Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems

arXiv:2604.14774 · eess.SY, cs.SY · Submitted 2026-04-16 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: We're looking at the paper: "Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems".

Dev: Here's the gist of the paper. Encrypted control employs homomorphic encryption (HE) to protect both the computation and communication stages, making it a promising approach for secure networked control systems This paper proposes a co-design framework for cryptographic parameters and delay-aware feedback gain to address performance degradation caused by encryption-induced delays in encrypted control systems.

Rosa: Problem Formulation The research considers a continuous-time linear plant described by the equation x˙(t) = Ax(t) + Bu(t), x(0) = xini, (2) To stabilize this system, a discrete-time static state feedback controller u

k: = Kx

k: is used with a sampling time of Ts > 0 for k ∈ Z≥0.

Dev: The networked control system implementation integrates an LWE based scheme to protect the feedback gain and sampled state against eavesdropping The overall configuration involves encrypting the controller gain K as K = Enc′K s mod q ∈ (C′)m×n, (4) and transmitting the sampled state x

k: as x

k: = Enc x

k: r mod q ∈ Cn, (5) to a remote server.

Rosa: Quantification of Total Delay The total delay τθ is decomposed as τθ = τ Comm θ + τ Enc θ + τ Dec θ + τ ⊗ θ, (8) where the terms represent delays due to communication, sensor-side encryption, actuator-side decryption, and server-side homomorphic multiplication.

Title and authors: Rosa: So this paper is called "Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems". It sounds like they're tackling the problem where you use encryption, right?

Dev: Yeah, essentially they're trying to figure out how to pick the right encryption settings and the feedback gain at the same time so that everything stays stable when there are delays introduced by that encryption.

Rosa: That makes sense for networked control systems, because if you just encrypt everything and then design a controller for the perfect system, those delays can really mess things up.

Dev: Exactly. They look at how much delay there is from communication and how much comes from the actual encryption steps, like encryption and decryption on the sensor side or homomorphic multiplication on the server side.

Taro: I wonder if this means we can design a system that doesn't just survive delays but actually performs better when it's running under those constraints.

Rosa: Right, because they are focusing on finding a stabilizing feedback gain that works for all those potential delays, not just one specific delay value.

Dev: That’s the core idea of this paper, so they aren't just looking at one fixed delay; they want a solution robust to the range of possible delays.

The paper's summary: Rosa: So, what does the paper actually propose? They seem to be suggesting a way to jointly pick those cryptographic parameters and that feedback gain.

Dev: They set up an outer loop for picking the cryptographic parameters based on security needs, and an inner loop where they test different feedback gains against a stability condition.

Rosa: The stability condition is really interesting because it boils down to checking if a certain matrix inequality holds for all possible delays within a certain range.

Dev: They characterize the total delay as an upper bound, which is important because they give us a concrete limit on how much delay we can expect based on those cryptographic parameters.

Taro: I'm interested in that bound part, because if you know the maximum possible delay, you can design a control law to handle it.

Rosa: Right. They use this upper bound to define a set of matrices where they check for stability using linear matrix inequalities, which means they turn an infinite search space into something finite and solvable.

Dev: That's what makes it practical for real-time systems, because we don't have time to test every possible combination of parameters and gains.

The paper's improvements: Rosa: One big improvement they point out is the co-design framework itself, which ties the security requirements directly into the control design.

Dev: They show how this lets you simultaneously satisfy both a desired security level for your LWE based scheme and maintain stability given the encryption delays.

Rosa: And they move away from just picking a gain that works for an ideal delay-free case, which is what they call K1, to something that works for the actual encrypted implementation.

Dev: The results show that K2 stabilizes the system even when the total delay reaches three-quarters of the upper bound, whereas K1 fails at three-quarters of that same bound.

Taro: So if you're designing a system, this suggests you need to account for those encryption delays proactively in your controller design rather than treating them as an afterthought.

Rosa: Exactly. The simulation data for state norm shows K2 keeping the system stable across different delay values compared to K1 which becomes unstable when the total delay hits three-quarters of its bound.

