arXiv subjects

Sangwook Park

Publications and source records attributed to Sangwook Park.

At least 19 recordsLinked to original sources

arXiv2026

Ultra-Lightweight Network for Ship-Radiated Sound Classification on Embedded Deployment

This letter presents ShuffleFAC, a lightweight acoustic model for ship-radiated sound classification in resource-constrained maritime monitoring systems. ShuffleFAC integrates Frequency-Aware convolution into an efficiency-oriented backbone using separable convolution, point-wise group convolution, and channel shuffle, enabling frequency-sensitive feature extraction with low computational cost. Experiments on the DeepShip dataset show that ShuffleFAC achieves competitive performance with substantially reduced complexity. In particular, ShuffleFAC ($γ=16$) attains a macro F1-score of 71.45 $\pm$ 1.18% using 39K parameters and 3.06M MACs, and achieves an inference latency of 6.05 $\pm$ 0.95ms on a Raspberry Pi. Compared with MicroNet0, it improves macro F1-score by 1.82 % while reducing model size by 9.7x and latency by 2.5x. These results indicate that ShuffleFAC is suitable for real-time embedded UATR.

cs.SD↗

arXiv2026

Hidden costs for inference with deep network on embedded system devices

This study evaluates the inference performance of various deep learning models under an embedded system environment. In previous works, Multiply-Accumulate operation is typically used to measure computational load of a deep model. According to this study, however, this metric has a limitation to estimate inference time on embedded devices. This paper poses the question of what aspects are overlooked when expressed in terms of Multiply-Accumulate operations. In experiments, an image classification task is performed on an embedded system device using the CIFAR-100 dataset to compare and analyze the inference times of ten deep models with the theoretically calculated Multiply-Accumulate operations for each model. The results highlight the importance of considering additional computations between tensors when optimizing deep learning models for real-time performing in embedded systems.

cs.CC↗

arXiv2024

Latest Evolution of the X-Ray Remnant of SN 1987A: Beyond the Inner Ring

Based on our Chandra imaging-spectroscopic observations, we present the latest evolution of the X-ray remnant of SN 1987A. Recent changes in the electron temperatures and volume emission measures suggest that the blast wave in SN 1987A is moving out of the dense inner ring structure, also called the equatorial ring (ER). The 0.5-2.0 keV X-ray light curve shows a linearly declining trend (by $\sim$4.5 % yr$^{-1}$) between 2016 and 2020, as the blast wave heats the hitherto unknown circumstellar medium (CSM) outside the ER. While the peak X-ray emission in the latest 0.3-8.0 keV image is still within the ER, the radial expansion rate in the 3.0-8.0 keV images suggests an increasing contribution of the X-ray emission from less dense CSM since 2012, at least partly from beyond the ER. It is remarkable that, since 2020, the declining soft X-ray flux has stabilized around $\sim$7 $\times$ 10$^{-12}$ erg s$^{-1}$ cm$^{-2}$, which may signal a contribution from the reverse-shocked outer layers of ejecta as predicted by the 3-D magneto-hydrodynamic (MHD) models. In the latest ACIS spectrum of supernova remnant (SNR) 1987A in 2022 we report a significant detection of the Fe K line at $\sim$6.7 keV, which may be due to changing thermal conditions of the X-ray emitting CSM and/or the onset of reverse shock interactions with the Fe-ejecta.

astro-ph.HE↗

arXiv2023

JWST NIRCam Observations of SN 1987A: Spitzer Comparison and Spectral Decomposition

JWST NIRCam observations at 1.5-4.5 $μ$m have provided broad and narrow band imaging of the evolving remnant of SN 1987A with unparalleled sensitivity and spatial resolution. Comparing with previous marginally spatially resolved Spitzer IRAC observations from 2004-2019 confirms that the emission arises from the circumstellar equatorial ring (ER), and the current brightness at 3.6 and 4.5 $μ$m was accurately predicted by extrapolation of the declining brightness tracked by IRAC. Despite the regular light curve, the NIRCam observations clearly reveal that much of this emission is from a newly developing outer portion of the ER. Spots in the outer ER tend to lie at position angles in between the well-known ER hotspots. We show that the bulk of the emission in the field can be represented by 5 standard spectral energy distributions (SEDs), each with a distinct origin and spatial distribution. This spectral decomposition provides a powerful technique for distinguishing overlapping emission from the circumstellar medium (CSM) and the supernova (SN) ejecta, excited by the forward and reverse shocks respectively.

