betaFeedback

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 R\mathbb{R} and C\mathbb{C}. For a complex scalar xx, x∗x^{*} and ∣x∣\lvert x\rvert denote its conjugate and magnitude. For vectors or matrices, (⋅)T(\cdot)^{T} and (⋅)H(\cdot)^{H} indicate the transpose and conjugate transpose. The nn-th entry of a vector x\mathbf{x} is [x]n[\mathbf{x}]_{n}, and ∣⋅∣\lvert\cdot\rvert denotes the Euclidean norm. The operators diag⁡(⋅)\operatorname{diag}(\cdot) and blkdiag⁡(⋅)\operatorname{blkdiag}(\cdot) form diagonal and block diagonal matrices, while Tr⁡(⋅)\operatorname{Tr}(\cdot) and rank⁡(⋅)\operatorname{rank}(\cdot) denote the trace and rank. Finally, E{⋅}\mathbb{E}\{\cdot\} represents expectation.

System model

We consider a downlink PASS system. The base station (BS) is connected to MM transmit dielectric waveguides of length LL, each containing NN 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 D=Dx×Dy m2D=D_x\times D_y\ \mathrm{m}^{2}. 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 xx-axis with respect to the x−o−yx-o-y plane at a height dd. The mm-th waveguide starts at the feed point ψm,0=[0,ym,0,d]T\boldsymbol{\psi}_{m,0}=[0,y_{m,0},d]^{T} and ends at [L,ym,0,d]T[L,y_{m,0},d]^{T}. In particular, the position of the nn-th PA on the mm-th waveguide can be denoted by ψm,n=[xm,np,ym,np,d]T\boldsymbol{\psi}_{m,n}=[x_{m,n}^{p},y_{m,n}^{p},d]^{T}. The xx-axis coordinates of all PAs can be stacked into a matrix X=[x1,x2,…,xM]∈RN×M\mathbf{X}=[\mathbf{x}_{1},\mathbf{x}_{2},\ldots,\mathbf{x}_{M}]\in\mathbb{R}^{N\times M}, where xm=[xm,1p,xm,2p,…,xm,Np]T\mathbf{x}_{m}=[x_{m,1}^{p},x_{m,2}^{p},\ldots,x_{m,N}^{p}]^{T} represents the xx-axis coordinate of the PAs on the mm-th waveguide. The positions of the CU and the sensing target are denoted ψξ=[xξ,yξ,zξ]T\boldsymbol{\psi}^{\xi}=[x^{\xi},y^{\xi},z^{\xi}]^{T} with ξ∈{c,s}\xi\in\{c,s\}. Similarly, the AP position is ψAP=[xAP,yAP,zAP]T\boldsymbol{\psi}^{\mathrm{AP}}=[x^{\mathrm{AP}},y^{\mathrm{AP}},z^{\mathrm{AP}}]^{T}.

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 mm-th waveguide to the sensing AP or CU is

hξ,m(xm)=[ηe−j2πλDm,1ξDm,1ξ,…,ηe−j2πλDm,NξDm,Nξ]H,(1)\mathbf{h}_{\xi,m}(\mathbf{x}_{m})= \left[ \frac{\sqrt{\eta}e^{-j\frac{2\pi}{\lambda}D_{m,1}^{\xi}}}{D_{m,1}^{\xi}}, \ldots, \frac{\sqrt{\eta}e^{-j\frac{2\pi}{\lambda}D_{m,N}^{\xi}}}{D_{m,N}^{\xi}} \right]^{H}, \tag*{(1)}

where λ=cfc\lambda=\frac{c}{f_{c}} denotes the wavelength; η=c216π2fc2\eta=\frac{c^{2}}{16\pi^{2}f_{c}^{2}} with cc and fcf_{c} being the speed of light and carrier frequency, respectively; Dm,nξD_{m,n}^{\xi} denotes the propagation distance associated with the nn-th PA on the mm-th waveguide. For the communication link, Dm,nc=∥ψc−ψm,n∥D_{m,n}^{c}=\lVert\boldsymbol{\psi}^{c}-\boldsymbol{\psi}_{m,n}\rVert, while for the sensing link, Dm,ns=∥ψs−ψm,n∥+∥ψs−ψAP∥D_{m,n}^{s}=\lVert\boldsymbol{\psi}^{s}-\boldsymbol{\psi}_{m,n}\rVert+\lVert\boldsymbol{\psi}^{s}-\boldsymbol{\psi}^{\mathrm{AP}}\rVert. Moreover, the overall channel vector is given by

hξ(X)=[hξ,1H(x1),…,hξ,MH(xM)]H∈CMN×1.(2)\mathbf{h}_{\xi}(\mathbf{X})= \left[ \mathbf{h}_{\xi,1}^{H}(\mathbf{x}_{1}),\ldots,\mathbf{h}_{\xi,M}^{H}(\mathbf{x}_{M}) \right]^{H}\in\mathbb{C}^{MN\times1}. \tag*{(2)}

