The phase of minimum shift keying (MSK) exhibits special memory that existing SNR estimators rarely exploit. Building on this, this letter proposes a phase-domain moment-based estimator built upon a bounded, carrier-independent observable derived from adjacent normalized samples. We prove that this observable is marginally sufficient for a single phase increment under any symmetric phase-noise law, and that its first moment is strictly increasing under the Tikhonov model, giving a unique lookup-table inverse. The estimator retains above 95% Godambe efficiency per independent increment over [−5, 20] dB, achieving comparable accuracy to M2M4 with only half the hardware resources.
Traditional model-driven passive localization for non-cooperative targets suffers severe performance degradation in multipath-rich environments, such as mountainous or island regions. Deep learning (DL) has been widely applied to localization in indoor multipath environments due to its strong nonlinear fitting capability, but these methods predominantly rely on Channel State Information (CSI), which is difficult to acquire in non-cooperative scenarios. Therefore, we learn channel and location information directly from raw multi-station in-phase and quadrature (I/Q) data. To handle the high dimensionality and complex temporal correlations of raw I/Q samples, we propose MambaLoc, utilizing Convolutional Neural Network (CNN) to extract local features and Mamba for efficient long-sequence modeling. A task-aware positional encoding model jointly captures receiver identity and temporal information to improve spatial-temporal reasoning. Experiments demonstrate that MambaLoc achieves superior localization accuracy, stronger robustness at low signal-to-noise ratios (SNRs) with long signal lengths, and lower computational cost than Transformer base-lines.
Protocol data units contain inherent redundancy that can serve as prior information for decoding. When self-synchronizing scrambling (SSS) precedes low-density paritycheck (LDPC) encoding, however, the recursive scrambling relation couples each protocol bit to multiple scrambled bits, thereby precluding the direct incorporation of protocol priors into conventional LDPC decoding. This letter recasts the problem of conveying protocol-derived priors to the LDPC decoder through the SSS descrambling relation as a joint decoding problem over an equivalent composite code. Factorization of the joint posterior yields a unified LDPC-like factor graph integrating protocol priors, SSS recursion constraints, and LDPC parity checks, from which a soft-input soft-output joint decoder based on the sum-product algorithm (SPA) is derived. With known-header priors covering 5%–10% of the pre-scrambling bits, protograph extrinsic information transfer (PEXIT) analysis predicts threshold gains of approximately 0.28–0.51 dB over conventional cascaded decoding. Finite-length simulations corroborate the predicted performance improvement, while the proposed decoder incurs only modest additional complexity.
Stacked intelligent metasurfaces (SIMs) can realize multiple-input multiple-output (MIMO) processing in the wave domain by configuring multilayer transmission coefficients. In this letter, we investigate noise-aware interference-free channel fitting for multiuser SIM-assisted MIMO multicasting, and formulate it as a user-averaged diagonal-matrix fitting problem that minimizes the mismatch between each noisy equivalent response and a desired diagonal target. This problem is challenging because the shared multilayer SIM phase variables are coupled through cascaded propagation and constrained to have unit modulus. To address this coupled unit-modulus design, we propose a alternating direction method of multipliers-based algorithm that separates the quadratic fitting update from the unit-circle feasibility step, yielding closed-form diagonal-vector updates followed by unit-circle projection and scaled-dual updates for each active layer. Numerical results under the tested settings show lower normalized mean-squared error (NMSE) and a more favorable measured runtime–NMSE tradeoff than Adam-based and explicit-gradient baselines.
This letter proposes a channel estimation algorithm for noise-based communication over a quasi-static memory finite impulse response channel. A Gaussian Quasi-Maximum Likelihood (QML) problem is formulated for the tap powers of the current realization, on which the received powers solely depend. Sufficient conditions for their identifiability are derived, both the QML estimator and a closed-form baseline are shown to be consistent, and the QML asymptotic covariance is obtained in closed form. The QML estimator roughly halves the baseline’s mean square error, whereas both give the same bit error probability, since detection depends only on the sum of the tap powers.
