Bistatic Integrated Sensing and Communications

• Bistatic ISAC is a dual-functional architecture where a spatially separated transmitter and sensing receiver jointly support communications and radar sensing.
• The framework leverages shared OFDM waveforms and coordinated resource allocation to overcome self-interference and meet stringent synchronization requirements.
• Key challenges include precise delay–Doppler estimation, waveform optimization, and managing trade-offs between high data rates and robust sensing performance.

Bistatic integrated sensing and communications (ISAC) denotes an ISAC configuration in which the transmitter that illuminates the scene and the sensing receiver that estimates target state are physically separated, in contrast to monostatic sensing, where both functions are co-located. In practical settings, bistatic sensing may be required either due to inherent system constraints or as a means to mitigate the strong self-interference encountered in monostatic configurations. In B5G/6G settings, bistatic ISAC is typically realized with shared waveforms and infrastructure so that communication and sensing are evaluated on the same propagation, synchronization, and resource-allocation stack rather than as two independent subsystems (Jiao et al., 2023, Brunner et al., 2024, Luo et al., 2024).

1. Architectural forms and operating assumptions

A canonical bistatic ISAC model comprises an ISAC transmitter that sends a message to a communication receiver and simultaneously probes a time-varying state sequence by broadcasting, while a sensing receiver at a different location observes its own signal and forms an estimate of the state. In the discrete information-theoretic formulation, the communication receiver observes $Y^n$ and knows the state sequence $S^n$ perfectly, whereas the sensing receiver observes $Z^n$ only; rate-splitting into a common part $W_0$ and a private part $W_1$ yields a broadcast-channel view with degraded message sets (Jiao et al., 2023).

In practical OFDM realizations, the same frame usually supports communication and sensing. A representative cellular architecture uses two cooperating 5G/6G gNodeBs at known, static locations: gNB #1 acts as an OFDM transmitter for both communication and sensing, and gNB #2 as a bistatic radar receiver. Path $p=0$ is a line-of-sight or strong reference link used for over-the-air synchronization and to define zero bistatic range and Doppler bias, while the remaining paths correspond to radar targets with unknown bistatic range and Doppler (Giroto et al., 22 Jan 2026). Closely related proof-of-concept systems process an incoming OFDM-based ISAC signal through over-the-air synchronization based on preamble symbols and pilots, and then perform bistatic radar processing using either only pilot subcarriers or the full OFDM frame; the full-frame approach requires estimation of the originally transmitted frame based on communication processing and therefore error-free communication, which can be achieved via appropriate channel coding (Brunner et al., 2024).

The architecture is not restricted to large-array or high-cost installations. Low-complexity deployments with a single-antenna transmitter and a single-antenna receiver are also considered, but such SISO bistatic operation must cope with clock asynchrony and Doppler-mirroring ambiguity in CSI, which cannot be mitigated using conventional multi-antenna methods (Wang et al., 18 Aug 2025). At the opposite end of the design space, satellite-borne transmitters, separate radar receivers, hybrid analog/digital arrays, and distributed multi-receiver sensing nodes all appear in current formulations of bistatic ISAC (Park et al., 2024, Mao et al., 17 Feb 2025, Liu et al., 2024).

2. Propagation and channel modeling

A major distinction between monostatic and bistatic ISAC lies in the channel model. An extension of 3GPP TR 38.901 for bistatic sensing preserves the communication channel structure while adding sensing-specific features. The same channel impulse response must carry all “strong” paths needed for comms and additional “weak” paths needed for radar-style sensing of distant or low-RCS targets. The NLoS part is written as

$$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$

where $\mathcal T$ indexes sensing-target clusters and $\mathcal E$ pure-environment clusters (Luo et al., 2024).

The extension modifies cluster retention rather than the large-scale laws. In baseline 3GPP modeling, clusters more than $25$ dB below the strongest are removed; the bistatic ISAC extension reduces that threshold to, for example, $S^n$0 dB or more, scenario-dependent, so that weak echoes from small targets or clutter remain. By generating clusters exactly as in TR 38.901 except for $S^n$1 and lowering the removal threshold, the identical large-scale, angular-spread, polarization and Doppler laws are retained. Environment clusters need no further change, while sensing clusters may be drawn from the weak end of the same pool and then either statistically expanded or deterministically inserted (Luo et al., 2024). This suggests that backward compatibility with communication-oriented simulators can be preserved while enabling sensing evaluation from the full $S^n$2.

