Joint Transmit–Receive Weight Optimization for FDA Radar to Balance Active Detection and RF Stealth Authors: Haoliang Guan, Shunsheng Zhang, Wen-Qin Wang Published: 2026 ABSTRACT Existing studies on frequency diverse array (FDA) radar sensing systems have primarily focused on radio-frequency (RF) stealth characteristics with limited attention to the balance between RF stealth and active detection performance. To address this issue, this paper proposes a joint transmit–receive weight optimization scheme for FDA radar systems to achieve an effective balance between active detection and RF stealth. The resulting optimization problem is non-convex, and a block coordinate descent (BCD)-based alternating optimization method with a carefully designed initialization strategy is developed to solve it efficiently. Simulation results demonstrate that the proposed method achieves improved RF stealth performance while maintaining comparable active detection capability, compared with conventional FDA radar and representative existing optimization-based benchmark methods. These results demonstrate the effectiveness of the proposed method for balancing active detection and RF stealth performance in FDA radar sensing systems. 1. Introduction The frequency diverse array (FDA) radar, as a novel radar sensing paradigm, has attracted increasing attention in recent years. It was first introduced by Antonik et al. at the U.S. Air Force Research Laboratory [1]. By introducing slight frequency offsets across array elements, FDA radar produces a range–angle–time-dependent transmit beam pattern and provides additional degrees of freedom for sensing-oriented signal design and performance regulation. As a result, FDA radar has been widely investigated in various applications, including target detection [2,3], parameter estimation [4,5], mainlobe jamming suppression [6,7], and radar imaging and sensing [8,9]. In many radar sensing applications, such as airborne or maritime surveillance, the radar system is required not only to actively detect and track targets, but also to reduce the probability of being intercepted by external passive receivers. This requires the radar sensing system to simultaneously maintain reliable active detection performance and low interceptability [10,11]. These requirements make FDA radar particularly attractive for sensing scenarios that require both reliable target sensing and low interceptability [1,2,8]. To the best of the authors’ knowledge, research on the RF stealth performance of FDA radar remains limited. Existing studies in the FDA framework have mainly investigated this topic from two perspectives, namely LPI/RF stealth characteristic analysis and waveform/beamforming design. For example, some studies have examined the LPI and RF-stealth-related characteristics of FDA-transmitted signals and their applications in passive localization deception and counter-DOA estimation [12]. In addition, FDA and FDA-MIMO-related studies have also considered low-probability-of-intercept transmit beamforming, waveform design, and joint transmit waveform–receive beamforming design for improving RF stealth performance [13,14,15]. However, despite these efforts, how to effectively coordinate active detection and RF stealth for FDA radar remains insufficiently explored. Motivated by this gap, this paper formulates a joint transmit–receive weight optimization problem for FDA radar systems to balance active detection and RF stealth performance. The probability of detection and probability of interception are adopted to characterize the radar-side detection capability and intercept-side exposure risk, respectively. A block coordinate descent (BCD)-based alternating optimization method with a carefully designed initialization strategy is then developed to solve the resulting non-convex optimization problem. In addition, the direction-of-arrival (DOA) estimation accuracy at the intercept receiver is further evaluated in the simulation section to characterize the adversarial parameter estimation capability. 