betaFeedback

WorksEngineering

Electromagnetically Consistent Monostatic ISAC Beamforming: Mutual Coupling, Active Mismatch and Tx-Rx Leakage in Compact Dipole Arrays

Abstract

Beamforming for integrated sensing and communication (ISAC) is usually designed on ideal array manifolds with a norm constraint on the transmit signal. In a compact monostatic transceiver this hides three antenna effects: distorted embedded patterns, beam-dependent active mismatch at the power amplifiers, and transmit-to-receive leakage that raises the radar noise floor. We build a multiport scattering model of a shared-aperture dipole array directly from method-of-moments simulations and use it to formulate a Cramer-Rao bound minimizing ISAC beamformer with multiuser SINR constraints, per-port active reflection constraints and a leakage budget, solved as a semidefinite program with a one-dimensional budget search. Full-wave evaluation shows that coupling-unaware designs miss their SINR targets by up to 6.2 dB and drive amplifiers to an equivalent VSWR above 3.3 at small spacings. For spacings from 0.3 to 0.5 wavelengths, the proposed design meets all targets with an equivalent VSWR below 2 and lowers the root Cramer-Rao bound by 32 to 53 percent for 80 dB of self-interference cancellation and by a factor of 3.4 to 5.6 for 70 dB. An independent finite-difference time-domain model of the final layout in openEMS confirms that the leakage null and the ranking of the designs are preserved.

Introduction

Integrated sensing and communication (ISAC) is a defining feature of 6G, where a base station uses the same spectrum, waveform and hardware to serve users and to sense its surroundings [1], [2]. Multi-antenna transmit beamforming is the main tool for trading off the two functions, and a mature literature designs it to satisfy communication SINR constraints while shaping a radar beampattern [3] or minimizing the Cramér–Rao bound (CRB) of target parameters [4].

Almost all of these designs share an electromagnetic abstraction: each antenna is an isotropic point with an ideal steering phase, the antennas do not interact, and the transmit power is ∥x∥2\lVert\mathbf{x}\rVert^2. For half-wavelength arrays this is a reasonable first approximation. Three trends make it fragile. First, compact and dense arrays are attractive for FR3 and handheld or vehicular ISAC nodes, and mutual coupling (MC) grows rapidly when the spacing drops below λ/2\lambda/2 [5], [6]. Second, a coupled array is a multiport network: the impedance seen by each power amplifier (PA) depends on the excitations of all other ports [7], so a beamformer that ignores this can push PAs into strong mismatch. Third, a monostatic ISAC transceiver receives while it transmits, and the Tx–Rx leakage, which in a compact shared aperture is dominated by near-field coupling between neighbouring elements, must be suppressed by 100 dB or more to reach the noise floor [8], [9].

Recent works have started to bring MC into ISAC. Robust beamforming against residual MC uncertainty in patch arrays is studied in [10], and MC in reconfigurable holographic surfaces is modelled with coupled dipole approximations to suppress sidelobes in [11]. In metasurface antennas, strong MC has been shown to increase the control over the radiation pattern [12]. These works focus on the radiated pattern. To the best of our knowledge, the joint effect of MC on the radiated pattern, on the active mismatch presented to the PAs and on the Tx–Rx leakage that sets the sensing noise floor has not been addressed in ISAC beamforming, nor validated with full-wave simulation.

This paper closes that gap with a hardware-grounded approach. Our contributions are:

  • We build a port-level scattering model of a shared-aperture Tx/Rx dipole array directly from NEC-2 method-of-moments simulations [13], giving realized embedded patterns, active reflections and Tx–Rx leakage as linear functions of the PA waves. We validate it against direct steering phase, the antennas do not interact, and the transmit power is ∥x∥2\lVert\mathbf{x}\rVert^2. For half-wavelength arrays this is a reasonable first approximation. Three trends make it fragile. First, compact and dense arrays are attractive for FR3 and handheld or vehicular ISAC nodes, and mutual coupling (MC) grows rapidly when the spacing drops below λ/2\lambda/2 [5], [6]. Second, a coupled array is a multiport network: the impedance seen by each power amplifier (PA) depends on the excitations of all other ports [7], so a beamformer that ignores this can push PAs into strong mismatch. Third, a monostatic ISAC transceiver receives while it transmits, and the Tx–Rx leakage, which in a compact shared aperture is dominated by near-field coupling between neighbouring elements, must be suppressed by 100 dB or more to reach the noise floor [8], [9].

Recent works have started to bring MC into ISAC. Robust beamforming against residual MC uncertainty in patch arrays is studied in [10], and MC in reconfigurable holographic surfaces is modelled with coupled dipole approximations to suppress sidelobes in [11]. In metasurface antennas, strong MC has been shown to increase the control over the radiation pattern [12]. These works focus on the radiated pattern. To the best of our knowledge, the joint effect of MC on the radiated pattern, on the active mismatch presented to the PAs and on the Tx–Rx leakage that sets the sensing noise floor has not been addressed in ISAC beamforming, nor validated with full-wave simulation.

This paper closes that gap with a hardware-grounded approach. Our contributions are:

  • We build a port-level scattering model of a shared-aperture Tx/Rx dipole array directly from NEC-2 method-of-moments simulations [13], giving realized embedded patterns, active reflections and Tx–Rx leakage as linear functions of the PA waves. We validate it against direct

full-wave runs.

  • We formulate a CRB-minimizing multiuser ISAC beamformer that, in addition to SINR and power constraints, limits the per-port active reflection (VSWR) and the Tx–Rx leakage, and we link the leakage to the radar noise floor through a self-interference cancellation (SIC) budget. The problem is a semidefinite program (SDP) with an outer one-dimensional search.

