WorksEngineering
Distributed Pinching-Antenna Architecture for Communication and Target Positioning
Abstract
We propose a pinching-antenna system (PASS)-assisted integrated sensing and communication (ISAC) framework for joint downlink communication and target positioning. The dielectric waveguides enable flexible signal radiation through activated pinching antennas (PAs), while the reflected echo signals are collected by a sensing receiver. A unified waveform is adopted to support both data transmission and sensing-echo processing. In particular, a two-stage procedure is developed for target localization using time-sum-of-arrival (TSoA) measurements, where the signal propagation distances are first estimated through de-chirp processing and beat frequency extraction. Then, the target position is obtained based on TSoA measurements using the weighted linear least squares(WLLS) technique. To characterize the fundamental limits of the PASS-assisted ISAC system, a Cramér-Rao lower bound (CRB) is derived, which captures the impact of PA activation. Furthermore, the PA activation problem is formulated to jointly optimize the communication spectral efficiency (SE) and sensing performance. To address this, a low-complexity heuristic greedy selection algorithm is proposed to determine the activated PA pattern. Numerical results demonstrate that the proposed scheme outperforms random PA selection, achieving gains of up to 81% improvement in communication SE.
Introduction
ISAC has emerged as a promising paradigm for next-generation wireless networks, as it enables communication and sensing functionalities to share spectrum, hardware, and signal resources. Meanwhile, flexible-antenna technologies, such as fluid and movable antennas, have been developed to improve spatial adaptability, but they still face inherent limitations in mitigating severe large-scale fading [1].
To overcome these limitations, PASS architectures have attracted increasing attention for improving the performance of ISAC systems, thanks to their capability of flexibly creating line-of-sight (LoS) links for both reliable communication and high-precision sensing [2]. By activating different PAs along the dielectric waveguides, a PASS can reshape the effective propagation paths and provide additional spatial degrees of freedom for joint communication and sensing.
Recent studies have explored PASS variants for different ISAC configurations and performance objectives. In [3], a wireless sensing architecture based on PASS was proposed, where leaky coaxial cables were used for collecting echo signals, and the waveform and PA positions were jointly optimized to improve multi-target sensing accuracy. In particular, recent work investigated PA-assisted positioning from a CRB perspective, showing that the reconfigurable and flexible array configuration of PAs can improve localization accuracy and support flexible user-centric positioning [4]. A received signal strength-based localization framework for PASS was developed in [5], where stochastic geometry was used to characterize target localizability and CRB performance. By exploiting spatial-phase diversity through non-uniform PA placement and in-waveguide propagation, PASS-ISAC can significantly tighten the sensing bounds compared with a uniform linear array [6]. The study in [7] investigated a PASS-ISAC framework with asymmetric sensing propagation, derived the sensing degrees of freedom and angle estimation CRB, and optimized the PA locations and transmit power to balance sensing accuracy and power consumption.
Although existing studies have advanced the understanding of PASS-assisted ISAC systems, several challenges remain underexplored for joint communication and localization. First, due to the limited number of radio-frequency (RF) chains in practical PASS architectures, PAs connected to the same dielectric waveguide typically transmit the same signal, necessitating a simple unified waveform that simultaneously supports communication and sensing [8]. Second, PA activation affects not only sensing-link quality but also the localization degrees of freedom determined by the sensing geometry. Thus, its impact on localization accuracy should be explicitly characterized from a geometry-based perspective. Third, the role of PA activation in balancing communication performance and localization accuracy remains insufficiently explored.