The framework supports both statistical and deterministic target models. In the statistical model, one chooses point- or extended-target, selects up to two bounces per ray, splits cluster power $S^n$3 into per-ray powers $S^n$4, and generates $S^n$5 using either an angle-priority construction based on an equivalent reflection point on the Tx–Rx ellipse or a position-priority construction that samples the reflection point first. In the deterministic model, entire clusters or subsets of clusters are replaced by real or ray-traced ray sets $S^n$6, which already carry time/angle coherence from measurement or simulation (Luo et al., 2024).

The small-scale representation distinguishes environment and target rays. The environment-cluster term keeps the 3GPP form,

$S^n$7

while the target-cluster term adds an extra Doppler-phase factor

$S^n$8

to enforce time coherence under target motion (Luo et al., 2024). The complete CIR is then

$S^n$9

Validation in ray tracing and measurements emphasizes the sensing role of weak paths. In an indoor office at $Z^n$0 GHz with $Z^n$1 MHz bandwidth, target echoes were observed at $Z^n$2 to $Z^n$3 dBm, i.e. at least $Z^n$4 dB below LoS, and the study showed that without lowering the cluster-removal threshold one would lose these sensing-critical paths. In the communication comparison, a standard 3GPP channel with $Z^n$5 and $Z^n$6 dB removal and an ISAC channel with $Z^n$7, $Z^n$8 dB removal, and $Z^n$9 random target clusters produced BER curves that coincide within $W_0$0 dB. For sensing, with three point targets at $W_0$1, $W_0$2, and $W_0$3 dB below LoS, a CFAR detector with $W_0$4 and $W_0$5 dB coherent gain reliably detected all targets at channel-SNR $W_0$6 dB, and the strongest target alone at $W_0$7 dB, with range RMSE $W_0$8 m once detected (Luo et al., 2024).

3. Synchronization, delay–Doppler estimation, and sensing beyond classical limits

Because transmitter and sensing receiver are separate, synchronization is a first-order problem. An OFDM-based over-the-air synchronization framework uses a preamble with Schmidl–Cox repetition for coarse timing and CFO, Tsai symbol pairs for SFO estimation, pilot-assisted residual SFO estimation from the pilot-delay profile, and a final residual FO correction derived from the delay–Doppler profile. Residual TO introduces phase rotation proportional to subcarrier index and hence range bias; residual FO introduces phase rotation proportional to symbol index and hence Doppler bias; residual SFO induces range/Doppler migration, amplitude roll-off, and ICI (Brunner et al., 2024). In a $W_0$9 GHz proof-of-concept bistatic setup with unsynchronized clocks, the reported post-synchronization constellation had $W_1$0, and the measured full-frame radar image showed three peaks corresponding to LoS, a static target, and a moving target (Brunner et al., 2024).

Several receiver designs explicitly address the bistatic CP limitation. In a single-target OFDM-based bistatic ISAC system, a sliding-window sensing receiver enumerates delay hypotheses $W_1$1, removes CP, performs an $W_1$2-point DFT on the aligned window, computes LS estimates on pilot positions, and evaluates a 2D periodogram

$W_1$3

The decision metric $W_1$4 identifies the coarse delay index; 1D quadratic interpolation then refines the bistatic range and velocity estimates. With $W_1$5 GHz, $W_1$6 kHz, $W_1$7, $W_1$8, $W_1$9, and $p=0$0 m, the construction extends the ISI-free sensing range from one CP-block to $p=0$1, concretely from $p=0$2 m to $p=0$3 m, and numerical results show that RMSE$p=0$4 and RMSE$p=0$5 closely track the ensemble-averaged CRB at high SNR (Ozturk et al., 17 May 2025).