astro-ph.SR↗

arXiv2023

Near-Infrared and Optical Observations of Type Ic SN 2021krf: Luminous Late-time Emission and Dust Formation

We present near-infrared (NIR) and optical observations of the Type Ic supernova (SN Ic) SN 2021krf obtained between days 13 and 259 at several ground-based telescopes. The NIR spectrum at day 68 exhibits a rising $K$-band continuum flux density longward of $\sim$ 2.0 $μ$m, and a late-time optical spectrum at day 259 shows strong [O I] 6300 and 6364 Å emission-line asymmetry, both indicating the presence of dust, likely formed in the SN ejecta. We estimate a carbon-grain dust mass of $\sim$ 2 $\times$ 10$^{-5}$ M$_{\odot}$ and a dust temperature of $\sim$ 900 - 1200 K associated with this rising continuum and suggest the dust has formed in SN ejecta. Utilizing the one-dimensional multigroup radiation hydrodynamics code STELLA, we present two degenerate progenitor solutions for SN 2021krf, characterized by C-O star masses of 3.93 and 5.74 M$_{\odot}$, but with the same best-fit $^{56}$Ni mass of 0.11 M$_{\odot}$ for early times (0-70 days). At late times (70-300 days), optical light curves of SN 2021krf decline substantially more slowly than that expected from $^{56}$Co radioactive decay. Lack of H and He lines in the late-time SN spectrum suggests the absence of significant interaction of the ejecta with the circumstellar medium. We reproduce the entire bolometric light curve with a combination of radioactive decay and an additional powering source in the form of a central engine of a millisecond pulsar with a magnetic field smaller than that of a typical magnetar.

astro-ph.HE↗

arXiv2022

The 3-D X-ray Ejecta Structure of Tycho's Supernova Remnant

We present our velocity measurements of 59 clumpy, metal-rich ejecta knots in the supernova remnant (SNR) of SN 1572 (Tycho). We use our 450 ks Chandra High Energy Transmission Grating Spectrometer observation to measure the Doppler shift of the He-like Si K$α$ line-center wavelength emitted from these knots to find their line-of-sight (radial) velocities ($v_r$). We find $v_r$ up to $\sim$ 5500 km s$^{-1}$, with roughly consistent speeds between blueshifted and redshifted ejecta knots. We also measure the proper motions (PMs) for our sample based on archival Chandra Advanced CCD Imaging Spectrometer data taken from 2003, 2009, and 2015. We estimate PMs up to 0$"$.35 yr$^{-1}$, which corresponds to a transverse velocity of about 5800 km s$^{-1}$ for the distance of 3.5 kpc to Tycho. Our $v_r$ and transverse velocity measurements imply space velocities of $\sim$ 1900 - 6000 km s$^{-1}$ for the ejecta knots in Tycho. We estimate a new expansion center of R.A.(J2000) = 00$^h$25$^m$18$^s$.725 $\pm$ 1$^s$.157 and decl.(J2000) = +64$^{\circ}$08$'$02$"$.5 $\pm$ 11$"$.2 from our PM measurements, consistent to within $\sim$ 13$"$ of the geometric center. The distribution of space velocities throughout the remnant suggests that the southeast quadrant generally expands faster than the rest of the SNR. We find that blueshifted knots are projected more in the northern shell, while redshifted knots are more in the southern shell. The previously estimated reverse shock position is consistent with most of our estimated ejecta distribution, however some ejecta show deviations from the 1-D picture of the reverse shock.

astro-ph.HE↗

arXiv2021

Spectral Evolution of the X-Ray Remnant of SN 1987A: A High-Resolution $Chandra$ HETG Study