Motivated by these challenges, this paper proposes a PASS-assisted ISAC localization framework with PA acti-
This work was supported by the U.K. Engineering and Physical Sciences Research Council (EPSRC) grant (EP/X04047X/2) for TI-TAN Telecoms Hub, the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 101001331), and the MULTIPLY-6G project that has received funding from the Smart Networks and Services Joint Undertaking (SNS JU) under the European Union’s Horizon Europe research and innovation programme under Grant Agreement No 101293106. The work of J. He was supported by the HORIZON-MSCA through the project ECO-ISAC under Grant 101274088. The work of H. Q. Ngo and M. Matthaiou was supported by a research grant from the Department for the Economy Northern Ireland under the US-Ireland R&D Partnership Programme. vation design. The main contributions of this paper are as follows:
A unified waveform supports both communication and sensing under a RF chain-limited PASS architecture. This design reduces waveform complexity and avoids dedicated sensing signals. The transmitted signal can support data transmission and echo-based range estimation, which is suitable for PASS architectures with shared RF chains.
We reveal that the PA positions affect not only the effective communication and sensing channels, but also the sensing geometry for target localization. Based on this observation, a geometry-based CRB analysis is developed to characterize the achievable localization accuracy under different PA activation patterns.
We formulate a PA activation optimization problem that balances between communication SE and sensing performance while ensuring localization feasibility. To solve this discrete and nonconvex problem efficiently, a low-complexity heuristic greedy selection algorithm is proposed to determine the activated PA pattern.
Notations: Scalars, vectors, and matrices are denoted by lowercase, bold lowercase, and bold uppercase letters. The sets of real and complex numbers are and . For a complex scalar , and denote its conjugate and magnitude. For vectors or matrices, and indicate the transpose and conjugate transpose. The -th entry of a vector is , and denotes the Euclidean norm. The operators and form diagonal and block diagonal matrices, while and denote the trace and rank. Finally, represents expectation.
System model
We consider a downlink PASS system. The base station (BS) is connected to transmit dielectric waveguides of length , each containing PAs. The PASS serves a single-antenna communication user (CU) and senses a point-like target. Both the CU and the target are located in a region of size . The sensing receiver is a single-antenna access point (AP) that observes the reflected sensing signals. The dielectric waveguides are assumed to be parallel to the -axis with respect to the plane at a height . The -th waveguide starts at the feed point and ends at . In particular, the position of the -th PA on the -th waveguide can be denoted by . The -axis coordinates of all PAs can be stacked into a matrix , where represents the -axis coordinate of the PAs on the -th waveguide. The positions of the CU and the sensing target are denoted with . Similarly, the AP position is .
Channel Model
We consider standard indoor LoS-only scenarios in which the CU and target are in LoS with the PAs. The channel vector from the -th waveguide to the sensing AP or CU is
where denotes the wavelength; with and being the speed of light and carrier frequency, respectively; denotes the propagation distance associated with the -th PA on the -th waveguide. For the communication link, , while for the sensing link, . Moreover, the overall channel vector is given by
We define as the wavelength of the carrier in the waveguide, where is the effective refractive index [9]. Following [2], the in-waveguide attenuation can be neglected under mild conditions, as it induces a negligible loss in the average rate. The in-waveguide channel vector for the -th waveguide can be expressed as
Therefore, the overall in-waveguide channel matrix can be represented by a block-diagonal matrix [10], which is given by
PA Configuration Strategies
We consider a binary activation strategy for the candidate PAs deployed on the waveguides. The PA activation vector is
where indicates that the -th PA on the -th waveguides is active, and otherwise. It is assumed that a total of PAs are activated. Accordingly, the corresponding PA activation matrix is defined as . With this activation strategy, only active PAs participate in the transmission and sensing processes.
Figure 1. System model of the PASS-assisted ISAC.
Signal Model
Linear frequency modulation (LFM) is adopted since it enables efficient sensing delay estimation by mapping delay to the beat frequency. In addition, the LFM signal provides a flexible carrier for communication symbol embedding [11]. The adopted LFM signal can be expressed as
where is the initial frequency and is the chirp rate. The LFM signal sweeps linearly over bandwidth within the temporal duration with a rate of . The LFM waveform serves as the communication carrier, and the data sequence is mapped to the quadrature amplitude modulation (QAM) symbols in each time slot [12]. Thus, the integrated waveform is written as
where and are the amplitude and phase of the QAM symbols, respectively. For 4-QAM, the constellation has a constant modulus and thus yields a constant-envelope waveform with , which can also enhance the sensing performance [13]. In contrast, higher-order QAM (e.g., 16-QAM) is multi-modulus and yields a non-constant envelope, which may improve the communication performance but can degrade the sensing accuracy [12]. We focus on the constant-envelope case and set over .