2. Signal Model and Problem Formulation 2.1. Signal Model As shown in Figure 1, the considered FDA radar system is used for target sensing, while a potential intercept receiver is located at (RI,θI). The FDA radar system is equipped with a uniform linear transmit array of M elements and a uniform linear receive array of N elements. Let dT and dR denote the inter-element spacings of the transmit and receive arrays, respectively. In the conventional FDA scheme, a fixed frequency increment is applied between adjacent transmit elements. Accordingly, the carrier frequency of the mth transmit element is given by(1)fm=f0+mΔf,m=0,1,…,M−1 where f0 is the reference carrier frequency and Δf is the fixed frequency increment. This scheme is referred to as FDAconv in the simulations. Let φm(t) denote the baseband waveform transmitted by the mth element over one pulse duration Tp, and let(2)wT=[w0,w1,…,wM−1]T∈CM×1 denote the transmit weight vector. Then, the transmitted signal of the mth element can be written as(3)sm(t)=wmφm(t)ej2πfmt,0≤t≤Tp. The transmitted baseband waveforms are assumed to be orthogonal, i.e.,(4)∫0Tpφm(t)φm′*(t)dt=δmm′, where δmm′ denotes the Kronecker delta. For a far-field point target located at (r,θ), the reference round-trip delay is(5)τ(r)=2rc, where c is the speed of light. The transmit-side and receive-side angular delays are respectively given by(6)τT,m(θ)=mdTsinθc,τR,n(θ)=ndRsinθc, for m=0,1,…,M−1 and n=0,1,…,N−1. Under the far-field and narrowband assumptions, the echo received at the nth receive element due to the mth transmit element can be approximated as(7)xn,m(t;r,θ)≈αswmφmt−τ(r)ej2πfm(t−τ(r))e−j2πfmτT,m(θ)e−j2πf0τR,n(θ), where αs∈C denotes the effective complex scattering coefficient of the target. Define(8)bm(r,θ)≜e−j2πfmτ(r)e−j2πfmτT,m(θ),m=0,1,…,M−1, and stack them into(9)b(r,θ)=[b0(r,θ),b1(r,θ),…,bM−1(r,θ)]T∈CM×1. Then, the effective transmit vector is(10)u(r,θ)=wT⊙b(r,θ)∈CM×1, where ⊙ denotes the Hadamard product. Under the narrowband receive-array approximation, the receive steering vector is expressed as(11)aR(θ)=1,e−j2πf0cdRsinθ,…,e−j2πf0c(N−1)dRsinθT∈CN×1. After multi-channel mixing and matched filtering, ideal channel separation is achieved under the orthogonal waveform assumption, and the output of the (n,m)th channel can be approximated as(12)yn,m≈αswme−j2πfmτ(r)e−j2πfmτT,m(θ)e−j2πf0τR,n(θ)+ηn,m, where ηn,m denotes the matched-filter output noise. Stacking all matched-filter outputs yields(13)Y≈αsaR(θ)uT(r,θ)+N, where Y,N∈CN×M, and N=[ηn,m] denotes the noise matrix. Let(14)z≜vec(Y)∈CMN×1,n≜vec(N)∈CMN×1, where vec(·) denotes the vectorization operator. Then,(15)z≈αsu(r,θ)⊗aR(θ)+n, where ⊗ denotes the Kronecker product. The vectorized noise in the matched-filter output domain is assumed to satisfy(16)n∼CN(0,σn2IMN). based on the above signal model, the receive-side output SNR and the intercept-side SNR are defined in the next subsection for joint optimization. Unlike the radar receiver, which exploits waveform orthogonality through matched filtering, the potential intercept receiver is assumed to observe the superposed transmitted waveform directly without channel separation. Accordingly, the intercept-side SNR is defined based on the average intercepted signal power over one pulse. 2.2. Joint Optimization Problem Consider the desired target located at (rT,θT). Define the equivalent target response vector as(17)gT(wT)≜wT⊙b(rT,θT)⊗aR(θT)∈CMN×1. Then, the vectorized received signal can be rewritten as(18)z≈αsgT(wT)+n, and the receive processor output is(19)y=wRHz. Accordingly, the radar-side output SNR is given by(20)SNR(wT,wR)=|αs|2wRHgT(wT)2σn2∥wR∥22. For an intercept receiver located at (RI,θI), the propagation delay is given by τI=RI/c. Accordingly, the corresponding noiseless intercepted signal is expressed as(21)sI(t;θI,RI)=αIaTT(θI)(Φ(t−τI)wT)⊙eI(t−τI), and the corresponding intercepted signal is(22)yI(t;θI,RI)=sI(t;θI,RI)+nI(t), where αI∈C denotes the complex intercepted-signal amplitude, nI(t)∼CN(0,σI2) denotes the intercept-side noise, and(23)aT(θI)=1,e−j2πf0cdTsinθI,…,e−j2πf0c(M−1)dTsinθIT∈CM×1,(24)Φ(t)=diag{φ0(t),φ1(t),…,φM−1(t)}∈CM×M, and(25)eI(t)=1,ej2πΔft,…,ej2π(M−1)ΔftT∈CM×1. The average intercepted signal power over one pulse is defined as(26)Ps,I(wT)≜1Tp∫0TpsI(t;θI,RI)2dt. after expansion, it can be written in quadratic form as(27)Ps,I(wT)=|αI|2wTHRIwT, where(28)RI≜diagaT(θI)HRϕ(Δf,φ)diagaT(θI)∈CM×M, and the equivalent waveform-coupling matrix is defined