betaFeedback

WorksEngineering

From Choirs to Solos: Reducing Multi-Cell Cooperation via Pinching Antennas

Abstract

This article investigates the potential of the pinching-antenna system (PASS) to reduce multi-cell cooperation requirements. Flexible pinching-antenna (PA) placement can enhance desired channel gains and reduce the required transmit power, which consequently alleviates inter-cell interference and the need for cooperative transmission. A two-cell PASS is considered, where the cooperation cost is quantified by the number of additional base station-user data associations. The cooperation cost is minimized subject to user signal-to-interference-plus-noise ratio requirements and individual transmit-power constraints. Joint PA placement and transmission designs are developed using alternating optimization and second-order cone programming, while low-complexity zero-forcing-based alternatives are also proposed. Numerical results demonstrate that PASS substantially reduces cooperation requirements compared with the conventional fixed-antenna system. These findings highlight the potential of PASS to transform cooperative multi-cell *choirs*'' into independent solos''.

Introduction

The pinching-antenna system (PASS) has recently attracted considerable attention due to its ability to enhance wireless channel gains [1]. PASS employs dielectric waveguides to convey radio-frequency signals and pinching antennas (PAs) attached to the waveguides to radiate these signals into free space [2]. By flexibly adjusting the PA positions along the waveguides, PASS provides a much larger spatial reconfiguration range than conventional reconfigurable antenna systems. In particular, a PA can be placed close to its intended user to establish a favorable line-of-sight (LoS) link and mitigate propagation loss and signal blockage [1, 2].

These advantages have motivated extensive research on PASS, as summarized in recent surveys [3, 4, 5]. Most existing studies have focused on single-cell systems and investigated various physical-layer transmission designs. More recently, researchers have extended PASS to multi-cell networks. For example, stochastic-geometry-based studies have characterized network throughput [6, 7], while other works have investigated joint precoding, PA placement, and base station (BS)-user association [8, 9, 10]. Despite these advances, the potential of PASS to reduce inter-cell cooperation requirements remains largely unexplored.

In conventional fixed-antenna systems, serving cell-edge users often requires high transmit power due to large propagation distances. The resulting inter-cell interference (ICI) further increases the power needed to satisfy users’ quality-of-service (QoS) requirements in neighboring cells. A conventional solution is cooperative transmission, where multiple BSs jointly transmit user data through cooperative precoding.

However, this approach incurs additional overhead associated with user-data sharing, cooperative signal processing, and inter-BS synchronization. PASS offers an alternative through flexible PA placement. By placing PAs close to their intended users, PASS shortens the wireless propagation distances and establishes stronger LoS links [1]. The enhanced desired channel gains reduce the required transmit power [11] and consequently alleviate ICI [12]. Moreover, flexible PA placement can reshape the interference channels between neighboring cells. These advantages motivate an important question: Can PASS reduce or even eliminate the need for cooperative transmission while satisfying users’ QoS requirements?

Motivated by this question, we investigate cooperation reduction in multi-cell PASS through joint PA placement and transmission design. We consider a representative two-cell system, where each cell employs one waveguide with one active PA to serve its associated user. We quantify the cooperation cost by the number of additional BS-user data associations and aim to minimize this cost subject to user signal-to-interference-plus-noise ratio (SINR) and individual BS power constraints. To this end, we investigate three transmission modes: independent transmission, one-sided cooperation, and full cooperation. For each mode, we jointly optimize PA placement and transmit power or cooperative precoding to minimize the required peak transmit power. We develop an alternating optimization (AO) framework with second-order cone programming (SOCP) for the cooperative designs and further propose low-complexity zero-forcing (ZF)-based alternatives. Numerical results demonstrate that PASS substantially reduces cooperation requirements compared with the conventional fixed-antenna system and enables independent per-cell transmission even under stringent QoS requirements. These findings highlight the potential of PASS to transform multi-cell cooperative “choirs” into independent “solos”.

System Model

Consider the two-cell PASS illustrated in Figure 1. Each cell k∈{1,2}k \in\{1,2\} has length DxD_x along the xx-axis and DyD_y along the yy-axis. The center of cell kk is located at (xkC,ykC)(x_k^{\mathrm{C}},y_k^{\mathrm{C}}), and its service region is defined as Ck≜[xkC−Dx2,xkC+Dx2]×[ykC−Dy2,ykC+Dy2]\mathcal{C}_k \triangleq[x_k^{\mathrm{C}}-\frac{D_x}{2},x_k^{\mathrm{C}}+\frac{D_x}{2}] \times[y_k^{\mathrm{C}}-\frac{D_y}{2},y_k^{\mathrm{C}}+\frac{D_y}{2}]. Each BS is equipped with a dielectric waveguide deployed along the xx-axis at height H>0H>0 above the user plane. The waveguide spans the entire cell length, and one PA is activated on each waveguide. Let tk∈[−Dx2,Dx2]≜Tt_k \in[-\frac{D_x}{2},\frac{D_x}{2}] \triangleq\mathcal{T} denote the PA position relative to the center of cell kk. Its three-dimensional location is given by tk(tk)=[xkC+tk,ykC,H]T\mathbf{t}_k(t_k)=[x_k^{\mathrm{C}}+t_k,y_k^{\mathrm{C}},H]^{\mathrm{T}}. User kk is located at uk=[xkU,ykU,0]T\mathbf{u}_k=[x_k^{\mathrm{U}},y_k^{\mathrm{U}},0]^{\mathrm{T}} in cell kk and is primarily associated with BS kk. Both users are simultaneously served over the same time-frequency resource. A central controller has perfect knowledge of the user locations and position-dependent channel coefficients and jointly determines the PA positions and transmission configurations of both cells. The user locations and propagation environment remain unchanged during each scheduling period.