We define λg=λne\lambda_{g}=\frac{\lambda}{n_{e}} as the wavelength of the carrier in the waveguide, where nen_{e} 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 gm(xm)∈CN×1\mathbf{g}_{m}(\mathbf{x}_{m})\in\mathbb{C}^{N\times1} for the mm-th waveguide can be expressed as

gm(xm)=[e−j2πλg∥ψm,1−ψm,0∥,…,e−j2πλg∥ψm,N−ψm,0∥]T.(3)\mathbf{g}_{m}(\mathbf{x}_{m})= \left[ e^{-j\frac{2\pi}{\lambda_{g}}\lVert\boldsymbol{\psi}_{m,1}-\boldsymbol{\psi}_{m,0}\rVert}, \ldots, e^{-j\frac{2\pi}{\lambda_{g}}\lVert\boldsymbol{\psi}_{m,N}-\boldsymbol{\psi}_{m,0}\rVert} \right]^{T}. \tag*{(3)}

Therefore, the overall in-waveguide channel matrix can be represented by a block-diagonal matrix G∈CMN×M\mathbf{G}\in\mathbb{C}^{MN\times M} [10], which is given by

G(X)=blkdiag⁡(g1(x1),g2(x2),…,gM(xM))=[g1(x1)0⋯00g2(x2)⋯0⋮⋮⋱⋮00⋯gM(xM)].(4)\begin{aligned} \mathbf{G}(\mathbf{X})=\operatorname{blkdiag}(\mathbf{g}_{1}(\mathbf{x}_{1}),\mathbf{g}_{2}(\mathbf{x}_{2}),\ldots,\mathbf{g}_{M}(\mathbf{x}_{M})) \\ &= \begin{bmatrix} \mathbf{g}_{1}(\mathbf{x}_{1}) & 0 & \cdots& 0\\ 0 & \mathbf{g}_{2}(\mathbf{x}_{2}) & \cdots& 0\\ \vdots& \vdots& \ddots& \vdots\\ 0 & 0 & \cdots& \mathbf{g}_{M}(\mathbf{x}_{M}) \end{bmatrix}. \tag*{(4)} \end{aligned}

PA Configuration Strategies

We consider a binary activation strategy for the candidate PAs deployed on the waveguides. The PA activation vector is

α=[α1,α2,…,αm,n,…,αMN]T∈{0,1}MN×1,(5)\boldsymbol{\alpha}=[\alpha_{1},\alpha_{2},\ldots,\alpha_{m,n},\ldots,\alpha_{MN}]^{T}\in\{0,1\}^{MN\times1}, \tag*{(5)}

where αm,n=1\alpha_{m,n}=1 indicates that the nn-th PA on the mm-th waveguides is active, and αm,n=0\alpha_{m,n}=0 otherwise. It is assumed that a total of KK PAs are activated. Accordingly, the corresponding PA activation matrix is defined as A=diag⁡(α)∈RMN×MN\mathbf{A}=\operatorname{diag}(\boldsymbol{\alpha})\in\mathbb{R}^{MN\times MN}. With this activation strategy, only K≤MNK\leq MN 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

sL(t)=exp⁡[jπ(2f0t+ut2)](6)s_{\mathrm{L}}(t)=\exp\left[j\pi\left(2f_0t+ut^2\right)\right] \tag*{(6)}

where f0f_0 is the initial frequency and uu is the chirp rate. The LFM signal sweeps linearly over bandwidth BB within the temporal duration TT with a rate of u=B/Tu=B/T. 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 TsT_s [12]. Thus, the integrated waveform is written as

s(t)=a(t)exp⁡[jπ(2f0t+ut2)+jϕ(t)](7)s(t)=a(t)\exp\left[j\pi\left(2f_0t+ut^2\right)+j\phi(t)\right] \tag*{(7)}

where a(t)a(t) and ϕ(t)\phi(t) 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 a(t)=1a(t)=1, 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 a(t)=1a(t)=1 over t∈[0,T]t\in[0,T].