Based on observations with the $Chandra$ X-ray Observatory, we present the latest spectral evolution of the X-ray remnant of SN 1987A (SNR 1987A). We present a high-resolution spectroscopic analysis using our new deep ($\sim$312 ks) $Chandra$ HETG observation taken in March 2018, as well as archival $Chandra$ gratings spectroscopic data taken in 2004, 2007, and 2011 with similarly deep exposures ($\sim$170 - 350 ks). We perform detailed spectral model fits to quantify changing plasma conditions over the last 14 years. Recent changes in electron temperatures and volume emission measures suggest that the shocks moving through the inner ring have started interacting with less dense circumstellar material, probably beyond the inner ring. We find significant changes in the X-ray line flux ratios (among H- and He-like Si and Mg ions) in 2018, consistent with changes in the thermal conditions of the X-ray emitting plasma that we infer based on the broadband spectral analysis. Post-shock electron temperatures suggested by line flux ratios are in the range $\sim$0.8 - 2.5 keV as of 2018. We do not yet observe any evidence of substantial abundance enhancement, suggesting that the X-ray emission component from the reverse-shocked metal-rich ejecta is not yet significant in the observed X-ray spectrum.

astro-ph.HE↗

arXiv2021

A Spectroscopic Study of Supernova Remnants with the Infrared Space Observatory

We present far-infrared (FIR) spectroscopy of supernova remnants (SNRs) based on the archival data of the Infrared Space Observatory ($ISO$) taken with the Long Wavelength Spectrometer (LWS). Our sample includes previously unpublished profiles of line and continuum spectra for 20 SNRs in the Galaxy and Magellanic Clouds. In several SNRs including G21.5-0.9, G29.7-0.3, the Crab Nebula, and G320.4-1.2, we find evidence for broad [O I], [O III], [N II], and [C II] lines with velocity dispersions up to a few 10$^3$ km s$^{-1}$, indicating that they are associated with high-velocity SN ejecta. Our detection of Doppler-broadened atomic emission lines and a bright FIR continuum hints at the presence of newly formed dust in SN ejecta. For G320.4-1.2, we present the first estimate of an ejecta-dust mass of 0.1 - 0.2 M$_\odot$, which spatially coincides with the broad line emission, by applying a blackbody model fit with components of the SNR and background emission. Our sample includes raster maps of 63, 145 $μ$m [O I] and 158 $μ$m [C II] lines toward SNRs Kes 79, CTB 109, and IC 443. Based on these line intensities, we suggest interacting shock types in these SNRs. Finally, we compare our LWS spectra of our sample SNRs with the spectra of several HII regions, and discuss their FIR line intensity ratios and continuum properties. Follow-up observations with modern instruments (e.g. $JWST$ and $SOFIA$) with higher spatial and spectral resolution are encouraged for an extensive study of the SN ejecta and the SN dust.

astro-ph.GA↗

arXiv2021

Cross-Referencing Self-Training Network for Sound Event Detection in Audio Mixtures

Sound event detection is an important facet of audio tagging that aims to identify sounds of interest and define both the sound category and time boundaries for each sound event in a continuous recording. With advances in deep neural networks, there has been tremendous improvement in the performance of sound event detection systems, although at the expense of costly data collection and labeling efforts. In fact, current state-of-the-art methods employ supervised training methods that leverage large amounts of data samples and corresponding labels in order to facilitate identification of sound category and time stamps of events. As an alternative, the current study proposes a semi-supervised method for generating pseudo-labels from unsupervised data using a student-teacher scheme that balances self-training and cross-training. Additionally, this paper explores post-processing which extracts sound intervals from network prediction, for further improvement in sound event detection performance. The proposed approach is evaluated on sound event detection task for the DCASE2020 challenge. The results of these methods on both "validation" and "public evaluation" sets of DESED database show significant improvement compared to the state-of-the art systems in semi-supervised learning.

