TL;DR
This paper tackles the critical issue of self-interference (SI) in full-duplex Integrated Sensing and Communication (ISAC) systems. By leveraging Symbol-Level Precoding (SLP), the authors develop a method to cancel residual interference at the transmitter, significantly boosting target detection probability while maintaining high-quality service for communication users.
What Problem Does This Paper Address?
Integrated Sensing and Communication (ISAC) is a cornerstone of 6G, allowing a single base station (BS) to perform radar tasks and data transmission simultaneously. To do this efficiently, the BS often operates in full-duplex (FD) mode—it listens for weak radar echoes from a target while simultaneously blasting high-power signals to communication users.
The "Achilles' heel" of this setup is Self-Interference (SI). Even with standard analog and digital cancellation techniques, the leakage from the transmitter can be orders of magnitude stronger than the target echo, "blinding" the radar. While previous works used traditional transmit beamforming to mitigate this, they often relied on simple Signal-to-Interference-plus-Noise Ratio (SINR) metrics. This paper argues that we need a more direct approach: optimizing for the actual probability of detection and using the finer control offered by symbol-level processing.
Methods
The authors propose a novel SLP-based Self-Interference Cancellation (SIC) framework. Unlike traditional precoding that changes every frame, SLP adapts the signal for every individual symbol, exploiting "constructive interference" to meet communication requirements more flexibly.
Key technical steps include:
- Metric Derivation: They derive the target detection probability ($P_D$) in the presence of SI using a Generalized Likelihood Ratio Test (GLRT).
- Problem Formulation: They set up an optimization problem to maximize $P_D$ subject to a total power budget and specific Quality of Service (QoS) constraints for communication users (ensuring symbols land in the "correct" region of the constellation map).
- Algorithm Design: The resulting problem is a "non-convex fractional program," which is notoriously difficult to solve. The authors developed a penalty-based Block Coordinate Descent (BCD) algorithm. By using a quadratic transform, they broke the complex problem into three sub-problems (for variables $y$, $X$, and $\Lambda$) that each have efficient, closed-form solutions.
Datasets
As a theoretical signal processing paper, no external datasets were used. Instead, the authors evaluated their approach using Monte Carlo simulations of a MIMO-OFDM system:
- Setup: A BS with 32 antennas serving 8 single-antenna users.
- Channel Model: Rayleigh fading with large-scale path loss models.
- Benchmarks: They compared their proposed "SIC" method against two baselines: "Without SI" (the ideal, unattainable upper bound) and "With SI but no SIC" (the current state-of-the-art limitation).
Key Insights and Ideas
- Beyond SINR: Most radar-communication papers optimize for SINR. By deriving and optimizing the detection probability directly, the authors provide a more robust guarantee for radar performance.
- Exploiting Symbol Structure: SLP provides more degrees of freedom than block-level precoding. By treating interference as a resource (constructive interference) for communication, the system frees up more "spatial room" to cancel the interference leaking into the sensing receiver.
- Computational Efficiency: The "aha moment" in the optimization section is the use of the quadratic transform and penalty method. This allows the algorithm to avoid heavy iterative solvers for each sub-step, instead using direct formulas (closed-form updates), making it much more practical for real-time systems.
Key Results
- Detection Performance: In Receiver Operating Characteristic (ROC) curves, the proposed SLP-based SIC method performed nearly as well as the "ideal" scenario where no self-interference exists.
- Sensitivity to Distance: As the target moves further away (from 80m to 120m), the detection probability naturally drops. However, the proposed method maintained a much higher accuracy than the baseline, which failed almost completely beyond 100m due to SI.
- DoA Accuracy: The Root Mean Square Error (RMSE) of the Direction of Arrival (DoA) estimation was significantly lower with the proposed method, matching the performance of an interference-free system.
Limitations and Future Work
The primary limitation acknowledged is the assumption that the residual SI channel is identically and independently distributed (i.i.d.) Gaussian. In real-world hardware, residual interference often has more complex, non-Gaussian structures or hardware impairments.
The authors suggest that robust optimization—designing the precoder to handle uncertainty in the SI channel state information—is the next logical step for this research. Additionally, extending this from QPSK to higher-order modulation schemes (like 64-QAM) would be necessary for high-speed 6G applications.
Summary updated