betaFeedback

WorksEngineering

When Does Inter-BS Interference Hurt Sensing in ISAC? An Exact Fisher Information Analysis under Finite-Alphabet Interference

Abstract

In networked integrated sensing and communication (ISAC), a base station (BS) performing sensing using its own downlink signals is subject to interference from the downlink transmissions of neighboring BSs. This paper investigates the impact of the signal constellation and waveform on the sensing performance in the presence of such inter-BS interference. For any linear receiver, we show that the mean-squared cross-ambiguity between two random ISAC signals is invariant to the modulation bases and constellations. We then derive the exact Fisher information under finite-alphabet interference with a flat-fading link, which equals the interference-free Fisher information scaled by 1-γ\,\mmse_{\C}(γ), where γγ denotes the interference-to-noise ratio (INR) and \mmse_{\C} is the minimum mean-square error (MMSE) in estimating an interfering data symbol from its noise-corrupted observation at the sensing receiver. Consequently, Gaussian interference is the worst case and corresponds to the CRB loss of the matched filtering, whereas finite-alphabet interference induces a non-monotonic CRB loss that peaks at intermediate INR and vanishes at high INR. The loss is waveform-independent for the flat-fading BS-BS link, and an expectation-maximization (EM) interference-cancellation receiver approaches the theoretical bound for delay estimation.

Introduction

Integrated sensing and communication (ISAC) has attracted significant attention as a key enabler of next-generation wireless networks, in which the downlink (DL) data signals of cellular base stations (BSs) are reused for radar sensing [1], [2]. In a dense, TDD-synchronized network, a BS that performs full-duplex monostatic sensing inevitably receives the DL transmissions of its neighbors on the same time-frequency resources [3], [4]. Since BS-to-BS links are typically line-of-sight (LoS), this inter-BS interference can exceed the thermal noise by tens of decibels, and it fundamentally limits the sensing performance of network-level ISAC [5].

Since communication payloads are random, the sensing performance of a data-bearing signal depends on its modulation constellation and basis. OFDM achieves the lowest average ranging sidelobe under sub-Gaussian constellations [6], while the sidelobes induced by high-order QAM can be suppressed by mismatched filtering receiver [7]. Constellation shaping or selection balances this randomness against the communication rate [8]–[10], and the delay-estimation accuracy of OFDM-ISAC under matched filtering (MF) and reciprocal filtering (RF) has been characterized via the fourth and inverse second moments of the constellation [11]. However, all of these results concern the auto-ambiguity of a BS’s own signal.

For inter-BS (inter-cell) interference, constellation selection and power allocation in multi-cell OFDM-ISAC have been studied under MF and RF processing [12], and blind estimation under co-channel OFDM sources has also been investigated [13], which characterize the interference by its constellation moments and bound the performance with a Gaussian surrogate, respectively. On the communication side, it is well known that discrete interference causes less performance degradation than Gaussian interference [14]. Despite these advances, an exact estimation-theoretic characterization of sensing under finite-alphabet inter-BS interference remains unexplored, particularly regarding how interference strength and modulation determine the achievable sensing accuracy.

To address this gap, we develop an estimation-theoretic framework for sensing under finite-alphabet inter-BS interference. First, we first show that the mean-squared cross-ambiguity is invariant to the modulation bases and constellations, so that no constellation or waveform design benefits a linear receiver. Then, we derive the exact Fisher information matrix (FIM) under finite-alphabet interference via a complex Brown identity, revealing that Gaussian interference is the worst case and that finite-alphabet interference yields a non-monotonic loss that vanishes at high INR. We further prove that the loss is waveform-independent for a flat-fading and synchronized BS-BS link, and show that an expectation-maximization (EM)-based soft interference-cancellation receiver approaches the bound.

System Model

Transmit Signal Model

As illustrated in Figure 1, BS0_0 performs full-duplex (FD) monostatic sensing with its DL data signal, while a neighboring BS1_1 transmits independent DL data over the same time-frequency resources. To cover a broad class of waveforms, we consider a generic block transmission, in which each BS maps NN data symbols onto a length-NN block through a unitary modulation basis and appends a cyclic prefix (CP), i.e.,

ISAC under Inter-BS interference

Figure 1. ISAC under Inter-BS interference: BS0_0 senses with its own DL data while receiving the DL data of BS1_1 on the same time-frequency resources.

x0=U0s0,x1=U1s1,(1)\mathbf{x}_0 = \mathbf{U}_0\mathbf{s}_0,\qquad\mathbf{x}_1 = \mathbf{U}_1\mathbf{s}_1, \tag*{(1)}

where s0\mathbf{s}_0 and s1\mathbf{s}_1 are the symbol vectors of BS0_0 and BS1_1, and U0,U1∈CN×N\mathbf{U}_0,\mathbf{U}_1 \in\mathbb{C}^{N\times N} are unitary. This model includes single carrier (SC, U1=I\mathbf{U}_1=\mathbf{I}), CP-OFDM (U1=WNH\mathbf{U}_1=\mathbf{W}_N^H with the unitary DFT matrix WN\mathbf{W}_N), and OTFS (U1=WNH⊗IN/ND\mathbf{U}_1=\mathbf{W}_{N}^{H}\otimes\mathbf{I}_{N/N_D} with NDN_D Doppler bins), as well as DFT-spread OFDM and AFDM. The symbols are i.i.d., zero-mean, and unit-power, and the interfering symbols are drawn from the constellation S\mathcal{S}. We define the fourth moment (kurtosis) and the inverse second moment of a constellation as μ4=E[∣s∣4]\mu_4=\mathbb{E}[|s|^4] and ν−2=E[∣s∣−2]\nu_{-2}=\mathbb{E}[|s|^{-2}].

Sensing Received Signal Model

After self-interference suppression and CP removal, the received signal of BS0_0 is

r=μ(η)+As1+n,(2)\mathbf{r}=\boldsymbol{\mu}(\boldsymbol{\eta})+\mathbf{A}\mathbf{s}_1+\mathbf{n}, \tag*{(2)}
A=HIU1,n∼CN(0,σ2I),(3)\mathbf{A}=\mathbf{H}_{I}\mathbf{U}_1,\qquad\mathbf{n}\sim\mathcal{CN}(\mathbf{0},\sigma^2\mathbf{I}), \tag*{(3)}

where μ(η)\boldsymbol{\mu}(\boldsymbol{\eta}) is the target echo of BS0_0’s own signal x0\mathbf{x}_0, e.g., μ(η)=αJτ,νx0\boldsymbol{\mu}(\boldsymbol{\eta})=\alpha\mathbf{J}_{\tau,\nu}\mathbf{x}_0 for a single target with complex amplitude α\alpha, delay τ\tau, Doppler ν\nu, and the delay-Doppler shift operator Jτ,ν\mathbf{J}_{\tau,\nu}, and η\boldsymbol{\eta} collects the real-valued sensing parameters. HI\mathbf{H}_I is the circulant matrix of the BS1_1-to-BS0_0 channel hI\mathbf{h}_I, and we define the per-sample INR as γ=∥hI∥2/σ2\gamma=\|\mathbf{h}_I\|^2/\sigma^2. For a CP-OFDM interferer, A=WNHdiag⁡(Hk)\mathbf{A}=\mathbf{W}_N^H\operatorname{diag}(H_k), which yields the conventional per-subcarrier model r~k=μ~k(η)+Hks1,k+n~k\tilde{r}_k=\tilde{\mu}_k(\boldsymbol{\eta})+H_k s_{1,k}+\tilde{n}_k after the DFT.

