Rate-Reliability Tradeoff for Deterministic Identification over Gaussian Channels

We extend the recent analysis of the rate-reliability tradeoff in deterministic identification (DI) to general linear Gaussian channels, marking the first such analysis for channels with continuous output. Because DI provides a framework that can substantially enhance communication efficiency, and since the linear Gaussian model underlies a broad range of physical communication systems, our results offer both theoretical insights and practical relevance for the performance evaluation of DI in future networks. Moreover, the structural parallels observed between the Gaussian and discrete-output cases suggest that similar rate-reliability behaviour may extend to wider classes of continuous channels.

💡 Research Summary

The paper investigates deterministic identification (DI) over general linear Gaussian channels, providing the first thorough analysis of the rate‑reliability trade‑off for channels with continuous output. In the DI paradigm a receiver only needs to decide whether a particular message was transmitted, which can dramatically reduce the required resources compared to conventional Shannon‑style transmission. The authors consider a channel model Yⁿ = A Xⁿ + Zⁿ where A is a full‑rank deterministic matrix, Zⁿ is zero‑mean Gaussian noise with covariance Σ, and a per‑codeword power constraint ‖Xⁿ‖² ≤ nP is imposed.

A DI code consists of N pairs (u_i, D_i) where u_i ∈ ℝⁿⁿ is a codeword and D_i ⊂ ℝⁿ is a decoding region. The two error probabilities are λ₁ (missed identification) and λ₂ (false identification). Both are expressed exponentially as λ_i = e^{‑nE_i(n)} with error exponents E_i(n) ≥ Ω(1/n). The central goal is to relate these exponents to the identification rate R(n) = (1/n) log N.

Symmetric error regime – Theorem 2 treats the case where the two error exponents are of the same order. By bounding the total variation distance between the output distributions of two different codewords and converting it to the fidelity (Bhattacharyya coefficient), the authors obtain a closed‑form expression for the fidelity of two displaced Gaussians with identical covariance: F = exp{‑½ ΔᵀΣ⁻¹Δ} = exp{‑½ ‖x_i − x_j‖²_M}, where Δ = A(x_i − x_j) and M = AᵀΣ⁻¹A. Using the relation between fidelity and the error exponents yields a minimum Euclidean distance requirement ‖x_i − x_j‖ ≥ 2 r with r = √{(2 ln 2 · ν_max)/(n E(n))}, ν_max being the largest eigenvalue of M.

This distance constraint translates into a sphere‑packing problem in ℝⁿ: all codewords must lie inside the power sphere of radius √{nP} and be separated by non‑overlapping balls of radius r. By comparing the volume of a larger sphere that contains all small balls with the volume of a single small ball, the authors derive an upper bound on the number of codewords N ≤ (√{nP}+r)ⁿ / rⁿ ≤ 2(√{nP}/r)ⁿ. Consequently, the identification rate satisfies
R(n) ≤ ½ log(8 ν_max P E(n)).

If the error exponents are constant (E(n)=Θ(1)), the bound reduces to R(n)=O(1), i.e., log N grows only linearly with n. If the exponents decay as Ω(1/n), the bound becomes R(n)≈½ log n + O(1), recovering the well‑known linear‑logarithmic scaling (log N ≈ ½ n log n). Thus, for Gaussian channels the same phenomenon observed for discrete channels holds: exponential decay of the errors forces the identification rate to drop to a constant level.

Asymmetric error regime – When λ₁ and λ₂ differ dramatically (e.g., one is exponentially small while the other only decays slowly), the fidelity‑based argument becomes loose. The authors develop a separate analysis that yields distinct distance lower bounds for each error exponent. They show that both exponents must be at least Ω(1/n) to achieve the linear‑logarithmic scaling; if one exponent is Θ(1) while the other is Ω(1/n), the overall rate collapses to O(1). This confirms that the minimum of the two exponents dominates in the symmetric case, but in the asymmetric case both exponents matter.

Capacity implication – Plugging the symmetric‑error upper bound into the definition of the linear‑logarithmic DI capacity,
Ċ_DI(G) = sup_{E_i→0, nE_i→∞} liminf_{n→∞} (log N)/(log n),
yields Ċ_DI(G) ≤ ½. Combined with known lower bounds (½ ≤ Ċ_DI(G) ≤ ½), the paper re‑establishes that the DI capacity of a general linear Gaussian channel equals ½.

Practical relevance – The model encompasses AWGN, slow/fast fading, ISI, and MIMO scenarios through appropriate choices of A and Σ, making the results applicable to a broad class of modern communication systems (e.g., 6G, sensor networks, tactile internet). The derived trade‑off informs system designers how stringent reliability requirements (small error exponents) limit the number of identifiable messages, and conversely how relaxing reliability can dramatically increase identification capacity.

Conclusion – The authors provide a rigorous, unified treatment of the rate‑reliability trade‑off for deterministic identification over continuous‑output Gaussian channels. By establishing tight upper bounds for both symmetric and asymmetric error settings, they demonstrate that the linear‑logarithmic scaling is attainable only when errors vanish slower than exponentially, mirroring the behavior known for discrete channels. The work bridges a gap between theory and practice, offering concrete guidelines for employing DI in realistic Gaussian communication environments.