Clock asynchrony and receiver motion can be treated jointly rather than separately. A narrowband model with asynchronous transmitter and moving receiver writes the discrete CIR as

$p=0$6

with common phase nuisance $p=0$7, target Doppler $p=0$8, and receiver-motion Doppler $p=0$9. Subtracting the LoS phase cancels $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$0; time differencing cancels static path phases; known AoAs give $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$1, producing a nonlinear least-squares problem in $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$2. At least $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$3 static paths are required. In simulation, the median normalized Doppler-estimation error was below $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$4 for SNR $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$5 dB with the minimum $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$6 static scatterers, improving to $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$7 when $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$8; longer windows up to $$H_{u,s}^{\rm NLOS}(\tau,t) = \sum_{n\in\mathcal T}\sum_{m=1}^{M_n} H^{\rm tar}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}) + \sum_{n\in\mathcal E}\sum_{m=1}^{M} H^{\rm env}_{u,s,n,m}(t)\,\delta(\tau-\tau_{n,m}),$$9 ms further reduced variance (Ventura et al., 2024).

At the SISO end of the spectrum, self-referencing cross-correlation (SRCC) removes symbol-indexed random phase in CSI by correlating CSI with a delay-windowed reconstruction that shares the same unknown phase distortion, and delay-domain beamforming with MVDR suppresses Doppler mirroring. The resulting delay–Doppler–time tensor enables lightweight inference, and on a Raspberry Pi 4B the reported feature-extraction latency is $\mathcal T$0 ms with standard deviation $\mathcal T$1 ms; a MobileViT-XXS with $\mathcal T$2M parameters is then used for downstream sensing (Wang et al., 18 Aug 2025). A common misconception is that bistatic sensing fundamentally requires multi-antenna synchronization structure; the SISO formulation shows that clock-asynchronous single-antenna sensing is possible, but only after explicit phase-nuisance and ambiguity suppression (Wang et al., 18 Aug 2025).

4. Waveform, resource, beam, and interference design

A defining design tension in bistatic ISAC is that the communication link aims to transmit higher modulation order symbols to maximize throughput, whereas lower modulation order is preferable for sensing to achieve a higher signal-to-noise ratio in the radar image. One response is a hybrid resource-allocation scheme for OFDM data channels: choose a sensing grid

$\mathcal T$3

transmit QPSK on $\mathcal T$4, and use 16-QAM elsewhere. In the reported setup with $\mathcal T$5 GHz, $\mathcal T$6 kHz, $\mathcal T$7, $\mathcal T$8, $\mathcal T$9, $\mathcal E$0, $\mathcal E$1, $\mathcal E$2, and code-rate $\mathcal E$3, the hybrid scheme closes the gap to a genie-aided bound by up to $\mathcal E$4 dB in target SNR around an SNR of $\mathcal E$5 dB pre-radar, improves $\mathcal E$6 by an order of magnitude in the $\mathcal E$7–$\mathcal E$8 dB region, and incurs a minor $\mathcal E$9 loss in spectral efficiency, from $25$0 to $25$1 bits/s/Hz (Henninger et al., 16 Jan 2026).

A second line of work formulates OFDM waveform optimization directly over sensing and communication subcarriers. With sensing-assignment vector $25$2 and per-subcarrier powers $25$3, the communication data rate is

$25$4

while the delay CRB depends on a squared effective bandwidth term formed by the indices of sensing subcarriers. The resulting optimization shows that the achievable communication data rate is determined by the number of communication subcarriers, whereas the delay sensing accuracy is governed by the index distribution of sensing subcarriers. After quadratic transformation and dual decomposition, a subcarrier is allocated for sensing if and only if its Fisher information gain exceeds the corresponding communication rate loss, and the power allocation for communication subcarriers exhibits a bounded water-filling structure (Du et al., 9 Mar 2026).

Array-constrained implementations replace fully digital precoding by hybrid beamforming. In a $25$5-D bistatic mmWave configuration with two half-duplex DFRC base stations, OFDM signaling, and a closed-form position error bound derived from AOA/ToA estimation, one can optimize analog and digital beamformers to maximize achievable spectral efficiency while ensuring a predefined PEB threshold. Two algorithms are reported: a Riemannian trust-region approach, which achieves superior performance in terms of global optima and convergence speed compared to conventional gradient-based methods, and an orthogonal matching pursuit alternative, which offers lower complexity with reasonable spectral efficiency while maintaining the PEB constraint (Mao et al., 17 Feb 2025). The underlying trade-off is explicit: relaxing the PEB threshold improves spectral efficiency, but even at large $25$6 the integrated-sensing constraint prevents reaching the fully digital communication-only benchmark (Mao et al., 17 Feb 2025).

