This letter proposes a pulse shaping filter scheme for Zak-Orthogonal Time Frequency Space (Zak-OTFS) transmission. Inspired by the construction of Wilson bases in the Gabor framework, a Gaussian-Riesz (GR) function is introduced, which achieves shift-orthogonality on a subsampled lattice while retaining time-frequency localization comparable to that of the Gaussian window. This strong localization is pivotal, as it effectively mitigates inter-Zak interference. Simulation results demonstrate that, by significantly reducing interference, the proposed scheme achieves favorable performance in terms of normalized mean square error and bit error rate, indicating accurate channel estimation and robust communication performance in Zak-OTFS systems.
We propose a latent maximum a posteriori (MAP) channel estimator for one-bit massive multiple-input multiple-output (MIMO) systems using a mixture of factor analyzers (MFA) as a generative prior. Rather than directly estimating the high-dimensional channel, the proposed method performs MAP inference over low-dimensional latent variables while incorporating the high-dimensional residual into the one-bit likelihood. We develop a Newton-based algorithm for componentwise latent MAP estimation and a safe pruning rule that exploits strong convexity to enable efficient online inference. Simulation results show that the proposed estimator outperforms existing Gaussian mixture-based estimators with significantly fewer online floating-point operations and lower memory requirements.
In single-input multiple-output (SIMO) free-space optical (FSO) systems, coherent detection requires instantaneous channel state information for every receive aperture, whose acquisition overhead grows with the number of apertures. Motivated by this, this letter investigates noncoherent detection and signal design for SIMO FSO systems employing intensity modulation and direct detection (IM/DD) over Gamma-Gamma turbulence channels. Assuming that the receiver knows only the channel statistics and not the instantaneous fading gains, we derive an exact, symbol-by-symbol, noncoherent maximum-likelihood (ML) detector by averaging over the Gamma-Gamma distribution. We further exploit the derived ML detector to optimize the positive pulse amplitude modulation (PAM) intensity levels under an average optical intensity constraint. Numerical results show that the proposed noncoherent ML detector, together with the corresponding optimized PAM constellation, significantly improves the symbol error rate (SER) performance compared to conventional equispaced PAM as well as PAM constellations optimized for the channel-mean Euclidean detector.
Pinching-antenna systems reshape short-range line-of-sight links, yet their network-level integrated sensing and communication (ISAC) behavior is not yet characterized when many waveguides coexist. This letter develops a finite line-segment Cox model in which communication pinching antennas and sensing transmit-receive pairs, activated by individual users and targets, are offspring of a shared random access-point field. We derive the coupled probability generating and Laplace functionals, obtain the exact joint coverage, including its Palm form for a serving access point drawn from the field, and prove that the shared parent induces positive communication-sensing coupling and, at fixed node density, a finite optimal waveguide length governed by placement gain against Cox-field crowding.
Deploying vision-language models (VLMs) on low Earth orbit (LEO) edge nodes requires energy-efficient inference under strict latency deadlines. This research studies a source-satellite workflow where raw visual context remains onboard and only prompts and results traverse access and inter-satellite links. The proposed joint container allocation and frequency scaling (JCAFS) scheme uses a structure-exploiting optimizer to jointly select active VLM containers and GPU frequencies. A workload-aware M/G/k model captures prompt, task, and output-token variability from Moondream-DOTA profiling. Simulations show zero deadline violations, 5.86–7.59% savings over greedy joint scaling, 25.61–34.96% over P-A DVFS, and about 8× shorter decision time than dense-grid search.