In this study, we adopt the following assumptions:

  • (A1) s1\mathbf{s}_1 has i.i.d. entries uniform on S\mathcal{S} and is independent of η\boldsymbol{\eta}, s0\mathbf{s}_0, and n\mathbf{n}.

  • (A2) The propagation delay of the BS1_1 signal is within the CP, and the residual carrier frequency offset (CFO) is negligible.

  • (A3) BS0_0 knows its own data, the static HI\mathbf{H}_I (estimated from the reference signals of BS1_1), U1\mathbf{U}_1, and S\mathcal{S}.

  • (A4) S\mathcal{S} is invariant to 90∘90^\circ rotations, and thus E[s12]=0\mathbb{E}[s_1^2]=0.

Main Results

Linear Receivers: Second-Order Invariance

With the unitary delay-Doppler shift Jℓ,q=diag⁡(ej2πqn/N)Πℓ\mathbf{J}_{\ell,q}=\operatorname{diag}(e^{j2\pi qn/N})\mathbf{\Pi}^{\ell}, where Π\mathbf{\Pi} is the cyclic shift matrix, the periodic cross-ambiguity function between the signals in (1) is defined as

χ(ℓ,q)=x1HJℓ,qx0=s1HGℓ,qs0,(4)\chi(\ell,q)=\mathbf{x}_1^H\mathbf{J}_{\ell,q}\mathbf{x}_0=\mathbf{s}_1^H\mathbf{G}_{\ell,q}\mathbf{s}_0, \tag*{(4)}
G=U1HJℓ,qU0.(5)\mathbf{G}= \mathbf{U}_1^H\mathbf{J}_{\ell,q}\mathbf{U}_0. \tag*{(5)}

With this in hand, we first show that the second moment of χ\chi is invariant to the modulation bases and constellations for any NN, and so is its fourth moment for large NN.

Proposition 1. For all (ℓ,q)(\ell,q), all unitary bases U0,U1\mathbf{U}_0,\mathbf{U}_1, and all unit-power constellations with E[s02]=E[s12]=0\mathbb{E}[s_0^2]=\mathbb{E}[s_1^2]=0, E[∣χ(ℓ,q)∣2]=N\mathbb{E}[|\chi(\ell,q)|^2]=N. Moreover, with the kurtosis μ4,s0\mu_{4,s_0} and μ4,s1\mu_{4,s_1} of s0s_0 and s1s_1,

E[∣χ∣4]=2N2+2N(μ4,s0+μ4,s1−3)+(μ4,s0−2)(μ4,s1−2)∥G∥44,(6)\mathbb{E}[|\chi|^4]=2N^2+2N(\mu_{4,s_0}+\mu_{4,s_1}-3) +(\mu_{4,s_0}-2)(\mu_{4,s_1}-2)\|\mathbf{G}\|_4^4, \tag*{(6)}

where ∥G∥44=∑i,j∣Gij∣4∈[1,N]\|\mathbf{G}\|_4^4=\sum_{i,j}|G_{ij}|^4\in[1,N].

Proof. Please refer to Appendix A.

The upper and lower bounds of ∥G∥44\|\mathbf{G}\|_4^4 are attained iff G\mathbf{G} is monomial (e.g., OFDM-OFDM) and iff ∣Gij∣2=1/N|G_{ij}|^2=1/N (e.g., OFDM-SC), respectively. Since E[∣χ(ℓ,q)∣2]=N\mathbb{E}[|\chi(\ell,q)|^2]=N for all (ℓ,q)(\ell,q), the interference after MF forms a flat floor over the delay-Doppler map approximately 1/N1/N relative to the main lobe E[∣x0Hx0∣2]=N2+N(μ4,s0−1)\mathbb{E}[|\mathbf{x}_0^H\mathbf{x}_0|^2]=N^2+N(\mu_{4,s_0}-1) regardless of the basis and constellation of the interferer unlike the auto-ambiguity sidelobes [6]. Moreover, (6) yields E[∣χ∣4]/(E[∣χ∣2])2=2+c/N\mathbb{E}[|\chi|^4]/(\mathbb{E}[|\chi|^2])^2=2+c/N with cc bounded in NN, implying that the floor fluctuates as Gaussian noise for large NN.

Remark 1. Linear filters such as MF, RF, and LMMSE depend on the interference only through its covariance E[(As1)(As1)H]=HIHIH\mathbb{E}[(\mathbf{A}\mathbf{s}_1)(\mathbf{A}\mathbf{s}_1)^H]=\mathbf{H}_I\mathbf{H}_I^H, which is independent of U1\mathbf{U}_1 and S\mathcal{S}, and also of s0\mathbf{s}_0 by (A1). Hence, this holds even when the linear filter is designed for each realization of the known s0\mathbf{s}_0. Therefore, the choice of the waveform and constellation of the interferer cannot affect linear receivers, and only the own constellation enters through ν−2\nu_{-2} for RF processing [11, 12]. Exploiting the interference structure thus requires non-linear processing.

Fisher Information under Finite-Alphabet Interference

Since η\boldsymbol{\eta} enters (2) only through μ(η)\boldsymbol{\mu}(\boldsymbol{\eta}), the FIM is governed by the Fisher information of w=As1+n\mathbf{w}=\mathbf{A}\mathbf{s}_1+\mathbf{n} with respect to a location shift. For finite-alphabet s1\mathbf{s}_1, w\mathbf{w} is a Gaussian mixture with ∣S∣N|\mathcal{S}|^N components whose Fisher information admits no closed form, while a Gaussian surrogate yields only the worst case [15]. Instead, we use Brown’s identity [16, 17], which expresses this Fisher information with the minimum mean-square error (MMSE) in estimating an interfering data symbol from its noise-corrupted observation at the sensing receiver, which is efficiently computable. Its complex vector form is stated as follows.

