A global view on star formation: The GLOSTAR Galactic plane survey. VI. Radio Source Catalog II: $28^\circ<\ell<36^\circ$ and $|b|<1^\circ$, VLA B-configuration

As part of the GLOSTAR survey we have used the VLA in its B-configuration to observe the part of the Galactic plane between longitudes of 28d and 36d and latitudes from -1d to +1d at the C-band (4--8 GHz). To reduce the contamination of extended sources that are not well recovered by our coverage of the (u, v)-plane we discarded short baselines that are sensitive to emission on angular scales $<4"$. The resulting radio continuum images have an angular resolution of 1.0", and sensitivity of $\sim60 \mu$Jy~beam$^{-1}$; making it the most sensitive radio survey covering a large area of the Galactic plane with this angular resolution. An automatic source extraction algorithm was used in combination with visual inspection to identify a total of 3325 radio sources. A total of 1457 radio sources are $\geq7\sigma$ and comprise our highly reliable catalog; 72 of these are grouped as 22 fragmented sources, e.g., multiple components of an extended and resolved source. To explore the nature of the catalogued radio sources we searched for counterparts at millimeter and infrared wavelengths. Our classification attempts resulted in 93 HII region candidates, 104 radio stars, 64 planetary nebulae, while most of the remaining radio sources are suggested to be extragalactic sources. We investigated the spectral indices ($\alpha$, $S_\nu\propto\nu^\alpha$) of radio sources classified as HII region candidates and found that many have negative values. This may imply that these radio sources represent young stellar objects that are members of the star clusters around the high mass stars that excite the HII regions, but not these HII regions themselves. By comparing the peak flux densities from the GLOSTAR and CORNISH surveys we have identified 49 variable radio sources, most of them with an unknown nature. Additionally, we provide the list of 1866 radio sources detected within 5 to 7$\sigma$ levels.


Introduction
The Global view on star formation (GLOSTAR) survey is presently the most sensitive radio survey (∼ 60 µJy beam −1 ) of the northern hemisphere of the Galactic plane at the C-band (4 to 8 GHz) Medina et al. 2019). Taking full use of the capabilities of the Karl G. Jansky Very Large Array (VLA), a distinction of GLOSTAR compared to previous surveys is that it simultaneously observes radio continuum and spectral line emission. GLOSTAR is indeed complemen-tary to the wealth of Galactic plane surveys at infrared and submillimeter wavelengths that address star formation in the Galaxy, some of which are described in subsection 3.7.
The primary goal of the GLOSTAR survey is to localize signposts of massive star formation (MSF) activity (see Brunthaler et al. 2021, for a detailed overview of the survey). Towards this goal, in the continuum mode, the survey mainly observes compact, ultra-and hyper-compact HII regions, that trace different early phases of MSF activity (e.g., Medina et al. 2019;Nguyen et al. 2021). Thus, the GLOSTAR survey complements previous radio surveys by providing a powerful and comprehensive radio-wavelength survey of the ionized gas in the Galactic Plane with an unprecedented sensitivity. However, the survey area is also populated with radio sources related to post-main sequence stars (e.g. Wolf-Rayet stars, pulsars, etc.), planetary nebulae, supernova remnants and extragalactic radio sources (Medina et al. 2019;Chakraborty et al. 2020;Dokara et al. 2021). In spectral line mode, it traces the radio recombination lines from regions ionized by massive stars and the methanol maser line at 6.7 GHz (Ortiz-León et al. 2021;Nguyen et al. 2022), both of which are related to massive star formation. The formaldehyde absorption line at 4.8 GHz is also observed. It traces neutral molecular gas and its radial velocity information with respect to the local standard of rest (LSR) radial can help to solve distance ambiguities. The final GLOSTAR images will cover the Galactic plane between Galactic longitudes, , of −2 • and 60 • and latitudes, b, from −1 • to +1 • and the Cygnus X region. The final dataset will consist of low resolution images (∼ 20 ), using the VLA in the D configuration, and high resolution images (∼ 1. 0), using the B configuration. These VLA data sets could also be combined for optimal sensitivity of the intermediate spatial ranges, such images will be presented in future works. The VLA observations will be complemented with very low resolution (∼ 150 ) images of single dish observations from the Effelsberg radio telescope, to recover the most extended emission and solve the "missing short spacing" issue affecting interferometer-only images. The overview of the full GLOSTAR capabilities are described in detail by Brunthaler et al. (2021).
Previously, we have reported the radio source catalog of the low resolution VLA images covering the area 28 • < < 36 • and −1 • < b < +1 • (Medina et al. 2019). Complementary infraredand sub-millimeter-wavelength data were examined towards the radio source positions with the goal to elucidate the nature of the radio sources. In this paper we report the compact (∼ 1. 0) radio sources from the same region using the high resolution observations obtained with the VLA in B-configuration. The interesting part of this new catalog is that we can identify the most compact, and probably youngest, (hypercompact) HII regions.