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 w=[w1,…,wM]T∈CM×1\mathbf{w}=[w_1,\ldots,w_M]^{\mathsf{T}}\in\mathbb{C}^{M\times1} denote the transmit beamforming vector for the CU, designed using maximum ratio transmission (MRT). The received signal at CU at time tt is given by

yc(t)=Pt∑m=1M∑n=1Nαm,n[hc,m]n∗[gm]nwms(t−τm,nc)+nc(t)(8)y_{\mathrm{c}}(t)=\sqrt{P_t}\sum_{m=1}^{M}\sum_{n=1}^{N}\alpha_{m,n}\left[\mathbf{h}_{\mathrm{c},m}\right]_n^*\left[\mathbf{g}_m\right]_nw_ms\left(t-\tau_{m,n}^{\mathrm{c}}\right)+n_{\mathrm{c}}(t) \tag*{(8)}

where PtP_t is the transmit power fed into the waveguides, τm,nc\tau_{m,n}^{\mathrm{c}} represents the propagation delay determined by the distance between the nn-th PA on the mm-th waveguide and the CU, and nc(t)n_{\mathrm{c}}(t) denotes the additive white Gaussian noise (AWGN) at time slot tt, whose power is σc2\sigma_{\mathrm{c}}^2. 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

SNRc=Pt∣hcHAGw∣2σc2.\mathrm{SNR}_{\mathrm{c}}=\frac{P_t\left|\mathbf{h}_{\mathrm{c}}^{\mathsf{H}}\mathbf{A}\mathbf{G}\mathbf{w}\right|^2}{\sigma_{\mathrm{c}}^2}.

For PA activation design, we adopt the Gaussian-input SE benchmark SE=log⁡2(1+SNRc)\mathrm{SE}=\log_2(1+\mathrm{SNR}_{\mathrm{c}}).

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

zc(t)=yc(t)sL∗(t)=∑m=1M∑n=1Nξm,ncejϕ(t−τm,nc)e−j2πuτm,nct+n~c(t),(9)\begin{aligned} z_{\mathrm{c}}(t)=y_{\mathrm{c}}(t)s_{\mathrm{L}}^*(t) \\ &=\sum_{m=1}^{M}\sum_{n=1}^{N}\xi_{m,n}^{\mathrm{c}}e^{j\phi(t-\tau_{m,n}^{\mathrm{c}})}e^{-j2\pi u\tau_{m,n}^{\mathrm{c}}t}+\tilde{n}_{\mathrm{c}}(t), \tag*{(9)} \end{aligned}

where ξm,nc=Ptαm,n[hc,m]n∗[gm]nwm\xi_{m,n}^{\mathrm{c}}=\sqrt{P_t}\alpha_{m,n}[\mathbf{h}_{\mathrm{c},m}]_n^*[\mathbf{g}_m]_nw_m is the complex gain and phase terms of the cascaded in-waveguide and free-space propagation, while n~c(t)=nc(t)sL∗(t)\tilde{n}_{\mathrm{c}}(t)=n_{\mathrm{c}}(t)s_{\mathrm{L}}^*(t). The data phase remains unchanged during each symbol interval of duration TsT_s. Let I=TTsI=\frac{T}{T_s} denote the number of symbol intervals within one LFM waveform duration, and i∈{1,…,I}i\in\{1,\ldots,I\} denote the symbol index. Since the data phase is constant within each symbol interval and τm,nc≪Ts\tau_{m,n}^{\mathrm{c}}\ll T_s, the delayed phase term ϕ(t−τm,nc)\phi(t-\tau_{m,n}^{\mathrm{c}}) can be approximated by the same symbol phase ϕic\phi_i^{\mathrm{c}} over the ii-th interval. Since the propagation delay satisfies τm,nc≪Ts\tau_{m,n}^{\mathrm{c}}\ll T_s, the delayed phase term is still the same symbol. Therefore, for the ii-th symbol interval, it can be approximated as a constant phase term, denoted by ϕic\phi_i^{\mathrm{c}}. We rewrite (9) as

zc(t)=∑m=1M∑n=1Nξm,ncejϕice−j2πuτm,nct+n~c(t),(10)z_{\mathrm{c}}(t)=\sum_{m=1}^{M}\sum_{n=1}^{N}\xi_{m,n}^{\mathrm{c}}e^{j\phi_i^{\mathrm{c}}}e^{-j2\pi u\tau_{m,n}^{\mathrm{c}}t}+\tilde{n}_{\mathrm{c}}(t), \tag*{(10)}

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 ψm,n\boldsymbol{\psi}_{m,n} as ∣Γm,n∣2∈[0,1]|\Gamma_{m,n}|^2\in[0,1], given by [15, 16]: ∣Γm,n∣=ϵaθm,n+ϵb|\Gamma_{m,n}|=\epsilon_a\theta_{m,n}+\epsilon_b, where the incident angle θm,n\theta_{m,n} is computed by θm,n=12arccos⁡(∥ψm,n−ψAP∥2−∥ψm,n−ψS∥2−∥ψS−ψAP∥2−2∥ψm,n−ψS∥∥ψS−ψAP∥)\theta_{m,n}=\frac{1}{2}\arccos\left(\frac{\|\boldsymbol{\psi}_{m,n}-\boldsymbol{\psi}^{\mathrm{AP}}\|^2-\|\boldsymbol{\psi}_{m,n}-\boldsymbol{\psi}^{\mathrm{S}}\|^2-\|\boldsymbol{\psi}^{\mathrm{S}}-\boldsymbol{\psi}^{\mathrm{AP}}\|^2}{-2\|\boldsymbol{\psi}_{m,n}-\boldsymbol{\psi}^{\mathrm{S}}\|\|\boldsymbol{\psi}^{\mathrm{S}}-\boldsymbol{\psi}^{\mathrm{AP}}\|}\right). It is worth noting that ϵa\epsilon_a and ϵb\epsilon_b 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