Lemma 1. Let sw(w)=∇w∗ln⁡pw(w)s_{\mathbf{w}}(\mathbf{w}) = \nabla_{\mathbf{w}^{*}} \ln p_{\mathbf{w}}(\mathbf{w}) denote the Wirtinger score of w=As1+n\mathbf{w} = A\mathbf{s}_{1} + \mathbf{n}. Then, sw(w)=−σ−2E[n∣w]s_{\mathbf{w}}(\mathbf{w}) = -\sigma^{-2}\mathbb{E}[\mathbf{n}|\mathbf{w}] and

Jw≜E[swswH]=1σ2(I−1σ2E[cov⁡(As1∣w)]).(7)J_{\mathbf{w}} \triangleq\mathbb{E}[s_{\mathbf{w}}s_{\mathbf{w}}^{H}] = \frac{1}{\sigma^{2}}\left(I-\frac{1}{\sigma^{2}}\mathbb{E}\left[\operatorname{cov}(A\mathbf{s}_{1}|\mathbf{w})\right]\right). \tag*{(7)}

Under (A1) and (A4), E[swswT]=0\mathbb{E}[s_{\mathbf{w}}s_{\mathbf{w}}^{T}] = 0.

Proof. Please refer to Appendix B.

Theorem 1. The FIM of η\boldsymbol{\eta} in (2) is given by

[F]ij=2ℜ{∂μH∂ηiJw∂μ∂ηj}.(8)[F]_{ij}=2\Re\left\{\frac{\partial\boldsymbol{\mu}^{H}}{\partial\eta_i}J_{\mathbf{w}}\frac{\partial\boldsymbol{\mu}}{\partial\eta_j}\right\}. \tag*{(8)}

Moreover, Jw⪰(σ2I+AAH)−1J_{\mathbf{w}} \succeq(\sigma^{2}I+AA^{H})^{-1} and the equality holds for Gaussian s1\mathbf{s}_{1}.

Proof. Please refer to Appendix C.

To evaluate (7), consider the scalar channel Y=γS1+NY=\sqrt{\gamma}S_{1}+N with S1S_{1} uniform on S\mathcal{S} and N∼CN(0,1)N\sim\mathcal{CN}(0,1). Then, we define

MMSE⁡S(γ)=E[∣S1−E[S1∣Y]∣2].(9)\operatorname{MMSE}_{\mathcal{S}}(\gamma)=\mathbb{E}\left[|S_{1}-\mathbb{E}[S_{1}|Y]|^{2}\right]. \tag*{(9)}

When the interference decouples into such scalar channels, Theorem 1 reduces to the following closed form.

Corollary 1. Let us define ρS(γ)=1−γMMSE⁡S(γ)\rho_{\mathcal{S}}(\gamma)=1-\gamma\operatorname{MMSE}_{\mathcal{S}}(\gamma). If AA has orthogonal columns, i.e., A=QΛA=Q\Lambda with unitary QQ and Λ=diag⁡(λ1,…,λN)\Lambda=\operatorname{diag}(\lambda_{1},\ldots,\lambda_{N}), then

Jw=1σ2Qdiag⁡(ρS(γ1),…,ρS(γN))QH,γi=∣λi∣2σ2.(10)J_{\mathbf{w}}=\frac{1}{\sigma^{2}}Q\operatorname{diag}\left(\rho_{\mathcal{S}}(\gamma_{1}),\ldots,\rho_{\mathcal{S}}(\gamma_{N})\right)Q^{H},\qquad\gamma_i=\frac{|\lambda_i|^{2}}{\sigma^{2}}. \tag*{(10)}

If γi=γ\gamma_i=\gamma for all ii, then F=ρS(γ)F0F=\rho_{\mathcal{S}}(\gamma)F_{0}, where F0F_{0} is the interference-free FIM. Hence, the entire CRB matrix is inflated by the same factor, which we define as the CRB loss

LS(γ)=1ρS(γ)=11−γMMSE⁡S(γ).(11)L_{\mathcal{S}}(\gamma)=\frac{1}{\rho_{\mathcal{S}}(\gamma)}=\frac{1}{1-\gamma\operatorname{MMSE}_{\mathcal{S}}(\gamma)}. \tag*{(11)}

Proof. Please refer to Appendix D.

Note that FF is conditioned on the known s0\mathbf{s}_{0} and marginalized over s1\mathbf{s}_{1}. Since F=ρS(γ)F0F=\rho_{\mathcal{S}}(\gamma)F_{0} holds for every realization of s0\mathbf{s}_{0}, LS(γ)L_{\mathcal{S}}(\gamma) is identical whether or not the CRB is further averaged over s0\mathbf{s}_{0}.

Proposition 2. The CRB loss LS(γ)L_{\mathcal{S}}(\gamma) satisfies the following properties:

(i) 1≤LS(γ)≤1+γ1\leq L_{\mathcal{S}}(\gamma)\leq1+\gamma, where the upper bound holds with equality for Gaussian S1S_{1} and equals the loss of MF, i.e., treating interference as noise (TIN).

(ii) lim⁡γ→0LS(γ)−1γ=1\lim_{\gamma\to0}\frac{L_{\mathcal{S}}(\gamma)-1}{\gamma}=1. Hence, at low INR, finite-alphabet interference is as harmful as Gaussian interference.

(iii) LS(γ)→1L_{\mathcal{S}}(\gamma)\to1 as γ→∞\gamma\to\infty for any finite S\mathcal{S}.

(iv) LS(γ)L_{\mathcal{S}}(\gamma) is maximized at a finite INR γ⋆\gamma^{\star}.

Proof. Please refer to Appendix E.

Remark 2: Notably, Proposition 2 reveals that, unlike the TIN loss LTIN(γ)=1+γL_{\mathrm{TIN}}(\gamma)=1+\gamma growing without bound, the loss under finite-alphabet interference is non-monotonic: it follows the Gaussian loss at low INR, peaks at γ⋆\gamma^{\star}, and vanishes at high INR, where the interfering symbols can be reliably detected and canceled as shown in Figure 2.

Figure 2. CRB loss LS(γ)L_{\mathcal{S}}(\gamma) in (11) versus INR. The dotted line corresponds to Gaussian interference, which equals the loss of MF (TIN).

Dependence of the FIM on the Interferer Waveform

By Theorem 1, the waveform of the interferer affects the FIM only through A=HIU1A=H_{I}U_{1}, and Corollary 1 yields a closed form when AA has orthogonal columns. The following proposition gives a sufficient condition for waveform independence.

Proposition 3. If the BS-BS channel is flat-fading, i.e., HI=hIH_{I}=h_{I}, then Jw=ρS(γ)I/σ2J_{\mathbf{w}}=\rho_{\mathcal{S}}(\gamma)I/\sigma^{2} for every unitary U1U_{1}. Hence, LS(γ)L_{\mathcal{S}}(\gamma) is identical for every interferer waveform (SC, CP-OFDM, OTFS, DFT-s-OFDM, AFDM, etc.), and F=ρS(γ)F0F=\rho_{\mathcal{S}}(\gamma)F_{0} holds for any waveform U0U_{0}.