  • Full-wave evaluation over element spacings 0.2λ0.2\lambda to 0.5λ0.5\lambda quantifies the SINR loss, PA mismatch and sensing loss of coupling-unaware and pattern-only coupling-aware designs, and the gains of the proposed design.

  • We rebuild the final layout as strip dipoles in an independent FDTD solver (openEMS) and show that the leakage null and the ranking of the designs survive the change of solver and feed model, while the active reflection is the most model-sensitive quantity.

Notation: Bold lower (upper) case letters denote vectors (matrices); (⋅)T(\cdot)^{\mathsf T}, (⋅)H(\cdot)^{H} and (⋅)∗(\cdot)^{*} denote transpose, Hermitian transpose and conjugate; [⋅]nn[\cdot]_{nn} is the nn-th diagonal entry; f˙=∂f/∂θ\dot{f}=\partial f/\partial\theta.

Electromagnetic System Model

Multiport Model from Full-Wave Simulation

Consider the transceiver of Figure 1: N=NT+NRN=N_{\mathrm T}+N_{\mathrm R} identical zz-directed thin-wire dipoles, centre-fed, placed along the xx axis with spacing dd. The first NTN_{\mathrm T} ports are driven by PAs and the last NRN_{\mathrm R} ports feed low-noise amplifiers (LNAs); all are referenced to Z0Z_0. We focus on the azimuth plane θel=90∘\theta_{\mathrm{el}}=90^\circ and denote azimuth by ϕ\phi, measured from the array axis.

Monostatic ISAC transceiver with a shared-aperture collinear array of z-directed dipoles

Figure 1. Monostatic ISAC transceiver with a shared-aperture collinear array of zz-directed dipoles. The Tx subarray serves KK users and illuminates a target; the Rx subarray receives the echo together with near-field Tx–Rx leakage. Azimuth ϕ\phi is measured from the array (xx) axis.

We characterize the array by NN method-of-moments runs in NEC-2 [13]. In run mm, port mm is driven by a 1 V gap source and all other ports are short-circuited. The feed currents of all ports give the mm-th column of the short-circuit admittance matrix YY, and the far field gives the mm-th column of Gsc(ϕ)∈C1×NG_{\mathrm{sc}}(\phi)\in\mathbb{C}^{1\times N}, the field per unit port voltage. By linearity, for any port state, Z=Y−1Z=Y^{-1}, v=Ziv=Zi, and the far field is Gsc(ϕ)vG_{\mathrm{sc}}(\phi)v. With the power-wave definitions, the scattering matrix is

S=(Z−Z0I)(Z+Z0I)−1=[STTSTRSRTSRR].(1)S=(Z-Z_0I)(Z+Z_0I)^{-1} = \begin{bmatrix} S_{\mathrm{TT}} & S_{\mathrm{TR}}\\ S_{\mathrm{RT}} & S_{\mathrm{RR}} \end{bmatrix}. \tag*{(1)}

Let x∈CNTx\in\mathbb{C}^{N_{\mathrm T}} collect the forward waves launched by the PAs, scaled so that ∥x∥2\lVert x\rVert^2 is the incident power, while the Rx ports are matched (zero incident wave). The port currents are then i=(Z+Z0I)−1[2Z0xT,0T]Ti=(Z+Z_0I)^{-1}[\sqrt{2\sqrt{Z_0}x^{\mathsf T}},0^{\mathsf T}]^{\mathsf T}, and the radiation intensity U(ϕ)U(\phi) satisfies 4πU(ϕ)=∣u(ϕ)x∣24\pi U(\phi)=\lvert u(\phi)x\rvert^2 with the realized embedded pattern row vector

u(ϕ)=κ[Gsc(ϕ)Z(Z+Z0I)−1]1:NT,κ=4Z0η0,(2)u(\phi)=\kappa\left[G_{\mathrm{sc}}(\phi)Z(Z+Z_0I)^{-1}\right]_{1:N_{\mathrm T}},\qquad \kappa=4\sqrt{\frac{Z_0}{\eta_0}}, \tag*{(2)}

where η0\eta_0 is the free-space impedance. Thus ∣u(ϕ)x∣2/∥x∥2\lvert u(\phi)x\rvert^2/\lVert x\rVert^2 is the realized gain, which includes coupling, mismatch and absorption in the Rx loads. By reciprocity, the receive manifold v(ϕ)∈CNRv(\phi)\in\mathbb{C}^{N_{\mathrm R}} is given by the last NRN_{\mathrm R} columns of the same expression.

The model exposes three hardware quantities that are linear in xx:

  • Active reflection. The wave reflected back into PA nn is [STTx]n[S_{\mathrm{TT}}x]_n, so the active reflection coefficient is Γn=[STTx]n/xn\Gamma_n=[S_{\mathrm{TT}}x]_n/x_n [7]. It depends on the whole beamformer, not only on the element.

  • Tx–Rx leakage. The wave delivered to the LNAs is SRTxS_{\mathrm{RT}}x, with power PL=∥SRTx∥2P_{\mathrm L}=\lVert S_{\mathrm{RT}}x\rVert^2.

  • Radiated power For lossless wires, Prad=∥x∥2−∥STTx∥2−∥SRTx∥2P_{\mathrm{rad}}=\lVert x\rVert^2-\lVert S_{\mathrm{TT}}x\rVert^2-\lVert S_{\mathrm{RT}}x\rVert^2, which differs from ∥x∥2\lVert x\rVert^2.