ys(t)=∑m=1M∑n=1Nξm,nss(t−τm,ns)+ns(t).(11)y_{\mathrm{s}}(t)=\sum_{m=1}^{M}\sum_{n=1}^{N}\xi_{m,n}^{\mathrm{s}}s\left(t-\tau_{m,n}^{\mathrm{s}}\right)+n_{\mathrm{s}}(t). \tag*{(11)}

Here, ξm,ns=Ptαm,n[hs,m]n∗[gm]nwm∣Γm,n∣2\xi_{m,n}^{\mathrm{s}}=\sqrt{P_t}\alpha_{m,n}[\mathbf{h}_{\mathrm{s},m}]_n^*[\mathbf{g}_m]_nw_m|\Gamma_{m,n}|^2 is the sensing effective cascaded gain, while ns(t)n_{\mathrm{s}}(t) is the AWGN with power σs2\sigma_{\mathrm{s}}^2 at time slot tt. Moreover, τm,ns\tau_{m,n}^{\mathrm{s}} is the round trip propagation delay associated with the path from the nn-th PA on the mm-th waveguide to the target and then to the AP. The sensing SNR at AP is SNRs=∑m=1M∑n=1N∣ξm,ns∣2σs2\mathrm{SNR}_{\mathrm{s}}=\frac{\sum_{m=1}^{M}\sum_{n=1}^{N}|\xi_{m,n}^{\mathrm{s}}|^2}{\sigma_{\mathrm{s}}^2}.

Stage I: Delay Estimation

The received echo is mixed with the local reference to generate the beat frequency signal, which is

b(t)=ys(t)s∗(t).(12)b(t)=y_{\mathrm{s}}(t)s^*(t). \tag*{(12)}

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 s(t−τm,ns)s(t-\tau_{m,n}^{\mathrm{s}}) into (12), the beat signal can be expanded as

2(xAP−xkp)xs+2(yAP−ykp)ys+2(zAP−d)zs+2rkdAP+mTSoA,k=rk2+(xAP)2+(yAP)2+(zAP)2−(xkp)2−(ykp)2−d2(13)2(x^{\mathrm{AP}}-x_k^p)x^s+2(y^{\mathrm{AP}}-y_k^p)y^s+2(z^{\mathrm{AP}}-d)z^s+2r_kd_{\mathrm{AP}}+m_{\mathrm{TSoA},k}=r_k^2+(x^{\mathrm{AP}})^2+(y^{\mathrm{AP}})^2+(z^{\mathrm{AP}})^2-(x_k^p)^2-(y_k^p)^2-d^2 \tag*{(13)}
b(t)=∑m=1M∑n=1Nξm,nsejϕm,nce−j2πfb,m,ntejϑm,nc+n~(t),(14)b(t)=\sum_{m=1}^{M}\sum_{n=1}^{N}\xi_{m,n}^{s}e^{j\phi_{m,n}^{c}}e^{-j2\pi f_{b,m,n}t}e^{j\vartheta_{m,n}^{c}}+\tilde{n}(t), \tag*{(14)}

where the de-chirped noise is n~(t)=ns(t)s∗(t)\tilde{n}(t)=n_s(t)s^{*}(t) and ϑm,nc≜π((τm,ns)2−2f0τm,ns)\vartheta_{m,n}^{c}\triangleq\pi\left((\tau_{m,n}^{s})^2-2f_0\tau_{m,n}^{s}\right) denotes the delay-dependent constant phase term, and fb,m,n≜uτm,nsf_{b,m,n}\triangleq u\tau_{m,n}^{s} 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 fbf_b. Therefore, performing a fast Fourier transform (FFT) and peak detection on the beat signal still enables robust estimation of fb,m,nf_{b,m,n} and τ^m,n=f^b,m,n/u\hat{\tau}_{m,n}=\hat{f}_{b,m,n}/u. Finally, we obtain the sensing link propagation distance rm,n=τ^m,ncr_{m,n}=\hat{\tau}_{m,n}c.