Proof. Since A=hU1A=hU_{1}, Corollary 1 applies with Q=U1Q=U_{1} and λi=h\lambda_i=h, i.e., γi=∣h∣2/σ2=γ\gamma_i=|h|^{2}/\sigma^{2}=\gamma for all ii. Thus, Jw=ρS(γ)I/σ2J_{\mathbf{w}}=\rho_{\mathcal{S}}(\gamma)I/\sigma^{2} and does not depend on U1U_{1}, and F=ρS(γ)F0F=\rho_{\mathcal{S}}(\gamma)F_{0} follows for any μ(η)\boldsymbol{\mu}(\boldsymbol{\eta}), i.e., any U0U_{0}.

Under a frequency-selective channel, HI=WNHdiag⁡(Hk)WNH_{I}=W_{N}^{H}\operatorname{diag}(H_k)W_{N}, so that AA has orthogonal columns for CP-OFDM (Q=WNHQ=W_{N}^{H}, γk=∣Hk∣2/σ2\gamma_k=|H_k|^{2}/\sigma^{2}), whose loss averages the scalar losses over the channel-induced INR spread. In contrast, AHA=U1HWNHdiag⁡(∣Hk∣2)WNU1A^{H}A=U_{1}^{H}W_{N}^{H}\operatorname{diag}(|H_k|^{2})W_{N}U_{1} is not diagonal for SC and OTFS, so that (7) involves the joint MMSE attainable only by sequence-level processing. Since BS-BS links are typically LoS-dominated with a small delay spread, the flat-channel assumption of Proposition 3 is a reasonable approximation in practice.

EM-Based Soft Interference Cancellation

The exact FIM in Theorem 1 provides an asymptotic benchmark for maximum-likelihood (ML) estimation under standard regularity conditions. To approach this benchmark, we employ an expectation-maximization (EM)-based receiver that iteratively estimates and suppresses the finite-alphabet interference [18]. Given the current estimate η^\hat{\boldsymbol{\eta}}, the E-step reduces to computing the posterior mean sˉ1=E[s1∣r−μ(η^)]\bar{\mathbf{s}}_{1}=\mathbb{E}[\mathbf{s}_{1}|\mathbf{r}-\boldsymbol{\mu}(\hat{\boldsymbol{\eta}})]. In the separable case of Corollary 1, it decouples into per-symbol computations on v=QH(r−μ(η^))\mathbf{v}=Q^{H}(\mathbf{r}-\boldsymbol{\mu}(\hat{\boldsymbol{\eta}})) as

sˉ1,i=∑c∈S ⁣ce−∣vi−λic∣2/σ2∑c∈Se−∣vi−λic∣2/σ2.(12)\bar{s}_{1,i}=\frac{\sum_{\mathbf{c}\in\mathcal{S}}\!c e^{-|v_i-\lambda_i c|^{2}/\sigma^{2}}}{\sum_{\mathbf{c}\in\mathcal{S}}e^{-|v_i-\lambda_i c|^{2}/\sigma^{2}}}. \tag*{(12)}

whereas the non-separable case requires the joint posterior over all ∣S∣N|\mathcal{S}|^{N} candidates of s1\mathbf{s}_{1}. Since the expected complete-data log-likelihood equals −∥r−Asˉ1−μ(η)∥2/σ2-\|\mathbf{r}-A\bar{\mathbf{s}}_{1}-\boldsymbol{\mu}(\boldsymbol{\eta})\|^{2}/\sigma^{2} up to a constant, the M-step is the conventional interference-free estimator applied to the soft-cancelled signal r−Asˉ1\mathbf{r}-A\bar{\mathbf{s}}_{1}, similar to soft interference cancellation as in iterative multiuser detection [19]. In the separable case, each iteration requires only N∣S∣N|\mathcal{S}| distance evaluations, without backhaul data sharing or decoding. As summarized in Algorithm 1, the receiver is initialized interference-first with μ^=0\hat{\boldsymbol{\mu}}=\mathbf{0} in the first E-step, so that it remains robust when strong interference makes the initial MF estimate unreliable.

Algorithm 1 EM-Based Soft Interference Cancellation

Require: r\boldsymbol{r}, own data s0s_0, A=QΛ\boldsymbol{A}=\boldsymbol{Q}\boldsymbol{\Lambda}, S\mathcal{S}, σ2\sigma^2, and number of iterations TT

  1. Initialize μ^←0\hat{\boldsymbol{\mu}}\leftarrow\boldsymbol{0} (interference-first)

  2. for t=1,…,Tt=1,\ldots,T do

  3. v←QH(r−μ^)\boldsymbol{v}\leftarrow\boldsymbol{Q}^{H}(\boldsymbol{r}-\hat{\boldsymbol{\mu}})

  4. E-step: compute sˉ1,i\bar{s}_{1,i}, i=1,…,Ni=1,\ldots,N, via (12)

  5. Soft cancellation: r~←r−Asˉ1\tilde{\boldsymbol{r}}\leftarrow\boldsymbol{r}-\boldsymbol{A}\bar{\boldsymbol{s}}_1

  6. M-step: η^←arg⁡min⁡η∥r~−μ(η)∥2\hat{\eta}\leftarrow\arg\min_{\eta}\|\tilde{\boldsymbol{r}}-\boldsymbol{\mu}(\eta)\|^2 (global search at t=1t=1, local refinement for t>1t>1)

  7. μ^←μ(η^)\hat{\boldsymbol{\mu}}\leftarrow\boldsymbol{\mu}(\hat{\eta})

  8. end for

Ensure: η^\hat{\eta}

Numerical Results

Unless otherwise stated, both BSs employ CP-OFDM over a flat-fading BS-BS channel, for which Corollary 1 holds with γi=γ\gamma_i=\gamma. The MMSE in (11) has no closed form but is an expectation over the Gaussian noise, MMSE⁡S(γ)=1∣S∣∑c∈SEN[∣c−s^1(γc+N)∣2]\operatorname{MMSE}_{\mathcal{S}}(\gamma)=\frac{1}{|\mathcal{S}|}\sum_{c\in\mathcal{S}}\mathbb{E}_{N}[|c-\hat{s}_1(\sqrt{\gamma}c+N)|^2] with s^1(y)=E[S1∣Y=y]\hat{s}_1(y)=\mathbb{E}[S_1\mid Y=y] given in closed form by (12) with λi=γ\lambda_i=\sqrt{\gamma} and σ2=1\sigma^2=1. We evaluate it deterministically by Gauss-Hermite quadrature instead of Monte Carlo simulation.

CRB Loss versus INR

Figure 2 evaluates the CRB loss in (11) for various constellations. All constellations follow the Gaussian curve at low INR, reach a peak, and then collapse to 0 dB. The peak losses of QPSK, 16QAM, 64QAM, 256QAM, and 1024QAM are 2.8, 5.7, 8.7, 11.6, and 14.5 dB at γ⋆=2\gamma^\star=2, 8, 14, 19.5, and 25.5 dB, respectively. That is, each fourfold increase of the QAM order raises the peak by approximately 2.9 dB and shifts it by approximately 6 dB. PSK exhibits a lower peak of approximately 3.5 dB owing to its constant amplitude. For 64QAM at γ=20\gamma=20 dB, MF suffers a 20 dB CRB loss, whereas the optimal receiver can lose only 2 dB.