Dev: That means we have a specific gain choice that handles the worst-case scenario within the bounds they derived, and it’s much more robust than a general setting.

Conclusion: Rosa: So, wrapping up this paper on "Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems", the main thing is this co-design framework that links cryptographic parameters to the feedback gain.

Dev: It provides a way to find a set of parameters that meet security needs while simultaneously finding a stabilizing feedback gain, which is much more robust than previous methods.

Taro: It suggests we should stop treating encryption delays as just noise and start designing controllers specifically aware of how those delays grow with the security level.

Rosa: And they showed that this approach works in simulation, demonstrating that the gain K2 stabilizes the system across different total delays, even up to three-quarters of the derived upper bound.

Dev: It’s a concrete set of results showing that for real networked control systems, you need to design your system aware of those encryption-induced delays.

Taro: It gives us a clear path on how to handle these types of problems gracefully when the world misbehaves and introduces latency through cryptography.

Seoul National University

eess.SY, cs.SY

Submitted: 2026-04-16

Updated: 2026-07-28

Comments: 6 pages, 3 figures, 2 tables, accepted to SICE FES 2026

DOI: 10.23919/SICEFES71276.2026.11739283

Code: https://github.com/tuneinsight/lattigo

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 77/100

The gist: Encrypted control employs homomorphic encryption (HE) to protect both the computation and communication stages…

Key concepts

Homomorphic Encryption (HE)
HE is a cryptographic technique that allows computations to be performed directly on encrypted data without decrypting it first. In this context, it protects both the control signals and the state information during transmission and processing.
Total Delay ($ au_ heta$)
This measures the cumulative time delay introduced by various stages of encryption, communication, and decryption in a networked control system. The paper quantifies this delay using specific lemmas to establish an upper bound ($ar{ au}_ heta$), which is crucial for stability analysis.
Delay-Aware Gain Design
This involves designing the feedback gain ($K$) specifically considering the maximum possible total delay ($ au_ heta$). Instead of a fixed gain, the system seeks a gain that ensures stability across all delays within a defined range.
Outer-Inner Co-design Framework
This is an iterative design strategy. The outer loop chooses cryptographic parameters ($ heta$) based on security and delay constraints. The inner loop then searches for a stabilizing feedback gain ($K$) that works with the chosen parameters, ensuring both security and performance goals are met simultaneously.

Terminology

Summary

Encrypted control employs homomorphic encryption (HE) to protect both the computation and communication stages, making it a promising approach for secure networked control systems This paper proposes a co-design framework for cryptographic parameters and delay-aware feedback gain to address performance degradation caused by encryption-induced delays in encrypted control systems. p. 0

Where the paper says it

Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems Yeongjun Jang1† 1ASRI, Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea (Tel: +82-2-880-9053; E-mail: jangyj0512@snu. kr) Abstract: Encrypted control employs homomorphic encryption (HE) to protect both the computation and communication stages, making it a promising approach for secure networked control systems.

Problem Formulation

The research considers a continuous-time linear plant described by the equation x˙(t) = Ax(t) + Bu(t), x(0) = xini, (2) To stabilize this system, a discrete-time static state feedback controller u[k] = Kx[k] is used with a sampling time of Ts > 0 for k ∈ Z≥0. The networked control system implementation integrates an LWE based scheme to protect the feedback gain and sampled state against eavesdropping The overall configuration involves encrypting the controller gain K as K = Enc′K s mod q ∈ (C′)m×n, (4) and transmitting the sampled state x[k] as x[k] = Enc x[k] r mod q ∈ Cn, (5) to a remote server. p. 1

Where the paper says it

To protect the feedback gain and sampled state against eavesdropping on the communication channels and the server, we integrate the LWE based scheme introduced in Section 2. The overall configuration is illustrated in Fig.

