System Identification with Quantized Observations

System identification with quantized observations addresses the problem of inferring dynamic models when measurements are available only in coarse, discrete form. Such quantization may arise from limited sensor resolution, data compression or event-triggered transmission in networked settings. The loss of amplitude detail and introduction of nonlinear measurement mapping pose challenges in estimator design, convergence analysis and uncertainty characterisation. Contemporary approaches embed quantization effects into probabilistic frameworks, employing recursive algorithms, convex optimisation or Bayesian inference to recover model parameters and latent states. Practical applications span remote monitoring of industrial processes, communications channel estimation and control of resource-constrained cyber-physical systems. By exploiting prior knowledge of noise statistics, threshold structure and system linearity or nonlinearity, researchers have developed methods that balance identification accuracy against communication or computational cost, ensuring robust performance even under packet loss or harsh noise conditions.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent studies have advanced identification of discrete-time systems under constrained communications. A novel congruential summation-triggered scheme reduces transmission rates by sending updates only when cumulative deviations exceed thresholds. This approach incorporates stochastic packet loss into parameter estimation for finite impulse response systems and demonstrates strong convergence under unknown drop probabilities, balancing communication budget and identification accuracy.

Another stream employs probabilistic filtering to reconstruct states from quantized outputs. By approximating the quantized output likelihood with Gaussian mixtures via Gauss-Legendre quadrature, two-filter algorithms achieve accurate smoothing and filtering with reduced computational load, proving effective for low-resolution sensor networks.

Work on channel identification explores higher-order cumulants, binary measurements and kernel-based estimators. Comparative analysis reveals cumulant methods are robust under blind scenarios, while kernel and binary-measurement approaches excel in moderate noise settings. These insights guide algorithm selection for diverse signal-to-noise and data-availability regimes.

System Identification with Quantized Observations publication trend

The graph below shows the total number of articles in system identification with quantized observations across all publications each year (not limited to Nature Index journals).

Technical terms

Quantization: Conversion of continuous measurements into a discrete set of levels, introducing information loss.

Finite impulse response (FIR) system: A linear filter whose response to an impulse input is of finite duration.

Event-triggered mechanism: A sampling strategy that transmits data only when predefined conditions on system behaviour are met.

Gauss-Legendre quadrature: A numerical integration method that approximates integrals by weighted sums at optimally chosen points.

Cumulant: A statistical measure related to moments, used to characterise non-Gaussian features of signals.

Reproducing kernel: A function enabling nonlinear mapping of data into high-dimensional feature spaces for estimation tasks.

References

1. Congruential Summation-Triggered Identification of FIR Systems under Binary Observations and Uncertain Communications. Applied Sciences (2024).
2. A Two-Filter Approach for State Estimation Utilizing Quantized Output Data. Sensors (2021).
3. Channel Identification Based on Cumulants, Binary Measurements, and Kernels. Systems (2021).