Receiver Performance

Figure 3 evaluates delay estimation with N=256N=256 subcarriers and 8 OFDM symbols, where BS0_0 transmits QPSK data with

Figure 3. Delay estimation RMSE versus INR under 16QAM and 64QAM.

a per-sample echo SNR of −10-10 dB and BS1_1 transmits 16QAM or 64QAM data. The EM receiver runs 10 iterations with a FFT-based delay estimator in the M-step, and the RMSE is averaged over 1000 trials. Since Corollary 1 scales the entire FIM, the delay CRB is inflated by the same factor LS(γ)L_{\mathcal{S}}(\gamma). The RMSE of MF follows the TIN bound, which equals 1+γ\sqrt{1+\gamma} times the interference-free bound, and breaks down at γ≈10\gamma\approx10 dB, where the integrated SINR falls below the detection threshold. In contrast, the EM receiver follows the finite-alphabet bound and enables sensing at INRs where TIN fails completely.

Waveform Comparison

Figure 4 compares SC, CP-OFDM, and OTFS (ND=2N_D=2) interferers with QPSK over a flat-fading channel and two frequency-selective channels (a two-tap channel with min⁡k∣Hk∣2=0.023\min_k |H_k|^2=0.023 and a three-tap channel with min⁡k∣Hk∣2=0.001\min_k |H_k|^2=0.001), where γ\gamma denotes the mean INR. We evaluate (7) with 1000 Monte Carlo samples of w\boldsymbol{w} per point. For a non-separable A\boldsymbol{A}, this requires the joint posterior of s1\boldsymbol{s}_1 over ∣S∣N|\mathcal{S}|^N hypotheses, whose number grows exponentially with NN. Therefore, we set N=8N=8 with 48=65,5364^8=65{,}536 hypotheses. Moreover, since Jw\boldsymbol{J}_{\boldsymbol{w}} is no longer a scaled identity, the CRB loss depends on the echo derivative in (8). We thus provide the average Fisher-information (FI) loss 1/ρeff1/\rho_{\mathrm{eff}} with ρeff=σ2tr⁡(Jw)/N\rho_{\mathrm{eff}}=\sigma^2\operatorname{tr}(\boldsymbol{J}_{\boldsymbol{w}})/N, which reduces to LS(γ)L_{\mathcal{S}}(\gamma) under Proposition 3. Over the flat-fading channel in Figure 4(a), all waveforms coincide, confirming Proposition 3. Over the frequency-selective channels, SC and OTFS with joint processing remain close to the flat curve, since each symbol spreads over all subcarriers and thus experiences the full channel energy. In contrast, CP-OFDM exhibits a lower peak but a residual tail, since deep-fade subcarriers remain in the harmful INR band as shown in Figure 4(b) and Figure 4(c).

Figure 4. Average Fisher-information loss for a QPSK interferer over (a) a flat-fading channel, (b) a two-tap channel, and (c) a three-tap channel.

Conclusion

In this paper, we have presented an estimation-theoretic framework for ISAC sensing under finite-alphabet inter-BS interference. For linear receivers, the impact of the inter-BS interference is invariant to the constellation and waveform. The exact FIM is scaled by 1−γMMSE⁡S(γ)1-\gamma\operatorname{MMSE}_{\mathcal{S}}(\gamma), so that the loss peaks at intermediate INR, vanishes at high INR, is waveform-independent for flat-fading and synchronized BS-BS links, and is approached by EM-based soft cancellation. Hence, inter-BS coordination should avoid intermediate INRs, where increasing the interference power may even improve sensing.

Appendix A: Proof of Proposition 1

Since E[s1s1H]=IN\mathbb{E}[\boldsymbol{s}_1\boldsymbol{s}_1^H]=I_N and GG is unitary, the second moment is

E[∣χ∣2]=E[s0HGHE[s1s1H]Gs0]=E[∥s0∥2]=N.(13)\mathbb{E}[|\chi|^2]=\mathbb{E}[\boldsymbol{s}_0^H G^H\mathbb{E}[\boldsymbol{s}_1\boldsymbol{s}_1^H]G\boldsymbol{s}_0]=\mathbb{E}[\|\boldsymbol{s}_0\|^2]=N. \tag*{(13)}

For the fourth moment, let g=Gs0\boldsymbol{g}=G\boldsymbol{s}_0 so that χ=∑i=1Ngis1,i∗\chi=\sum_{i=1}^{N}g_i s_{1,i}^*. For i.i.d. proper symbols with unit power and kurtosis μ4\mu_4, the only non-zero fourth-order moments are E[si∗sksl∗sl]=μ4\mathbb{E}[s_i^*s_k s_l^*s_l]=\mu_4 for i=j=k=li=j=k=l and 1 for {i,j}={k,l}\{i,j\}=\{k,l\} with i≠ji\ne j, since E[si2]=0\mathbb{E}[s_i^2]=0. Conditioning on g\boldsymbol{g} yields

E[∣χ∣4∣g]=2∥g∥4+(μ4,s1−2)∑i=1N∣gi∣4.(14)\mathbb{E}[|\chi|^4\mid\boldsymbol{g}]=2\|\boldsymbol{g}\|^4+(\mu_{4,s_1}-2)\sum_{i=1}^{N}|g_i|^4. \tag*{(14)}

Since ∥g∥=∥s0∥\|\boldsymbol{g}\|=\|\boldsymbol{s}_0\|, the first term gives

E[∥s0∥4]=∑iE[∣s0,i∣4]+∑i≠jE[∣s0,i∣2]E[∣s0,j∣2]=N2+N(μ4,s0−1).(15)\mathbb{E}[\|\boldsymbol{s}_0\|^4]=\sum_i\mathbb{E}[|s_{0,i}|^4]+\sum_{i\ne j}\mathbb{E}[|s_{0,i}|^2]\mathbb{E}[|s_{0,j}|^2]=N^2+N(\mu_{4,s_0}-1). \tag*{(15)}

Applying (14) again to gi=∑jGijs0,jg_i=\sum_jG_{ij}s_{0,j} and using ∑j∣Gij∣2=1\sum_j|G_{ij}|^2=1 for each row of the unitary GG,

∑i=1NE[∣gi∣4]=∑i=1N(2+(μ4,s0−2)∑j∣Gij∣4)=2N+(μ4,s0−2)∥G∥44.(16)\begin{aligned} \sum_{i=1}^{N}\mathbb{E}[|g_i|^4] &=\sum_{i=1}^{N}\left(2+(\mu_{4,s_0}-2)\sum_j|G_{ij}|^4\right)\\ &=2N+(\mu_{4,s_0}-2)\|G\|_4^4. \tag*{(16)} \end{aligned}

Substituting (15) and (16) into (14) yields (6). Finally, since each row of the unitary GG has unit norm, 1/N≤∑j∣Gij∣4≤11/N\le\sum_j|G_{ij}|^4\le1 with equality iff ∣Gij∣2=1/N|G_{ij}|^2=1/N and iff the row has a single non-zero entry, respectively, which gives ∥G∥44∈[1,N]\|G\|_4^4\in[1,N], completing the proof.

