nqn_q using measurements from multiple receivers. For a given tuple n=(n1,…,nq)\mathbf{n} = (n_1, \ldots, n_q), our proposed multilateration with ghosts can be formulated as:

min⁡p,n∑q=1Q(∥p−pTx∥+∥p−pRxq∥−Bq)2.(6)\min_{\mathbf{p},\mathbf{n}} \sum_{q=1}^{Q} \left(\lVert\mathbf{p}-\mathbf{p}_{\mathrm{Tx}}\rVert+\lVert\mathbf{p}-\mathbf{p}_{\mathrm{Rx}_q}\rVert-B_q\right)^2 . \tag*{(6)}

where pTx\mathbf{p}_{\mathrm{Tx}} and pRxq\mathbf{p}_{\mathrm{Rx}_q} denote the known positions of the transmitter (Tx) and receiver (Rx) qq, respectively.

By leveraging multiple independent measurements, the correct combination of ambiguity indices leads to a consistent geometric intersection. In the absence of ambiguities, i.e., when nq=0n_q = 0 for all qq, the problem reduces to conventional multi-static multilateration, as seen in the right plot of Figure 3.

Block diagram of the proposal and a trilateration illustration

Figure 3. The Block diagram of the proposal and a trilateration illustration.

In contrast, when ambiguities are present, each receiver gives rise to a family of candidate ellipses rather than a single one. In this case, the integer-valued ambiguity indices nqn_q must be jointly estimated together with the target position in order to identify the geometrically consistent ellipse corresponding to the true target.

Except in edge cases, the proposed method reliably distinguishes true targets from ghost targets. However, it increases the computational complexity of the multilateration step, which now scales with the total number of integer combinations across all possible ellipses, given by ∏q=1QNq\prod_{q=1}^{Q} N_q, where NqN_q is the maximum index of nqn_q for receiver qq. Fortunately, in practice, NqN_q is naturally bounded by the size of the region of interest, making the approach computationally feasible. Nevertheless, a key limitation arises in multi-target scenarios, where correctly associating measurements across different receivers becomes challenging. To address both the association problem and the overall computational cost, we propose a hybrid strategy: multilateration with ghost elimination is applied only to newly detected targets, while the successive estimation method is used for already tracked targets, as detailed in the following section.

Successive Estimation

While ambiguity-aware multilateration enables reliable ghost elimination, its computational complexity increases with the number of ambiguity hypotheses and becomes more challenging in multi-target scenarios due to data association across receivers. To address these limitations, we propose a successive estimation strategy that exploits temporal consistency between consecutive sensing updates. The key idea is to apply the ambiguity-aware multilateration procedure only when a new target is detected. Once the position of a target has been estimated and validated, its subsequent updates are tracked using a successive estimation approach that relies on the previously estimated position as a prior. This prior information significantly constrains the set of admissible ambiguity indices, thereby reducing both the search space and the risk of incorrect data association.

The successive estimation process consists of two main components:

  • Peak Clustering: Upon receiving a new echo, each receiver detects all peaks in the bistatic range profile. These peaks are then clustered based on the predicted ghost spacing DD, such that all peaks separated by integer multiples of DD are grouped into the same cluster. Each cluster therefore corresponds to a set of candidate measurements associated with a single physical target (e.g., two such clusters are illustrated in Figure 4).

  • Matching Process: For each target stored in the historical database, the closest detected peak within the corresponding cluster is selected as the true measurement, while the remaining peaks in that cluster are classified as ghost targets. This matching step also enables the identification of targets that have exited the sensing region, as well as newly appearing targets, which are subsequently processed using the ambiguity-aware multilateration procedure.