Communication Receiver Processing and Demodulation
CU Received Signal Model
In the downlink, the CU receives a superposition of delayed integrated waveforms radiated by the PAs. Let denote the transmit beamforming vector for the CU, designed using maximum ratio transmission (MRT). The received signal at CU at time is given by
where is the transmit power fed into the waveguides, represents the propagation delay determined by the distance between the -th PA on the -th waveguide and the CU, and denotes the additive white Gaussian noise (AWGN) at time slot , whose power is . We point out that communication and sensing are supported by the same integrated waveform and no additional sensing interference is imposed on the CU. Accordingly, the SNR for the CU is
For PA activation design, we adopt the Gaussian-input SE benchmark .
Signal De-chirp and Restoration Processing
Since the carrier parameters of the LFM waveform are known at the CU, the local reference signal in (6) can be generated for de-chirp processing with (8) by
where is the complex gain and phase terms of the cascaded in-waveguide and free-space propagation, while . The data phase remains unchanged during each symbol interval of duration . Let denote the number of symbol intervals within one LFM waveform duration, and denote the symbol index. Since the data phase is constant within each symbol interval and , the delayed phase term can be approximated by the same symbol phase over the -th interval. Since the propagation delay satisfies , the delayed phase term is still the same symbol. Therefore, for the -th symbol interval, it can be approximated as a constant phase term, denoted by . We rewrite (9) as
which is a standard multi-path continuous-time signal model. With de-chirp processing, the delay-dependent phase can be estimated and compensated within each symbol interval, after which standard QAM demodulation can be performed based on the restored symbol phase [14].
Sensing Signal Processing and Two-Stage Localization
AP Received Echo and De-chirp
We define the target reflection loss from as , given by [15, 16]: , where the incident angle is computed by . It is worth noting that and are the reflection coefficients determined by the permittivity of the reflecting surface. The validity of the reflection loss model has been verified in [17]. Furthermore, the target-reflected echo signal received at the AP is given by
Here, is the sensing effective cascaded gain, while is the AWGN with power at time slot . Moreover, is the round trip propagation delay associated with the path from the -th PA on the -th waveguide to the target and then to the AP. The sensing SNR at AP is .
Stage I: Delay Estimation
The received echo is mixed with the local reference to generate the beat frequency signal, which is
Although the received echo is a superposition of the signals from all activated PAs, the per-PA components can be distinguished during localization due to their different propagation delays [15]. By substituting the expression of into (12), the beat signal can be expanded as
where the de-chirped noise is and denotes the delay-dependent constant phase term, and represents the corresponding beat frequency determined by the propagation delay. Within one symbol duration, the first exponential term is constant. The modulation phase mainly introduces a constant phase rotation and slight spectral spreading, while the signal energy remains concentrated around . Therefore, performing a fast Fourier transform (FFT) and peak detection on the beat signal still enables robust estimation of and . Finally, we obtain the sensing link propagation distance .
Remark 1. In our considered activation pattern, the delay differences between adjacent activated PAs are well above the minimum resolvable delay. Therefore, echoes from different activated PAs can be reliably resolved as distinct FFT peaks.