The ideal (coupling-unaware) model used by conventional designs instead takes uid(ϕ)=g0(ϕ)[ejkx1cos⁡ϕ,…,ejkxNTcos⁡ϕ]u_{\mathrm{id}}(\phi)=g_0(\phi)[e^{\mathrm{j}kx_1\cos\phi},\ldots,e^{\mathrm{j}kx_{N_{\mathrm T}}\cos\phi}], with g0g_0 the realized pattern of an isolated matched dipole and xnx_n the element positions, and implicitly assumes S=0S=0.

Communication and Sensing Signals

The transmitted waves are x=∑k=1Kwksk+xsx=\sum_{k=1}^{K}w_ks_k+x_{\mathrm s}, where sks_k are unit-power user symbols and xsx_{\mathrm s} is a dedicated sensing signal with covariance RsR_{\mathrm s}. The total covariance is R=∑kWk+RsR=\sum_kW_k+R_{\mathrm s}, Wk=wkwkHW_k=w_kw_k^H, and tr⁡(R)\operatorname{tr}(R) is the total PA output power. User kk at azimuth ϕk\phi_k receives over a line-of-sight channel hk=βkv(ϕk)h_k=\sqrt{\beta_k}v(\phi_k), where βk\beta_k is the path gain, so that

SINR⁡k=hkWkhkH∑j≠khkWjhkH+hkRshkH+σ2.(3)\operatorname{SINR}_k= \frac{h_kW_kh_k^H} {\sum_{j\ne k}h_kW_jh_k^H+h_kR_{\mathrm s}h_k^H+\sigma^2}. \tag*{(3)}

For sensing, a point target at azimuth θ\theta produces at the Rx ports, over LL snapshots,

Ys=αA(θ)X+NSI+N,A(θ)=v(θ)u(θ),(4)Y_{\mathrm s}=\alpha A(\theta)X+N_{\mathrm{SI}}+N,\qquad A(\theta)=v(\theta)u(\theta), \tag*{(4)}

where ∣α∣2=λ2σRCS/((4π)3r4)\lvert\alpha\rvert^2=\lambda^2\sigma_{\mathrm{RCS}}/((4\pi)^3r^4) follows the radar equation and NN is thermal noise of power σ2\sigma^2 per port. The leaked signal SRTxS_{\mathrm{RT}}x is first reduced by analog and digital SIC by CC dB [8]. We model the residual NSIN_{\mathrm{SI}} as white with power 10−C/10PL/NR10^{-C/10}P_{\mathrm L}/N_{\mathrm R} per port, so that the effective noise power is

σR2(R)=σ2+10−C/10tr⁡(SRTRSRTH)NR.(5)\sigma_{\mathrm R}^2(R)=\sigma^2+10^{-C/10}\frac{\operatorname{tr}(S_{\mathrm{RT}}RS_{\mathrm{RT}}^H)}{N_{\mathrm R}}. \tag*{(5)}

For large LL and sample covariance RR, the CRB of θ\theta is [4]

CRB⁡(θ)=σR2(R)2∣α∣2L1tr⁡(A˙HA˙R)−∣tr⁡(A˙HAR)∣2tr⁡(AHAR).(6)\operatorname{CRB}(\theta)= \frac{\sigma_{\mathrm R}^2(R)}{2\lvert\alpha\rvert^2L} \frac{1}{\operatorname{tr}(\dot A^H\dot A R)-\frac{\lvert\operatorname{tr}(\dot A^H A R)\rvert^2}{\operatorname{tr}(A^H A R)}}. \tag*{(6)}

with A˙=v˙u+vu˙\dot A=\dot v u+v\dot u. In (6), coupling enters through uu, vv and their derivatives, and leakage enters through σR2\sigma_{\mathrm R}^2.

Coupling-, Mismatch- and Leakage-Aware Beamforming

We minimize the CRB subject to the per-user SINR target γ\gamma, the PA power budget PTP_{\mathrm{T}}, a per-port active reflection limit Γmax⁡\Gamma_{\max} (VSWR limit) and a Tx–Rx leakage budget η\eta. Since σR2\sigma_{\mathrm{R}}^{2} in (5) is non-decreasing in PLP_{\mathrm{L}}, for a fixed η\eta we have σR2≤σc2+10−C/10η/NR\sigma_{\mathrm{R}}^{2} \le\sigma_{\mathrm{c}}^{2} + 10^{-C/10}\eta/N_{\mathrm{R}}, and minimizing the resulting upper bound on the CRB amounts to maximizing the Fisher term, i.e., the denominator in (6). Using the Schur complement as in [4], the per-η\eta problem is