Illustration of a two-cell PASS

Figure 1. Illustration of a two-cell PASS.

Channel Model

PASS is particularly attractive for high-frequency communications [2], where LoS propagation typically dominates [13]. We thus consider LoS propagation between the PAs and users. The free-space channel coefficient from the PA in cell k′k' to user kk is modeled as follows [13]:

hk′,kLoS(tk′)=ηe−jk0∥tk′(tk′)−uk∥∥tk′(tk′)−uk∥,h^{\mathrm{LoS}}_{k',k}(t_{k'}) = \frac{\sqrt{\eta}e^{-\mathrm{j}k_0\lVert\mathbf{t}_{k'}(t_{k'})-\mathbf{u}_k\rVert}}{\lVert\mathbf{t}_{k'}(t_{k'})-\mathbf{u}_k\rVert},

where η≜λ216π2\eta\triangleq\frac{\lambda^2}{16\pi^2}, λ\lambda is the free-space wavelength, and k0≜2πλk_0 \triangleq\frac{2\pi}{\lambda} is the free-space wavenumber. In addition to free-space propagation, the signal propagates along the dielectric waveguide from the feed point to the activated PA. Let rk′Feed\mathbf{r}^{\mathrm{Feed}}_{k'} denote the feed-point location of waveguide k′k'. The corresponding in-waveguide channel coefficient is modeled as follows:

hk′In(tk′)=e−jneffk0∥tk′(tk′)−rk′Feed∥,h^{\mathrm{In}}_{k'}(t_{k'}) = e^{-\mathrm{j}n_{\mathrm{eff}}k_0\lVert\mathbf{t}_{k'}(t_{k'})-\mathbf{r}^{\mathrm{Feed}}_{k'}\rVert},

where neffn_{\mathrm{eff}} is the effective refractive index of the dielectric waveguide. The end-to-end channel coefficient from BS k′k' to user kk is therefore given by hk′,k(tk′)=hk′In(tk′)hk′,kLoS(tk′)h_{k',k}(t_{k'})=h^{\mathrm{In}}_{k'}(t_{k'})h^{\mathrm{LoS}}_{k',k}(t_{k'}). Accordingly, the channel power gain is expressed as follows:

gk′,k(tk′)≜∣hk′,k(tk′)∣2=ηqk′,k+(tk′−ξk′,k)2,g_{k',k}(t_{k'}) \triangleq\left|h_{k',k}(t_{k'})\right|^2 = \frac{\eta}{q_{k',k}+\left(t_{k'}-\xi_{k',k}\right)^2},

where qk′,k≜(ykU−yk′C)2+H2>0q_{k',k}\triangleq(y_k^{\mathrm{U}}-y_{k'}^{\mathrm{C}})^2+H^2>0 and ξk′,k≜xkU−xk′C\xi_{k',k}\triangleq x_k^{\mathrm{U}}-x_{k'}^{\mathrm{C}}.

Cooperative Transmission

Let sks_k denote the unit-power information symbol intended for user kk. When cooperation is enabled, each BS may also transmit the neighboring user’s data, as illustrated in Figure 2. Let wk′,kw_{k',k} denote the digital precoding coefficient applied to sks_k at BS k′k'. The transmitted signal from BS k′k' is given by xk′=∑k=12wk′,kskx_{k'}=\sum_{k=1}^{2}w_{k',k}s_k. In particular, wk′,k=0w_{k',k}=0 for k′≠kk'\ne k indicates that BS k′k' does not transmit user kk’s data. The received signal at user kk is expressed as follows:

Illustration of two-cell cooperation

Figure 2. Illustration of two-cell cooperation.

yk=∑k′=12hk′,k(tk′)∑j=12wk′,jsj+nk,y_k=\sum_{k'=1}^{2}h_{k',k}(t_{k'})\sum_{j=1}^{2}w_{k',j}s_j+n_k,

where nk∼CN(0,σ2)n_k\sim\mathcal{CN}(0,\sigma^2) denotes the additive noise with power σ2\sigma^2. Coherent inter-BS phase synchronization is assumed whenever joint transmission is activated. The SINR of user kk is given by

Γk=∣∑k′=12hk′,k(tk′)wk′,k∣2∑k′=12∣hk′,k(tk′)wk′,j∣2+σ2,\Gamma_k=\frac{\left|\sum_{k'=1}^{2}h_{k',k}(t_{k'})w_{k',k}\right|^2}{\sum_{k'=1}^{2}\left|h_{k',k}(t_{k'})w_{k',j}\right|^2+\sigma^2},

where j≠kj\ne k. In conventional fixed-antenna systems, serving cell-edge users often requires high transmit power, which can cause substantial ICI. Cooperative precoding can mitigate ICI but incurs additional overhead due to inter-BS data sharing, signal processing, and synchronization.

