Sensing with Random Signals

Shihang Lu, Fan Liu, Fuwang Dong, Yifeng Xiong, Jie Xu, Ya-Feng Liu

TL;DR

This paper tackles the "deterministic-random tradeoff" in Integrated Sensing and Communications (ISAC) by introducing a new metric: Ergodic Linear Minimum Mean Square Error (ELMMSE). The authors prove that random communication signals inherently degrade sensing performance compared to deterministic radar signals and propose a practical, offline-trainable precoding algorithm (SGP) to minimize this degradation in short-frame scenarios.

What Problem Does This Paper Address?

In the push toward 6G, researchers are looking at Integrated Sensing and Communications (ISAC), where the same hardware and spectrum are used to both transmit data and "see" the environment like a radar.

The fundamental conflict is that radar works best with deterministic signals (carefully designed waveforms), while communication requires random signals (to carry unpredictable information). Most existing research assumes that if you look at a long enough sequence of data, the randomness "averages out" (the Law of Large Numbers).

However, in modern massive MIMO systems or high-mobility scenarios, we often use short frames. In these cases, the randomness of the data cannot be ignored—it creates a performance gap. This paper addresses how to design signal precoders (the "filters" that shape the transmitted signal) when the signal itself is a random variable, specifically focusing on minimizing the average estimation error.

Methods

The authors approach the problem through three main contributions:

  1. A New Metric (ELMMSE): They define the Ergodic LMMSE, which measures the average sensing error across all possible random signal realizations, rather than just a single snapshot.
  2. Theoretical Proof: Using Jensen’s Inequality, they mathematically prove that random signals will always result in a higher (worse) sensing error than deterministic signals.
  3. Two Precoding Strategies:
    • Data-Dependent Precoding: This scheme optimizes the precoder for every single realization of the data. While it offers the best performance, it is computationally "expensive" because you have to redesign the precoder in real-time as data changes. They solve this using a Successive Convex Approximation (SCA) algorithm.
  4. Data-Independent Precoding (SGP): This is the more practical approach. They design a single precoding matrix that works well on average for any random data. They propose a Stochastic Gradient Projection (SGP) algorithm that can be trained offline using signal samples, significantly reducing real-time overhead.

Datasets

The paper does not use external real-world datasets. Instead, it relies on Monte Carlo simulations with the following parameters:

  • MIMO Setup: 64 transmit antennas ($N_t$) and 32 receive antennas ($N_r$).
  • Channel Model: The Target Impulse Response (TIR) matrix eigenvalues follow a uniform distribution on the interval [1, 10] to represent spatial channel correlation.
  • Signal Samples: $N=100$ samples of random ISAC signals used for stochastic approximation.

Key Insights and Ideas

  • The Randomness Penalty: The "aha moment" is the formal proof that randomness is a bug, not a feature, for sensing. Even with perfect precoding, a random communication signal will never beat a deterministic radar signal of the same power.
  • Short Frames Matter: The gap between random and deterministic sensing is most pronounced when the frame length ($L$) is small. As $L$ increases, the performance of random signals asymptotically approaches that of deterministic ones.
  • Water-Filling is Suboptimal: The classical "water-filling" precoding method (standard in communications) fails in short-frame sensing because it ignores the rank-deficient nature of short random signal matrices, leading to an "error floor."
  • Offline Training for Online Efficiency: By using the SGP algorithm, engineers can "pre-calculate" a robust precoder offline that accounts for signal randomness, providing a "favorable performance-complexity tradeoff."

Key Results

  • Convergence: The proposed SGP algorithm converges quickly, typically within tens of iterations.
  • Performance Gain: In simulations with a frame length of $L=32$, the SGP-based data-independent precoder significantly outperformed the classical water-filling method, which hit a performance floor due to its inability to handle signal randomness.
  • The Gold Standard: Data-dependent precoding (SCA) achieved the lowest error, confirming it as the theoretical performance limit, though SGP remained a close and more practical runner-up.
  • Asymptotic Behavior: As frame length increased from $L=32$ to $L=2048$, the sensing error of random signals decreased, confirming that the "randomness penalty" vanishes as the Law of Large Numbers takes over.

Limitations and Future Work

  • Monostatic Assumption: The study assumes a monostatic setup where the transmitter knows exactly what signal was sent. In bistatic or multistatic systems (where the receiver is elsewhere), the receiver might not have perfect knowledge of the random communication data, adding another layer of complexity.
  • Computational Complexity of SCA: While the data-dependent SCA algorithm provides a benchmark, its high complexity makes it currently unsuitable for real-time 6G applications.
  • Channel Knowledge: The model assumes the statistical correlation of the channel ($R_H$) is known. Future work could explore scenarios where this information is imperfect or must be estimated simultaneously.

Summary updated

Abstract

Radar systems typically employ well-designed deterministic signals for target sensing. In contrast to that, integrated sensing and communications (ISAC) systems have to use random signals to convey useful information, potentially causing sensing performance degradation. In this paper, we define a new sensing performance metric, namely, ergodic linear minimum mean square error (ELMMSE), accounting for the randomness of ISAC signals. Then, we investigate a data-dependent precoding scheme to minimize the ELMMSE, which attains the optimized sensing performance at the price of high computational complexity. To reduce the complexity, we present an alternative data-independent precoding scheme and propose a stochastic gradient projection (SGP) algorithm for ELMMSE minimization, which can be trained offline by locally generated signal samples. Finally, we demonstrate the superiority of the proposed methods by simulations.
BibTeX
@inproceedings{lu2024sensing,
  author = {Shihang Lu and Fan Liu and Fuwang Dong and Yifeng Xiong and Jie Xu and Ya-Feng Liu},
  title = {{Sensing with Random Signals}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2024},
  doi = {10.1109/icassp48485.2024.10446784},
}

Loading related content...

Figure