by(29)Rϕ(Δf,φ)m,m′≜1Tp∫0Tpφm(t)φm′*(t)ej2π(m−m′)Δftdt,m,m′=0,1,…,M−1. Unlike the matched-filter output at the radar receiver, the matrix Rϕ(Δf,φ) is retained here in its general form and is not approximated by the identity matrix. Therefore, the intercept-side SNR is given by(30)SNRI(wT)=Ps,I(wT)σI2=|αI|2σI2wTHRIwT. The intercept-side SNR provides the basis for evaluating the interception probability and is also related to the adversarial parameter estimation capability after successful interception. In the simulation section, the DOA estimation accuracy at the intercept receiver is further examined as a representative example. To balance active detection and RF stealth performance, the following weighted objective function is adopted:(31)Γ(wT,wR)=α1SNRI,0SNRI(wT)+(1−α1)SNR(wT,wR)SNRD,0,0≤α1≤1 where SNRI,0=SNRI(wT(0)) and SNRD,0=SNR(wT(0),wR(0)) are the fixed reference SNRs obtained from the initial feasible FDA configuration and expressed in linear scale, so both terms in (31) become dimensionless and are scaled to a similar order of magnitude. The weighting factor α1 controls the trade-off between RF stealth and active detection performance, thereby enabling a flexible balance between the two objectives. Notably, the reciprocal structure of the intercept-side term acts as a stealth-oriented objective component by converting the reduction in the intercept-side SNR into a maximization-oriented design criterion. Compared with a simple linear combination of the radar-side and intercept-side SNRs, this formulation more explicitly captures the opposite design requirements of active detection enhancement and RF stealth improvement. Accordingly, the joint transmit–receive weight optimization problem is formulated as(32)maxwT,wRΓ(wT,wR)s.t.∥wT−wT0∥22≤ε,∥wT∥22=M. here, wT0∈CM×1 denotes the target-oriented reference transmit weight vector, and ε controls the allowable deviation from this reference. Under the transmit-power constraint in (32), the constraint yields the following conservative lower bound on the normalized target-direction power gain η:(33)ε≤2M(1−η) thus, ε can be selected according to the allowable target-direction gain loss. In the simulations, M=16 and ε=2 correspond to a worst-case gain loss bounded by approximately 0.56 dB. Problem (32) is non-convex and generally difficult to solve globally due to the nonlinear objective and the coupling between the transmit and receive weight vectors. Under the considered signal model, active detection performance is mainly preserved by the target-oriented transmit initialization, the similarity constraint, and the receive-side SNR maximization, whereas RF stealth is primarily controlled through transmit weight design. Therefore, an efficient alternating optimization strategy is developed in the next section to obtain a high-quality suboptimal solution. 3. Proposed Optimization Method To solve (32), a block coordinate descent (BCD)-based alternating optimization method is adopted. Specifically, the optimization variables are partitioned into two blocks, namely the transmit weight vector wT and the receive weight vector wR. In each iteration, the receive weight vector is updated by maximizing the radar-side output SNR for the current transmit weight vector, followed by a relaxation step for numerical stability, while the transmit weight vector is updated by reducing the intercept-side SNR subject to the similarity and power constraints. The proposed method consists of transmit weight initialization, transmit weight update, receive weight update, and the corresponding alternating optimization algorithm. 3.1. Transmit Weight Initialization Since the performance of the BCD algorithm is sensitive to initialization, a target-oriented reference transmit weight vector is first constructed. Specifically, before introducing the RF stealth requirement into the transmit weight design, the transmit weight vector is initialized by maximizing the coherent transmit gain toward the desired target location (rT,θT) under a fixed transmit-power constraint. Let(34)bT≜b(rT,θT)∈CM×1, where b(r,θ) is defined in the previous subsection. The reference transmit weight vector is obtained by solving(35)maxwTbTTwT2s.t.