Stage II: TSoA Localization
For notational simplicity, we use a single index to denote the -th activated PA. The corresponding TSoA observation can then be expressed as
where represents the distance between the -th activated PA and the sensing target, which is given by
while denotes the distance between the sensing target and the AP, given by
Accordingly, the least-squares estimator for the target estimation can be formulated as
It is observed that the nonlinearity of the TSoA-based localization problem arises from the two square root distance terms. By rearranging the range measurement in (15), the PA-target distance is computed by
We square both sides to derive a tractable linearized representation as
For compact notation, we define the noise-induced residual as
Accordingly, the squared measurement equation is
Then, by substituting the geometric expressions of and , and collecting the terms related to the unknown target position, the linearized measurement model in (13), shown at the top of the page. Let be the augmented unknown vector. Then, the linearized TSoA equation can be written as , where
Since the TSoA observations associated with different activated PAs generally have unequal estimation accuracy, a weighted linear least squares estimator (WLLS)1 is adopted as
Here, , where represents the normalized weight associated with the -th activated PA, obtained by dividing its sensing SNR by the maximum sensing SNR among all activated PAs. Finally, the sensing target position is obtained from the first three elements of .
CRB Analysis
By stacking the observations of all activated PAs into a vector, we have
Moreover, the noise is assumed to be a zero-mean Gaussian vector with covariance matrix , where the corresponding measurement noise variance is given by [18]: , where is the effective bandwidth, is the number of transmitted symbols and denotes the sensing SNR corresponding to the -th PA link. In the CRB derivation, is assumed to be known and is evaluated from the measurement stage. Therefore, the probability density function (PDF) of is given by
The Fisher information matrix (FIM) is then
while the entries of are
The CRB for estimating the target position is
Thus, we obtain the corresponding position error bound (PEB):
which represents the theoretical lower bound on the target localization error. To evaluate the practical localization performance, the root mean square error (RMSE) is used, and is defined as .
Remark 2. For the TSoA measurement vector in (27) and a given sensing geometry, the measurement noise variance is inversely proportional to the per-link sensing SNR defined in (14). Thus, a higher sensing SNR generally leads to a lower CRB/PEB, motivating its adoption as a tractable sensing metric.
Optimization Problem Formulation
In this section, we maximize a weighted combination of the communication and sensing metrics through PA activation. Since the activated PA positions significantly affect localization accuracy, a collinear activation pattern provides insufficient geometric diversity, leading to poor localization feasibility [19]. To address this issue, we impose a waveguide activation constraint and introduce the binary variable to indicate whether the -th waveguide is active. Following Remark 2, the sensing SNR is adopted as a tractable sensing metric [20]. The activation constraints promote geometric diversity, while WLLS feasibility additionally requires .
The objective function is the weighted SE and sensing SNR. The constraint (39b) represents the number of activated PAs. Constraints (39c)-(39e) promote geometric diversity by requiring PA activation on at least three waveguides, thereby reducing the risk of rank deficiency. Finally, constraints (39f) and (39g) impose binary requirements on the PA activation variables and waveguide activation indicators, respectively.
It is noted that the PA positions are strongly coupled with both the dielectric waveguide propagation and the free-space channel, which makes the objective function non-convex. Moreover, the binary activation variables introduce a discrete combinatorial structure, which makes the problem intractable for exhaustive search when the system dimension becomes large. Therefore, a heuristic greedy selection (HGS) algorithm is adopted to efficiently determine the PA activation pattern with low computational complexity. Specifically, one PA is first selected from each waveguide to form an initial feasible set, where all cross-waveguide three-PA combinations are examined and the best one is retained. Then, the remaining PAs are added one by one according to the largest objective improvement until PAs are activated. The resulting computational complexity of the HGS algorithm is approximately , which is much lower than that of the exhaustive search. The steps defined in the greedy procedure are summarized in Algorithm 1.
