Robust Control Design for LTI Systems with Sampling-Time Jitter
summary
The gist
Sampling-time jitter makes a sampled-data control loop time-varying, so standard LTI robust-control synthesis does not apply directly The exact zero-order-hold model is decomposed into a nominal…
In short
The paper addresses time-varying control loops caused by sampling-time jitter in sampled-data systems. It decomposes the system into a nominal plant and a memoryless uncertainty block representing jitter. This allows standard robust control tools to be used, proving stability for every possible jitter sequence using matrix norms, while enabling synthesis of controllers that optimize either nominal performance or weighted robustness.
Key concepts
- Sampling-Time Jitter
- This refers to the small timing errors that occur when a continuous-time system is sampled. It makes the control loop time-varying because the exact time between samples is not perfectly constant. This jitter acts as an unknown, time-varying disturbance in the system dynamics.
- Zero-Order Hold (ZOH) Decomposition
- The paper breaks down the exact ZOH model into two parts: a stable, nominal LTI plant and a memoryless uncertainty block. This decomposition isolates the effect of jitter as a static gain between stacked signals, simplifying the time-varying problem into a standard robustness test.
- Structured HÛ√ Synthesis
- This synthesis method is used to find a controller that achieves optimized nominal performance while guaranteeing robust stability against the jitter. It minimizes a specific norm of the closed-loop system, ensuring good performance within the bounds of known jitter uncertainties.
- µ-Synthesis (mu Synthesis)
- This technique is used to achieve robust performance by considering a set of uncertainty bounds. It finds a controller that satisfies a scaled small-gain condition across the entire envelope of possible jitter sequences, guaranteeing performance even under worst-case jitter conditions.
Terminology used across episodes
This episode discusses
The paper
Robust Control Design for LTI Systems with Sampling-Time Jitter · Read on arXiv
Dieter Schwarzmann, Jan Lunze
Institute for Systems Theory and Automatic Control, University of Stuttgart · Faculty of Electrical Engineering and Information Technology, Ruhr-University Bochum
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: We're looking at the paper: "Robust Control Design for LTI Systems with Sampling-Time Jitter".
Dev: Here's the gist of the paper. Sampling-time jitter makes a sampled-data control loop time-varying, so standard LTI robust-control synthesis does not apply directly The exact zero-order-hold model is decomposed into a nominal time-invariant plant and a memoryless, time-varying uncertainty block with a norm bound representing the sampling-time jitter As the block is memoryless, the operator small-gain condition for the timevarying loop reduces to a standard robustness test with a matrixnorm bound It is shown that the same test also guarantees uniform exponential stability for every admissible jitter sequence Embedding the decomposition into a generalized plant with performance channels serves two synthesis objectives with standard robust-control tools: Robust stability with optimized nominal performance through structured H∞ synthesis, and robust performance through µ synthesis with constant scalings Problem Formulation and Decomposition The paper reformulates the time-varying design problem as a normbounded robust-control problem The exact ZOH decomposition into nominal LTI dynamics and a memoryless jitter block is known The state equation becomes xk+1 = Ad,0 xk + Bd,0 uk + ∆A(εk) xk + ∆B(εk) uk The jitter-driven perturbation is ∆A(εk)xk+∆B(εk)uk Both matrices depend only on the current εk and carry no internal dynamics, so the perturbation acts as a static gain between the stacked signals xk and uk Robust Stability Certification The operator smallgain theorem is applied as a robustness test for this feedback loop and prove that this test reduces to a matrix-norm condition on the memoryless jitter block This condition ensures stability for every jitter sequence in E(¯ε) The operator norm condition (18) therefore simplifies to the matrix norm condition on the memoryless jitter block Synthesis Workflow The synthesis workflow has three steps: Choose the jitter bound ε¯ from a timing specification or from measurements, compute the uncertainty bound ¯δ from (12), and construct the generalized plant P including the performance weights Produce a fixed controller with one of the two synthesis tools: Structured H∞ synthesis yields robust stability with optimized nominal performance, and µ synthesis yield robust performance Verify the certificate of the selected controller with the conditions below, which differ in the property they certify Performance Objectives For robust stability with optimized nominal performance, the design problem is min K∥N22(K)∥H∞ subject to ¯δ∥N11(K)∥H∞ < 1 over a prescribed controller structure with an internally stable nominal loop For a target induced gain γ, the scaled small-gain condition is DoNbγ(K)D−1i H∞ < 1 with constant scalings Design Parameter Tuning The bound ε¯ can therefore be viewed as a tuning knob: Choosing a design jitter bound εdes ≤ ε¯ and ¯δdes = maxε≤εdes σ¯ ∆(ε) ≤ ¯δ trades certified jitter robustness for performance, since a larger∥N11∥H∞ becomes admissible The certificate covers E(εdes) alone, so ruling out the discarded tail εk > εdes is left to the engineer’s timing measurements Time-Domain Validation The simulations show the distinction between the certificates: A constant-jitter check does not prevent instability under alternating jitter, and robust stability by itself leaves performance degradation untouched The µ design trades nominal speed for a guarantee on weighted performance over the whole jitter envelope E(¯ε) The certificate is what distinguishes Kµ from a merely detuned design Structured H∞ synthesis then optimizes nominal performance under a jitter-stability constraint, and constant-scaled µ-synthesis certifies robust performance as well The guarantees hold for every sequence in the jitter envelope E(¯ε), including uniform exponential stability The full block, together with the requirement to cover every sequence, is conservative relative to physical jitter Measured stochastic jitter models and explicit inter-sample performance analysis would relax the design The certificate covers jitter only Appendix Details
Paper summary: Rosa: So, we're looking at this paper called "Robust Control Design for LTI Systems with Sampling-Time Jitter". The main idea here is that the sampling-time jitter makes a sampled data control loop time-varying, which means the standard robust control methods you usually use don't apply directly.