Appendix B: Proof of Lemma 1

The density of w\boldsymbol{w} is pw(w)=Es1[pn(w−As1)]p_{\boldsymbol{w}}(\boldsymbol{w})=\mathbb{E}_{s_1}[p_{\boldsymbol{n}}(\boldsymbol{w}-A s_1)] with pn(n)=(πσ2)−Ne−∥n∥2/σ2p_{\boldsymbol{n}}(\boldsymbol{n})=(\pi\sigma^2)^{-N}e^{-\|\boldsymbol{n}\|^2/\sigma^2}. Since ∇w∗pn(w−As1)=−σ−2(w−As1)pn(w−As1)\nabla_{\boldsymbol{w}^*}p_{\boldsymbol{n}}(\boldsymbol{w}-A s_1)=-\sigma^{-2}(\boldsymbol{w}-A s_1)p_{\boldsymbol{n}}(\boldsymbol{w}-A s_1), Bayes’ rule gives

sw(w)=∇w∗pw(w)pw(w)=−Es1[(w−As1)pn(w−As1)]σ2pw(w)=−σ−2E[n∣w].(17)\boldsymbol{s}_{\boldsymbol{w}}(\boldsymbol{w})=\frac{\nabla_{\boldsymbol{w}^*}p_{\boldsymbol{w}}(\boldsymbol{w})}{p_{\boldsymbol{w}}(\boldsymbol{w})}=-\frac{\mathbb{E}_{s_1}[(\boldsymbol{w}-A s_1)p_{\boldsymbol{n}}(\boldsymbol{w}-A s_1)]}{\sigma^2p_{\boldsymbol{w}}(\boldsymbol{w})}=-\sigma^{-2}\mathbb{E}[\boldsymbol{n}\mid\boldsymbol{w}]. \tag*{(17)}

By the law of total covariance with E[n]=0\mathbb{E}[\boldsymbol{n}]=0, En[nnH]=Ew[cov⁡(n∣w)]+Ew[E[n∣w]E[n∣w]H]\mathbb{E}_{\boldsymbol{n}}[\boldsymbol{n}\boldsymbol{n}^H]=\mathbb{E}_{\boldsymbol{w}}[\operatorname{cov}(\boldsymbol{n}\mid\boldsymbol{w})]+\mathbb{E}_{\boldsymbol{w}}[\mathbb{E}[\boldsymbol{n}\mid\boldsymbol{w}]\mathbb{E}[\boldsymbol{n}\mid\boldsymbol{w}]^H], and hence

Jw=1σ4E[E[n∣w]E[n∣w]H]=1σ4(σ2I−E[cov⁡(n∣w)]).(18)J_{\boldsymbol{w}}=\frac{1}{\sigma^4}\mathbb{E}[\mathbb{E}[\boldsymbol{n}\mid\boldsymbol{w}]\mathbb{E}[\boldsymbol{n}\mid\boldsymbol{w}]^H]=\frac{1}{\sigma^4}\left(\sigma^2I-\mathbb{E}[\operatorname{cov}(\boldsymbol{n}\mid\boldsymbol{w})]\right). \tag*{(18)}

Since n=w−As1\boldsymbol{n}=\boldsymbol{w}-A s_1 and w\boldsymbol{w} is given, cov⁡(n∣w)=cov⁡(As1∣w)\operatorname{cov}(\boldsymbol{n}\mid\boldsymbol{w})=\operatorname{cov}(A s_1\mid\boldsymbol{w}), and substituting this into (18) yields (7).

Under (A1) and (A4), the circularity of n\boldsymbol{n} implies that jwj\boldsymbol{w} has the same distribution as w\boldsymbol{w}, satisfying pw(jw)=pw(w)p_{\boldsymbol{w}}(j\boldsymbol{w})=p_{\boldsymbol{w}}(\boldsymbol{w}). The Wirtinger chain rule with ∂(jw)∗/∂w∗=−jI\partial(j\boldsymbol{w})^*/\partial\boldsymbol{w}^*=-jI yields sw(jw)=jsw(w)\boldsymbol{s}_{\boldsymbol{w}}(j\boldsymbol{w})=j\boldsymbol{s}_{\boldsymbol{w}}(\boldsymbol{w}), and therefore

E[swswT]=E[sw(jw)sw(jw)T]=−E[swswT],(19)\mathbb{E}[\boldsymbol{s}_{\boldsymbol{w}}\boldsymbol{s}_{\boldsymbol{w}}^T]=\mathbb{E}[\boldsymbol{s}_{\boldsymbol{w}}(j\boldsymbol{w})\boldsymbol{s}_{\boldsymbol{w}}(j\boldsymbol{w})^T]=-\mathbb{E}[\boldsymbol{s}_{\boldsymbol{w}}\boldsymbol{s}_{\boldsymbol{w}}^T], \tag*{(19)}

which implies E[swswT]=0\mathbb{E}[\boldsymbol{s}_{\boldsymbol{w}}\boldsymbol{s}_{\boldsymbol{w}}^T]=0. This completes the proof.

Appendix C: Proof of Theorem 1

The log-likelihood is given by ln⁡pw(r−μ(η))\ln p_{\boldsymbol{w}}(\boldsymbol{r}-\boldsymbol{\mu}(\boldsymbol{\eta})). Let ai=∂μ/∂ηi\boldsymbol{a}_i=\partial\boldsymbol{\mu}/\partial\eta_i. Since ln⁡pw\ln p_{\boldsymbol{w}} is real-valued, ∇wln⁡pw=sw∗\nabla_{\boldsymbol{w}}\ln p_{\boldsymbol{w}}=\boldsymbol{s}_{\boldsymbol{w}}^*, and the chain rule gives

∂∂ηiln⁡pw(r−μ)=−swHai−swTai∗=−2ℜ{swHai}.(20)\frac{\partial}{\partial\eta_i}\ln p_{\boldsymbol{w}}(\boldsymbol{r}-\boldsymbol{\mu})=-\boldsymbol{s}_{\boldsymbol{w}}^H\boldsymbol{a}_i-\boldsymbol{s}_{\boldsymbol{w}}^T\boldsymbol{a}_i^*=-2\Re\{\boldsymbol{s}_{\boldsymbol{w}}^H\boldsymbol{a}_i\}. \tag*{(20)}