Observations
The Karl G. Jansky Very Large Array (VLA) of the National Radio Astronomy Observatory 1 was used in its B-configuration to observe the (4-8 GHz) continuum emission. We follow the same instrumental setups and calibration as for the D-configuration data. We refer the reader to the overview paper ) for a detailed description of the data management that we summarize in the following subsections.

Observation Strategy
The correlator setup consisted of two 1 GHz wide sub-bands, centered at 4.7 and 6.9 GHz. Each sub-band was divided into eight spectral windows of 128 MHz, and each spectral window comprising of 64 channels with a channel width of 2 MHz 2 . The chosen setup avoids strong persistent radio frequency interference (RFI) seen (most prominently) at 6.3 and 4.1 GHz and allows estimation of source spectral indices.
For each epoch the total observing time was five hours. At the beginning of the observations, the amplitude/bandpass calibrator 3C 286 is observed for ten minutes. Then, the phase calibrator, J1804+0101, is observed for one minute, followed by pointings on target fields for eight minutes, after which the phase calibrator is observed for another minute. The observation cycle, consisting of phase calibrator-targets scans, is repeated over the full five hours. During this time, an area of 2 • × 1 • is covered with phase centers for 676 target fields in a semi-mosaic mode (see Brunthaler et al. 2021, for a detailed overview of the observation strategy). Each pointing was observed twice for 11 seconds which, after considering the slewing time, yields a total integration time of 15 seconds per field. The theoretical noise level in brightness (or peak flux density) from these observations is 90 µJy beam −1 per field and per sub-band. The noise is improved after combining the fields and both sub-bands by a factor of ∼ 2. We observed a total of eight epochs, or a total of 1 The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. 2 Higher frequency resolution correlator windows were used to cover prominent methanol maser emission line at 6.7 GHz, formaldehyde absorption at 4.8 GHz, and seven radio recombination lines. The full results of these line observations will be reported in forthcoming papers.

Calibration, Data Reduction and Imaging
The data were calibrated, edited and imaged using the Obit software environment (Cotton 2008), which inter-operates with the classic Astronomical Image Processing Software package (AIPS) (Greisen 2003). We have written calibration scripts that handle the GLOSTAR data edition and calibration . The calibration follows the standard procedures of editing and calibrating interferometric data. These include bandpass, amplitude and phase calibration. The calibrated data were imaged using the Obit task MFImage. The CLEANing process from MFImage divides the observed band into nine frequency bins, that are narrow enough to perform a spectral deconvolution, addressing the effects of variable spectral index and antenna pattern variations. In the end, we obtained an image representing the data for the full frequency band and images of each individual frequency bin. The images were obtained from the data that had projected baselines with distances in the (u, v)-plane larger than 50 kλ; i.e., discarding all radio emission on angular scales larger than ≈ 4. 13 at all observed frequencies. This choice rejects emission from poorly mapped extended structures that introduced artifacts in the images, with a minor impact in the overall sensitivity. The images were convolved with a circular beam 1. 0 of size, and with pixel size of 0. 25. The mosaics are constructed to obtain images of 35000 × 30000 pixels, containing the 1 • × 2 • angular area surveyed by epoch, following the schemes described by Brunthaler et al. (2021). The mean measured noise in the resulting images is 60 µJy beam −1 ; though it can be significantly higher in some areas in which extended emission was not properly recovered or around very bright radio sources that produce imperfectly cleaned sidelobe emission.

Catalog construction
In this section we discuss the procedure used for constructing the catalogues presented in this work. First, we have extracted the sources from the images, selected the sources that are real, identified candidate radio sources, and discarded image artifacts. We have investigated the astrometry, flux density (S ν ), spectral indices (α; S ν ∝ ν α ), and have searched for counterparts at other wavelengths. Based on the counterpart information, we attempted a classification of the radio sources. The final catalog is presented in Table 1, with the reliable radio sources, their properties and the source classification.