Remark 1. PASS can substantially shorten the wireless propagation distances through flexible PA placement, which enhances the desired channel gains and reduces the required transmit power [11]. The reduced transmit power alleviates ICI [12]. Consequently, PASS may satisfy the same QoS requirements with less cooperative transmission than conventional fixed-antenna systems.

Define the cooperative precoding matrix as follows:

W≜[w1,1w1,2w2,1w2,2]∈C2×2.\mathbf{W}\triangleq\begin{bmatrix}w_{1,1}&w_{1,2}\\w_{2,1}&w_{2,2}\end{bmatrix}\in\mathbb{C}^{2\times2}.

Under independent transmission, each BS transmits only its own user’s data, and W\mathbf{W} is diagonal. When wk′,k≠0w_{k',k}\ne0 for k′≠kk'\ne k, user kk’s data must additionally be shared with BS k′k'. We therefore define the cooperation cost as follows:

C(W)≜1{∣w1,2∣>0}+1{∣w2,1∣>0}∈{0,1,2},C(\mathbf{W})\triangleq\mathbb{1}\{|w_{1,2}|>0\}+\mathbb{1}\{|w_{2,1}|>0\}\in\{0,1,2\},

where 1{⋅}\mathbb{1}\{\cdot\} denotes the indicator function. Moreover, C=0C=0, C=1C=1, and C=2C=2 correspond to independent transmission, one-sided cooperation, and two-sided cooperation, respectively, as illustrated in Figure 2.

Remark 2. The cooperation cost CC quantifies the number of additional cross-cell user-data associations beyond the primary BS-user associations. For an equal scheduled payload of LL bits per user, the associated additional user-data delivery requirement is LCLC bits per scheduling interval. In particular, C=0C=0 eliminates additional cross-cell user-stream delivery and the requirement for coherent joint transmission. The proposed metric therefore captures a concrete component of cooperation-related operational burden.

Problem Formulation

Our objective is to minimize the cooperation cost required to satisfy both users’ QoS requirements under a per-BS transmit power budget Pmax⁡P_{\max}. Let t=[t1,t2]\mathbf{t}=[t_1,t_2] and let γ\gamma denote the common SINR target. The optimization problem can be formulated as follows:

min⁡t,WC(W)(1)\min_{\mathbf{t},\mathbf{W}}\quad C(\mathbf{W}) \tag*{(1)}
s.t.tk∈T, Γk≥γ,k=1,2,(2)\mathrm{s.t.}\quad t_k\in\mathcal{T},\ \Gamma_k\ge\gamma,\quad k=1,2, \tag*{(2)}
∑k′=12∣wk′,k∣2≤Pmax⁡,k=1,2.(3)\sum_{k'=1}^{2}|w_{k',k}|^2\le P_{\max},\quad k=1,2. \tag*{(3)}

To solve problem (1), we consider three cooperation levels c∈{0,1,2}c\in\{0,1,2\} and define the corresponding feasible precoding sets as follows:

W0≜{W:w1,2=w2,1=0},\mathcal{W}_0\triangleq\{\mathbf{W}:w_{1,2}=w_{2,1}=0\},
endsmid=1{ends_mid=1}

W1≜{W:w1,2=0 or w2,1=0}\mathcal{W}_1 \triangleq\{\mathbf{W}: w_{1,2}=0\text{ or }w_{2,1}=0\}, W2≜C2×2\mathcal{W}_2 \triangleq\mathbb{C}^{2\times2}.

For each cooperation level cc, the minimum required peak transmit power is defined as follows:

Qc⋆≜min⁡t,W∈Wc max⁡k=1,2∑k′=12∣wk,k′∣2Q_c^\star\triangleq\min_{\mathbf{t},\mathbf{W}\in\mathcal{W}_c}\ \max_{k=1,2}\sum_{k'=1}^{2}|w_{k,k'}|^2
s.t.W∈Wc, Γk≥γ, tk∈T, k=1,2.\text{s.t.}\quad\mathbf{W}\in\mathcal{W}_c,\ \Gamma_k\ge\gamma,\ t_k\in\mathcal{T},\ k=1,2.

We set Qc⋆=∞Q_c^\star=\infty if the corresponding problem is infeasible. Since W0⊆W1⊆W2\mathcal{W}_0\subseteq\mathcal{W}_1\subseteq\mathcal{W}_2, the minimum required powers satisfy Q2⋆≤Q1⋆≤Q0⋆Q_2^\star\le Q_1^\star\le Q_0^\star. Consequently, the minimum cooperation cost is determined by

C⋆=min⁡{c∈{0,1,2}:Qc⋆≤Pmax⁡},C^\star=\min\{c\in\{0,1,2\}:Q_c^\star\le P_{\max}\},

provided that Q2⋆≤Pmax⁡Q_2^\star\le P_{\max}. Otherwise, the QoS requirements cannot be satisfied even under full cooperation.