cs.SD↗

arXiv2021

High-cadence Dispersed Spectral Analysis of Supernova Remnant 1987A

We present an analysis of the dispersed spectral data from 11 epochs (March 2011 to September 2018) of supernova remnant (SNR) 1987A observations performed with Chandra. These observations were performed with the High Energy Transmission Grating (HETG) as part of our ongoing Chandra monitoring campaign of SNR 1987A, whose 1st-order dispersed spectrum provides a significantly greater energy resolution than the previously-published 0th-order spectrum. Our data sets with moderate exposure times of ~50-70ks per epoch cover the time period between deep Chandra HETG observations (with individual exposures >200ks) taken in March 2011 and March 2018. These data have a much higher cadence than the widely-spaced deep high-resolution spectra, at the expense of total exposure time. While statistical uncertainties are large due to low photon count statistics in the observed 1st-order spectra, we find that spectral model parameters are generally in line with the shock wave propagating into the medium beyond the dense inner ring, as suggested by Frank et al. 2016. As the reverse shock begins ionizing the heavier elements of the supernova ejecta interior to the equatorial ring, spectral fit parameters are expected to change as the chemical makeup and physical properties of the shocked gas evolve. Based on our broadband spectral model fits, we find that abundance values appear to be constant in this time period. While our results are somewhat limited due to photon statistics, we demonstrate the utility of the dispersed HETG spectral analysis that can be performed with our regular Chandra monitoring observations of SNR 1987A.

astro-ph.HE↗

arXiv2020

Radiative Supernova Remnants and Supernova Feedback

Supernova (SN) explosions are a major feedback mechanism regulating star formation in galaxies through their momentum input. We review the observations of SNRs in radiative stages in the Milky Way to validate the theoretical results on the momentum/energy injection from a single SN explosion. For seven SNRs where we can observe fast-expanding, atomic radiative shells, we show that the shell momentum inferred from HI 21 cm line observations is in the range of (0.5--4.5)$\times 10^5$ $M_\odot$ km s$^{-1}$. In two SNRs (W44 and IC 443), shocked molecular gas with momentum comparable to that of the atomic SNR shells has been also observed. We compare the momentum and kinetic/thermal energy of these seven SNRs with the results from 1D and 3D numerical simulations. The observation-based momentum and kinetic energy agree well with the expected momentum/energy input from an SN explosion of $\sim 10^{51}$ erg. It is much more difficult to use data/model comparisons of thermal energy to constrain the initial explosion energy, however, due to rapid cooling and complex physics at the hot/cool interface in radiative SNRs. We discuss the observational and theoretical uncertainties of these global parameters and explosion energy estimates for SNRs in complex environments.

astro-ph.GA↗

arXiv2020

Element Stratification in the Middle-Aged Type Ia Supernova Remnant G344.7-0.1

Despite their importance, a detailed understanding of Type Ia supernovae (SNe Ia) remains elusive. X-ray measurements of the element distributions in supernova remnants (SNRs) offer important clues for understanding the explosion and nucleosynthesis mechanisms for SNe Ia. However, it is challenging to observe the entire ejecta mass in X-rays for young SNRs, because the central ejecta may not have been heated by the reverse shock yet. Here we present over 200 kilosecond Chandra observations of the Type Ia SNR G344.7-0.1, whose age is old enough for the reverse shock to have reached the SNR center, providing an opportunity to investigate the distribution of the entire ejecta mass. We reveal a clear stratification of heavy elements with a centrally peaked distribution of the Fe ejecta surrounded by intermediate-mass elements (IMEs: Si, S, Ar Ca) with an arc-like structure. The centroid energy of the Fe K emission is marginally lower in the central Fe-rich region than in the outer IME-rich regions, suggesting that the Fe ejecta were shock-heated more recently. These results are consistent with the prediction for standard SN Ia models, where the heavier elements are synthesized in the interior of an exploding white dwarf. We find, however, that the peak location of the Fe K emission is slightly offset to the west with respect to the geometric center of the SNR. This apparent asymmetry is likely due to the inhomogeneous density distribution of the ambient medium, consistent with our radio observations of the ambient molecular and neutral gas.