The peak clustering procedure is simple and computationally efficient. First, all peaks are detected from the bistatic range profile and grouped according to their separation by integer multiples of the ambiguity period DD. Specifically, clustering starts from the minimum detected peak value, and all other peaks whose separation from it is an integer multiple of DD are assigned to the same cluster. This procedure is repeated iteratively for the remaining unassigned peaks until all detected peaks are clustered. Analytically, consider a set of detected peaks with bistatic range estimates {Bi}i=1Npeaks\{B_i\}_{i=1}^{N_{\mathrm{peaks}}}. The first cluster contains any peak j∈{1,…,Npeaks}j \in\{1,\ldots,N_{\mathrm{peaks}}\} such that the offset ratio given by

OffsetRatioj=Bj−min⁡(Bi)D,i∈{1,…,Npeaks}(7)\mathrm{OffsetRatio}_j = \frac{B_j-\min(B_i)}{D}, \qquad i \in\{1,\ldots,N_{\mathrm{peaks}}\} \tag*{(7)}

is sufficiently close to an integer, within a predefined threshold ϵ\epsilon. After forming the first cluster, its elements are removed from the set {Bi}\{B_i\}, and the same procedure is applied recursively to construct the remaining clusters. Once clustering is completed, the matching process between the detected clusters and previously stored (historical) target estimates is performed. For each historical bistatic range estimate Bhistory,kB_{\mathrm{history},k}, with k∈{1,…,Nhistory}k \in\{1,\ldots,N_{\mathrm{history}}\}, the cluster containing the closest detected peak is selected. The peak corresponding to this minimum distance is identified as the true target measurement, while the remaining peaks within the same cluster are classified as ghost targets and discarded. This matching process allows ghost elimination for previously detected targets without invoking ambiguity-aware multilateration.

Formally, a historical estimate Bhistory,kB_{\mathrm{history},k} is matched to the detected peak BiB_i, where

i^=arg⁡min⁡i∈{1,…,Npeaks}∣Bhistory,k−Bi∣,∀k∈{1,…,Nhistory}.(8)\hat{i}=\underset{i\in\{1,\ldots,N_{\mathrm{peaks}}\}}{\arg\min}\left|B_{\mathrm{history},k}-B_i\right|,\quad\forall k\in\{1,\ldots,N_{\mathrm{history}}\}. \tag*{(8)}

If no detected cluster lies within a predefined tolerance of a given historical estimate—where the tolerance may be chosen based on the maximum expected target velocity—the corresponding target is removed from the history, indicating that it has exited the sensing region. Conversely, any detected cluster that is not associated with a historical estimate is declared as a new target and is processed using the ambiguity-aware multilateration method described in Subsection III-B.

The left-hand side of Figure 3 illustrates the proposed processing flow alongside the State-of-the-Art (SotA) benchmark. In the benchmark approach, multilateration is directly applied to all detected peaks. In contrast, the proposed method first performs peak clustering and matching, followed by history updating. Standard multilateration without ambiguities is then applied to previously detected targets, while ambiguity-aware multilateration is invoked only for newly detected targets. The complete processing chain is summarized in Algorithm 1.

Sufficient Conditions of the Matching Process

We now examine the practical conditions under which the proposed matching strategy reliably identifies real targets. For a given target previously estimated, the matching process may result in an error if a target moves significantly between two sensing intervals. In such cases, the matching algorithm may associate the target with a previous position rather than its new shifted location. The

Algorithm 1 The Multi-static PRS-Only Sensing Algorithm

  1. QQ receivers perform sensing using PRS only and share detected peaks with the fusion anchor.

  2. Peaks Clustering: The fusion anchor clusters the peaks following Eq. (7).

  3. Matching Process: The fusion anchor matches the historical values with the targets’ clusters following Eq. (8).

  4. if (There is a new target) then Perform multilateration with ghosts to jointly estimate p\mathbf{p} and n\mathbf{n} for the new target following Eq. (6).

  5. Using n\mathbf{n}, update the history values for the QQ receivers.

  6. end if

  7. if (There is one or more old targets) then Perform standard multilateration (i.e., Eq. (6) for nq=0n_q=0) to estimate p\mathbf{p} using the validated bistatic distances.

  8. end if

likelihood of such errors depends the distance between ghost peaks DD, the velocity of the target vv, the direction of movement, and the sensing refresh rate, denoted Δt\Delta t. Intuitively, the longer the system waits before refreshing the sensing data, or the faster the target moves, the higher the risk of mismatch. To address this, we define a worst-case scenario and derive a sufficient condition on the maximum allowable target velocity that ensures reliable matching. This condition enables us to select optimal values for KcombK_{\mathrm{comb}} and the sensing refresh rate Δt\Delta t, which are two critical degrees of freedom in balancing the efficiency and robustness of the ISAC system.