Joint PA Placement and Transmission Design

No Cooperation: C = 0

We first consider independent transmission, where w1,2=w2,1=0w_{1,2}=w_{2,1}=0. Let pk=∣wk,k∣2p_k=|w_{k,k}|^2 denote the transmit power of BS kk. For fixed PA positions t\mathbf{t}, the SINR constraints reduce to

p1g1,1(t1)γ≥p2g2,1(t2)+σ2,p2g2,2(t2)γ≥p1g1,2(t1)+σ2.\frac{p_1g_{1,1}(t_1)}{\gamma}\ge p_2g_{2,1}(t_2)+\sigma^2,\qquad\frac{p_2g_{2,2}(t_2)}{\gamma}\ge p_1g_{1,2}(t_1)+\sigma^2.

The componentwise minimum feasible power solution is obtained when both inequalities hold with equality. For notational simplicity, the arguments of the channel power gains are omitted hereafter. Define Δ0≜g1,1g2,2−γ2g1,2g2,1\Delta_0\triangleq g_{1,1}g_{2,2}-\gamma^2g_{1,2}g_{2,1}. When Δ0>0\Delta_0>0, the minimum required transmit powers are

p1⋆(t)=γσ2(g2,2+γg2,1)Δ0,p2⋆(t)=γσ2(g1,1+γg1,2)Δ0.p_1^\star(\mathbf{t})=\frac{\gamma\sigma^2(g_{2,2}+\gamma g_{2,1})}{\Delta_0},\qquad p_2^\star(\mathbf{t})=\frac{\gamma\sigma^2(g_{1,1}+\gamma g_{1,2})}{\Delta_0}.

Otherwise, no finite feasible power solution exists. The minimum required peak transmit power under independent transmission is thus given by Q0⋆=min⁡t1,t2∈TΦ0(t)Q_0^\star=\min_{t_1,t_2\in\mathcal{T}}\Phi_0(\mathbf{t}), where

Φ0(t)≜{max⁡{p1⋆(t),p2⋆(t)},Δ0>0,∞,otherwise.\Phi_0(\mathbf{t})\triangleq \begin{cases} \max\{p_1^\star(\mathbf{t}),p_2^\star(\mathbf{t})\}, & \Delta_0>0,\\ \infty, & \text{otherwise}. \end{cases}

To address the coupling between the two PA positions, we adopt an element-wise AO method. Specifically, each PA position is optimized while the other is fixed:

min⁡tk∈TΦ0(t),k∈{1,2}.\min_{t_k\in\mathcal{T}}\Phi_0(\mathbf{t}),\qquad k\in\{1,2\}.

Since each subproblem involves only one variable over a bounded interval, we employ a one-dimensional search. Specifically, the feasible interval T=[−Dx/2,Dx/2]\mathcal{T}=[-D_x/2,D_x/2] is discretized into a QQ-point grid Q={−Dx/2+(q−1)DxQ−1}q=1Q\mathcal{Q}=\left\{-D_x/2+\frac{(q-1)D_x}{Q-1}\right\}_{q=1}^{Q}. The PA position is then updated as follows:

tk←arg⁡min⁡t∈Q∪{tk} Φ0(t∣tk=t),t_k\leftarrow\underset{t\in\mathcal{Q}\cup\{t_k\}}{\arg\min}\ \Phi_0(\mathbf{t}|_{t_k=t}),

where t∣tk=t\mathbf{t}|_{t_k=t} denotes the PA-position vector with its kkth entry replaced by tt. Candidate positions that do not yield finite feasible power solutions are discarded. The two PA positions are updated sequentially until convergence.

One-Sided Cooperation: C = 1

Under one-sided cooperation, at most one cross-cell data stream is allowed, i.e., either w1,2=0w_{1,2}=0 or w2,1=0w_{2,1}=0. We optimize the two possible cooperation directions separately and select the one with the smaller required peak transmit power. For either direction, the joint design is addressed through AO between digital precoding and PA placement.

Digital Precoding Optimization

For fixed PA positions t\mathbf{t}, the minimum peak-power precoding problem is formulated as follows:

min⁡W,P Ps.t.Γk≥γ, ∑k′=12∣wk,k′∣2≤P, k=1,2,(4)\min_{\mathbf{W},P}\ P\quad\text{s.t.}\quad\Gamma_k\ge\gamma,\ \sum_{k'=1}^{2}|w_{k,k'}|^2\le P,\ k=1,2, \tag*{(4)}

together with the corresponding zero-entry constraint on W\mathbf{W}. By rotating the phase of each user-stream precoder, the SINR constraints can be equivalently expressed as follows:

∣hkHwj∣2+σ2≤ℜ{hkHwk}γ,ℑ{hkHwk}=0, j≠k,\sqrt{|\mathbf{h}_k^{\mathsf{H}}\mathbf{w}_j|^2+\sigma^2}\le\frac{\Re\{\mathbf{h}_k^{\mathsf{H}}\mathbf{w}_k\}}{\sqrt{\gamma}},\qquad\Im\{\mathbf{h}_k^{\mathsf{H}}\mathbf{w}_k\}=0,\ j\ne k,

where hk=[h1,k(t1),h2,k(t2)]T\mathbf{h}_k=[h_{1,k}(t_1),h_{2,k}(t_2)]^{\mathsf{T}} and wk=[w1,k,w2,k]T\mathbf{w}_k=[w_{1,k},w_{2,k}]^{\mathsf{T}}. Thus, the precoding problem can be reformulated as an SOCP and solved optimally for fixed t\mathbf{t} using convex optimization solvers such as CVX.

