A reproducible, quasi-stationary edge localized mode (ELM)-suppressed scenario was obtained over a wide range of plasma parameters by continuous injection of boron (B) powder into an upper-single null discharge in the experimental advanced superconducting tokamak [Sun et al., Nucl. Fusion 61, 014002 (2021)]. This powder-induced ELM-absent regime is associated with an edge harmonic mode (EHM) that provides continuous particle exhaust to maintain constant density without confinement degradation and impurity accumulation, the latter of which is often observed in ELM-free regimes. A flow rate threshold of B powder injection, leading to a threshold intensity of the EHM, is necessary for full ELM suppression. The fundamental harmonic of the EHM exhibits a toroidal mode number n = 1. The mode is observable in the entire poloidal cross section with a peak near the upper X-point in an upper-single null configuration. The EHM spans radially across the pedestal and scrape-off layer, peaking inside the separatrix. The EHM appears to be insensitive to q95, heating power, plasma toroidal rotation, and pedestal collisionality.
multi-pass enhanced photoacoustic spectroscopy (PAS) sensors achieve high sensitivity but suffer from detrimental laser interference fringe noise (Etalon effect), which limits the measurement precision and the detection limit. Traditional suppression methods introduce trade-offs in complexity, cost, and effectiveness, especially when laser interference fringe characteristics overlap gas absorption features. This study presents a novel PAS sensor integrating a multi-pass cell with a deep learning Residual Network (ResNet) model. The ResNet is specifically designed to suppress complex, dynamic interference fringes without distorting the underlying photoacoustic signal. Experimental results demonstrate significant noise reduction (up to approximately 81.5-fold lower standard deviation), improved signal-to-noise ratio (SNR enhancement factors of 11.6 to 63.1), and dramatically enhanced long-term stability and precision compared to raw data and Kalman filtering. Most notably, Allan deviation analysis shows an approximately 1700-fold improvement at 0.59 s and an approximately 4667-fold improvement in the minimum detection limit (from 1.4 ppmv to 0.3 ppbv), breaking the sensitivity barrier imposed by laser interference fringe noise.