Channel-Anchored Seed Establishment for Keyed Artificial Noise

Wireless physical-layer security has an awkward secret of its own. Several of its most effective techniques — injecting artificial noise on chosen subcarriers, or a diffusion model — a generator trained by gradually denoising random noise into samples — that synthesizes masking signals on pilot symbols — require the legitimate receiver to already hold a shared secret key or random seed, so it can subtract the junk energy before decoding. Secrecy Wireless Information and Power Transfer in OFDMA Systems and Generative AI based Secure Wireless Sensing for ISAC Networks both take that seed for granted. But in ad-hoc and industrial-IoT networks you cannot pre-distribute symmetric keys to every possible receiver, so the seed has to be established over the air — and the lightweight, certificate-free cryptography you would naturally use for that is precisely the cryptography with the worst attack record in the recent literature. An active adversary can swap in a fake public key and forge signatures for any identity (On the Security of an Efficient and Robust Certificateless Signature Scheme for IIoT Environments); observing a single valid signature suffices for universal forgery via a simple ratio rescaling (On the security of an enhanced signature scheme); and even a well-formed ciphertext offers no protection if the encryption implementation is subverted, because ciphertext generation that ignores the recipient's public key lets an eavesdropper with a subversion key recover session keys from just two consecutive ciphertexts (Subvert KEM to Break DEM: Practical Algorithm-Substitution Attacks on Public-Key Encryption). The physical-layer community assumes the seed exists; the cryptographic community assumes an ideal channel. Nobody closes the loop.

This proposal closes it in the opposite direction from what a pure cryptographer would expect: instead of fixing the algebra and then doing wireless, we use the wireless channel itself to patch the algebra. The core mechanism is a protocol we call Channel-Anchored Seed Establishment (CASE), in which the artificial-noise machinery that the physical layer already deploys — mandatory pilot symbols, key-driven noise sequences, null-space projection — is turned into a physical proof-of-possession and a second entropy source for the session key. A forger can imitate a signature, but it cannot cancel noise derived from a private key it does not hold; a subverted encryptor can chain coins across ciphertexts, but the session key no longer depends on those coins alone.

Core Claim

The claim is that keyed artificial noise can bootstrap its own key establishment and, in doing so, neutralize all three cryptographic attack classes at once. CASE has two ingredients. First, the AN seed is established with a lightweight certificateless key-encapsulation mechanism, giving up-front computational security without certificates. Second, the seed immediately drives an AN-cancellation round on pilot symbols: the responder must transmit pilots masked with noise derived from a value only the genuine private-key holder can compute, and the initiator must observe clean channel state in return. This round simultaneously acts as a proof of possession — defeating public-key replacement and single-signature forgery, since On the security of an enhanced signature scheme shows a forger scales signature component $V$ by a ratio of hashes without ever learning the key, and a key-replacer as in On the Security of an Efficient and Robust Certificateless Signature Scheme for IIoT Environments cancels unknowns in the verification equation with a crafted public key, but neither can produce the private-key-derived noise — and it anchors the final session key to the physical channel, since an eavesdropper at a different spatial location observes an independent channel and thus an independent anchoring factor, the same independence that Multiple-Input Multiple-Output Gaussian Broadcast Channels with Confidential Messages exploits when it projects noise into the legitimate receiver's null space.

Concretely, the symmetric key becomes

$$K_{\text{sym}} = \mathrm{HKDF}\big(k_{\text{KEM}} \,\|\, \hat{h}_{AB}\big),$$

where $k_{\text{KEM}}$ is the secret shared by the encapsulation mechanism and $\hat{h}_{AB}$ is the channel state measured during the AN-cancellation round. The design goal is a stated, testable property: an adversary holding the subversion key of Subvert KEM to Break DEM: Practical Algorithm-Substitution Attacks on Public-Key Encryption and any two (or all) consecutive ciphertexts still cannot recover $K_{\text{sym}}$, because the encapsulated secret is no longer the whole key. The chaining of coins across sessions becomes insufficient rather than merely discouraged.

Research Program

We will build and analyze three things. The first is the CASE protocol itself, specified for a two-node initiator–responder link and, in an extension, for the multi-receiver setting of Secrecy Wireless Information and Power Transfer in OFDMA Systems, where per-subcarrier seeds must be provisioned individually. The design deliberately reuses overhead the physical layer already pays: pilots are mandatory for channel estimation, and Generative AI based Secure Wireless Sensing for ISAC Networks already demonstrates that safeguarding signals can be modulated onto those pilots, so the proof-of-possession rides on transmissions that would happen anyway. This is our answer to the cost concern that Chen et al. raise about proofs of possession generally — that they reintroduce handshakes or non-interactive zero-knowledge proofs that negate the lightweight, non-interactive advantage of certificate-free design.