PA Placement Optimization

For the optimized precoding matrix W\mathbf{W}, define vk=wkp^k=[v1,k,v2,k]T\mathbf{v}_k=\frac{\mathbf{w}_k}{\sqrt{\hat{p}_k}}=[v_{1,k},v_{2,k}]^{\mathsf{T}}, where p^k=∥wk∥2\hat{p}_k=\|\mathbf{w}_k\|^2 denotes the total power allocated to user stream kk. Let V=[v1,v2]\mathbf{V}=[\mathbf{v}_1,\mathbf{v}_2] denote the normalized precoding directions. For fixed V\mathbf{V}, define the effective channel gains g^k′,k≜∑j=12∣vj,k′hj,k(tj)∣2\hat{g}_{k',k}\triangleq\sum_{j=1}^{2}|v_{j,k'}h_{j,k}(t_j)|^2. The SINR constraints then reduce to

p^1g^1,1γ≥p^2g^2,1+σ2,p^2g^2,2γ≥p^1g^1,2+σ2.\frac{\hat{p}_1\hat{g}_{1,1}}{\gamma}\ge\hat{p}_2\hat{g}_{2,1}+\sigma^2,\qquad\frac{\hat{p}_2\hat{g}_{2,2}}{\gamma}\ge\hat{p}_1\hat{g}_{1,2}+\sigma^2.

Define Δ^0≜g^1,1g^2,2−γ2g^1,2g^2,1\hat{\Delta}_0\triangleq\hat{g}_{1,1}\hat{g}_{2,2}-\gamma^2\hat{g}_{1,2}\hat{g}_{2,1}. For Δ^0>0\hat{\Delta}_0>0, the minimum required stream powers are given by

p^1⋆=γσ2(g^2,2+γg^2,1)Δ^0,p^2⋆=γσ2(g^1,1+γg^1,2)Δ^0.(5)\hat{p}_1^\star=\frac{\gamma\sigma^2(\hat{g}_{2,2}+\gamma\hat{g}_{2,1})}{\hat{\Delta}_0},\qquad\hat{p}_2^\star=\frac{\gamma\sigma^2(\hat{g}_{1,1}+\gamma\hat{g}_{1,2})}{\hat{\Delta}_0}. \tag*{(5)}

Accordingly, for fixed precoding directions V\mathbf{V}, the minimum required peak BS power is expressed as follows:

Φ(t;V)=max⁡k=1,2∑k′=12p^k⋆∣vk,k′∣2,\Phi(\mathbf{t};\mathbf{V})=\max_{k=1,2}\sum_{k'=1}^{2}\hat{p}_k^\star|v_{k,k'}|^2,

where Φ(t;V)=∞\Phi(\mathbf{t};\mathbf{V})=\infty if Δ^0≤0\hat{\Delta}_0\le0. We then optimize the two PA positions sequentially to minimize Φ(t;V)\Phi(\mathbf{t};\mathbf{V}) using the element-wise AO and one-dimensional search described in Section III-A. After convergence, the stream powers are updated by (3), and the precoding matrix is reconstructed as W=Vdiag⁡(p^1⋆,p^2⋆)\mathbf{W}=\mathbf{V}\operatorname{diag}(\sqrt{\hat{p}_1^\star},\sqrt{\hat{p}_2^\star}). The digital precoding and PA placement updates are repeated until convergence. The procedure is performed for both cooperation directions, and the solution with the smaller required peak power is selected.

Full Cooperation: C = 2

Under full cooperation, both BSs may transmit both users’ data streams, and the precoding matrix W∈C2×2\mathbf{W}\in\mathbb{C}^{2\times2} is unrestricted. The AO algorithm developed for one-sided cooperation directly applies to this case, except that the zero-entry constraint on W\mathbf{W} is removed from (2).

ZF-Based Low-Complexity Design

To reduce the computational complexity of the SOCP-based design, we further develop a ZF-based approach that admits explicit peak-power expressions. Define the channel matrix

H(t)≜[abcd]=[h1,1(t1)h2,1(t2)h1,2(t1)h2,2(t2)]\mathbf{H}(t) \triangleq\begin{bmatrix} a & b \\ c & d \end{bmatrix} = \begin{bmatrix} h_{1,1}(t_1) & h_{2,1}(t_2) \\ h_{1,2}(t_1) & h_{2,2}(t_2) \end{bmatrix}

where Δ≜ad−bc\Delta\triangleq ad-bc and ζ≜γσ2\zeta\triangleq\sqrt{\gamma\sigma^2}. We consider Δ≠0\Delta\ne0 in the following ZF designs.