max⁡t,{Wk},Rsts.t.Wk⪰0, Rs⪰0,(7)\begin{aligned} \max_{t,\{\mathbf{W}_k\},\mathbf{R}_{\mathrm{s}}}\quad&t \\ \text{s.t.}\quad&\mathbf{W}_k \succeq0,\ \mathbf{R}_{\mathrm{s}} \succeq0, \tag*{(7)} \end{aligned}
\begin{bmatrix} \operatorname{tr}(\dot{\mathbf{A}}^{H}\mathbf{A}\dot{\mathbf{R}})-t\operatorname{tr}(\dot{\mathbf{A}}^{H}\mathbf{A}\mathbf{R}) & \operatorname{tr}(\dot{\mathbf{A}}^{H}\dot{\mathbf{R}}) \\ \operatorname{tr}(\mathbf{A}^{H}\mathbf{A}\dot{\mathbf{R}}) & \operatorname{tr}(\mathbf{A}^{H}\mathbf{A}\mathbf{R}) \end{bmatrix} \succeq0, \tag*{(8)}\end{bmatrix}
(1+γ−1)hkWkhkH≥hkRhkH+σc2,∀k,(9)(1+\gamma^{-1})\mathbf{h}_k\mathbf{W}_k\mathbf{h}_k^{H} \ge\mathbf{h}_k\mathbf{R}\mathbf{h}_k^{H}+\sigma_{\mathrm{c}}^{2},\quad\forall k, \tag*{(9)}
tr⁡(R)≤PT,(10)\operatorname{tr}(\mathbf{R})\le P_{\mathrm{T}}, \tag*{(10)}
[STTRSTTH]nn≤Γ‾n2[R]nn,n=1,…,NT,(11)[\mathbf{S}_{\mathrm{TT}}\mathbf{R}\mathbf{S}_{\mathrm{TT}}^{H}]_{nn} \le\overline{\Gamma}_{n}^{2}[\mathbf{R}]_{nn},\quad n=1,\ldots,N_{\mathrm{T}}, \tag*{(11)}
tr⁡(SRTRSRTH)≤η.(12)\operatorname{tr}(\mathbf{S}_{\mathrm{RT}}\mathbf{R}\mathbf{S}_{\mathrm{RT}}^{H})\le\eta. \tag*{(12)}

Constraint (11) bounds the power-averaged active reflection at each PA, Γ‾n2=[STTRSTTH]nn/[R]nn\overline{\Gamma}_{n}^{2}=[\mathbf{S}_{\mathrm{TT}}\mathbf{R}\mathbf{S}_{\mathrm{TT}}^{H}]_{nn}/[\mathbf{R}]_{nn}, i.e., the ratio of mean reflected to mean incident power, and is linear in R\mathbf{R}; so is (12). When rank⁡(R)>1\operatorname{rank}(\mathbf{R})>1, the instantaneous coefficient [STTx]n/xn[\mathbf{S}_{\mathrm{TT}}\mathbf{x}]_{n}/x_{n} varies with the data, so (1+Γ‾n)/(1−Γ‾n)(1+\overline{\Gamma}_{n})/(1-\overline{\Gamma}_{n}) is an equivalent VSWR that sets the mean reflected power seen by the PA, not a bound on the peak VSWR of every symbol; a per-symbol guarantee would require symbol-level constraints. Hence (7) is an SDP after dropping rank⁡(Wk)=1\operatorname{rank}(\mathbf{W}_k)=1. As in [3], a rank-one solution is recovered from any optimal {Wk,Rs}\{\mathbf{W}_k,\mathbf{R}_{\mathrm{s}}\} by wk=WkhkH/hkWkhkH\mathbf{w}_k=\mathbf{W}_k\mathbf{h}_k^{H}/\sqrt{\mathbf{h}_k\mathbf{W}_k\mathbf{h}_k^{H}} and Rs←Rs−∑kwkwkH⪰0\mathbf{R}_{\mathrm{s}}\leftarrow\mathbf{R}_{\mathrm{s}}-\sum_k\mathbf{w}_k\mathbf{w}_k^{H}\succeq0. This leaves R\mathbf{R} unchanged, and therefore (8), (10)–(12) and every SINR are preserved.

The budget η\eta trades the Fisher information that can be delivered by the Tx subarray against the noise floor at the Rx subarray. A tight η\eta forces the Tx beam to place weak fields on the Rx elements, which costs degrees of freedom, while a loose η\eta lets residual SI dominate (5). Since (7) is solved in a fraction of a second, we select η\eta by a one-dimensional grid search that evaluates the true CRB (6), including the actual leakage of each solution.

Remarks. (a) Constraint (11) is scale invariant: replacing R\mathbf{R} by cRc\mathbf{R}, c>0c>0, leaves Γ‾n\overline{\Gamma}_{n} unchanged. Hence mismatch cannot be repaired by power backoff; it is a property of the beam direction, not of its power. (b) Since PL=∥SRTx∥2P_{\mathrm{L}}=\lVert\mathbf{S}_{\mathrm{RT}}\mathbf{x}\rVert^{2}, leakage-free transmission requires x∈ker⁡SRT\mathbf{x}\in\ker\mathbf{S}_{\mathrm{RT}}. In our array the singular values of SRT\mathbf{S}_{\mathrm{RT}} decay very fast: at d=0.3λd=0.3\lambda they are −4.5-4.5, −21.1-21.1, −41.6-41.6 and −66.6-66.6 dB for the first four modes, so nulling the two strongest leakage modes costs only two of the eight transmit degrees of freedom; this is why deep leakage budgets remain feasible for d≥0.3λd\ge0.3\lambda.

Benchmarks. We compare with: CU, the coupling-unaware design that solves (8)–(10) on uid\mathbf{u}_{\mathrm{id}}; PA, a pattern-aware design that solves (8)–(10) on the true embedded patterns (2), which captures the radiation-side MC compensation of recent works; and VSWR-only, which adds (11) but not (12). All designs are evaluated on the same NEC-2-derived model.