Theorem 1. Let vv denote the target’s velocity. The matching process correctly identifies the real targets if:

ΔtKcomb<c4vΔf.(9)\Delta t K_{\mathrm{comb}} < \frac{c}{4v\Delta f}. \tag*{(9)}

Proof. Over one sensing interval Δt\Delta t, the worst-case two-way range displacement is

Δd≤2vΔt.(10)\Delta d \leq2v\Delta t. \tag*{(10)}

Reliable matching requires the displacement to be smaller than half of the distance between adjacent ghost peaks (two-way):

2vΔt<D/2.(11)2v\Delta t < D/2. \tag*{(11)}

Substituting DD and rearranging yields (9). □

Theorem 1 expression offers a practical design guideline. For a given maximum target velocity vv and subcarrier spacing Δf\Delta f, the system designer can jointly choose the largest possible values of KcombK_{\mathrm{comb}} and Δt\Delta t that satisfy the inequality. Larger values are preferred as they reduce the frequency of sensing operations and the reference signal overhead, thus achieving better resource utilization. Table I illustrates the maximum distance (i.e., vΔtv\Delta t) a target should not exceed with respect to the different KcombK_{\mathrm{comb}} and SCS. An analysis to the effect of the target velocity will be discussed in section IV.

comb 2comb 4comb 6comb 12
SCS = 60 kHz614312208104
SCS = 120 kHz31215610452

Table I. Maximum allowable target displacement vΔtv\Delta t derived from the sufficient condition in Eq. (9) for different PRS comb sizes KcombK_{\mathrm{comb}} and subcarrier spacings (SCS).

A key advantage of this threshold (Eq. (9)) is its ability to significantly reduce sensing overhead. By enabling the use of higher values for both KcombK_{\mathrm{comb}} and Δt\Delta t, while taking into account the velocity of the target, the system can allocate more resources for communication and reduce the frequency of sensing operations. Assuming that a given baseline is using a fixed values of KcombK_{\mathrm{comb}} and Δt\Delta t, and that our method selects a more efficient configuration using max⁡(Kcomb)\max(K_{\mathrm{comb}}) and max⁡(Δt)\max(\Delta t) (if the target’s velocity allows), the relative gain in resource efficiency can be expressed as:

Gain=[1−Kcomb⋅Δtmax⁡(Kcomb)⋅max⁡(Δt)]×100.(12)\mathrm{Gain} = \left[1 - \frac{K_{\mathrm{comb}}\cdot\Delta t}{\max(K_{\mathrm{comb}})\cdot\max(\Delta t)}\right]\times100. \tag*{(12)}

This simplified expression compares the average number of sensing resource slots used per unit time in our method versus the baseline, assuming that the only considered metrics here are vv and Δt\Delta t. A higher gain indicates better resource utilization and more capacity left for communication. Table II illustrates the achievable gain for Δt∈{20,80,160}\Delta t \in\{20, 80, 160\} ms and different KcombK_{\mathrm{comb}}.

comb 2comb 4comb 6comb 12
Δt=20\Delta t = 2097.9%95.8%93.75%87.5%
Δt=80\Delta t = 8091.6%83.3%75%50
Δt=160\Delta t = 16083.3%66.6%50%0

Table II. Percentage gain (Eq. (12)) in resource usage achieved by the proposed method compared to SotA baseline, for different Δt\Delta t and KcombK_{\mathrm{comb}}.

Numerical Results

In this section, we evaluate the performance of the proposed PRS-only multi-static sensing framework and compare it against the SotA benchmark based on joint PRS–DMRS processing. The numerical results aim to assess the effectiveness of ghost target elimination, and the robustness of the proposed successive estimation strategy under target motion.

All results are obtained using MATLAB simulations and its 5G toolboxes, ensuring compatibility with the 3GPP standards.

A single gNB transmitting a 5G NR signal with a carrier frequency of 28 GHz and an SCS of 120 kHz toward a sensing area containing multiple targets, and three UEs are used to perform bistatic range estimation. Multiple targets are considered, with targets being introduced sequentially until restarting when reaching a predefined maximum number (check Table III). The receiver processes the measurements either using the benchmark fused method, i.e, PRS+DMRS, or the proposed PRS-only method.