Using ℜ{u}ℜ{v}=12ℜ{uv∗}+12ℜ{uv}\Re\{u\}\Re\{v\}=\frac{1}{2}\Re\{uv^*\}+\frac{1}{2}\Re\{uv\} with u=swHaiu=\boldsymbol{s}_{\boldsymbol{w}}^H\boldsymbol{a}_i and v=swHajv=\boldsymbol{s}_{\boldsymbol{w}}^H\boldsymbol{a}_j,

[F]ij=4E[ℜ{swHai}ℜ{swHaj}]=2ℜ{ajHJwai}+2ℜ{aiTE[sw∗swH]aj}.(21)[F]_{ij}=4\mathbb{E}\left[\Re\{\boldsymbol{s}_{\boldsymbol{w}}^H\boldsymbol{a}_i\}\Re\{\boldsymbol{s}_{\boldsymbol{w}}^H\boldsymbol{a}_j\}\right] =2\Re\{\boldsymbol{a}_j^H J_{\boldsymbol{w}}\boldsymbol{a}_i\}+2\Re\{\boldsymbol{a}_i^T\mathbb{E}[\boldsymbol{s}_{\boldsymbol{w}}^*\boldsymbol{s}_{\boldsymbol{w}}^H]\boldsymbol{a}_j\}. \tag*{(21)}

where the second term vanishes by Lemma 1, and ℜ{aiHJwaj}=ℜ{aiHJwaj}\Re\{\boldsymbol{a}_i^H J_{\boldsymbol{w}}\boldsymbol{a}_j\}=\Re\{\boldsymbol{a}_i^H J_{\boldsymbol{w}}\boldsymbol{a}_j\} since JwJ_{\boldsymbol{w}} is Hermitian, which gives (8). For the bound, the MMSE error covariance is dominated by that of the LMMSE estimator, i.e., E[cov⁡(As1∣w)]⪰\mathbb{E}[\operatorname{cov}(A s_1\mid\boldsymbol{w})]\succeq B−B(B+σ2I)−1BB-B(B+\sigma^2I)^{-1}B with B=AAHB=AA^H, with equality for Gaussian s1s_1. Substituting this into (7) yields

Jw⪰1σ2(I−B(B+σ2I)−1)=(σ2I+AAH)−1.(22)J_w \succeq\frac{1}{\sigma^2}\left(I-B(B+\sigma^2I)^{-1}\right)=(\sigma^2I+AA^H)^{-1}. \tag*{(22)}

Appendix D: Proof of Corollary 1

If A=QΛA=Q\Lambda, the sufficient statistic v=QHw=Λs1+QHnv=Q^Hw=\Lambda s_1+Q^Hn consists of NN independent scalar channels vi=λis1,i+n~iv_i=\lambda_i s_{1,i}+\tilde n_i, since QHn∼CN(0,σ2I)Q^Hn\sim\mathcal{CN}(0,\sigma^2I) and the entries of s1s_1 are i.i.d. Normalizing by σ\sigma and noting that the MMSE is invariant to the known phase of λi\lambda_i, we have E[cov⁡(s1∣w)]=diag⁡(MMSE⁡S(γi))\mathbb{E}[\operatorname{cov}(s_1|w)]=\operatorname{diag}(\operatorname{MMSE}_S(\gamma_i)), and thus

E[cov⁡(As1∣w)]=QΛdiag⁡(MMSE⁡S(γi))ΛHQH=σ2Qdiag⁡(γiMMSE⁡S(γi))QH.(23)\mathbb{E}\left[\operatorname{cov}(As_1|w)\right]=Q\Lambda\operatorname{diag}(\operatorname{MMSE}_S(\gamma_i))\Lambda^HQ^H =\sigma^2Q\operatorname{diag}\left(\gamma_i\operatorname{MMSE}_S(\gamma_i)\right)Q^H. \tag*{(23)}

Substituting this into (7) yields (10). If γi=γ\gamma_i=\gamma, then Jw=ρS(γ)I/σ2J_w=\rho_S(\gamma)I/\sigma^2, so that (8) becomes ρS(γ)\rho_S(\gamma) times the interference-free FIM F0F_0 [20], and F−1=F0−1/ρS(γ)F^{-1}=F_0^{-1}/\rho_S(\gamma), completing the proof.

Appendix E: Proof of Proposition 2

(i) Consider the linear estimator S^1,L=γ1+γY\hat S_{1,L}=\frac{\sqrt{\gamma}}{1+\gamma}Y, whose mean-square error is 1/(1+γ)1/(1+\gamma). Since the MMSE estimator minimizes the mean-square error, we have 0≤MMSE⁡S(γ)≤1/(1+γ)0\leq\operatorname{MMSE}_S(\gamma)\leq1/(1+\gamma). Substituting these bounds into (11) yields 1≤LS(γ)≤1+γ1\leq L_S(\gamma)\leq1+\gamma. The upper bound is attained for Gaussian S1S_1, for which the conditional mean estimator is linear.

(ii) From (11), we have

LS(γ)−1γ=MMSE⁡S(γ)1−γMMSE⁡S(γ).(24)\frac{L_S(\gamma)-1}{\gamma}=\frac{\operatorname{MMSE}_S(\gamma)}{1-\gamma\operatorname{MMSE}_S(\gamma)}. \tag*{(24)}

As γ→0\gamma\to0, the observation Y=γS1+NY=\sqrt{\gamma}S_1+N becomes independent of S1S_1. Consequently, E[S1∣Y]\mathbb{E}[S_1|Y] converges to E[S1]=0\mathbb{E}[S_1]=0, implying MMSE⁡S(γ)→E[∣S1∣2]=1\operatorname{MMSE}_S(\gamma)\to\mathbb{E}[|S_1|^2]=1. Since γMMSE⁡S(γ)→0\gamma\operatorname{MMSE}_S(\gamma)\to0, it follows that lim⁡γ→0(LS(γ)−1)/γ=1\lim_{\gamma\to0}(L_S(\gamma)-1)/\gamma=1.