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 kk to denote the kk-th activated PA. The corresponding TSoA observation can then be expressed as

rk=∥ψs−ψAP∥+∥ψk−ψs∥+nTSoA,k=dkt+dAP+nTSoA,k,(15)r_k=\lVert\boldsymbol{\psi}^{s}-\boldsymbol{\psi}^{\mathrm{AP}}\rVert+\lVert\boldsymbol{\psi}_k-\boldsymbol{\psi}^{s}\rVert+n_{\mathrm{TSoA},k} =d_k^t+d_{\mathrm{AP}}+n_{\mathrm{TSoA},k}, \tag*{(15)}

where dktd_k^t represents the distance between the kk-th activated PA and the sensing target, which is given by

dkt=(xkp−xs)2+(ykp−ys)2+(d−zs)2,d_k^t=\sqrt{(x_k^p-x^s)^2+(y_k^p-y^s)^2+(d-z^s)^2},

while dAPd_{\mathrm{AP}} denotes the distance between the sensing target and the AP, given by

dAP=(xs−xAP)2+(ys−yAP)2+(zs−zAP)2.d_{\mathrm{AP}}=\sqrt{(x^s-x^{\mathrm{AP}})^2+(y^s-y^{\mathrm{AP}})^2+(z^s-z^{\mathrm{AP}})^2}.

Accordingly, the least-squares estimator for the target estimation can be formulated as

ψ^s=arg⁡min⁡ψs∑k=1K(rk−dkt−dAP)2.(16)\hat{\boldsymbol{\psi}}^s=\underset{\boldsymbol{\psi}^s}{\arg\min}\sum_{k=1}^{K}\left(r_k-d_k^t-d_{\mathrm{AP}}\right)^2. \tag*{(16)}

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

rk−dAP=dkt+nTSoA,k.(17)r_k-d_{\mathrm{AP}}=d_k^t+n_{\mathrm{TSoA},k}. \tag*{(17)}

We square both sides to derive a tractable linearized representation as

rk2−2rkdAP+dAP2=(dkt)2+nTSoA,k2+2dktnTSoA,k.(18)r_k^2-2r_kd_{\mathrm{AP}}+d_{\mathrm{AP}}^2=(d_k^t)^2+n_{\mathrm{TSoA},k}^2+2d_k^tn_{\mathrm{TSoA},k}. \tag*{(18)}

For compact notation, we define the noise-induced residual as

mTSoA,k=nTSoA,k2+2dktnTSoA,k.(19)m_{\mathrm{TSoA},k}=n_{\mathrm{TSoA},k}^2+2d_k^tn_{\mathrm{TSoA},k}. \tag*{(19)}

Accordingly, the squared measurement equation is

rk2−2rkdAP+dAP2=(dkt)2+mTSoA,k.(20)r_k^2-2r_kd_{\mathrm{AP}}+d_{\mathrm{AP}}^2=(d_k^t)^2+m_{\mathrm{TSoA},k}. \tag*{(20)}

Then, by substituting the geometric expressions of dAP2d_{\mathrm{AP}}^2 and (dkt)2(d_k^t)^2, and collecting the terms related to the unknown target position, the linearized measurement model in (13), shown at the top of the page. Let θs=[xs,ys,zs,dAP]T∈R4×1\boldsymbol{\theta}^s=[x^s,y^s,z^s,d_{\mathrm{AP}}]^{\mathrm{T}}\in\mathbb{R}^{4\times1} be the augmented unknown vector. Then, the linearized TSoA equation can be written as Lθs+q=b\mathbf{L}\boldsymbol{\theta}^s+\mathbf{q}=\mathbf{b}, where

L=2[xAP−x1pyAP−y1pzAP−dr1⋮⋮⋮⋮xAP−xkpyAP−ykpzAP−drk⋮⋮⋮⋮xAP−xKpyAP−yKpzAP−drK]∈RK×4,(21)\mathbf{L}=2\begin{bmatrix} x^{\mathrm{AP}}-x_1^p & y^{\mathrm{AP}}-y_1^p & z^{\mathrm{AP}}-d & r_1\\ \vdots& \vdots& \vdots& \vdots\\ x^{\mathrm{AP}}-x_k^p & y^{\mathrm{AP}}-y_k^p & z^{\mathrm{AP}}-d & r_k\\ \vdots& \vdots& \vdots& \vdots\\ x^{\mathrm{AP}}-x_K^p & y^{\mathrm{AP}}-y_K^p & z^{\mathrm{AP}}-d & r_K \end{bmatrix}\in\mathbb{R}^{K\times4}, \tag*{(21)}
q=[mTSoA,1,⋯ ,mTSoA,k,⋯ ,mTSoA,K]T∈RK×1,(22)\mathbf{q}=[m_{\mathrm{TSoA},1},\cdots,m_{\mathrm{TSoA},k},\cdots,m_{\mathrm{TSoA},K}]^{\mathrm{T}}\in\mathbb{R}^{K\times1}, \tag*{(22)}
b=[r12+(xAP)2+(yAP)2+(zAP)2−(x1p)2−(y1p)2−d2⋮rk2+(xAP)2+(yAP)2+(zAP)2−(xkp)2−(ykp)2−d2⋮rK2+(xAP)2+(yAP)2+(zAP)2−(xKp)2−(yKp)2−d2]∈RK×1.(23)\mathbf{b}=\begin{bmatrix} r_1^2+(x^{\mathrm{AP}})^2+(y^{\mathrm{AP}})^2+(z^{\mathrm{AP}})^2-(x_1^p)^2-(y_1^p)^2-d^2\\ \vdots\\ r_k^2+(x^{\mathrm{AP}})^2+(y^{\mathrm{AP}})^2+(z^{\mathrm{AP}})^2-(x_k^p)^2-(y_k^p)^2-d^2\\ \vdots\\ r_K^2+(x^{\mathrm{AP}})^2+(y^{\mathrm{AP}})^2+(z^{\mathrm{AP}})^2-(x_K^p)^2-(y_K^p)^2-d^2 \end{bmatrix}\in\mathbb{R}^{K\times1}. \tag*{(23)}