∥wT∥22=M. According to the Cauchy–Schwarz inequality, the optimal solution to (35) is given by(36)wT0=MbT*∥bT∥2. it is straightforward to verify that ∥wT0∥22=M. The obtained wT0 is then used as the reference transmit weight vector in the subsequent similarity constraint. 3.2. Transmit Weight Update Under the alternating optimization framework, the transmit weight vector is refined to reduce the intercept-side SNR while preserving the target-oriented transmit characteristic through the similarity constraint. Accordingly, the transmit weight update is carried out by solving(37)minwTSNRI(wT)=|αI|2σI2wTHRIwTs.t.∥wT−wT0∥22≤ε,∥wT∥22=M. The gradient of SNRI(wT) with respect to wT* is given by(38)∇wT*SNRI=|αI|2σI2RIwT. the descent direction at the kth iteration is defined as(39)dT(k)=∇wT*SNRIwT(k). For a candidate step size μ>0, construct the normalized trial point as(40)w˜T(μ)=MwT(k)−μdT(k)∥wT(k)−μdT(k)∥2, which automatically satisfies the power constraint. The step size is selected by backtracking line search. Specifically, let μ=μ0βℓ, where μ0>0, 0<β<1, and ℓ=0,1,2,…. Starting from ℓ=0, the smallest ℓ is chosen such that(41)∥w˜T(μ)−wT0∥22≤ε and(42)SNRIw˜T(μ)≤SNRIwT(k). The transmit weight vector is then updated as(43)wT(k+1)=w˜T(μk), where μk denotes the accepted step size at the kth iteration. 3.3. Receive Weight Update For a fixed transmit weight vector wT(k), define(44)gT(k)≜gT(wT(k))=wT(k)⊙b(rT,θT)⊗aR(θT). Then, the receive weight update is obtained by maximizing the radar-side output SNR, namely(45)maxwRwRHgT(k)2∥wR∥22. Under the white noise assumption, (45) is a standard Rayleigh quotient problem. Equivalently, it can be viewed as a special case of the minimum variance distortionless response (MVDR) filter when the interference-plus-noise covariance matrix reduces to a scaled identity matrix. Therefore, the intermediate receive weight vector is first obtained as(46)w^R(k+1)=gT(k)∥gT(k)∥2. To improve numerical stability and avoid abrupt receiver adaptation during the iterative process, a relaxation step is further introduced:(47)wR(k+1)=(1−ρ)wR(k)+ρw^R(k+1)(1−ρ)wR(k)+ρw^R(k+1)2,0<ρ≤1, where ρ is the relaxation factor. 3.4. BCD-Based Alternating Optimization Algorithm Based on the above two update steps, the overall alternating optimization algorithm is summarized as follows. 1.Initialize the transmit weight vector as wT(0)=wT0 using (36), and initialize the receive weight vector as(48)wR(0)=gT(wT(0))∥gT(wT(0))∥2.Set the relaxation factor ρ, the maximum iteration number Kmax, and the convergence threshold ξ, and let k=0.2.For the current transmit weight vector wT(k), compute the intermediate receive weight vector w^R(k+1) according to (46), and then update wR(k+1) according to (47).3.Update the transmit weight vector wT(k+1) according to (38)–(43).4.Evaluate the relative change in the intercept-side SNR. If(49)SNRI(wT(k+1))−SNRI(wT(k))SNRI(wT(k))≤ξ or k≥Kmax, terminate the iteration; otherwise, set k←k+1 and repeat Steps 2–4. 1.Initialize the transmit weight vector as wT(0)=wT0 using (36), and initialize the receive weight vector as(48)wR(0)=gT(wT(0))∥gT(wT(0))∥2.Set the relaxation factor ρ, the maximum iteration number Kmax, and the convergence threshold ξ, and let k=0. Initialize the transmit weight vector as wT(0)=wT0 using (36), and initialize the receive weight vector as(48)wR(0)=gT(wT(0))∥gT(wT(0))∥2. Set the relaxation factor ρ, the maximum iteration number Kmax, and the convergence threshold ξ, and let k=0. 2.For the current transmit weight vector wT(k), compute the intermediate receive weight vector w^R(k+1) according to (46), and then update wR(k+1) according to (47). For the current transmit weight vector wT(k), compute the intermediate receive weight vector w^R(k+1) according to (46), and then update wR(k+1) according to (47). 3.Update the transmit weight vector wT(k+1) according to (38)–(43). Update the transmit weight vector wT(k+1) according to (38)–(43). 4.Evaluate the relative change in the intercept-side SNR. If(49)SNRI(wT(k+1))−SNRI(wT(k))SNRI(wT(k))≤ξ or k≥Kmax, terminate the iteration; otherwise, set k←k+1 and repeat Steps 2–4. Evaluate the relative change in the intercept-side SNR. If(49)SNRI(wT(k+1))−SNRI(wT(k))SNRI(wT(k))≤ξ or k≥Kmax, terminate the iteration; otherwise, set k←k+1 and repeat Steps 2–4. The above iterative procedure yields a high-quality suboptimal solution that reduces the intercept-side SNR while preserving active detection capability through target-oriented transmit initialization, similarity-constrained transmit updating, and relaxation-enhanced receive-side SNR maximization. 