One-Sided Cooperation

We first consider w1,2=0w_{1,2}=0, where BS 2 additionally transmits user 1’s data. By imposing cw1,1+dw2,1=0cw_{1,1}+dw_{2,1}=0, the interference caused by user 1’s stream at user 2 is eliminated. The resulting partial-ZF precoder is

WL=[du/Δ0−cu/Δζ/d],\mathbf{W}_{\mathrm{L}}=\begin{bmatrix} du/\Delta& 0 \\ -cu/\Delta& \zeta/d \end{bmatrix},

where u=ζ1+γ∣b/d∣2u=\zeta\sqrt{1+\gamma\lvert b/d\rvert^2}. It follows that

HWL=[ubζ/d0ζ].\mathbf{H}\mathbf{W}_{\mathrm{L}}=\begin{bmatrix} u & b\zeta/d \\ 0 & \zeta\end{bmatrix}.

Thus, both users achieve the target SINR γ\gamma, and the required peak BS power can be written as follows:

ΦL(t)=γσ2∣Δ∣2max⁡{AL,∣c∣2AL+∣Δ∣2∣d∣2},\Phi_{\mathrm{L}}(t)=\frac{\gamma\sigma^2}{\lvert\Delta\rvert^2}\max\left\{A_{\mathrm{L}},\frac{\lvert c\rvert^2A_{\mathrm{L}}+\lvert\Delta\rvert^2}{\lvert d\rvert^2}\right\},

where AL≜∣d∣2+γ∣b∣2A_{\mathrm{L}}\triangleq\lvert d\rvert^2+\gamma\lvert b\rvert^2.

Similarly, for the opposite cooperation direction w2,1=0w_{2,1}=0, the partial-ZF precoder is

WU=[ζ/a−bv/Δ0av/Δ],\mathbf{W}_{\mathrm{U}}=\begin{bmatrix} \zeta/a & -bv/\Delta\\ 0 & av/\Delta\end{bmatrix},

where v=ζ1+γ∣c/a∣2v=\zeta\sqrt{1+\gamma\lvert c/a\rvert^2}. The corresponding required peak BS power can be written as follows:

ΦU(t)=γσ2∣Δ∣2max⁡{∣b∣2AU+∣Δ∣2∣a∣2,AU},\Phi_{\mathrm{U}}(t)=\frac{\gamma\sigma^2}{\lvert\Delta\rvert^2}\max\left\{\frac{\lvert b\rvert^2A_{\mathrm{U}}+\lvert\Delta\rvert^2}{\lvert a\rvert^2},A_{\mathrm{U}}\right\},

where AU≜∣a∣2+γ∣c∣2A_{\mathrm{U}}\triangleq\lvert a\rvert^2+\gamma\lvert c\rvert^2. Both partial-ZF constructions achieve the target SINRs with equality.

The minimum required peak power under one-sided partial-ZF transmission is therefore

Q1,pZF∗=min⁡t1,t2∈Tmin⁡{ΦL(t),ΦU(t)}.Q_{1,\mathrm{pZF}}^*=\min_{t_1,t_2\in\mathcal{T}}\min\{\Phi_{\mathrm{L}}(t),\Phi_{\mathrm{U}}(t)\}.

We optimize the two cooperation directions separately and select the better solution.

Full Cooperation

Under full cooperation, conventional ZF eliminates the inter-user interference at both users. The corresponding precoding matrix is given by

WF=ζH−1=ζΔ[d−b−ca].\mathbf{W}_{\mathrm{F}}=\zeta\mathbf{H}^{-1}=\frac{\zeta}{\Delta}\begin{bmatrix} d & -b \\ -c & a \end{bmatrix}.

Since HWF=ζI2\mathbf{H}\mathbf{W}_{\mathrm{F}}=\zeta\mathbf{I}_2, both users achieve the target SINR γ\gamma. The required peak BS power is given by

ΦF(t)=γσ2∣Δ∣2max⁡{∣d∣2+∣b∣2,∣c∣2+∣a∣2}.\Phi_{\mathrm{F}}(t)=\frac{\gamma\sigma^2}{\lvert\Delta\rvert^2}\max\{\lvert d\rvert^2+\lvert b\rvert^2,\lvert c\rvert^2+\lvert a\rvert^2\}.

The corresponding PA-placement problem is

Q2,ZF∗=min⁡t1,t2∈TΦF(t).Q_{2,\mathrm{ZF}}^*=\min_{t_1,t_2\in\mathcal{T}}\Phi_{\mathrm{F}}(t).

PA Placement Optimization

The above designs yield three explicit peak-power objectives: ΦL(t)\Phi_{\mathrm{L}}(t), ΦU(t)\Phi_{\mathrm{U}}(t), and ΦF(t)\Phi_{\mathrm{F}}(t). Each objective depends only on the two PA positions. We therefore apply the element-wise AO method developed in Section III-A. Specifically, for ν∈{L,U,F}\nu\in\{\mathrm{L},\mathrm{U},\mathrm{F}\}, the PA positions are updated sequentially as

tk←arg⁡min⁡t∈Q∪{tk}Φν(t∣tk=t).t_k\leftarrow\mathop{\arg\min}_{t\in\mathcal{Q}\cup\{t_k\}}\Phi_\nu(t\mid t_k=t).

