Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems
Yeongjun Jang 1† · 2026
Read Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems by Yeongjun Jang 1† (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-14774.txt.
Introduction
Jang addresses a snag in encrypted control: homomorphic encryption protects computation and communication, but its delays grow with the security level and can degrade feedback or destabilize it. Most designs fix the controller, then encrypt. Jang instead co-designs crypto settings with a delay-aware feedback gain. Jang first bounds encryption delay from the cryptographic parameters, then derives a stability condition as linear matrix inequalities, holding for all admissible delays. An outer loop picks settings meeting the security target while an inner loop seeks a stabilizing gain.
Catalog record
| CREATOR | Yeongjun Jang 1† |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26758 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.14774
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Co-Design of Cryptographic Parameters and Delay-Aware Feedback Gain for Encrypted Control Systems Yeongjun Jang 1† 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, which 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 characterize the encryption-induced delay as a function of the cryptographic parameters and derive a sufficient condition for the existence of a stabilizing delay-aware feedback gain, expressed as a finite set of linear matrix inequalities. This leads to a tractable outer-inner design procedure that searches over cryptographic parameters that satisfy a desired security level and, for each such parameter, seeks a stabilizing feedback gain. \affils 1 ASRI, Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea (Tel: +82-2-880-9053; E-mail: [email protected]) keywords Encrypted control, homomorphic encryption, security, privacy 1 Introduction Networked control systems allow resource-limited devices to efficiently outsource computationally intensive tasks to remote servers [ 1 ] . However, this architecture requires the transmission of data that may contain sensitive information, giving rise to privacy concerns [ 2 ] . In particular, communication channels are vulnerable to eavesdropping, and the remote servers may be semi-honest, correctly executing the prescribed protocol while attempting to infer private information from the received data. The leaked information can be exploited to design cyberattacks capable of causing significant physical damage, as demonstrated by the incidents in [ 3 , 4 ] . Recently, encrypted control has emerged as a promising framework for enhancing the security of networked control systems [ 5 , 6 , 7 ] . By employing homomorphic encryption (HE), which enables arithmetic operations to be performed directly on encrypted data without decryption, it protects both the communication and computation stages. More specifically, sensor measurements are encrypted and transmitted to an untrusted remote server, where the control input is computed over encrypted data and sent back to the actuator for decryption (see Fig. 1 ). Most related works begin with a controller pre-designed in the plaintext domain and then incorporate HE, while preserving the original control performance [ 8 , 9 , 10 ] . This approach, however, may be problematic in practice. Encrypted control systems inevitably introduce non-negligible delay due to the transmission of ciphertexts, whose size is typically much larger than that of plaintexts, as well as the computational burden of encryption, homomorphic evaluation, and decryption. Such delays may cause significant performance degradation and even destabilize the closed-loop system [ 11 ] . Although [ 12 ] presented a method for jointly designing the key length of a cryptosystem and control laws achieving a desired security level and control performance, it did not account for encryption-induced delays. Plant side Network ZOH Plant Sample T_{s} Scale Quantize {\mathsf{Dec}} {\mathsf{Enc}} Encrypted Controller u(t) x(t) {\mathsf{Dec}}({\mathbf{u}}[k]) u_{e}[k] x[k] \left\lceil\frac{x[k]}{{\mathsf{r}}}\right\rfloor {\mathbf{x}}[k] {\mathbf{u}}[k] Continuous signal Discrete signal Figure 1: Configuration of encrypted control system. In this paper, we propose a co-design framework for cryptographic parameters and delay-aware feedback gain for encrypted control systems. To this end, we first characterize the encryption-induced delay as a function of the cryptographic parameters. Then, for a fixed cryptographic parameters, we derive a sufficient condition for the existence of a stabilizing delay-aware feedback gain, given by a finite set of linear matrix inequalities (LMIs). This leads to a tractable outer-inner design procedure that searches over cryptographic parameters satisfying a desired security level and, for each such parameter, seeks a stabilizing feedback gain. Notation: Let {\mathbb{Z}} , {\mathbb{Z}}_{\geq 0} , {\mathbb{N}} , and {\mathbb{R}} denote the sets of integers, nonnegative integers, positive integers, and real numbers, respectively. 