astro-ph.HE↗

arXiv2020

NuSTAR Detection of Quiescent Hard X-Ray Emission from SGR 0526-66 in the Large Magellanic Cloud

The soft gamma-ray repeater (SGR) 0526-66 is the first-identified magnetar, and is projected within the supernova remnant N49 in the Large Magellanic Cloud. Based on our ~50 ks NuSTAR observation, we detect the quiescent-state 0526-66 for the first time in the 10-40 keV band. Based on the joint analysis of our NuSTAR and the archival Chandra ACIS data, we firmly establish the presence of the nonthermal component in the X-ray spectrum of 0526-66 in addition to the thermal emission. In the best-fit blackbody (BB) plus power law (PL) model, the slope of the PL component (photon index Gamma = 2.1) is steeper than those (Gamma > ~1.5) for other magnetars. The soft part of the X-ray spectrum can be described with a BB component with the temperature of kT = 0.43 keV. The best-fit radius (R = 6.5 km) of the X-ray-emitting area is smaller than the canonical size of a neutron star. If we assume an underlying cool BB component with the canonical radius of R = 10 km for the neutron star in addition to the hot BB component (2BB + PL model), a lower BB temperature of kT = 0.24 keV is obtained for the passively cooling neutron star'ssurface, while the hot spot emission with kT = 0.46 keV dominates the thermal spectrum (~85% of the thermal luminosity in the 0.5-5 keV band). The nonthermal component (Gamma ~ 1.8) is still required.

astro-ph.HE↗

arXiv2020

An Ejecta Kinematics Study of Kepler's Supernova Remnant with High-Resolution $Chandra$ HETG Spectroscopy

We report our measurements of the bulk radial velocity from a sample of small, metal-rich ejecta knots in Kepler's Supernova Remnant (SNR). We measure the Doppler shift of the He-like Si K$α$ line center energy in the spectra of these knots based on our $Chandra$ High-Energy Transmission Grating Spectrometer (HETGS) observation to estimate their radial velocities. We estimate high radial velocities of up to $\sim$ 8,000 km s$^{-1}$ for some of these ejecta knots. We also measure proper motions for our sample based on the archival $Chandra$ Advanced CCD Imaging Spectrometer (ACIS) data taken in 2000, 2006, and 2014. Our measured radial velocities and proper motions indicate that some of these ejecta knots are almost freely-expanding after $\sim$ 400 years since the explosion. The fastest moving knots show proper motions up to $\sim$ 0.2 arcseconds per year. Assuming that these high velocity ejecta knots are traveling ahead of the forward shock of the SNR, we estimate the distance to Kepler's SNR $d$ $\sim$ 4.4 to 7.5 kpc. We find that the ejecta knots in our sample have an average space velocity of $ v_{s} \sim$ 4,600 km s$^{-1}$ (at a distance of 6 kpc). We note that 8 out of the 15 ejecta knots from our sample show a statistically significant (at the 90$\%$ confidence level) redshifted spectrum, compared to only two with a blueshifted spectrum. This may suggest an asymmetry in the ejecta distribution in Kepler's SNR along the line of sight, however a larger sample size is required to confirm this result.

astro-ph.HE↗

arXiv2019

Final Spitzer IRAC Observations of the Rise and Fall of SN 1987A

Spitzer's final Infrared Array Camera (IRAC) observations of SN 1987A show the 3.6 and 4.5 $μ$m emission from the equatorial ring (ER) continues a period of steady decline. Deconvolution of the images reveals that the emission is dominated by the ring, not the ejecta, and is brightest on the west side. Decomposition of the marginally resolved emission also confirms this, and shows that the west side of the ER has been brightening relative to the other portions of the ER. The infrared (IR) morphological changes resemble those seen in both the soft X-ray emission and the optical emission. The integrated ER light curves at 3.6 and 4.5 $μ$m are more similar to the optical light curves than the soft X-ray light curve, though differences would be expected if dust is responsible for this emission and its destruction is rapid. Future observations with the James Webb Space Telescope will continue to monitor the ER evolution, and will reveal the true spectrum and nature of the material responsible for the broadband emission at 3.6 and 4.5 $μ$m. The present observations also serendipitously reveal a nearby variable source, subsequently identified as a Be star, that has gone through a multi-year outburst during the course of these observations.