Candidates with Δ=0\Delta=0 are excluded. Since the current PA position is included in each search, the objective value is non-increasing over successive updates.

Remark 3. Unlike the SOCP-based design, the proposed ZF-based approach only requires evaluations of closed-form power expressions during PA placement. However, the additional interference-nulling constraints may increase the required transmit power. Therefore, the ZF-based solutions provide low-complexity feasible designs rather than globally optimal solutions to the original problem.

Computational Complexity Analysis

Let I0I_0, IAI_{\mathrm{A}}, and IZI_{\mathrm{Z}} denote the numbers of AO iterations for independent transmission, SOCP-based cooperation, and ZF-based cooperation, respectively. Their computational complexities are O(I0Q)\mathcal{O}(I_0Q), O(IA(CSOCP+IPQ))\mathcal{O}(I_{\mathrm{A}}(C_{\mathrm{SOCP}}+I_{\mathrm{P}}Q)), and O(IZQ)\mathcal{O}(I_{\mathrm{Z}}Q), respectively, where IPI_{\mathrm{P}} is the number of inner PA-position AO iterations and CSOCP=O(n3log⁡(1/ϵ))C_{\mathrm{SOCP}}=\mathcal{O}(n^3\log(1/\epsilon)) is an interior-point complexity estimate for one SOCP, with n=7n=7 and n=9n=9 real optimization variables for one-sided and full cooperation, respectively, and ϵ\epsilon the solution accuracy. Thus, both cooperative designs scale linearly with QQ, while the ZF-based design avoids the additional SOCP complexity.

Numerical Results

In this section, we evaluate the cooperation cost of PASS and compare it with the conventional fixed-antenna system, where each antenna is deployed at the cell center. Unless otherwise stated, the carrier frequency is fc=28f_{\mathrm{c}}=28 GHz, the noise power is σ2=−90\sigma^2=-90 dBm, the effective refractive index is neff=1.4n_{\mathrm{eff}}=1.4, and the waveguide height is H=3H=3 m. Two adjacent cells are considered, each with dimensions Dx=80D_x=80 m and Dy=20D_y=20 m and one uniformly distributed user, as illustrated in Figure 1. The proposed SOCP-based design is initialized using the ZF-based solutions. Each one-dimensional PA search employs a uniform grid of Q=2001Q=2001 points. The target rate is Rth=log⁡2(1+γ)R_{\mathrm{th}}=\log_2(1+\gamma), where Gaussian signaling is assumed and interference is treated as noise. The power budget Pmax⁡P_{\max} is set to the 60th percentile of the minimum peak transmit power required by the conventional fixed-antenna system with ZF-based full cooperation. This value is estimated using 10510^5 Monte-Carlo realizations and recalculated as DxD_x or RthR_{\mathrm{th}} varies. The cooperation cost is capped at C∗=2C^*=2 for all realizations where the SINR requirements cannot be satisfied.

Figure 3(a) compares the average cooperation cost of the conventional fixed-antenna system, nearest-PA placement, and optimized PA placement under SOCP-based and ZF-based precoding. Here, nearest-PA placement positions each PA at the projection of its intended user onto the corresponding waveguide, i.e., tk=xkU−xkCt_k=x_k^{\mathrm{U}}-x_k^{\mathrm{C}} for k∈{1,2}k\in\{1,2\}. All four PASS schemes achieve substantially lower cooperation costs than their conventional counterparts. This improvement is primarily attributed to the spatial flexibility of PASS, which enhances the desired channels and alleviates ICI. The proposed optimized PA placement further outperforms nearest-PA placement, particularly for small cell lengths. For example, at Dx=20D_x = 20 m, the average cooperation cost is reduced from 0.69 to 0.57 under SOCP-based precoding. As DxD_x increases, the cooperation costs decrease for all schemes, while the performance gap between the two PA placement strategies becomes negligible. Furthermore, ZF-based precoding achieves performance close to that of SOCP-based precoding, which demonstrates its effectiveness for cooperation reduction. Figure 3(b) shows the average cooperation cost versus the target rate RthR_{\mathrm{th}}. As expected, the cooperation cost increases with RthR_{\mathrm{th}}, since higher SINR requirements make independent transmission more difficult. Nevertheless, PASS consistently outperforms the conventional fixed-antenna system. Remarkably, even at Rth=6R_{\mathrm{th}} = 6 bps/Hz, the proposed PASS achieves an average cooperation cost below 0.2, compared with approximately 1.6 for the conventional fixed-antenna system. The relatively small difference between nearest-PA and optimized PA placement further indicates that simple geometry-based PA placement can capture most of the cooperation reduction benefits. This highlights the practicality of PASS, as substantial cooperation savings can be achieved without sophisticated PA placement optimization.

Average cooperation cost

Figure 3. Average cooperation cost.