(iii) Let S^1,MD\hat S_{1,\mathrm{MD}} denote the minimum-distance detector, and let dmin⁡d_{\min} and dmax⁡d_{\max} be the minimum and maximum distances between distinct constellation points, respectively. Since the MMSE estimator minimizes the mean-square error, we have

MMSE⁡S(γ)≤E[∣S1−S^1,MD∣2]≤dmax⁡2Pe(γ),(25)\operatorname{MMSE}_S(\gamma)\leq\mathbb{E}[|S_1-\hat S_{1,\mathrm{MD}}|^2]\leq d_{\max}^2P_e(\gamma), \tag*{(25)}

where Pe(γ)=Pr⁡(S^1,MD≠S1)P_e(\gamma)=\Pr(\hat S_{1,\mathrm{MD}}\ne S_1). For an MM-ary constellation in complex AWGN, the union bound and Q(x)≤12e−x2/2Q(x)\leq\frac{1}{2}e^{-x^2/2} give

Pe(γ)≤(M−1)Q(γdmin⁡22)≤M−12e−γdmin⁡2/4.(26)P_e(\gamma)\leq(M-1)Q\left(\sqrt{\frac{\gamma d_{\min}^2}{2}}\right)\leq\frac{M-1}{2}e^{-\gamma d_{\min}^2/4}. \tag*{(26)}

Combining these inequalities yields

0≤γMMSE⁡S(γ)≤(M−1)dmax⁡22γe−γdmin⁡2/4.(27)0\leq\gamma\operatorname{MMSE}_S(\gamma)\leq\frac{(M-1)d_{\max}^2}{2}\gamma e^{-\gamma d_{\min}^2/4}. \tag*{(27)}

Since dmin⁡>0d_{\min}>0, the right-hand side converges to zero as γ→∞\gamma\to\infty. Thus, γMMSE⁡S(γ)→0\gamma\operatorname{MMSE}_S(\gamma)\to0, and (11) gives LS(γ)→1L_S(\gamma)\to1.

(iv) LS(γ)L_S(\gamma) is continuous and approaches unity as γ→0\gamma\to0 and γ→∞\gamma\to\infty, as established in (ii) and (iii), and exceeds unity for γ>0\gamma>0 since MMSE⁡S(γ)>0\operatorname{MMSE}_S(\gamma)>0. Therefore, there exist 0<a<b<∞0<a<b<\infty such that the global maximum is attained within [a,b][a,b]. By continuity, LS(γ)L_S(\gamma) attains its maximum at some finite γ⋆∈(0,∞)\gamma^\star\in(0,\infty), completing the proof.

References

  1. [1]F. Liu, Y.-F. Liu, Y. Cui, C. Masouros, J. Xu, T. X. Han, S. Buzzi, Y. C. Eldar, and S. Jin, “Sensing with communication signals: From information theory to signal processing,” IEEE Journal on Selected Areas in Communications, 2025.arxiv.org/abs/2502.10819
  2. [2]C. Sturm and W. Wiesbeck, “Waveform design and signal processing aspects for fusion of wireless communications and radar sensing,” Proceedings of the IEEE, vol. 99, no. 7, pp. 1236–1259, 2011.DOI
  3. [3]K. Han, K. Meng, X.-Y. Wang, and C. Masouros, “Network-level ISAC design: State-of-the-art, challenges, and opportunities,” IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation, 2025.arxiv.org/abs/2505.01295
  4. [4]K. Meng, K. Han, C. Masouros, and L. Hanzo, “Network-level ISAC: An analytical study of antenna topologies ranging from massive to cell-free MIMO,” IEEE Transactions on Wireless Communications, vol. 24, no. 12, pp. 10 003–10 018, 2025.DOI
  5. [5]K. Meng, C. Masouros, G. Chen, and F. Liu, “Network-level integrated sensing and communication: Interference management and BS coordination using stochastic geometry,” IEEE Transactions on Wireless Communications, vol. 23, no. 12, pp. 19 365–19 381, 2024.arxiv.org/abs/2311.09052
  6. [6]F. Liu, Y. Zhang, X. Xiong, S. Li, W. Yuan, F. Gao, S. Jin, and G. Caire, “CP-OFDM achieves the lowest average ranging sidelobe under QAM/PSK constellations,” IEEE Transactions on Information Theory, 2025.arxiv.org/abs/2407.06691
  7. [7]M. F. Keskin, M. M. Mojabedian, J. O. Lacruz, C. Marcus, M. Eriksson, A. Giorgetti, J. Widmer, and H. Wymeersch, “Fundamental trade-offs in monostatic ISAC: A holistic investigation toward 6G,” IEEE Transactions on Wireless Communications, vol. 24, no. 9, pp. 7856–7873, 2025.DOI
  8. [8]Z. Du, F. Liu, Y. Xiong, T. X. Han, Y. C. Eldar, and S. Jin, “Reshaping the ISAC tradeoff under OFDM signaling: A probabilistic constellation shaping approach,” IEEE Transactions on Signal Processing, vol. 72, pp. 4782–4797, 2024.arxiv.org/abs/2312.15941
  9. [10]K. Meng, K. Han, C. Masouros, and F. Liu, “Constellation selection and power control for OFDM-based ISAC: From theory to prototype,” IEEE Transactions on Signal Processing, 2026.arxiv.org/abs/2603.03895
  10. [11]K. Han, K. Meng, A. Chatzicharistou, and C. Masouros, “OFDM-ISAC over Data Payloads: MSE Analysis, Constellation Design, and Experimentation,” arXiv preprint arXiv:2608.15564, 2026.arxiv.org/abs/2608.15564
  11. [12]K. Meng, K. Han, C. Masouros, and L. Hanzo, “Constellation Selection and Power Allocation for Multi-Cell OFDM-ISAC: Managing Inter-Cell Interference and Sensing Sidelobes,” arXiv preprint arXiv:2607.21418, 2026.arxiv.org/abs/2607.21418
  12. [13]H. Pu, A. Musallam, H. Li, and L. Hanzo, “Blind OFDM-ISAC relying on asymmetric modem constellations,” arXiv:2604.26200, 2026.arxiv.org/abs/2604.26200
  13. [14]A. Dytso, D. Tuninetti, and N. Devroye, “Interference or noise: Friend or foe?” IEEE Transactions on Information Theory, vol. 62, no. 6, pp. 3561–3596, 2016.arxiv.org/abs/1506.02597
  14. [15]P. Stoica and P. Babu, “The Gaussian data assumption leads to the largest Cramér–Rao bound,” IEEE Signal Processing Magazine, vol. 28, no. 3, pp. 132–133, 2011.DOI
  15. [16]L. D. Brown, “A proof of the central limit theorem motivated by the Cramér–Rao inequality,” in Statistics and Probability: Essays in Honor of C. R. Rao. North-Holland, 1982, pp. 141–148.
  16. [17]M. Payaró and D. Palomar, “Hessian and concavity of mutual information, differential entropy, and entropy power in linear vector Gaussian channels,” IEEE Transactions on Information Theory, vol. 55, no. 8, pp. 3613–3628, 2009.arxiv.org/abs/0903.1945
  17. [18]A. P. Dempster, N. M. Laird, and D. B. Rubin, “Maximum likelihood from incomplete data via the EM algorithm,” Journal of the Royal Statistical Society Series B (Methodological), vol. 39, no. 1, pp. 1–38, 1977.DOI
  18. [19]X. Wang and H. V. Poor, “Iterative (turbo) soft interference cancellation and decoding for coded CDMA,” IEEE Transactions on Communications, vol. 47, no. 7, pp. 1046–1061, 1999.DOI
  19. [20]S. M. Kay, Fundamentals of Statistical Signal Processing: Estimation Theory. Upper Saddle River, NJ: Prentice Hall, 1993.

Paper details

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

views · 11

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