Full-Wave Results

Setup. The carrier is f=10f=10 GHz (FR3, λ=3\lambda=3 cm). Each dipole is a thin wire of radius λ/200\lambda/200 divided into 21 segments; its length 0.46λ0.46\lambda was tuned with NEC-2 to resonance, Zin=72.9+j1.0 ΩZ_{\mathrm{in}}=72.9+\mathrm{j}1.0\ \Omega, and we set Z0=73 ΩZ_{0}=73\ \Omega so that an isolated element is matched (realized gain 2.13 dBi). The array has NT=NR=8N_{\mathrm{T}}=N_{\mathrm{R}}=8 and a Tx–Rx gap equal to dd, and we sweep d∈[0.2,0.5]λd\in[0.2,0.5]\lambda. Simulations use the NEC-2 engine nec2c with a 0.25∘0.25^\circ azimuth grid, and CRB derivatives are taken by central differences. K=3K=3 users are located at ϕk∈{50∘,110∘,135∘}\phi_k\in\{50^\circ,110^\circ,135^\circ\} at 200 m under free-space propagation, with γ=12\gamma=12 dB. The target is at θ=75∘\theta=75^\circ, 60 m, σRCS=1 m2\sigma_{\mathrm{RCS}}=1\ \mathrm{m}^{2}, and L=256L=256. We set PT=30P_{\mathrm{T}}=30 dBm, bandwidth 100 MHz and noise figure 9 dB, giving σ2=σc2=−85\sigma^{2}=\sigma_{\mathrm{c}}^{2}=-85 dBm. The equivalent-VSWR limit is 2:12:1 (Γmax⁡=1/3\Gamma_{\max}=1/3), and all VSWR values reported below are equivalent VSWRs (1+Γ‾n)/(1−Γ‾n)(1+\overline{\Gamma}_{n})/(1-\overline{\Gamma}_{n}) and C∈{70,80,90}C\in\{70,80,90\} dB. Problem (7) is solved with CVXPY/Clarabel, and η\eta is searched over [−45,−10][-45,-10] dB relative to PTP_{\mathrm{T}} in 2.5 dB steps.

Model validation. The extracted Z\mathbf{Z} is reciprocal to within 10−510^{-5}. A direct NEC-2 run with all Tx ports driven through Z0Z_{0} and all Rx ports loaded agrees with the reconstruction to relative errors of 9×10−69\times10^{-6} (currents) and 8×10−58\times10^{-5} (far field), and its full-sphere radiated power matches the network value ∥x∥2−∥[S [xT,0T]T]∥2\lVert\mathbf{x}\rVert^{2}-\lVert[\mathbf{S}\,[\mathbf{x}^{T},\mathbf{0}^{T}]^{T}]\rVert^{2} to 3.5×10−43.5\times10^{-4}. Refining the mesh from 21 to 41 segments changes Z\mathbf{Z} by 4.3% (mainly the feed-gap reactance), but changes the CRB\sqrt{\mathrm{CRB}} of the designs in Table 1 by less than 0.4%, their SINR by less than 0.02 dB and the largest active reflection of the proposed design from 0.333 to 0.337.

ddDesignmin SINR (dB)max eq. VSWRPL/PTP_{\mathrm{L}}/P_{\mathrm{T}} (dB)Prad/PTP_{\mathrm{rad}}/P_{\mathrm{T}} (%)CRB\sqrt{\mathrm{CRB}} (deg)
0.3λ0.3\lambdaCU6.83.36−14.8-14.882.52.53
0.3λ0.3\lambdaPA12.03.57−16.5-16.582.22.11
0.3λ0.3\lambdaVSWR-only12.02.00−12.7-12.783.53.81
0.3λ0.3\lambdaProposed12.02.00−37.5-37.588.91.12
0.4λ0.4\lambdaCU9.52.41−19.2-19.289.20.88
0.4λ0.4\lambdaPA12.02.44−19.0-19.088.80.90
0.4λ0.4\lambdaVSWR-only12.02.00−19.4-19.489.40.87
0.4λ0.4\lambdaProposed12.02.00−35.0-35.090.20.41

Table 1. Full-wave metrics at C=80C = 80 dB (target SINR 12 dB)

Coupling. Fig. 2(a) shows that the adjacent-element coupling rises from −15.2-15.2 dB at 0.5λ0.5\lambda to −8.1-8.1 dB at 0.3λ0.3\lambda and −6.6-6.6 dB at 0.2λ0.2\lambda. The leakage of a simple broadside beam into the Rx subarray is −17.9-17.9 dB at 0.3λ0.3\lambda, so, without beam-domain control, a 30 dBm transmitter delivers about 12 dBm to the LNAs. Fig. 2(b) shows that at d=0.3λd=0.3\lambda the realized embedded patterns deviate from the isolated 2.13 dBi omnidirectional pattern by several dB and differ from element to element, with the edge element radiating preferentially towards endfire.

Patterns. Fig. 3 compares the patterns at d=0.3λd=0.3\lambda. The CU design looks correct on the ideal model (dotted), but on the real array its main beam is squinted by 1.7∘1.7^\circ and its gain towards the target drops by 1.9 dB. More importantly, its user SINRs fall to 10.9, 9.0 and 6.8 dB instead of 12 dB. The PA design restores the SINRs and the main beam. The proposed design has a gain towards the target is 3.7 dB lower than PA and it places a strong lobe around 145∘145^\circ, away from the Rx subarray (which lies towards ϕ=0∘\phi=0^\circ). This beam reduces the Tx–Rx leakage from −16.5-16.5 dB to −37.5-37.5 dB relative to PTP_{\mathrm{T}}, which brings the residual SI per Rx port from −75.5-75.5 dBm, 9.9 dB above thermal noise, to −96.5-96.5 dBm, below thermal noise.

Figure 2. (a) Coupling between adjacent elements, passive reflection of an inner element and total Tx–Rx leakage of a broadside beam versus spacing. (b) Realized embedded patterns of Tx elements 1 (edge), 4 and 8 (next to the Rx subarray) at d=0.3λd = 0.3\lambda, compared with the isolated dipole. Since the collinear array has no structure in yy, all azimuth patterns satisfy U(ϕ)=U(−ϕ)U(\phi) = U(-\phi) and are shown for 0∘≤ϕ≤180∘0^\circ\le\phi\le180^\circ.