To further characterize the potential of PASS for independent transmission, Figure 4 presents the probability of zero cooperation, Pr⁡(C=0)\Pr(C = 0). As shown in Figure 4(a), this probability increases with DxD_x for all schemes, but PASS achieves a substantially higher probability than the conventional fixed-antenna system. Figure 4(b) shows that Pr⁡(C=0)\Pr(C = 0) decreases with RthR_{\mathrm{th}} due to the increasingly stringent SINR requirements. Nevertheless, the proposed PASS maintains a probability above 0.8 even at Rth=6R_{\mathrm{th}} = 6 bps/Hz, whereas the conventional fixed-antenna system achieves less than 0.1. These results demonstrate that PASS can substantially reduce inter-BS data sharing and enable a transition from cooperative transmission

Probability of zero cooperation

Figure 4. Probability of zero cooperation Pr⁡(C=0)\Pr(C = 0).

to independent cell operation.

Conclusion

This article has demonstrated that flexible PA placement can substantially reduce multi-cell cooperation requirements and transform cooperative “choirs” into independent “solos”. Moreover, nearest-PA placement achieves performance close to that of the proposed joint design, which makes it an attractive low-complexity alternative.

References

References

  1. [1]Y. Liu, Z. Wang, X. Mu, C. Ouyang, X. Xu, and Z. Ding, “Pinching antenna systems (PASS): Architectures, opportunities, and outlook,” IEEE Commun. Mag., vol. 64, no. 1, pp. 190–196, Jan. 2026.arxiv.org/abs/2501.18409
  2. [2]A. Fukuda, H. Yamamoto, H. Okazaki, Y. Suzuki, and K. Kawai, “Pinching antenna: Using a dielectric waveguide as an antenna,” NTT DOCOMO Technical J., vol. 23, no. 3, pp. 5–12, Jan. 2022.
  3. [3]Y. Liu, H. Jiang, X. Xu, Z. Wang, J. Guo, C. Ouyang, X. Mu, Z. Ding, A. Nallanathan, G. K. Karagiannidis et al., “Pinching-antenna systems (PASS): A tutorial,” IEEE Trans. Commun., vol. 74, pp. 4881–4918, 2026.DOI
  4. [4]Y. Lu, W. Mao, X. Xie, S. Liao, X. Xu, Y. Lin, R. Zhang, B. Ai, X. Wang, and Z. Ding, “A survey on pinching-antenna systems: Optimization techniques, intelligent designs, and multifunctional applications,” IEEE Commun. Surveys Tuts., vol. 28, pp. 6592–6627, 2026.DOI
  5. [5]Y. Liu, H. Jiang, X. Gan, X. Xu, J. Guo, Z. Wang, C. Ouyang, X. Mu, Z. Ding, A. Nallanathan, O. A. Dobre, G. K. Karagiannidis, and R. Schober, “A survey on pinching-antenna systems (PASS),” IEEE Commun. Surveys Tuts., vol. 28, pp. 7250–7280, 2026.arxiv.org/abs/2601.18927
  6. [6]Y. Zhu, Z. Ding, and X. You, “Topological perspective of large-scale multi-cell deployment of excitable waveguide dielectrics,” IEEE Wireless Commun. Lett., vol. 15, pp. 151–155, 2026.DOI
  7. [7]Y. Sun, Z. Ding, and G. K. Karagiannidis, “A stochastic geometric analysis on multi-cell pinching-antenna systems under blockage effect,” IEEE Wireless Commun. Lett., vol. 15, pp. 1085–1089, 2026.arxiv.org/abs/2510.06972
  8. [8]Y. Chen, Q. Xue, M. Hua, B. Zhou, and S. Ma, “Pinching-antenna enabled multicell wireless systems,” IEEE Trans. Wireless Commun., vol. 25, pp. 24226–24240, 2026.DOI
  9. [9]H. Li, J. Lei, C. Zeng, Z. Hu, C. Dong, and Y. Liu, “Pinching antenna systems (PASS) for cell-free communications,” IEEE Trans. Veh. Tech., Early Access, 2026.arxiv.org/abs/2510.03628
  10. [10]Y. Jiang, J. Zhang, Z. Liu, C. Ouyang, J. Zheng, Y. Liu, A. Nallanathan, and B. Ai, “Pinching antennas-enabled cell-free massive MIMO: Two-stage AP–UE association and pinching beamforming optimization,” IEEE Trans. Veh. Tech., Early Access, 2026.DOI
  11. [11]Z. Ding, “Toward a quiet wireless world: Multi-cell pinching-antenna transmission,” IEEE Wireless Commun. Lett., vol. 15, pp. 3791–3795, 2026.
  12. [12]——, “Characterizing multi-cell pinching-antenna transmission: Revealing the other side of the coin,” arXiv preprint arXiv:2609.09585, 2026.arxiv.org/abs/2609.09585
  13. [13]C. Ouyang, Z. Wang, Y. Chen, X. Mu, and P. Zhu, “Cooperation for near-field communications for next-generation multiple access,” Proc. IEEE, vol. 112, no. 9, pp. 1527–1565, Sep. 2024.

Paper details

vlnc:W_xgc8x9ej
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

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