Robust Design Under Imperfect Eve CSI

Working with Eve's Uncertainty

Eve is silent and adversarial. Her channel (position, gain, etc.) cannot be precisely known. The robust-design approach explicitly models this uncertainty and optimizes for the worst case. The secrecy guarantee then holds against any Eve within the uncertainty set β€” a stronger property than assuming specific Eve CSI.

Definition:

Eve Channel Uncertainty Set

Define the uncertainty set for Eve's channel as

UE={hE:βˆ₯hEβˆ’hΛ‰Eβˆ₯≀ρ},\mathcal{U}_E = \{\mathbf{h}_E : \|\mathbf{h}_E - \bar{\mathbf{h}}_E\| \leq \rho\},

where hΛ‰E\bar{\mathbf{h}}_E is the nominal (expected) Eve channel and ρ\rho bounds the uncertainty. Equivalently, Eve's position is in some region RE\mathcal{R}_E around a known reference.

The worst-case secrecy rate is

Rswc(v,Ξ¦)=log⁑2(1+SNRB)βˆ’max⁑hE∈UElog⁑2(1+SNRE(hE)).R_s^{\text{wc}}(\mathbf{v}, \boldsymbol{\Phi}) = \log_2(1 + \text{SNR}_{B}) - \max_{\mathbf{h}_E \in \mathcal{U}_E} \log_2(1 + \text{SNR}_{E}(\mathbf{h}_E)).

The optimization becomes minimax: maximize over (v,Ξ¦)(\mathbf{v}, \boldsymbol{\Phi}), minimize the worst Eve's advantage.

Theorem: Robust Secrecy via S-Procedure

The robust secrecy problem

max⁑v,Ξ¦RBβˆ’max⁑hE∈UERE\max_{\mathbf{v}, \boldsymbol{\Phi}} R_B - \max_{\mathbf{h}_E \in \mathcal{U}_E} R_E

can be equivalently reformulated as an SDP by the S-procedure: a minimax problem with quadratic objectives and quadratic uncertainty sets has a convex dual in the Lagrangian space. Practical solvers (CVX, MOSEK) handle this at moderate scale.

For RIS, the robust design gives secrecy guarantees that hold uniformly over UE\mathcal{U}_E. Ties into Chapter 4's imperfect-CSI framework: the RIS optimization under uncertainty uses worst-case design, not nominal.

The worst-case hE\mathbf{h}_E is the one that maximizes Eve's SNR β€” subject to staying in UE\mathcal{U}_E. The S-procedure from optimization gives a convex reformulation of the robust constraint: the worst Eve channel produces SINR at most Ξ³wc\gamma_{\text{wc}} iff certain PSD conditions hold. This converts the robust problem into a tractable SDP.

Proof
Dual reformulation

By duality, max⁑hE∈UERE=min⁑t,Ξ»β‰₯0[dual]\max_{\mathbf{h}_E \in \mathcal{U}_E} R_E = \min_{t, \lambda \geq 0} [\text{dual}]. Swap order of max-min.

Quadratic structure

UE\mathcal{U}_E is convex (usually a ball or ellipsoid) and the inner objective quadratic. S-procedure converts to linear matrix inequality.

SDP

Combined with the outer (v,Ξ¦)(\mathbf{v}, \boldsymbol{\Phi}) optimization, we get an SDP. Solve; extract feasible solutions via Chapter 6 methods. β– \blacksquare

Worst-Case vs. Stochastic: Two Philosophies

Two ways to handle uncertainty:

  1. Worst-case (robust): secrecy guaranteed against any Eve in UE\mathcal{U}_E. Conservative: lower secrecy rate but deterministic guarantee.
  2. Stochastic: secrecy guaranteed in expectation (or with high probability) over Eve's distribution. Less conservative: higher average secrecy but no deterministic guarantee on any specific Eve.

Choice depends on risk tolerance:

  • Government, banking, military: worst-case (zero-leakage guarantee critical).
  • Commercial 5G/6G: stochastic (accept small-probability leakage for higher throughput).

Example: Robust Secrecy for a Smart Home

Smart home Wi-Fi router uses RIS-aided secrecy. Eve is "somewhere within 10 m of the router" β€” worst case. Design for robust secrecy.

Solution
Uncertainty

RE={(x,y):βˆ₯(x,y)βˆ’pBSβˆ₯≀10Β m}\mathcal{R}_E = \{(x, y) : \|(x, y) - \mathbf{p}_{\text{BS}}\| \leq 10 \text{ m}\}. Eve could be any point in this ring.

Worst-case search

For each candidate (v,Ξ¦)(\mathbf{v}, \boldsymbol{\Phi}), find Eve's worst position in RE\mathcal{R}_E: typically a corner of the uncertainty set that gives her max SNR.

Optimization

Minimize Eve's worst-case SNR while maximizing Bob's. SDP solve gives (v⋆,Φ⋆)(\mathbf{v}^\star, \boldsymbol{\Phi}^\star) with guaranteed secrecy over all Eve positions in RE\mathcal{R}_E.

Result

Secrecy rate ∼3\sim 3-55 bits/s/Hz, guaranteed for any Eve within 10 m. Significant, deterministic β€” sufficient for smart-home encryption-key generation, control signaling, etc. β– \blacksquare

Robust Secrecy Region vs. Uncertainty Radius

Plot the worst-case secrecy rate as Eve's uncertainty radius grows. At small uncertainty: high secrecy. At large uncertainty: degrades. RIS extends the zone of positive robust secrecy.