3.5. Computational Complexity and Hardware Feasibility Analysis Assuming that the iteration-independent quantities have been precomputed, the computational complexity of the proposed BCD algorithm is analyzed as follows. Here, K is the number of outer iterations, and Lls is the average number of backtracking line-search trials per transmit weight update. For the receive weight update, the main cost comes from constructing the equivalent target response vector gT(k), whose dimension is MN×1. Thus, the complexity of this step is O(MN). For the transmit weight update, the main cost comes from gradient evaluation and repeated objective-function evaluations during backtracking line search, which are dominated by operations over the M-dimensional transmit weight vector. Therefore, the complexity of this step is O(LlsM2). Consequently, the computational complexity per outer iteration is O(MN+LlsM2), and the total complexity after K iterations is OK(MN+LlsM2). During radar operation, the proposed optimization is performed over a dwell or update interval using coarse spatial information extracted from previous echoes. The updated wR and wT are then applied to receive processing and subsequent pulse transmission, respectively. Since the online computation mainly consists of matrix–vector multiplications, vector construction, normalization, projection, and backtracking line search, without semidefinite programming, eigenvalue decomposition, or large-scale matrix inversion, the algorithm is suitable for parallel implementation on SDR-, FPGA-, or RFSoC-based radar platforms, such as the NI Ettus USRP X410 and AMD ZCU111 RFSoC evaluation kit [16,17]. 4. Simulation Results In this section, the proposed method is evaluated through numerical simulations and measured-data validation. The simulations assess the detection and stealth balance, whereas the measured-data results verify the active detection capability using real radar echoes.In this section, the proposed method is evaluated through numerical simulations and measured-data validation. The simulations assess the detection and stealth balance, whereas the measured-data results verify the active detection capability using real radar echoes. The main simulation parameters are summarized in Table 1. 4.1. Joint Simulation Results of Active Detection and RF Stealth We first evaluate the detection and stealth balance achieved by the proposed method. Figure 2 shows the probability of detection versus the probability of interception under different values of the weighting factor α1. As α1 increases, the optimization places greater emphasis on RF stealth, resulting in a lower interception probability with a slight reduction in detection probability. These results indicate that α1 provides an effective mechanism for adjusting the balance between active detection and RF stealth according to different operational requirements. Figure 3 compares the detection and stealth balance of different radar schemes under the RF-stealth-emphasized setting, i.e., α1=0.8. For comparison, the methods in [15,18] are selected as two representative benchmarks, corresponding to FDA-MIMO low-intercept transmit beampattern optimization and transmit–receive joint LPI beamforming under a transmit-subaperturing MIMO architecture, respectively. Compared with the conventional FDA scheme and the two benchmark methods, the proposed method achieves a more favorable balance between active detection and RF stealth. Specifically, for a given detection probability, the proposed method yields a lower probability of interception, especially in the high-detection-probability region. These results demonstrate the effectiveness of the proposed transmit–receive optimization strategy in balancing active detection capability and RF stealth performance. To further verify the stability of the proposed optimization procedure, Figure 4 shows the convergence behavior of the proposed transmit–receive optimization algorithm under α1=0.8. It can be observed that the objective value decreases rapidly during the first several iterations and then