2 Preliminaries and Problem Formulation 2.1 LWE based cryptosystem This section briefly reviews the LWE based cryptosystem of [ 13 , 14 ] , focusing on its multiplicative homomorphic property and computational complexity. For a prime number q\in{\mathbb{N}} , we consider {\mathbb{Z}}_{q}:={\mathbb{Z}}\cap[-q/2,q/2) as the space of plaintexts. The modulo operation that maps {\mathbb{Z}} onto {\mathbb{Z}}_{q} is defined as \displaystyle a~{\mathrm{mod}}~q:=a-\left\lfloor\frac{a+q/2}{q}\right\rfloor q\in{\mathbb{Z}}_{q},~~~~\forall a\in{\mathbb{Z}}, which applies entrywise to vectors and matrices. The secret key {\mathsf{sk}}\in{\mathbb{Z}}_{q}^{N} , where N\in{\mathbb{N}} is a power-of-two, is drawn uniformly at random from the ternary set \{-1,0,1\}^{N} . Only an entity with access to it can perform both encryption and decryption. The LWE based scheme supports matrix-vector multiplication to be evaluated over encrypted data. Specifically, it employs two distinct encryption algorithms, denoted by {\mathsf{Enc}}^{\prime} and {\mathsf{Enc}} , for the multiplier and multiplicand, respectively: \displaystyle{\mathsf{Enc}}^{\prime} \displaystyle:{\mathbb{Z}}_{q}\to{\mathbb{Z}}_{q}^{(N+1)\times d(N+1)}=:{\mathcal{C}}^{\prime}, \displaystyle{\mathsf{Enc}} \displaystyle:{\mathbb{Z}}_{q}\to{\mathbb{Z}}_{q}^{N+1}=:{\mathcal{C}}, where d:=\lceil\log_{2}q\rceil . The corresponding decryption algorithm is denoted by {\mathsf{Dec}}:{\mathcal{C}}\to{\mathbb{Z}}_{q} . With a slight abuse of notation, {\mathsf{Enc}}^{\prime} , {\mathsf{Enc}} , and {\mathsf{Dec}} are applied entrywise to vectors and matrices. Then, given M\in{\mathbb{Z}}_{q}^{h\times l} and m\in{\mathbb{Z}}_{q}^{l} , there exists a homomorphic multiplication operation 1 1 1 For simplicity, we omit the dependence of \otimes on h and l . \otimes:({\mathcal{C}}^{\prime})^{h\times l}\times{\mathcal{C}}^{l}\to{\mathcal{C}}^{h} such that [ 14 ] \displaystyle{\mathsf{Dec}}({\mathsf{Enc}}^{\prime}(M)\otimes{\mathsf{Enc}}(m))\approx Mm~{\mathrm{mod}}~q. (1) The approximation error in ( 1 ) is caused by the “errors” injected during encryption, which are fundamental to the security of the LWE based scheme. These errors are typically drawn from a zero mean discrete Gaussian distribution with standard deviation \sigma>0 . We refer to \theta=(N,q,\sigma) as the cryptographic parameter set and denote the corresponding security level by \lambda(\theta) . In general, the security level increases with N and \sigma/q [ 15 ] , and can be estimated using a tool called the LWE estimator [ 16 ] . For the scalar case, i.e., when h=l=1 , the computational complexities of {\mathsf{Enc}}^{\prime} , {\mathsf{Enc}} , {\mathsf{Dec}} , and \otimes are summarized in Table 1 using the big-O notation. It can be seen that the complexities primarily depend on the parameters N and q , indicating an inherent tradeoff between security level and computational complexity. Table 1: Asymptotic computational complexity of cryptographic operations in the scalar case ( h=l=1 ). Operation {\mathsf{Enc}}^{\prime} {\mathsf{Enc}} {\mathsf{Dec}} \otimes Complexity {\mathcal{O}}(N^{2}d) {\mathcal{O}}(Nd) {\mathcal{O}}(Nd) {\mathcal{O}}(Nd\log N) 2.2 Problem formulation Consider the continuous-time linear plant written by \dot{x}(t)=Ax(t)+Bu(t),~~~~x(0)=x_{{\mathsf{ini}}}, (2) where x(t)\in{\mathbb{R}}^{n} is the state with the initial value x_{\mathsf{ini}}\in{\mathbb{R}}^{n} and u(t)\in{\mathbb{R}}^{m} is the control input. To stabilize ( 2 ), suppose that a discrete-time static state feedback controller has been designed as \displaystyle u[k]=Kx[k], (3) where x[k]:=x(kT_{s}) is the state sampled with a sampling time T_{s}>0 for k\in{\mathbb{Z}}_{\geq 0} , and K\in{\mathbb{R}}^{m\times n} is the feedback gain. Ideally, the signal u[k] is fed-back to the plant ( 2 ) through a zero-order hold (ZOH), as \displaystyle u(t)=u[k],~~~~kT_{s}\leq t<(k+1)T_{s}. Now, we consider a networked control system implementation of ( 3 ) in which the control input is computed by a remote server residing in the network layer. In particular, we consider a clock-driven sensor that samples the state periodically, whereas the controller/actuator are event-driven and computes/applies the control input as soon as the required data becomes available. To protect the feedback gain and sampled state against eavesdropping on the communication channels and the server, we integrate the