Interference management is further complicated when the information messages are unknown to the sensor and the channel between the transmitters and the sensor is unknown to the transmitters. For bistatic ISAC models with heterogeneous coherence times or heterogeneous connectivity, blind interference alignment and topological interference management create non-trivial communication–sensing degrees-of-freedom tradeoff points that outperform time-sharing. In the $25$7-user SISO full-connectivity model, blind interference alignment plus zero-forcing achieves

$25$8

which lies above the time-sharing line $25$9; in simulation for $S^n$00, the sensor’s channel-estimation MSE gains $S^n$01 dB over treating interference as noise across SNR in $S^n$02 dB, while BER at communication receivers remains essentially unchanged (Liu et al., 2024). This directly contradicts the common assumption that unknown communication symbols can only be suppressed by sacrificing communication degrees of freedom.

5. Detection theory and capacity–distortion structure

Beyond architecture and algorithms, bistatic ISAC has a distinct information-theoretic formulation. For a state-dependent discrete memoryless two-receiver broadcast channel, the fundamental multi-letter capacity–distortion tradeoff is

$S^n$03

Single-letter achievability introduces an auxiliary $S^n$04 and a partial-decoding-based estimation rule $S^n$05, producing the region

$S^n$06

When $S^n$07 is physically degraded, this characterization is exact (Jiao et al., 2023). The numerical examples in that formulation show that partial decoding lies above blind estimation and full decoding, and is very close to the genie-aided outer bound in a non-degraded binary-state example (Jiao et al., 2023).

Because the distortion constraints are non-convex in $S^n$08, specialized numerical machinery has also been developed. Extended Arimoto–Blahut algorithms introduce auxiliary variables to transform non-convex squared-error and log-loss distortion constraints into linear constraints, prove equivalence to the original problem, and then alternate between density updates, closed-form estimator updates, and a one-dimensional root search for the Lagrange multiplier. The resulting algorithm provides a tractable way to calculate the rate-distortion trade-off in bistatic ISAC systems (Jiao et al., 11 Aug 2025).

At the sensing receiver, a parallel development addresses detection with mixed deterministic and stochastic signaling. In a bistatic downlink with a multi-antenna BS, a separate sensing receiver, deterministic sensing waveform $S^n$09, and Gaussian information stream $S^n$10, the Neyman–Pearson detector uses both the covariance of the Gaussian component and the known deterministic illumination. The resulting test statistic depends on

$S^n$11

and the analysis states that both signal components contribute to improving the overall detection performance. Beamforming then maximizes the detection probability subject to a minimum communication SINR and total power budget, using SDR and SCA. A higher communication-rate threshold directs more transmit power to Gaussian information-bearing signals, thereby diminishing deterministic-signal power and weakening detection performance (Song et al., 14 Nov 2025). The same sensing–communication trade-off appears in a low-altitude-economy UAV surveillance formulation, where the objective is to maximize the minimum detection probability over the surveillance region under a minimum authorized-UAV SINR constraint (Song et al., 21 Apr 2026). A recurring misconception is that Gaussian information signals are inevitably interference at the sensing receiver; these detector constructions show that, when modeled statistically, they can increase the noncentrality of the NP test rather than simply degrade it (Song et al., 14 Nov 2025, Song et al., 21 Apr 2026).

6. Platforms, angular sampling, propagation modifiers, and deployment issues

Bistatic ISAC is now being instantiated across several platform classes. In LEO satellite systems, a bistatic configuration separates the radar receiver from the satellite transmitter to mitigate the severe monostatic echo path loss associated with satellite altitude. A rate-splitting multiple access formulation optimizes dual-functional precoders to maximize the minimum user rate subject to radar CRB constraints, using SDR, SROCR, and SCA. Numerical results show that RSMA-ISAC outperforms SDMA-ISAC by $S^n$12 in minimum rate, and the common stream plays three vital roles: beamforming towards the radar target, interference management between communications and radar, and interference management among communication users (Park et al., 2024).