Figure 3. Transmit patterns at d=0.3λd = 0.3\lambda, C=80C = 80 dB (gain per total PA power), shown for 0∘≤ϕ≤180∘0^\circ\le\phi\le180^\circ by symmetry. Dashed lines: users; dash-dotted line: target.

Figure 4. Root-CRB of the proposed design versus leakage budget η\eta at d=0.3λd = 0.3\lambda for three SIC budgets, compared with the PA design.

Hardware metrics. Table I summarizes the full-wave metrics. Both CU and PA drive at least one PA into an equivalent VSWR of 3.4–3.6 at 0.3λ0.3\lambda, i.e., on average about 30% of its incident power is reflected back into it. The VSWR-only

design fixes the mismatch but, without (12), it moves energy into the leakage path and its CRB is the worst of all. The proposed design satisfies every constraint and also radiates more of the PA power (88.9% versus 82.2% for PA at 0.3λ0.3\lambda), because less power is reflected and less is absorbed by the Rx loads.

Leakage budget. Fig. 4 shows the CRB versus η\eta. For each SIC budget CC, the curve has a minimum: tightening η\eta first lowers the noise floor quickly and then starts to cost Fisher information. The optimum moves to tighter budgets as CC decreases (−25-25 dB for C=90C = 90 dB, −37.5-37.5 dB for C=80C = 80 dB and −45-45 dB, the edge of the grid, for C=70C = 70 dB). At C=70C = 70 dB the proposed design achieves 1.22∘1.22^\circ against 6.35∘6.35^\circ for PA, a 5.2× improvement. At C=90C = 90 dB the receiver is almost thermal-noise limited for all designs, and the proposed design is 12% worse than PA at d=0.3λd = 0.3\lambda. This is the price of the VSWR constraint, which PA violates; for d≥0.35λd \ge0.35\lambda the

@page 4/6 starts_mid=0 ends_mid=1

Figure 2 (PDF p. 4). (a) Coupling between adjacent elements, passive reflection of an inner element and total Tx–Rx leakage of a broadside beam versus spacing. (b) Realized embedded patterns of Tx elements 1 (edge), 4 and 8 (next to the Rx subarray) at d=0.3λd = 0.3\lambda, compared with the isolated dipole. Since the collinear array has no structure in yy, all azimuth patterns satisfy U(ϕ)=U(−ϕ)U(\phi) = U(-\phi) and are shown for 0∘≤ϕ≤180∘0^\circ\le\phi\le180^\circ.

Figure 3 (PDF p. 4). Transmit patterns at d=0.3λd = 0.3\lambda, C=80C = 80 dB (gain per total PA power), shown for 0∘≤ϕ≤180∘0^\circ\le\phi\le180^\circ by symmetry. Dashed lines: users; dash-dotted line: target.

Figure 4 (PDF p. 4). Root-CRB of the proposed design versus leakage budget η\eta at d=0.3λd = 0.3\lambda for three SIC budgets, compared with the PA design.

ddDesignmin SINR (dB)max eq. VSWRPL/PTP_{\mathrm{L}}/P_{\mathrm{T}} (dB)Prad/PTP_{\mathrm{rad}}/P_{\mathrm{T}} (%)CRB\sqrt{\mathrm{CRB}} (deg)
0.3λ0.3\lambdaCU6.83.36−14.8-14.882.52.53
0.3λ0.3\lambdaPA12.03.57−16.5-16.582.22.11
0.3λ0.3\lambdaVSWR-only12.02.00−12.7-12.783.53.81
0.3λ0.3\lambdaProposed12.02.00−37.5-37.588.91.12
0.4λ0.4\lambdaCU9.52.41−19.2-19.289.20.88
0.4λ0.4\lambdaPA12.02.44−19.0-19.088.80.90
0.4λ0.4\lambdaVSWR-only12.02.00−19.4-19.489.40.87
0.4λ0.4\lambdaProposed12.02.00−35.0-35.090.20.41

Table 1 (PDF p. 4). Full-wave metrics at C=80C = 80 dB (target SINR 12 dB)

Hardware metrics. Table I summarizes the full-wave metrics. Both CU and PA drive at least one PA into an equivalent VSWR of 3.4–3.6 at 0.3λ0.3\lambda, i.e., on average about 30% of its incident power is reflected back into it. The VSWR-only

design fixes the mismatch but, without (12), it moves energy into the leakage path and its CRB is the worst of all. The proposed design satisfies every constraint and also radiates more of the PA power (88.9% versus 82.2% for PA at 0.3λ0.3\lambda), because less power is reflected and less is absorbed by the Rx loads.

Leakage budget. Fig. 4 shows the CRB versus η\eta. For each SIC budget CC, the curve has a minimum: tightening η\eta first lowers the noise floor quickly and then starts to cost Fisher information. The optimum moves to tighter budgets as CC decreases (−25-25 dB for C=90C = 90 dB, −37.5-37.5 dB for C=80C = 80 dB and −45-45 dB, the edge of the grid, for C=70C = 70 dB). At C=70C = 70 dB the proposed design achieves 1.22∘1.22^\circ against 6.35∘6.35^\circ for PA, a 5.2× improvement. At C=90C = 90 dB the receiver is almost thermal-noise limited for all designs, and the proposed design is 12% worse than PA at d=0.3λd = 0.3\lambda. This is the price of the VSWR constraint, which PA violates; for d≥0.35λd \ge0.35\lambda the proposed design is still 2–14% better at C=90C = 90 dB.