astro-ph.SR↗

arXiv2019

High angular resolution ALMA images of dust and molecules in the SN 1987A ejecta

We present high angular resolution (~80 mas) ALMA continuum images of the SN 1987A system, together with CO $J$=2 $\!\rightarrow\!$ 1, $J$=6 $\!\rightarrow\!$ 5, and SiO $J$=5 $\!\rightarrow\!$ 4 to $J$=7 $\!\rightarrow\!$ 6 images, which clearly resolve the ejecta (dust continuum and molecules) and ring (synchrotron continuum) components. Dust in the ejecta is asymmetric and clumpy, and overall the dust fills the spatial void seen in H$α$ images, filling that region with material from heavier elements. The dust clumps generally fill the space where CO $J$=6 $\!\rightarrow\!$ 5 is fainter, tentatively indicating that these dust clumps and CO are locationally and chemically linked. In these regions, carbonaceous dust grains might have formed after dissociation of CO. The dust grains would have cooled by radiation, and subsequent collisions of grains with gas would also cool the gas, suppressing the CO $J$=6 $\!\rightarrow\!$ 5 intensity. The data show a dust peak spatially coincident with the molecular hole seen in previous ALMA CO $J$=2 $\!\rightarrow\!$ 1 and SiO $J$=5 $\!\rightarrow\!$ 4 images. That dust peak, combined with CO and SiO line spectra, suggests that the dust and gas could be at higher temperatures than the surrounding material, though higher density cannot be totally excluded. One of the possibilities is that a compact source provides additional heat at that location. Fits to the far-infrared--millimeter spectral energy distribution give ejecta dust temperatures of 18--23K. We revise the ejecta dust mass to $\mathrm{M_{dust}} = 0.2-0.4$M$_\odot$ for carbon or silicate grains, or a maximum of $<0.7$M$_\odot$ for a mixture of grain species, using the predicted nucleosynthesis yields as an upper limit.

astro-ph.HE↗

arXiv2019

Correlation Distance Skip Connection Denoising Autoencoder (CDSK-DAE) for Speech Feature Enhancement

Performance of learning based Automatic Speech Recognition (ASR) is susceptible to noise, especially when it is introduced in the testing data while not presented in the training data. This work focuses on a feature enhancement for noise robust end-to-end ASR system by introducing a novel variant of denoising autoencoder (DAE). The proposed method uses skip connections in both encoder and decoder sides by passing speech information of the target frame from input to the model. It also uses a new objective function in training model that uses a correlation distance measure in penalty terms by measuring dependency of the latent target features and the model (latent features and enhanced features obtained from the DAE). Performance of the proposed method was compared against a conventional model and a state of the art model under both seen and unseen noisy environments of 7 different types of background noise with different SNR levels (0, 5, 10 and 20 dB). The proposed method also is tested using linear and non-linear penalty terms as well, where, they both show an improvement on the overall average WER under noisy conditions both seen and unseen in comparison to the state-of-the-art model.

eess.AS↗

arXiv2019

Supernova Remnants in High Definition

Supernova remnants (SNRs) offer the means to study SN explosions, dynamics, and shocks at sub-parsec scales. X-ray observations probe the hot metals synthesized in the explosion and the TeV electrons accelerated by the shocks, and thus they are key to test recent, high-fidelity three-dimensional SN simulations. In this white paper, we discuss the major advances possible with X-ray spectro-imaging at arcsecond scales, with a few eV spectral resolution and a large effective area. These capabilities would revolutionize SN science, offering a three-dimensional view of metals synthesized in explosions and enabling population studies of SNRs in Local Group galaxies. Moreover, this future X-ray mission could detect faint, narrow synchrotron filaments and shock precursors that will constrain the diffusive shock acceleration process.

astro-ph.HE↗