Stochastic Aperiodic Control of Networked Systems with i.i.d. Time-Varying Communication Delays

Yohei Hosoe

This paper, published in early 2022, tackles a classic headache in control engineering: how to maintain stability when your controller and your robot (or plant) are talking to each other over a laggy, unpredictable network like the Internet.

TL;DR

The author derives a necessary and sufficient condition for stabilizing networked control systems (NCS) where communication delays are random and potentially unbounded. By framing the problem through stochastic stability, the paper provides a practical design tool—a Linear Matrix Inequality (LMI)—that allows engineers to guarantee stability even for unstable plants operating over unreliable networks.

What Problem Does This Paper Address?

In a perfect world, a controller receives sensor data and sends back commands instantly at fixed intervals. In the real world, signals travel over networks that introduce delays. If these delays are long or unpredictable, an unstable system (like a drone or a balancing robot) can quickly oscillate out of control.

Most existing research assumes that these delays are "bounded"—meaning they never exceed a certain maximum time. However, Internet traffic is notoriously "jittery" and can occasionally experience very long spikes. This paper addresses the gap in designing controllers for unstable plants where delays are stochastic (random) and unbounded. It specifically looks at "aperiodic" control, where the timing of the next sample is actually determined by how long the previous network delay took.

Methods

The author uses a multi-step mathematical approach to bridge the gap between continuous-time physics and discrete-time networking:

  1. Discretization with Input Delay: The continuous-time plant is modeled only at the "sampling points" (when data actually arrives). Because of the network structure, there is a natural one-step delay between sensing the state and the control action taking effect.
  2. Extended State Space: To handle this one-step delay, the author augments the system's "state" to include the previous control input. This turns a delayed system into a standard-looking (but larger) discrete-time system.
  3. Second-Moment Exponential Stability: Instead of requiring the system to be stable in every possible worst-case scenario (which is impossible with unbounded delays), the author aims for stability in a "mean-square" sense. This ensures that, on average, the system's energy/error decays exponentially over time.
  4. LMI Synthesis: The core contribution is Theorem 2, which provides a "Linear Matrix Inequality." This is a type of optimization problem that can be solved efficiently using standard software (like MATLAB's YALMIP) to find the optimal controller gains ($F_1$ and $F_2$).

Datasets

This is a theoretical control theory paper, so it does not use a traditional "Big Data" dataset. Instead, it validates the theory using:

  • A Linearized Inverted Pendulum Model: A classic "unstable" benchmark in control theory.
  • Synthetic Delay Distributions: The author uses Exponential Distributions to model network delays. This is a rigorous test because exponential distributions have "long tails," meaning there is a non-zero probability of experiencing a very long delay.
  • Monte Carlo Sampling: 1,000 random samples of the delay were used to approximate the mathematical expectations required to solve the LMI.

Key Insights and Ideas

  • Delays as Clock Cycles: A key insight is treating the network delay itself as the sampling interval. Instead of fighting the network to stay on a fixed schedule, the controller "waits" for the network, and the math accounts for this variable timing.
  • Unbounded is Manageable: The paper proves that you don't need to assume a "maximum delay" to guarantee stability. As long as the distribution of the delays is known (or can be sampled), you can design a controller that is robust to those rare, long-lag events.
  • Data-Driven Design: Because the stability conditions rely on expectations ($E[\cdot]$), the controller can be designed using real-world network trace data. You don't necessarily need a perfect formula for the network; you just need enough samples of the delay to calculate the mean behavior.

Key Results

  • Necessary and Sufficient Conditions: The paper provides a definitive mathematical test: if the LMI has a solution, the system is stable; if it doesn't, no such linear controller exists for that specific plant and delay profile.
  • Numerical Success: In simulations of an unstable inverted pendulum, the derived controller successfully stabilized the system with a stability factor ($\lambda$) of 0.7628.
  • Visual Convergence: The paper provides plots showing 100 different "random paths" of the system. Despite the random timing and delays, every single path converged to the equilibrium point (zero error).

Limitations and Future Work

  • I.I.D. Assumption: The paper assumes delays are "independent and identically distributed" (i.i.d.). In real networks, delays are often correlated (if the network is congested now, it’s likely to be congested in the next millisecond). Extending this to Markovian or "bursty" traffic models is a logical next step.
  • Full State Feedback: The current method assumes the controller knows the exact state of the plant. In reality, sensors are noisy. Future work would need to incorporate "output feedback" or observers to handle partial/noisy information.
  • Computational Complexity: For very high-dimensional systems, calculating the expectations in the LMI via sampling could become computationally expensive.

Reference: Hosoe, Y. (2022). Stochastic Aperiodic Control of Networked Systems with i.i.d. Time-Varying Communication Delays. arXiv:2203.03207v1 [eess.SY].

Summary updated

Abstract

This paper studies stochastic aperiodic stabilization of a networked control system (NCS) consisting of a continuous-time plant and a discrete-time controller. The plant and the controller are assumed to be connected by communication channels with i.i.d. (independent and identically distributed) time-varying delays. The delays are theoretically not required to be bounded even when the plant is unstable in the deterministic sense. In our NCS, the sampling interval is supposed to be determined directly by such communication delays. A necessary and sufficient inequality condition is presented for designing a state-feedback controller stabilizing the NCS at sampling points in a stochastic sense. The results are also illustrated numerically.
BibTeX
@inproceedings{hosoe2022stochastic,
  author = {Yohei Hosoe},
  title = {{Stochastic Aperiodic Control of Networked Systems with i.i.d. Time-Varying Communication Delays}},
  booktitle = {2022},
  year = {2022},
  doi = {10.1109/cdc51059.2022.9992797},
}

Loading related content...

Figure