Dev: Yeah, so what they claim is they can reformulate this whole dynamic output feedback design problem under bounded, unmeasured sample-to-sample jitter into a robust control problem.
Taro: But how do they handle that time variation? Does it mean the system dynamics are constantly shifting in a way that breaks things?
Rosa: They tackle this by decomposing the exact zero order hold model into two parts: a nominal time invariant plant and this memoryless, time varying uncertainty block.
Dev: That's interesting because if you can isolate the jitter part as something simple, like memoryless, it opens up a different way to look at the stability problem.
Taro: So they're saying we can treat this jitter not as a complex system that keeps changing its own internal structure, but more like a static gain between the signals at each sample time?
Rosa: Exactly. The paper shows that because this uncertainty block is memoryless, which means it doesn't have its own dynamics, the operator small gain condition for that time varying loop simplifies down to a standard robustness test using just a matrix norm bound.
Dev: That's the key simplification, right? So instead of dealing with every possible way the jitter sequence can behave over time, they only need to check one condition based on some maximum uncertainty bound.
Taro: And what does this mean for us in terms of actual system behavior? Does this guarantee stability even if the world misbehaves in a tricky way?
Rosa: They show that this single test—this matrix norm condition—actually guarantees uniform exponential stability for every admissible jitter sequence within a certain range.
Dev: That's strong because it means no matter what the jitter does, as long as it stays within that bound, the system won't become unstable.
Taro: But how do you actually use this to design a controller? Does this just give us a stability guarantee without telling us how to build the actual control law?
Paper summary: Rosa: They embed this whole decomposition into a generalized plant that includes performance channels. This setup lets you do two synthesis objectives at once: first, you get robust stability with optimized nominal performance using structured H infinity synthesis, and second, you get robust performance through mu synthesis with constant scalings.
Dev: So it gives us a choice there: either focus on getting the best possible control speed while keeping the jitter safe—that's the structured H infinity path—or you can use mu synthesis to guarantee performance over a whole range of jitter sequences, even if it means trading off some nominal speed.
Taro: I wonder about the trade-off part. If we pick a design jitter bound epsilon bar, and then choose a smaller design bound epsilon des for your actual controller—how does that affect the stability guarantee?
Rosa: Choosing a smaller design jitter boundepsilon des means you're trading off guaranteed jitter robustness for performance. The paper shows that if you pick epsilon des less than or equal to the larger bound epsilon bar, and you set the uncertainty bound delta delta as max of epsilon des sigma bar delta, then your certificate only covers the range up to epsilon des.
Dev: So if we want a tighter controller, we can't guarantee safety for jitter sequences that are bigger than that tighter limit.
Taro: That makes sense. It suggests that the certificate is only valid for the specific jitter envelope you design for, and anything outside of it is left up to us to handle with our own measurements or timing models.
Rosa: Right, so we're left with a practical tuning knob here—the bound epsilon bar—and then we have to rely on our physical understanding of the noise or jitter that's actually happening in the field.
Dev: When you look at the performance objectives, for robust stability with optimized nominal performance, you are minimizing a cost function involving N squared norm H infinity subject to this matrix norm condition being less than one over a prescribed controller structure.
Taro: And for the other path, mu synthesis, it's about finding constant scalings that satisfy the scaled small gain condition Do Nb gamma D minus i H infinity less than one. What does that scaling actually mean in practice?
