Asymptotic Properties of One-Bit Distributed Detection with Ordered Transmissions

Consider a sensor network made of remote nodes connected to a common fusion center. In a recent work Blum and Sadler [1] propose the idea of ordered transmissions -sensors with more informative samples deliver their messages first- and prove that optimal detection performance can be achieved using only a subset of the total messages. Taking to one extreme this approach, we show that just a single delivering allows making the detection errors as small as desired, for a sufficiently large network size: a one-bit detection scheme can be asymptotically consistent. The transmission ordering is based on the modulus of some local statistic (MO system). We derive analytical results proving the asymptotic consistency and, for the particular case that the local statistic is the log-likelihood (\ell-MO system), we also obtain a bound on the error convergence rate. All the theorems are proved under the general setup of random number of sensors. Computer experiments corroborate the analysis and address typical examples of applications including: non-homogeneous Poisson-deployed networks, detection by per-sensor censoring, monitoring of energy-constrained phenomenon.

💡 Research Summary

The paper investigates an extreme form of ordered transmission for binary hypothesis testing in wireless sensor networks (WSNs). Building on the concept introduced by Blum and Sadler—where sensors with more informative measurements access the communication channel earlier—the authors push the idea to its limit: only the first sensor that manages to transmit its local 1‑bit decision is allowed to decide the global hypothesis, and all subsequent transmissions are instantly suppressed. The transmission policy, termed Modulus Ordered (MO), schedules each sensor’s transmission time inversely proportional to the absolute value of a transformed local statistic T(Xi). Two concrete choices for T are considered: the identity (|Xi|) and the log‑likelihood ratio L(Xi). When T = L, the system is called ℓ‑MO.

The core theoretical contribution is a proof of asymptotic consistency for this one‑bit scheme under very general conditions, including random numbers of sensors and non‑homogeneous Poisson deployments. By employing extreme‑value theory (EVT), the authors show that the maximum absolute transformed observation Mₙ = max_i |T(Xi)|, after appropriate normalization, converges to a Gumbel or Fréchet distribution depending on tail behavior. Lemma 1 establishes the classic attraction to these limit laws for deterministic n, while Lemma 2 extends the result to a random sensor count N that scales with a parameter ν (N/ν → R in probability). If the right tail of the distribution under H₁ dominates the left tail, the decision is almost surely driven by the largest positive transformed sample; otherwise the most negative sample may dominate.

For the ℓ‑MO case, the log‑likelihood ratio is the optimal test statistic, and the authors derive explicit bounds on the error probabilities. They prove that both false‑alarm and miss probabilities decay exponentially with the number of sensors, i.e., P_e(n) ≤ C·exp(−κ n) for some constants C, κ > 0. This exponential convergence is substantially faster than the sub‑linear performance of schemes that require a fixed fraction of sensor reports.

The paper also presents extensive simulations to validate the theory. Three representative scenarios are examined: (i) sensors placed according to a non‑homogeneous Poisson process, (ii) per‑sensor censoring where each node transmits only if its statistic exceeds a local threshold, and (iii) energy‑constrained monitoring where only the absolute value ordering is used (identity transformation). In all cases, the empirical error curves match the analytical predictions, and the system remains robust to modest timing jitter and synchronization errors.

In conclusion, the authors demonstrate that, provided the network is sufficiently large, a single most informative sensor can determine the hypothesis with arbitrarily small error. This “winner‑takes‑all” approach yields dramatic reductions in communication overhead and energy consumption while preserving detection performance. The framework accommodates random sensor populations, heterogeneous deployments, and various forms of local processing, making it broadly applicable to practical WSNs. Future work is suggested on asynchronous transmission models, multi‑hypothesis extensions, and non‑parametric statistics where the log‑likelihood may be unavailable.