Abstract
Radar systems typically employ well-designed deterministic signals for target sensing. In contrast, integrated sensing and communications (ISAC) systems must use random signals to convey information, potentially degrading sensing performance. This paper defines a new metric, ergodic linear minimum mean square error (ELMMSE), to account for ISAC signal randomness. We investigate a data-dependent precoding scheme for optimal performance and a lower-complexity data-independent scheme using a stochastic gradient projection (SGP) algorithm. Simulations demonstrate the superiority of the proposed methods.
Introduction: The Challenge of Randomness in ISAC
In the evolution toward 6G, Integrated Sensing and Communications (ISAC) aims to unify hardware and spectrum for both data transmission and environmental sensing. A fundamental challenge is the "deterministic-random tradeoff": radar performs best with carefully designed deterministic waveforms, whereas communication relies on random signals to carry information.
While the Law of Large Numbers suggests that randomness might average out over long sequences, modern massive MIMO and high-mobility scenarios often utilize short frames. In these regimes, the randomness of the data cannot be ignored, leading to a significant performance gap in sensing. This paper addresses this gap by proposing a new performance metric and optimized precoding strategies.
Methodology: ELMMSE and Precoding Design
The authors introduce Ergodic Linear Minimum Mean Square Error (ELMMSE) as a metric to measure average sensing error across all possible random signal realizations. Using Jensen’s Inequality, they provide a theoretical proof that random signals inherently result in higher sensing errors than deterministic signals of equal power.
To mitigate this, two precoding strategies are proposed:
- Data-Dependent Precoding: Optimizes the precoder for every data realization using a Successive Convex Approximation (SCA) algorithm. This serves as a theoretical performance upper bound but is computationally intensive.
- Data-Independent Precoding (SGP): A more practical approach that designs a single robust precoder for any random data. The proposed Stochastic Gradient Projection (SGP) algorithm can be trained offline using signal samples, significantly reducing real-time overhead.
Experimental Results and Discussion
The proposed methods were evaluated using Monte Carlo simulations in a MIMO setup ($N_t=64, N_r=32$).
The Impact of Frame Length
The "randomness penalty" is most severe in short-frame scenarios. As shown in Figure 1, as the frame length ($L$) increases, the normalized error of random signals asymptotically approaches the performance of deterministic signals, eventually reaching Jensen's Lower Bound.
Precoding Performance Comparison
The simulations reveal that classical "water-filling" precoding, while standard for communication, is suboptimal for short-frame sensing. It fails to account for the rank-deficiency of short random signal matrices, resulting in a performance "error floor" at high SNR.
As seen in Figure 6, the SGP-based data-independent precoder significantly outperforms water-filling at high SNR, particularly when $L=32$. The data-dependent SCA scheme provides the best overall performance, confirming its role as the benchmark, while SGP offers a favorable tradeoff between complexity and accuracy.
Conclusion
This work formalizes the sensing performance degradation caused by signal randomness in ISAC systems. By introducing the ELMMSE metric and the SGP algorithm, the authors provide a framework for designing efficient precoders that are robust to data randomness, particularly in the critical short-frame regime of future 6G networks. Future research may extend these findings to bistatic systems or scenarios with imperfect channel state information.
Summary updated