Rosa: It means you're looking at how much performance degrades when you scale up or down the loop gains, and mu synthesis helps you find those optimal constant scaling factors to keep the performance bound tight over the whole jitter envelope.
Paper summary: Dev: This paper suggests that structured H infinity synthesis optimizes nominal performance under a jitter-stability constraint, while constant scaled mu synthesis certifies robust performance across that whole envelope.
Taro: So what's the bigger picture here for someone who just wants to run a system reliably in the real world? It seems like you have two ways to approach it depending on what you prioritize.
Rosa: It means if your top priority is getting the absolute best possible speed without worrying about jitter, structured H infinity synthesis is the way to go.
Dev: But if you need a guarantee that your performance stays within bounds even with some uncertainty in the sampling timing, mu synthesis gives you that robust performance certificate.
Taro: It sounds like we have tools for different kinds of real-world problems—one for optimization under known constraints, and one for guaranteeing behavior over an unknown, bounded set of conditions.
Rosa: Exactly. The paper on Robust Control Design for LTI Systems with Sampling-Time Jitter shows that by treating the jitter as a memoryless uncertainty block, we can use standard robust control tools to get provably correct controllers for all admissible jitter sequences within a defined bound.
Dev: It’s about using H infinity synthesis for optimized nominal performance when stability is guaranteed under those bounds, or using mu synthesis for robust performance guarantees across the entire jitter envelope.
Taro: The limitation they mention, and what I see from the setup, is that this approach relies on defining that bound epsilon bar upfront based on timing specifications or measurements.
Rosa: That means if you have stochastic jitter that’s truly unpredictable—not just bounded—this certificate might not cover everything, and we'd need to look at modeling that actual stochastic jitter instead of a simple worst-case bound.
Dev: So the paper is very solid for systems where we can reasonably define what the maximum possible sample-to-sample variation is likely to be.
Taro: It’s a powerful way to use existing robust control math to solve a problem that sounds like it should require a whole new set of time-varying control techniques.
Rosa: Yeah, it bridges the gap between theoretical stability guarantees and practical synthesis workflows using tools we already have in the toolbox.
Conclusion: Rosa: So we're wrapping up this paper, "Robust Control Design for LTI Systems with Sampling-Time Jitter," by the authors... it's about how to design control systems when the timing between samples isn't perfect, which makes everything time-varying.
Dev: Exactly. They show you can take that tricky time variation and turn it into a standard robust control problem by just looking at a matrix norm bound on some memoryless uncertainty block.
Rosa: It seems like the big idea is that they use these existing robust control tools, like structured H infinity synthesis or mu synthesis, to get either optimized performance or guaranteed performance over the whole jitter envelope.
Dev: Right. So you can choose between getting the best nominal speed under a stability constraint, or you can get robust performance across all possible jitter sequences by using constant scalings in mu synthesis.
Rosa: It boils down to picking a design jitter bound, and then tuning your actual controller's limit based on that. It’s basically a trade-off between how stable the system is under every possible timing error and how fast it runs when things are perfect.
Dev: And they make it clear that while the certificate works for every sequence within their defined bound, you still have to figure out what’s actually happening in your real world jitter outside that range.
Rosa: That's the catch. It gives you a way to build a controller that’s safe for a certain level of timing error, but it doesn't magically know every single random fluctuation you might see on the road or in the factory.
Dev: So what this means for us is that we can use these powerful tools to get provable guarantees when we have a reasonable idea of the worst-case timing variation, but we still need to model those actual stochastic jitters separately.
Rosa: Yeah, so this paper bridges that gap between theory and practice by giving us concrete synthesis workflows for handling sampling jitter in control systems.
More episodes
- 2608.13438-ContactGuard: Pre-Contact Execution Monitoring with Action-Conditioned Latent World Models
- 2610.34593-Future Prediction as Training-Only Video Supervision for Pretrained VLA Policies
- 2610.35385-EmbodiedLab: Beyond Manipulation, Benchmarking Embodied Agents That Act to Know
- 2610.36378-Data-Driven Hamiltonian for Neural Optimal Control
- 2610.37275-A Heuristic for Scheduling Autonomous Vehicles at Intersections
- 2610.37288-Dynamic Coordination of Autonomous Vehicles at Intersections
- 2610.37632-Real-Time Coordination of Autonomous Vehicles at Intersections
- 2611.13779-Softness is something you also need! Why compliance helps control: information metrics of a passive walker with compliant ankle
- 2611.13411-Residual-Corrected Physics-Informed Neural Network KKL Observers with Input-to-State Stability Guarantees
- 2611.14488-Is Your Robot's Memory Just a Clock? Temporal Shortcuts in Memory-Augmented VLA Policies