NodePosition (x, y, z) [m]
Tx (gNB)(0, 0, 0)
Rx₁; Rx₂; Rx₃(250, 200, 0); (450, 150, 0); (178, 405, 0)
Target₁ to Target₃(-310, 360, 0); (600, 500, 0); (-100, 433, 0)

Table III. Positions of the transmitter, receivers, and candidate targets

The PRS is configured to occupy 12 symbols. The DMRS uses configuration type 2, with three DMRS symbols located at symbol positions 3, 8, 12 within each slot. Additionally, the PDSCH employs mapping type ”A.” The code rate used to determine the transport block sizes is set to 490/1024, as specified in [15]. The Virtual Resource Block (VRB) bundle size is 4, and VRB to PRB interleaving is disabled, and we consider the 16 Quadrature Amplitude Modulation (16-QAM) for the PDSCH.

Figure 4 provides an illustrative comparison between the received echo when using PRS only and the benchmark PRS+DMRS method. The bottom plot corresponds to the PRS+DMRS approach, where the combination of both signals effectively suppresses most of the ghost components and yields dominant peaks at the true target positions. Nevertheless, small residual peaks can still be observed at distances spaced by DD, which may lead to incorrect detections when using threshold-based peak selection. The top plot shows the PRS-only case, where the periodic structure of the ghosts is clearly visible. Rather than attempting to remove these components at the waveform level, the proposed approach explicitly exploits their structured nature by grouping them into clusters associated with each physical target. Based on this clustering, our method distinguishes between newly detected targets and previously tracked ones. Ambiguity-aware multilateration is then applied only to newly detected targets, while standard, ghost-free multilateration is performed for tracked targets after the matching process.

To validate the performance of the proposed method, Figure 5 shows the perfect detection probability as a function of SNR for scenarios with a maximum of three targets. A perfect detection event corresponds to correctly identifying the number of targets without false positives or false negatives, and, in the PRS-only case, it requires correctly selecting the appropriate ellipse among the ambiguous candidates. Four curves are presented, corresponding to the two methods for two different comb values, Kcomb=2K_{\mathrm{comb}}=2 and Kcomb=4K_{\mathrm{comb}}=4. By averaging over 600 iterations, the results show that the PRS-only approach consistently outperforms the fused benchmark across the considered SNR values and for both KcombK_{\mathrm{comb}} configurations.

Figure 5. Comparative analysis of the perfect detection probability of our proposal and the SotA benchmark for different SNR values.

The degradation observed in the fused method is mainly due to residual peaks that persist even after combining PRS and DMRS, which can be falsely interpreted as additional targets. In contrast, the proposed method leverages the structured appearance of ghosts at distances separated by DD, allowing it to distinguish between true targets and spurious peaks. This result highlights a key limitation of the benchmark method, which does not exploit the available side information regarding the positions of the ghost components.

Figure 6 investigates the impact of target mobility on the matching process by evaluating the error probability as a function of the displacement between consecutive observations, i.e., vΔtv\Delta t, as predicted by Eq. (9). Using the same system parameters as before and focusing on a single-target scenario in a high-SNR regime, the performance is dominated by the ability of the matching process to correctly associate current measurements with historical estimates. Results are shown for four configurations corresponding to different combinations of subcarrier spacing (SCS) and PRS comb size KcombK_{\mathrm{comb}}. The results are obtained by averaging over 600 iterations, where each even iteration includes a random displacement of the target position with a maximum amplitude of vΔtv\Delta t. The results exhibit a clear threshold behavior, in agreement with Eq. (9). Specifically, we observe that for the green curve (corresponding to D=1249D=1249 m), errors remain negligible over the entire range of vΔtv\Delta t. When reducing DD to 624 m (red and black curves), the probability of error remains negligible for small displacements, and then increases once the displacement exceeds the threshold around vΔt=324v\Delta t=324 m. Further, reducing the ghost spacing to D=312D=312 m leads to a similar behavior, with the onset of errors occurring at even smaller displacements, around vΔt=156v\Delta t=156 m. This behavior confirms the robustness of the proposed method to mobility as long as the condition in Theorem 1 is satisfied. It also highlights the trade-off in system design, where larger ghost spacing improves robustness to mobility, while smaller ghost spacing may be sufficient in low-mobility sensing scenarios.