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

θ^s=(LTWsL)−1LTWsb.(24)\hat{\boldsymbol{\theta}}^s=(\mathbf{L}^{\mathrm{T}}\mathbf{W}_s\mathbf{L})^{-1}\mathbf{L}^{\mathrm{T}}\mathbf{W}_s\mathbf{b}. \tag*{(24)}

Here, Ws=diag⁡(ws,1,…,ws,K)\mathbf{W}_s=\operatorname{diag}(w_{s,1},\ldots,w_{s,K}), where ws,kw_{s,k} represents the normalized weight associated with the kk-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 θ^s\hat{\boldsymbol{\theta}}^s.

CRB Analysis

By stacking the observations of all activated PAs into a vector, we have

rTSoA=fTSoA(ψs)+nTSoA.(25)\mathbf{r}_{\mathrm{TSoA}}=\mathbf{f}_{\mathrm{TSoA}}(\boldsymbol{\psi}^s)+\mathbf{n}_{\mathrm{TSoA}}. \tag*{(25)}

Moreover, the noise is assumed to be a zero-mean Gaussian vector with covariance matrix C=diag⁡(σ12,σ22,…,σK2)\mathbf{C}=\operatorname{diag}(\sigma_1^2,\sigma_2^2,\ldots,\sigma_K^2), where the corresponding measurement noise variance is given by [18]: σk2=c28π2f0WbNsγs,k\sigma_k^2=\frac{c^2}{8\pi^2f_0W_bN_s\gamma_{s,k}}, where WbW_b is the effective bandwidth, NsN_s is the number of transmitted symbols and γs,k\gamma_{s,k} denotes the sensing SNR corresponding to the kk-th PA link. In the CRB derivation, σk2\sigma_k^2 is assumed to be known and is evaluated from the measurement stage. Therefore, the probability density function (PDF) of rTSoA\mathbf{r}_{\mathrm{TSoA}} is given by

p(rTSoA)=1(2π)K2∏k=1Kσkexp⁡(−12∑k=1K(rk−dkt−dAP)2σk2).(26)p(\mathbf{r}_{\mathrm{TSoA}})=\frac{1}{(2\pi)^{\frac{K}{2}}\prod_{k=1}^{K}\sigma_k}\exp\left(-\frac{1}{2}\sum_{k=1}^{K}\frac{(r_k-d_k^t-d_{\mathrm{AP}})^2}{\sigma_k^2}\right). \tag*{(26)}

The Fisher information matrix (FIM) is then

J(ψs)=(∂fTSoA∂ψs)TC−1(∂fTSoA∂ψs),(27)\mathbf{J}(\boldsymbol{\psi}^s)=\left(\frac{\partial\mathbf{f}_{\mathrm{TSoA}}}{\partial\boldsymbol{\psi}^s}\right)^{\mathrm{T}}\mathbf{C}^{-1}\left(\frac{\partial\mathbf{f}_{\mathrm{TSoA}}}{\partial\boldsymbol{\psi}^s}\right), \tag*{(27)}