The second is a combined attack model and security argument. Existing analyses are fragmented: Cryptanalysis on Two Certificateless Signature Schemes treats Type I key-replacement and Type II malicious-authority adversaries separately, and Subvert KEM to Break DEM: Practical Algorithm-Substitution Attacks on Public-Key Encryption treats subverted implementations separately again. We define a single game in which the adversary may replace public keys, observe one valid signature, and hold the subversion key simultaneously, and we prove that session-key recovery requires either breaking the encapsulation problem or distinguishing the channel anchor, with the latter bounded by the channel-independence argument of Multiple-Input Multiple-Output Gaussian Broadcast Channels with Confidential Messages. One genuine technical risk is mobility: channel state information at the transmitter becomes stale in fast-fading environments, and stale estimates cause noise to leak into the legitimate channel. We will handle this with a multi-slot variant adapting the retrospective alignment technique of On the secrecy degrees of freedom of multi-antenna wiretap channels with delayed CSIT, which recovers a secrecy degree of freedom of 2/3 even with delayed transmitter-side channel knowledge — though we must honestly report that the pure delayed-CSIT anchor is weaker than the instantaneous one, and our analysis will carry that loss through rather than assume ideal feedback. A second risk is that the certificateless literature has a documented history of flawed proofs — On the Security of an Efficient and Robust Certificateless Signature Scheme for IIoT Environments found four logical errors in a scheme with a published proof — which is itself an argument for a physical-layer check that does not depend on the correctness of any simulation argument.

The third deliverable is a reference implementation: the certificateless components on the Pairing-Based Cryptography library on an Intel Core i5-8500 (the platform used by Fully Secure Lightweight Certificateless Signature Scheme for IIoT), and the AN-cancellation round on a USRP software-radio testbed, reusing the pilot-modulation pipeline of Generative AI based Secure Wireless Sensing for ISAC Networks.

Baselines

The naive reference point is the do-nothing configuration used implicitly by every keyed-AN system today: the AN seed is simply assumed pre-shared, as in Secrecy Wireless Information and Power Transfer in OFDMA Systems and Generative AI based Secure Wireless Sensing for ISAC Networks, and the seed-distribution step is left out of the threat model entirely. Running this baseline means executing our attack implementations against it and measuring how often the attacker ends up holding the AN seed; it defines the floor the proposal must lift.

The strongest existing competitor is logical-layer-only repair: distribute the AN seed using the certificateless signature scheme of Fully Secure Lightweight Certificateless Signature Scheme for IIoT, which provably binds user secrets, authority partial keys, identities, and public parameters together and resists both Type I replacement and single-signature forgery in the standard model, at a reported 5.45 ms for signing plus verification on an Intel Core i5-8500 with the PBC library — nearly twice as fast as the previous Karati et al. scheme at 9.26 ms. This is the number to beat on overhead, because it represents the best published point on the security–cost frontier for exactly this setting. We will run it on identical hardware, identical channel traces, and an identical tuning budget, and we will count CASE as beating it only if it matches or undercuts its overhead while additionally surviving attacks that the pure signature fix does not address — specifically the algorithm-substitution recovery of session keys from two consecutive ciphertexts. The scheme of Fully Secure Lightweight Certificateless Signature Scheme for IIoT is a fair and serious baseline precisely because it solves two of our three attack classes; the honest framing of our contribution is the third class plus the physical check.

The key ablation isolates the channel anchor: CASE without the $\hat{h}_{AB}$ term in the key derivation, keeping the certificateless handshake and the AN round. If our security story is right, this ablated variant falls back to the two-ciphertext ASA of Subvert KEM to Break DEM: Practical Algorithm-Substitution Attacks on Public-Key Encryption — the subversion key plus two ciphertexts should again suffice — while the full protocol resists. A second ablation removes the AN-cancellation check and should again admit the single-signature ratio forgery and public-key replacement of Practical certificateless short signature scheme and On the Security of an Efficient and Robust Certificateless Signature Scheme for IIoT Environments. If either ablation survives its attack, the theory is wrong and we will say so.

Why not just…?