gradually approaches a stable value. This result indicates that the proposed alternating optimization algorithm can achieve stable convergence within a limited number of iterations, thereby supporting the effectiveness and practical applicability of the proposed method. Figure 5 compares the RF stealth performance of different radar schemes in terms of the DOA estimation RMSE at the intercept receiver. As the intercept-side SNR increases, the RMSE of all schemes decreases due to the improved observation quality at the intercept receiver. However, the proposed method consistently yields the largest RMSE over the considered SNR range, indicating that it can more effectively degrade the DOA estimation capability of the intercept receiver. This result further verifies the RF stealth advantage of the proposed method from the perspective of adversarial DOA estimation. 4.2. Complexity and Run-Time Analysis To further evaluate the computational efficiency of the proposed algorithm, the theoretical complexity and the measured run time are analyzed in this subsection. Table 2 summarizes the per-iteration computational complexity of different methods. The SDP-based method generally suffers from high computational complexity due to the use of an interior-point solver. The method in [18] involves matrix inversion or equivalent joint beamforming updates, resulting in a cubic-order computational burden. In contrast, the proposed method mainly relies on matrix–vector multiplications and a simple line-search procedure, whose complexity is approximately O(LlsM2+MN). Therefore, the proposed method can substantially reduce the computational burden. To evaluate the practical run-time performance, all methods are implemented under the same computational environment. The simulations are carried out in MATLAB R2022a on a computer equipped with an Intel Core i7-9700K CPU and 24 GB RAM. No GPU acceleration or parallel computing is used. Figure 6 shows the CPU time of the traditional SDP method, the MIMO-LPI benchmark in [18], and the proposed method as the number of transmit elements increases. It can be observed that the proposed method exhibits a slower growth in runtime than the baseline methods. This result is consistent with the theoretical complexity analysis in Table 2, further confirming the computational efficiency of the proposed algorithm. Hence, the proposed method has better potential for real-time implementation in practical FDA radar systems. 4.3. Measured-Data Validation of Active Detection Performance To further validate the active detection performance, measured data from the publicly available LSS-DAUR-1.0 dataset [19] were used. Specifically, the target Doppler complex data of a fixed-wing UAV were selected, and the corresponding time–Doppler response was used for evaluation. Figure 7 shows the time–Doppler response obtained from the measured TD data of the fixed-wing UAV. A distinct energy ridge can be observed in the nonzero Doppler region, indicating that the target echo exhibits a stable Doppler signature during the observation interval. This result is consistent with the associated track information and confirms that the adopted measured data contain reliable target motion information. Therefore, the measured-data resultprovides supplementary validation of the active detection performance considered in this work. It should be noted that the measured-data validation in this subsection is intended to provide supplementary evidence for the active detection performance. The RF stealth performance is evaluated based on the proposed intercept-side model and the corresponding numerical simulations. 5. Conclusions In this paper, a joint transmit–receive weight optimization scheme was developed for FDA radar to balance active detection and RF stealth performance. Simulation results show that the proposed method achieves a lower probability of interception than the conventional FDA scheme and the benchmark methods while maintaining comparable detection performance. The intercept-receiver DOA estimation results and run-time analysis further demonstrate its RF stealth advantage and computational feasibility. Overall, the proposed scheme provides an effective design approach for improving RF stealth capability in FDA radar systems.