while the entries of J(ψs)\mathbf{J}(\psi^{s}) are

J(ψs)=[JxxJxyJxzJyxJyyJyzJzxJzyJzz],(28)\mathbf{J}(\psi^{s}) = \begin{bmatrix} J_{xx} & J_{xy} & J_{xz} \\ J_{yx} & J_{yy} & J_{yz} \\ J_{zx} & J_{zy} & J_{zz} \end{bmatrix}, \tag*{(28)}
Jxx=∑k=1K1σk2(xs−xkpdkt+xs−xAPdAP)2,(29)J_{xx}=\sum_{k=1}^{K}\frac{1}{\sigma_k^2}\left(\frac{x^{s}-x_k^{p}}{d_k^{t}}+\frac{x^{s}-x^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right)^2, \tag*{(29)}
Jyy=∑k=1K1σk2(ys−ykpdkt+ys−yAPdAP)2,(30)J_{yy}=\sum_{k=1}^{K}\frac{1}{\sigma_k^2}\left(\frac{y^{s}-y_k^{p}}{d_k^{t}}+\frac{y^{s}-y^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right)^2, \tag*{(30)}
Jzz=∑k=1K1σk2(zs−ddkt+zs−zAPdAP)2,(31)J_{zz}=\sum_{k=1}^{K}\frac{1}{\sigma_k^2}\left(\frac{z^{s}-d}{d_k^{t}}+\frac{z^{s}-z^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right)^2, \tag*{(31)}
Jxy=Jyx=∑k=1K1σk2(xs−xkpdkt+xs−xAPdAP)(ys−ykpdkt+ys−yAPdAP),(32)J_{xy}=J_{yx}=\sum_{k=1}^{K}\frac{1}{\sigma_k^2}\left(\frac{x^{s}-x_k^{p}}{d_k^{t}}+\frac{x^{s}-x^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right)\left(\frac{y^{s}-y_k^{p}}{d_k^{t}}+\frac{y^{s}-y^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right), \tag*{(32)}
Jxz=Jzx=∑k=1K1σk2(xs−xkpdkt+xs−xAPdAP)(zs−ddkt+zs−zAPdAP),(33)J_{xz}=J_{zx}=\sum_{k=1}^{K}\frac{1}{\sigma_k^2}\left(\frac{x^{s}-x_k^{p}}{d_k^{t}}+\frac{x^{s}-x^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right)\left(\frac{z^{s}-d}{d_k^{t}}+\frac{z^{s}-z^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right), \tag*{(33)}
Jyz=Jzy=∑k=1K1σk2(ys−ykpdkt+ys−yAPdAP)(zs−ddkt+zs−zAPdAP).(34)J_{yz}=J_{zy}=\sum_{k=1}^{K}\frac{1}{\sigma_k^2}\left(\frac{y^{s}-y_k^{p}}{d_k^{t}}+\frac{y^{s}-y^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right)\left(\frac{z^{s}-d}{d_k^{t}}+\frac{z^{s}-z^{\mathrm{AP}}}{d_{\mathrm{AP}}}\right). \tag*{(34)}

The CRB for estimating the target position is

CRB=J−1(ψs).(35)\mathrm{CRB}=\mathbf{J}^{-1}(\psi^{s}). \tag*{(35)}

Thus, we obtain the corresponding position error bound (PEB):

PEB=Tr⁡(CRB),(36)\mathrm{PEB}=\sqrt{\operatorname{Tr}(\mathrm{CRB})}, \tag*{(36)}

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 RMSE=E{∥ψ^s−ψs∥2}\mathrm{RMSE}=\sqrt{\mathbb{E}\{\lVert\hat{\psi}^{s}-\psi^{s}\rVert^{2}\}}.

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 βm\beta_m to indicate whether the mm-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 rank⁡(L)=4\operatorname{rank}(\mathbf{L})=4.

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 KK PAs are activated. The resulting computational complexity of the HGS algorithm is approximately OHGS≈O(N4+KN2)O_{\mathrm{HGS}}\approx O(N^{4}+KN^{2}), 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 M=3M=3 waveguides, with N=10N=10 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 B=300B=300 MHz. The waveguide length, height, and refractive index are set to L=40L=40 m, d=5d=5 m, and ne=1.4n_e=1.4, respectively, while the waveguides are located at y=[5,10,15]y=[5,10,15] m. The communication and sensing noise powers are set to σc2=−90\sigma_c^2=-90 dBm and σs2=−100\sigma_s^2=-100 dBm, respectively. The weighting factor is set to μ=0.5\mu=0.5. The reflection loss parameters are chosen as ϵa=0.00572\epsilon_a=0.00572 and ϵb=0.2825\epsilon_b=0.2825 [15]. Moreover, all numerical results are averaged over 1,000 random realizations. In

max⁡αm,n,βmμSE+(1−μ)log⁡2(SNRs)(37)\max_{\alpha_{m,n},\beta_m}\quad\mu\mathrm{SE}+(1-\mu)\log_2(\mathrm{SNR}_s) \tag*{(37)}
s.t.∑m=1M∑n=1Nαm,n=K,(38)\text{s.t.}\quad\sum_{m=1}^{M}\sum_{n=1}^{N}\alpha_{m,n}=K, \tag*{(38)}
∑m=1Mβm≥3,(39)\sum_{m=1}^{M}\beta_m\ge3, \tag*{(39)}
αm,n≤βm,∀m,n,(40)\alpha_{m,n}\le\beta_m,\quad\forall m,n, \tag*{(40)}
∑n=1Nαm,n≥βm,∀m,(41)\sum_{n=1}^{N}\alpha_{m,n}\ge\beta_m,\quad\forall m, \tag*{(41)}
αm,n∈{0,1},∀m,n,(42)\alpha_{m,n}\in\{0,1\},\quad\forall m,n, \tag*{(42)}
βm∈{0,1},∀m.(43)\beta_m\in\{0,1\},\quad\forall m. \tag*{(43)}