Quantification of Total Delay

The total delay τθ is decomposed as τθ = τ Comm θ + τ Enc θ + τ Dec θ + τ ⊗ θ, (8) where the terms represent delays due to communication, sensor-side encryption, actuator-side decryption, and server-side homomorphic multiplication. Lemma 1 characterizes the communication delay as a bound: τ Comm θ ≤ (n + m)(N + 1)d / R + 2δ. p. 2

Where the paper says it

Based on this model, the following lemma characterizes τ Comm θ . Lemma 1: Under Assumption 2, the communication delay is upper bounded as τ Comm θ ≤ (n + m)(N + 1)d R + 2δ.

Lemma 2 provides an upper bound for the remaining components of the delay: τ Enc θ + τ Dec θ + τ ⊗ θ ≤ N d (cEncn + cDecm + c⊗mn log N). p. 2
Where the paper says it

The following lemma provides an explicit upper bound. Lemma 2: There exist positive constants cEnc, cDec, and c⊗ such that τ Enc θ + τ Dec θ + τ ⊗ θ ≤ N d (cEncn + cDecm + c⊗mn log N).

This leads to a total delay upper bound of τθ ≤ (n + m)(N + 1)d / R + 2δ + N dcEncn + cDecm + c⊗mn log N = ¯τθ. p. 3
Where the paper says it

A practical strategy is to execute each operation repeatedly and use the worst-case as an empirical estimate of those constants. Although the total delay τθ is unknown under Assumption 1, Lemmas 1 and 2 provide the following upper bound as a function of the cryptographic parameter θ: τθ ≤ (n + m)(N + 1)d R + 2δ + N dcEncn + cDecm + c⊗mn log N =: ¯τθ.

Delay-Aware Gain Design

The stability requirement (R3) is reduced to verifying the Schur stability of all matrices in the set Ψ(θ, K):= A¯(τθ, K) τθ ∈ [0, τ¯θ]. The augmented state dynamics are given by x[k + 1] = Φx[k]+Γ0(τθ)ue[k]+Γ1(τθ)ue[k − 1] (12), leading to the augmented state equation ξ[k + 1] = A¯(τθ, K) ξ[k], (13). p. 3

Where the paper says it

By exactly discretizing the closed-loop system (2) under the delayed input (7) at the sampling instants and using the relation ue[k] = u[k] = Kx[k], we obtain x[k + 1] = Φx[k]+Γ0(τθ)ue[k]+Γ1(τθ)ue[k − 1] (12) = (Φ + Γ0(τθ)K) x[k]+Γ1(τθ)Kx[k − 1], where Φ = exp(ATs) and Γ0(τθ) := Z Ts τθ exp(A(Ts − h)) dh B, Γ1(τθ) := Z τθ 0 exp(A(Ts − h)) dh B. By defining the augmented state ξ[k] := [x[k] ⊤, x[k − 1]⊤] ⊤ ∈ R 2n, we further obtain ξ[k + 1] = A¯(τθ, K) ξ[k], (13) where A¯(τθ, K) := Φ + Γ0(τθ)K Γ1(τθ)K In 0n×n .

The stability condition is satisfied if there exists a symmetric matrix P ≻ 0 such that Mη(θ, K)⊤PMη(θ, K)−P ≺ 0, ∀η ∈ (−, +) ν. p. 3
Where the paper says it

(18) Proposition 1: Suppose that a cryptographic parameter set θ and a feedback gain K ∈ R m×n are given. Under Assumptions 1 and 2, if there exists a symmetric matrix P ≻ 0 such that Mη(θ, K) ⊤PMη(θ, K)−P ≺ 0, ∀η ∈ {−, +} ν , (19

Co-design of Cryptographic Parameters and Gain

The co-design method follows an outer–inner procedure summarized in Algorithm 1. The outer loop selects cryptographic parameters θ from a discrete candidate set to satisfy the security and delay requirements (R1) and (R2). p. 4

Where the paper says it

Co-design of cryptographic parameters and gain The proposed co-design method for the cryptographic parameter set θ and the feedback gain K is formulated as an outer–inner procedure, summarized in Algorithm 1. In the outer loop, the cryptographic parameter set θ is selected from a discrete candidate set, so as to satisfy the security and the delay requirements (R1) and (R2).