Conclusions

This paper proposed a novel PRS-only sensing framework to address the challenges of ambiguity and false detection in 5G OFDM reference signals-based ISAC systems. By using the successive estimation technique and multilateration at the fusion anchor at each new target, our proposed method reduces reliance on multiple reference signals to resolve the ambiguity problem. A sufficient condition was established to ensure correct matching, allowing the selection of appropriate sensing parameters for a given mobility scenario. Numerical results demonstrated that the proposed method can significantly reduce sensing overhead and improve detection performance while maintaining full compatibility with 3GPP-compliant 5G NR signaling. Future work will focus on extending the proposed framework to more complex and dynamic scenarios to enhance its applicability.

Figure 6. Analysis of the perfect detection probability vs. target displacement showing the impact of ghost spacing DD and the sufficient condition of Eq. (9).

References

  1. [1]Z. Du et al., “Toward ISAC-empowered vehicular networks: Framework, advances, and opportunities,” IEEE Wireless Communications, vol. 32, no. 2, pp. 222–229, 2025.DOI
  2. [2]E. C. Strinati et al., “Distributed intelligent integrated sensing and communications: The 6G-DISAC approach,” in 2024 Joint European Conference on Networks and Communications & 6G Summit (EuCNC/6G Summit). IEEE, 2024, pp. 392–397.arxiv.org/abs/2402.18271
  3. [3]Z. Wei 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.arxiv.org/abs/2211.11488
  4. [4]M. Lyu et al., “Reference signal-based waveform design for integrated sensing and communications system,” in ICC 2025-IEEE International Conference on Communications. IEEE, 2025, pp. 6621–6626.
  5. [5]K. Khosroshahi, P. Sehier, and S. Mekki, “Leveraging PRS and PDSCH for integrated sensing and communication systems,” in GLOBECOM 2024. IEEE, 2024, pp. 4702–4707.
  6. [6]Z. Wei et al., “Multiple reference signals collaborative sensing for integrated sensing and communication system towards 5G-A and 6G,” IEEE Transactions on Vehicular Technology, 2024.DOI
  7. [7]K. Khosroshahi, P. Sehier, and S. Mekki, “Doppler ambiguity elimination using 5G signals in integrated sensing and communication,” in 2024 IEEE 100th VTC2024-Fall). IEEE, 2024, pp. 1–6.arxiv.org/abs/2408.00679
  8. [8]M. Golzadeh et al., “Joint sensing and UE positioning in 5G-6G: PRS range estimation with suppressed ambiguity,” in 2024 IEEE Radar Conference (RadarConf24). IEEE, 2024, pp. 1–6.DOI
  9. [9]G. Hakobyan, M. Ulrich, and B. Yang, “OFDM-MIMO radar with optimized nonequidistant subcarrier interleaving,” IEEE Transactions on Aerospace and Electronic Systems, vol. 56, no. 1, pp. 572–584, 2019.DOI
  10. [10]A. Al Khansa, G. Madhusudan, G. Larue, and L.-A. Dufrene, “Dynamic power allocation in OFDM ISAC for time of arrival estimation,” in 2024 IEEE CSCN. IEEE, 2024, pp. 249–254.
  11. [11]A. Al Khansa and Y. Bahaninis, “Adaptive dual-windowing strategies for multi-target detection in ofdm isac,” in 8th International Conference CommNet. IEEE, 2025, pp. 1–7.
  12. [12]A. Al Khansa, “Learning to count targets from dual-window: A cnn approach for ofdm isac,” in 2025 8th International Conference CommNet. IEEE, 2025, pp. 1–6.arxiv.org/abs/2511.22473
  13. [13]K. M. Braun, “OFDM radar algorithms in mobile communication networks,” Ph.D. dissertation, KIT, 2014.DOI
  14. [14]E. Widdison and D. G. Long, “A review of linear multilateration techniques and applications,” IEEE Access, vol. 12, p. 26251, 2024.DOI
  15. [15]3GPP, “NR; physical layer procedures for data,” 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 38.214, 2024, v18.2.0.

Paper details

vlnc:W_y346ytjr
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…
Contents