Why not just use the provably secure certificateless scheme of Rezaeibagha et al. and stop there? This is the most serious alternative. Their scheme, at 5.45 ms per sign-and-verify, plausibly closes the public-key-replacement and single-signature-forgery holes at the logical layer (Fully Secure Lightweight Certificateless Signature Scheme for IIoT). What it does not touch is the ciphertext side: an algorithm-substitution attack against the encapsulation mechanism is unaffected by how well the accompanying signature is designed, because the subverted implementation chains its own internal coins across sessions and the signature on the handshake verifies perfectly (Subvert KEM to Break DEM: Practical Algorithm-Substitution Attacks on Public-Key Encryption). It also inherits the field's proof-reliability problem, given that a peer scheme with a published proof was later shown to contain four logical errors (On the Security of an Efficient and Robust Certificateless Signature Scheme for IIoT Environments). Our first experiment makes this concrete: we will run the ASA against the Rezaeibagha-authenticated configuration and show the seed still leaks; if it does not, the proposal collapses to an integration study and we will report that.

Why not just add an explicit proof of possession? Chen et al. note that requiring users to prove knowledge of the private key behind a submitted public key does prevent key replacement — this works, and we do not dispute the algebra. The objection is cost and system shape: such proofs reintroduce round-trip handshakes or non-interactive zero-knowledge proofs that erase the lightweight, non-interactive advantage that motivated certificate-free design in the first place. CASE's cancellation check rides on pilot symbols that the receiver must transmit anyway for channel estimation, following the pilot-modulation pattern already demonstrated in Generative AI based Secure Wireless Sensing for ISAC Networks. We will settle this empirically by implementing both defenses on the same testbed and reporting millisecond overhead and radio resources side by side; if the explicit proof turns out cheaper in practice, that result stands.

Why not just pre-share symmetric keys for the AN seed? It trivially defeats all three attacks — and it is exactly the assumption the keyed-AN papers make without justification, which is the circularity this proposal exists to break. Pre-distribution does not scale to ad-hoc and multi-receiver networks where any node may need to become a legitimate receiver, and a compromised long-term symmetric key is catastrophic and permanent, whereas CASE derives a fresh anchoring factor per session from the physical channel. The pre-shared configuration remains in our experiments as the do-nothing baseline, so the reader can see exactly what mobility and scale cost it.

Proposed Experiments and Decisive Checks

Four experiments, each tied to resources and reported numbers in the literature. First, attack reproduction: we reimplement the ASA of Subvert KEM to Break DEM: Practical Algorithm-Substitution Attacks on Public-Key Encryption against a hash-proof-system-style encapsulation mechanism and confirm the two-consecutive-ciphertext session-key recovery, then repeat it against CASE and its no-anchor ablation. The decisive metric is session-key recovery success rate given the subversion key and the full transcript: 100% for the ablation, negligible for full CASE. Second, forgery under replacement: we reproduce the ratio attack $V^* = (\lambda^*/\lambda)V$ of On the security of an enhanced signature scheme and the offset-crafting replacement of On the Security of an Efficient and Robust Certificateless Signature Scheme for IIoT Environments, and measure end-to-end forgery acceptance — accepted at the logical layer but rejected at the AN-cancellation round. No reported timing figures exist for these attacks on Lacuna for any of the cryptanalysis papers, so we will report our own reference runtimes rather than compare against invented numbers.

Third, physical-layer performance must not regress: on the USRP N321 testbed of Generative AI based Secure Wireless Sensing for ISAC Networks, using its indoor channel-state-information dataset covering four activities (sitting down, walking, running, and falling) and packet rates from 100 to 600 packets per second, we require the unauthorized receiver's activity-recognition accuracy to remain degraded by roughly the 70% that paper reports, with an accuracy degradation rate near its reported 0.82, while the legitimate receiver's accuracy is statistically unchanged from the pre-shared-seed configuration. Fourth, cost: on the Intel Core i5-8500 with the PBC library, we compare sign-and-verify overhead against the 5.45 ms reported by Fully Secure Lightweight Certificateless Signature Scheme for IIoT and measure the incremental cost of the AN-cancellation round against an explicit zero-knowledge proof of possession. Success is defined narrowly and in advance: CASE matches the published logical-layer cost within a small margin, survives all three attack classes in simulation and on hardware, and its two ablations each fall to the attack they were supposed to block. Any other outcome is a negative result we would publish as such — the field's habit of discovering, after deployment, that a scheme's proof aborted on the adversary's winning query is exactly what this experimental discipline is meant to avoid.