Electromagnetically engineered propagation is also entering the bistatic ISAC design loop. Stacked intelligent metasurfaces at both transmitter and receiver can be tuned through a min–max steepest-ascent procedure that maximizes the weakest path gain across doubly dispersive channels, while radar parameter estimation is performed by a compressed-sensing-based probabilistic data association algorithm. In the reported $S^n$13 GHz setup with OFDM, OTFS, and AFDM waveforms, sensing-optimized SIM design yields a $S^n$14 dB gain in $S^n$15 and $S^n$16 over no-SIM, and still gives an approximately $S^n$17 dB BER gain (Ranasinghe et al., 29 Apr 2025). A sharply different result appears for a disco RIS with random and time-varying reflection coefficients: the DRIS induces active channel aging, significantly degrades communication SINR, increases $S^n$18 for AoD, and decreases $S^n$19 for AoA, so the same surface simultaneously disrupts communications, weakens AoD estimation, and enhances AoA estimation (Huang et al., 11 Apr 2026).

Angular-domain acquisition is itself a bistatic-specific problem. For azimuth-only scanning, separating azimuth operations of the two transmit and receive arrays is optimal in array-specific normalized angular frequency, enabling loss-less reconstruction of the angular domain by DFT-based interpolation rather than spline interpolation (Felix et al., 27 Apr 2025). For full azimuth–elevation bistatic sampling, the problem becomes four-dimensional, and the TX–RX elevation angles are coupled through the ortho-baseline coarray. The resulting BASIIS framework derives a minimal sampling and interpolation scheme that is near-lossless and realizable with any beamforming architecture. Monte Carlo simulations report that the proposed minimal acquisition essentially equalizes the detection accuracy of dense oversampled imaging while acquiring $S^n$20 to $S^n$21 times fewer TX–RX direction pairs; in the detailed comparison, BASIIS uses $S^n$22 fewer beams (Felix et al., 16 Jun 2026).

Vehicular and cellular deployments expose the remaining practical constraints. An automotive bistatic ISAC system based on orthogonal chirp division multiplexing uses the SUNDAE receiver, which first decodes communications and then reuses the decoded symbols for radar parameter estimation. In the reported $S^n$23 GHz, $S^n$24 MHz setting, BER loss from LS estimation plus linear interpolation is at most about $S^n$25 dB relative to perfect CSI, OCDM and OTFS outperform OFDM in high mobility, and the radar RMSE of range and velocity approaches the CRLB at high radar SNR (Bhattacharjee et al., 2021). For cellular OFDM, practical challenges include mutual coupling, beam-squint, PA nonlinearities, I/Q imbalance, phase noise, quantization, and sampling jitter; one $S^n$26G FR2-compliant evaluation with $S^n$27 GHz, $S^n$28 MHz, $S^n$29, $S^n$30, and processing gain $S^n$31 dB reports $S^n$32 m, maximum unambiguous range $S^n$33 km, ISI-free range about $S^n$34 m, and maximum unambiguous Doppler approximately $S^n$35 kHz (Giroto et al., 22 Jan 2026). Open challenges accordingly remain network-wide synchronization in multi-cell bistatic and multistatic ISAC, joint resource allocation and beam management, compensation of residual hardware impairments, high-resolution $S^n$36D angle estimation under hybrid or analog beamforming, inter-GNB interference cancellation, geometry calibration, distributed sensor fusion, and adaptive waveform design (Giroto et al., 22 Jan 2026).

Bistatic ISAC therefore emerges not as a minor variant of monostatic sensing, but as a family of architectures in which separated illumination and observation alter the channel model, synchronization requirements, angular sampling problem, interference structure, and capacity–distortion trade-off. Current results show that these difficulties are not merely implementation penalties: they also create distinctive opportunities, including weak-path-aware 3GPP-compatible channel generation, sensing beyond the CP limit, partial-decoding gains, RSMA-enabled dual-function precoding, and sampling-optimal angular acquisition (Luo et al., 2024, Ozturk et al., 17 May 2025, Jiao et al., 2023, Park et al., 2024, Felix et al., 16 Jun 2026).