Abstract This paper investigates the problem of signal amplitude estimation and detection when the temporal order of binary quantized measurements is lost or intentionally scrambled. Such scenarios, termed "unlabeled sensing," occur in distributed systems due to network delays or GPS spoofing attacks. We propose a Maximum Likelihood (ML) framework and an Alternating Maximization (AM) algorithm to jointly estimate the unknown signal amplitude and the hidden time-permutation. By reducing the permutation search to a sorting problem, we achieve computational efficiency. Theoretical analysis establishes scaling laws for the number of sensors required for perfect permutation recovery. Simulation results demonstrate that the proposed estimator's performance converges to the Cramér-Rao Lower Bound of the labeled case as the number of quantizers increases.
Introduction
In modern distributed sensing environments like smart grids and the Internet of Things (IoT), data fusion centers rely on accurate timestamps to reconstruct signals from multiple sensors. However, these timestamps are often unreliable due to network congestion, out-of-order delivery, or malicious GPS spoofing. This paper addresses the challenge of estimating a signal's scale parameter (amplitude) from binary quantized samples when their relative time order is unknown. Unlike previous work focusing on location parameters, this study tackles the more complex scale estimation problem ($x_i = h_i\theta + w_i$) under heavy data compression and channel noise.
System Model
The system considers a noiseless signal with known shape $h_i$ and unknown amplitude $\theta$. At each time step $i$, the signal is corrupted by noise and processed by $N$ binary quantizers. These quantized bits are transmitted over a noisy binary channel, potentially causing bit flips. Crucially, the resulting blocks of samples are permuted before reaching the fusion center, stripping them of their temporal labels.
Methodology
To solve the joint estimation problem of amplitude $\theta$ and permutation matrix $\Pi$, we utilize Maximum Likelihood estimation. Given the $O(K!)$ complexity of searching all permutations for a sequence of length $K$, we propose an Alternating Maximization (AM) algorithm. The algorithm iteratively optimizes one parameter while fixing the other. A key finding is that for a fixed amplitude, the optimal permutation can be found by simply sorting the observed frequencies of '1's, reducing the complexity to $O(K \log K)$.
To ensure convergence to the global optimum, we introduce a robust initialization strategy based on the Law of Large Numbers. Furthermore, a Generalized Likelihood Ratio Test (GLRT) is developed for signal detection in the unlabeled regime.
Theoretical Insights and Scaling Laws
The paper provides a rigorous analysis of identifiability, showing that certain symmetric signal shapes can lead to ambiguous solutions where different amplitudes and permutations produce identical probability distributions.
Additionally, we derive scaling laws for perfect permutation recovery. We prove that the number of quantizers $N$ needed to correctly recover the time-order of $K$ samples scales as $O(K^{2\alpha} \log K)$, where $\alpha$ depends on the signal shape and noise characteristics. This analytical bound provides a critical threshold for system design in unlabeled sensing environments.
Results and Discussion
Monte Carlo simulations on various signal types (ramp, sinusoidal, Gaussian) validate the theoretical findings. As the number of quantizers $N$ grows, the Mean Squared Error (MSE) of the unlabeled estimator approaches the performance of a perfectly labeled system, eventually reaching the Cramér-Rao Lower Bound (CRLB). Detection performance also converges to the labeled GLRT baseline. The empirical results for permutation recovery align closely with our derived analytical bounds, confirming the high sensor density required to overcome the loss of temporal information in practice.
Conclusion
We have presented an efficient ML-based approach for signal amplitude estimation and detection from unlabeled, binary quantized data. By transforming the permutation recovery into a sorting problem and providing analytical bounds for recovery, this work offers a theoretical and practical foundation for resilient sensing in environments prone to timing errors or spoofing.
Summary updated