Spacing. Figure 5 shows the CRB versus dd. For d≥0.3λd \ge0.3\lambda, the proposed design is consistently the best, reducing the root-CRB of the best benchmark by 32–53% at C=80C = 80 dB and by 3.4–5.6× at C=70C = 70 dB. Even at d=0.5λd = 0.5\lambda, where coupling is often considered negligible, CU misses the SINR target by 0.7 dB and the proposed design improves the root-CRB by 32% at C=80C = 80 dB, because leakage still limits the receiver. At d=0.2λd = 0.2\lambda, CU misses the target by 6.2 dB and both CU and PA reach VSWR values of 5.0–5.5. Below 0.3λ0.3\lambda the picture changes: at 0.25λ0.25\lambda the VSWR constraint can be met only with leakage budgets of −17.5 dB or looser, and the proposed design is worse than the (non-compliant) benchmarks, while at 0.2λ0.2\lambda no design meets both the SINR targets and the 2:1 VSWR limit. Here compactness must be paid for by matching or decoupling networks.

Figure 5. Root-CRB versus element spacing for C=70C = 70 dB (solid) and C=80C = 80 dB (dashed). Missing points: VSWR-compliant designs infeasible.

Independent FDTD Verification of the Final Layout

To test the conclusions against a different numerical method and feed model, we rebuilt the d=0.3λd = 0.3\lambda layout (Figure 6(a)) in the open-source finite-difference time-domain (FDTD) solver openEMS [14]. Each wire is replaced by a zero-thickness PEC strip of width w=4a=0.6w = 4a = 0.6 mm, the strip whose equivalent radius equals the NEC wire radius, and each dipole is fed by a lumped 73 Ω\Omega port across a 0.3 mm gap. The computational domain is terminated by 8-cell perfectly matched layers, the mesh is graded down to w/2w/2 at the strips and the gap, and a Gaussian pulse covers 7–13 GHz. Sixteen runs, one per driven port with all other ports terminated in 73 Ω\Omega, give the full 16×1616 \times16 scattering matrix and, through a near-to-far-field transformation, the realized embedded patterns in the same normalization as (2). No quantity is copied from the NEC model.

Independent openEMS (FDTD) model of the $d=0.3\lambda$ layout, showing the simulated geometry, scattering-matrix columns, and transmit patterns

Figure 6. Independent openEMS (FDTD) model of the d=0.3λd=0.3\lambda layout. (a) Simulated geometry: 16 PEC strip dipoles (Tx in blue, Rx in red) with lumped ports, inside the PML-terminated domain; the inset shows the feed gap and the local mesh. (b) Scattering-matrix columns for an inner Tx port and for the Tx port next to the Rx subarray. (c) Transmit patterns (0∘≤ϕ≤180∘0^\circ\le\phi\le180^\circ, by symmetry) of the NEC-designed proposed and PA beamformers on the NEC-2 and FDTD models, without re-optimization.

Agreement of the two solvers. The FDTD scattering matrix is reciprocal to within 0.25%. Figure 6(b) shows that the mutual terms agree with NEC-2 over a 35 dB range (to within 1.7 dB for all 240 mutual terms), from −8.1 versus −8.4 dB for the adjacent elements to −42.1 versus −42.8 dB for the most distant pair. The self terms differ: the strip dipoles with a finite gap are slightly detuned, and the passive reflection of an inner element rises from −13.9 dB (NEC-2) to −10.4 dB (FDTD). Over all 16 realized embedded patterns, wherever the gain exceeds −5 dBi, the two solvers differ by 0.28 dB rms and 0.68 dB at most.

Opt. onDesignmin SINR (dB)max eq. VSWRPL/PTP_L/P_T (dB)Prad/PTP_{\mathrm{rad}}/P_T (%)CRB\sqrt{\mathrm{CRB}} (deg)
NECCU6.72.69−15.385.52.35
NECPA11.93.03−16.985.41.97
NECVSWR-only11.72.21−13.285.03.49
NECProposed12.02.18−37.489.01.09
FDTDPA12.02.92−16.985.41.97
FDTDProposed12.02.00−37.588.90.87

Table 2. Metrics on the openEMS model (d=0.3λd = 0.3\lambda, C=80C = 80 dB).

Model mismatch. The upper part of Table II replays the beamformers designed on the NEC-2 model, without any change, on the FDTD model, which here plays the role of the “real” array. The proposed design keeps its SINRs within 0.05 dB of the target and its leakage at −37.4 dB, i.e., the beam-domain leakage null is robust to the change of solver and feed model, because it depends on the mutual terms, which both solvers agree on. Its root-CRB is 1.09°, against 1.97° for PA and 2.35° for CU, and Figure 6(c) shows that the patterns are practically unchanged. The quantity that moves is the active reflection, which depends on the self terms: the largest VSWR of the proposed design rises from 2.00 to 2.18, and the PA design still reaches 3.03. In practice this suggests a small margin on Γmax⁡\Gamma_{\max}, or re-optimization on the final model.

Where the leakage goes. Figure 7 shows the leakage constraint (7f) at work on the physical layout. The strip currents are obtained from the jump of the tangential magnetic field across the strip plane and superposed with the covariance R\mathbf{R} of each beamformer. The PA beamformer drives all Tx dipoles within about 3 dB of each other, including the one next to the gap, and the first Rx dipole carries a current only 10.2 dB below the strongest Tx current; in total the Rx strips carry −15.4 dB of the strip-current power. The proposed beamformer tapers the Tx currents towards the gap (the eighth dipole is 13 dB below the first) and phases them so that their fields cancel on the Rx side: the first Rx dipole drops to −29.8 dB and the Rx share to −35.1 dB. These values agree with the port-level leakage of Table II (−16.9 and −37.4 dB), so the leakage null is a near-field cancellation on the actual structure, not an artefact of the network model.