Algorithm 1 HGS algorithm for solving (39)

  1. Input: KK, NN, MM, hc\mathbf{h}_{\mathrm{c}}, hs\mathbf{h}_{\mathrm{s}}, G\mathbf{G}, μ\mu,

  2. Initialize: α⋆=0\boldsymbol{\alpha}^{\star}=0, the best objective value f⋆f^{\star} as −∞-\infty.

  3. for n1=1n_1=1 to NN do

  4. for n2=1n_2=1 to NN do

  5. for n3=1n_3=1 to NN do

  6. Construct α\boldsymbol{\alpha} by activating PAs (1,n1)(1,n_1), (2,n2)(2,n_2), and (3,n3)(3,n_3)

  7. Evaluate f(α)f(\boldsymbol{\alpha})

  8. if f(α)>f⋆f(\boldsymbol{\alpha})>f^{\star} then

  9. α⋆←α\boldsymbol{\alpha}^{\star}\leftarrow\boldsymbol{\alpha}, f⋆←f(α)f^{\star}\leftarrow f(\boldsymbol{\alpha})

  10. end if

  11. end for

  12. end for

  13. end for

  14. while ∑iαi⋆<K\sum_i \alpha_i^{\star}<K do

  15. floc=−∞f_{\mathrm{loc}}=-\infty, αloc=α⋆\boldsymbol{\alpha}_{\mathrm{loc}}=\boldsymbol{\alpha}^{\star}

  16. for each inactive PA kk do

  17. Activate PA kk temporarily and obtain α\boldsymbol{\alpha}

  18. Evaluate f(α)f(\boldsymbol{\alpha})

  19. if f(α)>flocf(\boldsymbol{\alpha})>f_{\mathrm{loc}} then

  20. floc←f(α)f_{\mathrm{loc}}\leftarrow f(\boldsymbol{\alpha}), αloc←α\boldsymbol{\alpha}_{\mathrm{loc}}\leftarrow\boldsymbol{\alpha}

  21. end if

  22. end for

  23. α⋆←αloc\boldsymbol{\alpha}^{\star}\leftarrow\boldsymbol{\alpha}_{\mathrm{loc}}

  24. end while

  25. Output: α⋆\boldsymbol{\alpha}^{\star}

each realization, the CU is randomly generated within xc∈[5,25]x_{\mathrm{c}}\in[5,25], yc∈[8,18]y_{\mathrm{c}}\in[8,18], and zc=1z_{\mathrm{c}}=1, while the sensing target is randomly generated within xs∈[10,30]x_{\mathrm{s}}\in[10,30], ys∈[5,15]y_{\mathrm{s}}\in[5,15], and zs=2z_{\mathrm{s}}=2. The AP location is fixed at ψAP=[32,8,1]T\boldsymbol{\psi}^{\mathrm{AP}}=[32,8,1]^{T}. We consider a random selection scheme as a benchmark, where KK 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 μ=1\mu=1 in (39), achieves the highest SE, whereas the sensing-only scheme (μ=0\mu=0) yields lower SE due to sensing-oriented PA activation. For the proposed HGS scheme, we set μ=0.5\mu=0.5. For the same KK, the proposed HGS consistently outperforms the benchmark, confirming the effectiveness of the PA activation design. The smaller SE gap between K=10K=10 and K=15K=15 than between K=5K=5 and K=10K=10 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 K=10K=10, 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 μ\mu 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 μ\mu enables a desirable tradeoff.

Figure 3. Average SE and sensing SNR versus the weighting factor μ\mu.

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 KK 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. [1]L. Zhou et al., “Fluid antenna-assisted ISAC systems,” IEEE Wireless Commun. Lett., vol. 13, pp. 3533–3537, Dec. 2024.DOI
  2. [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. [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. [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. [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. [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. [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. [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. [9]D. M. Pozar, Microwave Engineering: Theory and Techniques. John Wiley & Sons, 2021.
  10. [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. [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. [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. [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. [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. [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. [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. [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. [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. [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. [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

Paper details

vlnc:W_egfuwpx7
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