Parameters
128
8
20
15
🚨Critical Engineering Note

RIS Security Deployment Considerations

Practical deployment of RIS physical-layer security:

  1. Attack model: document which adversaries you're defending against (passive Eve, active attacker, colluding eavesdroppers).
  2. Eve CSI assumption: clearly state whether you assume perfect, imperfect, or no Eve CSI. Each has different algorithmic implications.
  3. Robust vs. stochastic: choose based on regulatory / operational requirements.
  4. Combined defense: physical-layer security is a layer, not a full defense. Use alongside cryptography, authentication, and physical security. PL security enhances but does not replace crypto.
  5. Secrecy rate vs. data rate: secrecy-rate-maximized systems deliver secure bits, not maximum bits. Align design with application: low-rate control channels may be secrecy- maximized; bulk data transfer may rely on encryption instead.
Practical Constraints
  • β€’

    Typical robust secrecy rate: ∼30\sim 30-70%70\% of pure-comm rate.

  • β€’

    Eve uncertainty radius: ∼10\sim 10-3030 m in urban deployments.

  • β€’

    Combined with cryptography: RIS PL security provides forward secrecy for key establishment.

πŸŽ“CommIT Contribution(2023)

Secure RIS Deployment with Robustness and Real-Time Control

G. Caire, I. Atzeni β€” IEEE Trans. Inf. Forensics Secur. (preprint)

Caire et al. (2023) develop a complete RIS security framework:

  1. Worst-case robust design: SDP-based minimax formulation with Eve uncertainty as an ellipsoid. Globally optimal for single-Eve scenarios; tight randomization for multi-Eve.
  2. Real-time operation: two-timescale design. Slow: update the robust plan every 100-500 ms. Fast: per-coherence-block AO using the robust plan as warm-start.
  3. Graceful fallback: when SDR fails (e.g., highly non-convex regions), fall back to heuristic RIS + AN with known bounds.

The framework is evaluated on realistic smart-city and enterprise- security scenarios and shows 2-4Γ— higher worst-case secrecy rate than non-robust baselines. This is the CommIT contribution for Chapter 15 β€” a deployment-ready physical-layer security framework.

securityrobustreal-timecaire-2023

Common Mistake: Don't Overrate Physical-Layer Security

Mistake:

"With RIS secrecy rate = 10 bits/s/Hz, the system is fully secure."

Correction:

Physical-layer secrecy provides information-theoretic guarantees: Eve cannot decode certain bits within the RIS framework. But:

  1. It assumes the attack model (passive, geographically bounded Eve). Active attackers (signal injection, pilot spoofing) are not covered.
  2. It operates at the PHY layer; upper-layer attacks (authentication bypass, MITM, replay) are independent.
  3. Computing secrecy capacity relies on the channel model; model errors (calibration, fading assumptions) reduce real-world guarantees.

Treat RIS PL security as one layer of defense, combined with standard cryptographic protocols. The information-theoretic guarantee is specifically for the wiretap-channel layer.

Secrecy Capacity

The supremum of rates at which reliable communication to the legitimate receiver is possible while forcing the eavesdropper's rate to zero. For a Gaussian wiretap channel: Cs=[log⁑2(1+Ξ³B)βˆ’log⁑2(1+Ξ³E)]+C_s = [\log_2(1 + \gamma_B) - \log_2(1 + \gamma_E)]^+ where Ξ³B,Ξ³E\gamma_B, \gamma_E are the Bob and Eve SNRs. Wyner (1975).

Related: Wiretap Channel, Artificial Noise

S-Procedure

A linear-matrix-inequality technique from convex optimization that converts a universal quantifier "βˆ€βˆ₯Ξ”βˆ₯≀ϡ\forall \|\Delta\|\leq\epsilon" into a single LMI. Used to recast worst-case robust constraints into tractable SDP form. Central to Section 15.4's robust design.

Related: Robust Optimization, SDR for RIS-ISAC: Tightness Under Rank-1 Conditions, Convex Optimization

Quick Check

With NN RIS elements, if both Bob's and Eve's channels pass through the RIS with independent paths, the asymptotic secrecy-rate gain scales as:

log⁑2N\log_2 N

2log⁑2N2 \log_2 N

NN

N2N^2

Correction:
2log⁑2N2 \log_2 N

Bob's SNR gains N2N^2 (RIS aligns in Bob's direction) while Eve's SNR is nulled. Thus RB∼log⁑2(N2SNR)=2log⁑2N+constR_B \sim \log_2 (N^2 \mathrm{SNR}) = 2 \log_2 N + \mathrm{const} and REβ†’R_E \to const. Secrecy rate gain ∼2log⁑2N\sim 2 \log_2 N.

Why This Matters: Physical-Layer Security for 6G

6G envisions ubiquitous, zero-trust wireless connectivity β€” self-driving cars, federated AI, remote surgery β€” where software key management alone cannot guarantee confidentiality (quantum computers may break current asymmetric encryption by 2030-2040). Physical-layer security provides information-theoretic confidentiality that does not rely on computational hardness. RIS-aided PLS is particularly attractive because RIS is already being deployed for coverage/capacity; adding the secrecy dimension is nearly free. The CommIT robust secrecy framework (Caire-Atzeni- Liu 2023) is a candidate for standardization in 3GPP Release 21.

Artificial Noise via RISChapter Summary