Numerical Results
In this section, the performance of the PASS is evaluated under different network conditions. Specifically, the PASS consists of waveguides, with candidate PAs deployed on each waveguide, resulting in 30 candidate PAs in total. The carrier frequency is set to 28 GHz and the chirp bandwidth is set to MHz. The waveguide length, height, and refractive index are set to m, m, and , respectively, while the waveguides are located at m. The communication and sensing noise powers are set to dBm and dBm, respectively. The weighting factor is set to . The reflection loss parameters are chosen as and [15]. Moreover, all numerical results are averaged over 1,000 random realizations. In
Algorithm 1 HGS algorithm for solving (39)
Input: , , , , , , ,
Initialize: , the best objective value as .
for to do
for to do
for to do
Construct by activating PAs , , and
Evaluate
if then
,
end if
end for
end for
end for
while do
,
for each inactive PA do
Activate PA temporarily and obtain
Evaluate
if then
,
end if
end for
end while
Output:
each realization, the CU is randomly generated within , , and , while the sensing target is randomly generated within , , and . The AP location is fixed at . We consider a random selection scheme as a benchmark, where PAs are randomly activated from the candidate PA set while satisfying the same activation constraints.
Figure 2 shows that the average SE increases with the transmit power for all schemes due to the improved received SNR. The communication-only scheme, corresponding to in (39), achieves the highest SE, whereas the sensing-only scheme () yields lower SE due to sensing-oriented PA activation. For the proposed HGS scheme, we set . For the same , the proposed HGS consistently outperforms the benchmark, confirming the effectiveness of the PA activation design. The smaller SE gap between and than between and indicates diminishing gains as more PAs are activated. This is mainly due to stronger phase coupling among PA signals, which may limit the coherent combining gain at the CU. The full activation scheme is also outperformed by optimized partial activation with , demonstrating that activating all PAs is not always optimal. Thus, the proposed HGS selects PAs with more favorable channel and phase conditions for communication.
Figure 2. Average SE versus the transmit power.
Figure 3 illustrates the tradeoff between communication and sensing performance under different values of the weighting factor. As increases, the average SE gradually improves, while the average sensing SNR correspondingly decreases, which is consistent with the communication-oriented preference of the objective function. In addition, a larger number of activated PAs generally provides better performance for both metrics. The figure also shows that a moderate enables a desirable tradeoff.
Figure 3. Average SE and sensing SNR versus the weighting factor .
Figure 4 compares the CDF of the WLLS-based localization RMSE and the corresponding PEB under different numbers of activated PAs. It is observed that both the PEB and RMSE curves shift to the left as increases, indicating that activating more PAs improves the localization accuracy by providing richer geometric diversity. Moreover, the RMSE curves remain above the corresponding PEB curves, which is consistent with the fact that the PEB serves as a theoretical lower bound while the practical estimator suffers from noise and linearization errors. The similar variation trend between RMSE and PEB also validates the effectiveness of the theoretical localization analysis.
Figure 4. CDF of the localization RMSE and PEB for different numbers of activated PAs.
Conclusion
A distributed PASS-assisted ISAC framework was proposed for joint downlink communication and target localization. A unified LFM-based waveform was designed for the PASS architecture. Subsequently, a WLLS estimator and the corresponding CRB were developed for target localization. To balance communication SE and sensing performance, a PA activation problem was formulated and efficiently solved using a low-complexity HGS algorithm. Numerical results revealed that the proposed PA activation scheme outperformed random selection in both communication and localization performance. The results also indicated that activating all PAs was unnecessary, as a selected subset can achieve comparable or even superior performance.