Figure 7. FDTD strip currents on the d=0.3λd=0.3\lambda layout for the NEC-designed PA (top) and proposed (bottom) beamformers: E∣I(z)∣2E|I(z)|^2 along each dipole, from the superposition of the eight Tx-port runs weighted by R\mathbf{R}, on a common normalization (strip width exaggerated for visibility).

Re-optimization. The lower part of Table II solves (7) directly on the FDTD data. The proposed design again meets every constraint, selects the same leakage budget (−37.5 dB) and reaches 0.87°, which is 56% below the PA design optimized on the same data; the latter still drives a PA to a VSWR of 2.92. The relative ranking of all designs is thus the same in both solvers, which supports the conclusions of § 4.

Conclusion

We showed that, in compact monostatic ISAC transceivers, the antenna array determines not only the radiated beam but also the load seen by the PAs and the noise floor of the radar receiver. A multiport model extracted from NEC-2 makes all three linear in the PA waves, so that VSWR and Tx–Rx leakage constraints fit naturally into a CRB-minimizing ISAC SDP. Full-wave evaluation shows that coupling-unaware designs violate SINR and VSWR requirements, and that controlling leakage in the beam domain can reduce the angle root-CRB by 32–53% at 80 dB SIC and by 3.4–5.6× at 70 dB, for spacings from 0.3λ0.3\lambda to 0.5λ0.5\lambda. An independent FDTD model of the final layout confirms that the leakage null is robust to the solver and feed model, whereas the active reflection depends on the feed details and calls for a small design margin. Future work will include dipoles over a ground plane on a substrate, wideband (OFDM) operation, robust designs against fabrication tolerances, and measurements on a prototype.

References

  1. [1]F. Liu, C. Masouros, A. P. Petropulu, H. Griffiths, and L. Hanzo, “Joint radar and communication design: Applications, state-of-the-art, and the road ahead,” IEEE Trans. Commun., vol. 68, no. 6, pp. 3834–3862, Jun. 2020.arxiv.org/abs/1906.00789
  2. [2]F. Liu, Y. Cui, C. Masouros, J. Xu, T. X. Han, Y. C. Eldar, and S. Buzzi, “Integrated sensing and communications: Toward dual-functional wireless networks for 6G and beyond,” IEEE J. Sel. Areas Commun., vol. 40, no. 6, pp. 1728–1767, Jun. 2022.DOI
  3. [3]X. Liu, T. Huang, N. Shlezinger, Y. Liu, J. Zhou, and Y. C. Eldar, “Joint transmit beamforming for multipurpose MIMO communications and MIMO radar,” IEEE Trans. Signal Process., vol. 68, pp. 3929–3944, 2020.DOI
  4. [4]F. Liu, Y.-F. Liu, A. Li, C. Masouros, and Y. C. Eldar, “Cramér-Rao bound optimization for joint radar-communication beamforming,” IEEE Trans. Signal Process., vol. 70, pp. 240–253, 2022.DOI
  5. [5]J. W. Wallace and M. A. Jensen, “Mutual coupling in MIMO wireless systems: A rigorous network theory analysis,” IEEE Trans. Wireless Commun., vol. 3, no. 4, pp. 1317–1325, Jul. 2004.DOI
  6. [6]M. T. Ivrlač and J. A. Nossek, “Toward a circuit theory of communication,” IEEE Trans. Circuits Syst. I, vol. 57, no. 7, pp. 1663–1683, Jul. 2010.DOI
  7. [7]D. M. Pozar, “The active element pattern,” IEEE Trans. Antennas Propag., vol. 42, no. 8, pp. 1176–1178, Aug. 1994.DOI
  8. [8]A. Sabharwal, P. Schniter, D. W. Bliss, S. Rangarajan, and R. Wichman, “In-band full-duplex wireless: Challenges and opportunities,” IEEE J. Sel. Areas Commun., vol. 32, no. 9, pp. 1637–1652, Sep. 2014.arxiv.org/abs/1311.0456
  9. [9]K. Han, C. Masouros, T. Riihonen, and M. G. Amin, “Next-generation MIMO transceivers for integrated sensing and communications: Unique security vulnerabilities and solutions,” arXiv:2511.20309, 2025.arxiv.org/abs/2511.20309
  10. [10]J. Maeng and K. Han, “Robust beamforming design for integrated sensing and communications with mutual coupling effect,” arXiv:2608.15556, 2026.arxiv.org/abs/2608.15556
  11. [11]S. Zeng, H. Zhang, D. Ji, H. Zhang, Z. Shao, Z. Han, H. V. Poor, and L. Song, “Holographic beamforming for integrated sensing and communication with mutual coupling effects,” IEEE J. Sel. Areas Commun., vol. 44, pp. 480–497, 2026.arxiv.org/abs/2509.08113
  12. [12]H. Prod’homme and P. del Hougne, “Mutual coupling in dynamic metasurface antennas: Foe, but also friend,” arXiv:2412.01002, 2024.
  13. [13]G. J. Burke and A. J. Poggio, “Numerical Electromagnetics Code (NEC): Method of moments,” Lawrence Livermore Lab., Tech. Rep. UCID-18834, 1981.
  14. [14]T. Liebig, A. Rennings, S. Held, and D. Erni, “openEMS: A free and open source equivalent-circuit (EC) FDTD simulation platform supporting cylindrical coordinates suitable for the analysis of traveling wave MRI applications,” Int. J. Numer. Model., vol. 26, no. 6, pp. 680–696, 2013.

Paper details

vlnc:W_ajq6zvw9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 13

downloads · 1

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