For each fixed candidate θ, the inner loop checks feasibility of the LMIs in (19) for each feedback gain K in a finite candidate set K. p. 4
Where the paper says it

The candidate set is discrete because the secret key length N and the modulus q are restricted to powers-of-two and prime numbers, respectively; in practice, such candidates are often precomputed in cryptographic libraries or can be chosen according to standard guidelines [15]. For each fixed candidate θ, the inner loop checks the feasibility of the LMIs in (19) for each feedback gain in a finite candidate set K.

This procedure jointly determines (θ, K) satisfying requirements (R1)–(R3). p. 2
Where the paper says it

Motivated by this issue, we propose a co-design framework for the cryptographic parameter set θ and the feedback gain K. Specifically, our goal is to jointly determine (θ, K) such that the following requirements are satisfied: (R1) Security requirement: For a desired security level λ ∗ , the condition λ(θ) ≥ λ ∗ holds.

Simulation Results

The simulation results demonstrate the effectiveness of the proposed method using a numerical example with specific plant parameters and encryption settings. The results show that gain K2 stabilizes the closed-loop system for all considered total delays, whereas the performance of the delay-unaware gain K1 deteriorates as τθ increases. p. 4

Where the paper says it

3 shows the corresponding state norm ∥x(t)∥. The results show that K2 stabilizes the closed-loop system for all considered total delays, whereas the performance of the delay-unaware gain K1 deteriorates as τθ increases.

In particular, the closed-loop system becomes unstable when τθ = 3¯τθ/4. p. 4
Where the paper says it

The results show that K2 stabilizes the closed-loop system for all considered total delays, whereas the performance of the delay-unaware gain K1 deteriorates as τθ increases. In particular, the closed-loop system becomes unstable when τθ = 3¯τθ/4.

This demonstrates both the effectiveness and necessity of the proposed co-design framework. p. 4
Where the paper says it

In particular, the closed-loop system becomes unstable when τθ = 3¯τθ/4. These results demonstrate both the effectiveness and necessity of the proposed codesign framework.

The simulation data for state norm x(t) shows K2 stabilizing the system across different delay values compared to K1. p. 4
Where the paper says it

We evaluated the closed-loop performance using the gains K1 and K2 for four different values of total delay τθ; Fig. 3 shows the corresponding state norm ∥x(t)∥.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Outer-Inner Co-design Framework for Cryptographic Parameters and Feedback Gain: The system can jointly select cryptographic parameters satisfying a desired security level while simultaneously seeking a stabilizing feedback gain that accounts for encryption-induced delays.

  2. Delay Quantification and Bound Derivation: The framework provides an upper bound on the total delay tau theta as a function of cryptographic parameters, using Lemma 1 (for communication delay) and Lemma 2 (for computation/encryption/decryption delays).

  3. Tractable Stability Condition via LMIs: The system derives a finite set of linear matrix inequalities (LMIs) in Proposition 1 which serve as a sufficient condition for the closed-loop system to remain stable for all admissible delays, replacing an infinite search space.

  4. Outer-Inner Design Procedure: The proposed algorithm can be implemented as an outer loop selecting cryptographic parameters and an inner loop searching over candidate gains K, where the feasibility check relies on solving standard semidefinite programming solvers given K.

Abstract

Encrypted control employs homomorphic encryption (HE) to protect both the computation and communication stages, making it a promising approach for secure networked control systems. Most existing results pre-design a controller in the plaintext domain and then implement it over encrypted data. However, this can be problematic because HE induces non-negligible communication and computation delays that typically increase with the security level, potentially degrading control performance and even destabilizing the closed-loop system. To address this issue, we propose a co-design framework for cryptographic parameters and delay-aware feedback gain. We first characterize an upper bound of the encryption-induced delay as a function of the cryptographic parameters. Then, for a given set of cryptographic parameters and feedback gain, we derive a sufficient condition under which the closed-loop system remains stable for all admissible delays, expressed as a finite set of linear matrix inequalities. This leads to a tractable outer-inner design procedure: the outer loop selects cryptographic parameters satisfying the desired security level, while the inner loop seeks a stabilizing delay-aware feedback gain.

Sources

Related papers