References
- [1]L. Zhou et al., “Fluid antenna-assisted ISAC systems,” IEEE Wireless Commun. Lett., vol. 13, pp. 3533–3537, Dec. 2024.DOI
- [2]M. Qian, X. Mu, L. You, and M. Matthaiou, “Pinching-antenna-based communications: Spectral efficiency analysis and deployment strategies,” IEEE Trans. Commun., vol. 74, pp. 4758–4771, 2026.arxiv.org/abs/2507.14831
- [3]Z. Wang, C. Ouyang, Y. Liu, and A. Nallanathan, “Wireless sensing via pinching-antenna systems,” IEEE Wireless Commun. Lett., vol. 14, pp. 3475–3479, Nov. 2025.DOI
- [4]Z. Ding, “Pinching-antenna assisted ISAC: A CRLB perspective,” npj Wireless Technology, vol. 1, p. 4, Oct. 2025.arxiv.org/abs/2504.05792
- [5]J. He, X. Mu, H. Q. Ngo, and M. Matthaiou, “Pinching-antenna system-assisted localization: A stochastic geometry perspective,” IEEE Wireless Commun. Lett., vol. 15, pp. 1737–1741, 2026.arxiv.org/abs/2511.15444
- [6]D. Bozanis, V. K. Papanikolaou, S. A. Tegos, and G. K. Karagianidis, “Cramér-Rao bounds for integrated sensing and communications in pinching-antenna systems,” in Proc. IEEE PIMRC, Sept. 2025.DOI
- [7]N. Xue, J. He, and M. Matthaiou, “Pinching-antenna-assisted ISAC based on the reflection law,” in Proc. IEEE ICC, May 2026.DOI
- [8]Z. Sui, Z. Liu, L. Musavian, L.-L. Yang, and L. Hanzo, “Generalized spatial modulation aided affine frequency division multiplexing,” IEEE Trans. Wireless Commun., vol. 25, pp. 4658–4673, Oct. 2026.arxiv.org/abs/2501.10865
- [9]D. M. Pozar, Microwave Engineering: Theory and Techniques. John Wiley & Sons, 2021.
- [10]X. Xu, X. Mu, Y. Liu, and A. Nallanathan, “Joint transmit and pinching beamforming for pinching antenna system (PASS): Optimization-based or learning-based?,” IEEE Trans. Wireless Commun., vol. 25, pp. 11449–11464, 2026.DOI
- [11]Z. Lyu et al., “Radar-centric photonic terahertz integrated sensing and communication system based on LFM-PSK waveform,” IEEE Trans. Microw. Theory Techn., vol. 71, pp. 5019–5027, Nov. 2023.DOI
- [12]Z. Lyu et al., “Multi-channel photonic THz-ISAC system based on integrated FLM-QAM waveform,” J. Lightw. Technol., vol. 42, no. 11, pp. 3981–3988, Jun. 2024.DOI
- [13]Y. Xiong et al., “On the fundamental tradeoff of integrated sensing and communications under Gaussian channels,” IEEE Trans. Inf. Theory, vol. 69, pp. 5723–5751, Sept. 2023.arxiv.org/abs/2204.06938
- [14]Z. Sui et al., “Multi-functional chip signalling for next-generation multi-carrier wireless networks: Communications, sensing and isac perspectives,” arXiv preprint arXiv:2508.06022, 2025.arxiv.org/abs/2508.06022
- [15]J. He et al., “RSS localization in cell-free massive MIMO: Algorithms, analysis, and implementation,” IEEE Trans. Wireless Commun., vol. 25, pp. 4021–4037, Sept. 2026.DOI
- [16]J. He et al., “RSS-based localization with a single receiver: Method and stochastic analysis,” IEEE Trans. Wireless Commun., vol. 25, pp. 15116–15133, 2026.DOI
- [17]Y. Xing, O. Kanhere, S. Ju, and T. S. Rappaport, “Indoor wireless channel properties at millimeter wave and sub-terahertz frequencies,” in Proc. IEEE GLOBECOM, Dec. 2019.arxiv.org/abs/1908.09765
- [18]J. He et al., “Stochastic analysis of Cramér-Rao lower bound for positioning in mmWave-THz HetNets,” IEEE Trans. Commun., vol. 74, pp. 9043–9060, 2026.DOI
- [19]J. He et al., “Cramér–Rao lower bound analysis for elliptic localization with random sensor placements,” IEEE Trans. Aerosp. Electron. Syst., vol. 60, pp. 5587–5595, Aug. 2024.DOI
- [20]Y. Hu, T. Zhang, X. Xu, and Y. Liu, “Pinching antenna-enabled integrated sensing and communication for low-altitude UAV,” IEEE Trans. Veh. Technol., vol. 75, pp. 11728–11733, Jun. 2026.DOI