Source extraction
The source extraction was performed following the procedures described by Medina et al. (2019), and we refer the reader to that work for the details, while here we will give a brief summary. Using the SExtractor (Bertin & Arnouts 1996) tool from the Graphical Astronomy and Image Analysis Tool package (GAIA 3 ), we first create a noise image. Then, we have used the BLOBCAT package (Hales et al. 2012) to extract the sources from each of the final images. Both the intensity map and the noise map are used as inputs by BLOBCAT. This package recognizes islands of pixels (blobs) representing sources with a minimum peak flux above N times the noise level in the area. Purcell et al. (2013) noted that with N < 4.5, large radio images will be dominated by spurious sources. Thus, to diminish the number of spurious detections in our catalog, for our extraction we have initially defined N = 5 and require blobs to consist of a minimum of 12 contiguous pixels. The minimum number of pixels was chosen to be the number of expected pixels in 50% of the beam area. The final number of extracted blobs is 3880 . After this first extraction we performed a visual inspection of all the blobs to identify clear image artifacts, such as sidelobes from very bright radio sources, and discarded 555 blobs. The spatial distribution of all the blobs, excluding the artifacts, is shown in Figure 1. We detected a total of 1457blobs with a signal-to-noise ratio (S/N) ≥ 7.0 and 1866 blobs with 5.0 ≤ S/N < 7.0.
It has been determined that in large radio surveys such as GLOSTAR the most reliable sources are those with peak flux values above 7σ noise as no spurious sources are expected at these levels (e.g., Purcell et al. 2013;Bihr et al. 2016;Wang et al. 2018;Medina et al. 2019). However, as some blobs with a brightness between 5-7σ noise could represent real sources, we have searched for counterparts inside a radius of 2 in the SIMBAD astronomical database 4 for all sources (see details below). This comprehensive database is the best option to look for counterparts at any wavelength in large areas of the sky but, admittedly, it can miss recent catalogs. We have also considered weak blobs (peak flux values < 7σ noise ) as real radio sources whose positions are consistent with the radio sources from our D-configuration catalog by Medina et al. (2019). The full number of weak blobs with a known counterpart in the SIMBAD database and our Dconfiguration catalog is 142. The remaining 1724 radio blobs having a S/N between 5 and 7 and have no known counterpart are referred to as candidate radio source detections. We give the list of these 1866 sources with S/N< 7 in Appendix A (Table  A.1) labeling those with SIMBAD or D-configuration counterparts, and we do not analyze them further.
We consider the remaining 1457 blobs as highly reliable radio sources. They are represented by the blue circles in Figure 1 and are listed in Table 1. The position, the SNR, the peak and the integrated flux density values (columns (2) to (8) in Table 1) are taken from the values determined by the BLOBCAT software (Hales et al. 2012). From now on, this paper will focus on the analysis of these sources. Considering the ratio between the integrated flux density (in units of Jy) and the peak flux density (in units of Jy beam −1 (here named as the Y-factor; Y = S ν,Int /S ν,Peak ) we can divide these sources into extended (Y > 2.0), compact (1.1 < Y ≤ 2.0) and point-like (Y ≤1.1) sources. The Y factor of each source is listed in column (9) of Table 1. Using this classification we obtain 100 extended, 455 compact, and 904 point-like sources. As expected for these images, the sources are dominated by compact and point-like sources as we have rejected extended structures in our imaging process.

Astrometry
For our previous catalog based on the VLA D-configuration observations we have estimated that the accuracy of the astrometry is of the order of 1 . The position errors of the extracted sources with BLOBCAT in the VLA B-configuration images have a mean value of 0. 06 in both Galactic longitude and latitude. The 0. 06 is a formal statistical error estimate from BLOBCAT and does not  EgC Notes: Only a small portion of the data is provided here, the full table is available in electronic form at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.125.5) or via http://cdsweb.u-strasbg.fr/cgi-bin/qcat?J/A&A/. Classification (17) see also section 3.8: EgC = Extragalctic radio source candidate, HII = HII region candidate, Radio-star, PN = planetary nebula, WR = Wolf-Rayet star, PDR = Photodissociation region, Cataclysmic variable, Pulsar, and Unclear = source with no clear classification.
reflect position errors resulting from imperfect phase calibration. As most of the observed radio sources are expected to be background extragalactic objects, the proper motion of these sources is expected to be zero. The only other recent Galactic plane survey that observed the same region at a similar radio frequency is the Co-Ordinated Radio 'N' Infrared Survey for High-mass star formation (CORNISH; Hoare et al. 2012;Purcell et al. 2013). The observations of CORNISH were obtained at 5 GHz using the VLA in its B-configuration. Because of the similarity of the frequency and the use of the same VLA array configuration the angular resolution of CORNISH is 1. 5, similar as that of the GLOSTAR survey B array data discussed here (1. 0).
A total of 257 compact and point-like sources are listed in both the GLOSTAR-B and CORNISH catalogs (Purcell et al. 2013) with a maximum angular separation of 1. 5, the COR-NISH angular resolution. In the upper panel of Figure 2 the measured offsets of these sources are plotted. We obtain mean and standard deviation for the position offsets of −0. 04 ± 0. 01 and 0. 17 in Galactic longitude direction, and 0. 03 ± 0. 01 and 0. 15 in Galactic latitude direction. The mean offsets in both directions are smaller than the mean position error of GLOSTAR radio sources, suggesting a good astrometry.
The most accurate positions of radio sources are obtained with the Very Long Baseline Interferometry (VLBI) technique. The Radio Fundamental catalog of extragalactic radio sources compiles the positions of ∼ 19, 000 sources that have been measured with VLBI 5 . We found in this catalog 15 sources detected both by GLOSTAR and CORNISH. In the lower panel of Figure 2 we show the offsets between the GLOSTAR and the VLBI positions as black crosses, and the offsets between the CORNISH and VLBI positions as blue crosses. The mean GLOSTAR−VLBI position offset and its standard deviation is −0. 05 ± 0. 02 and 0. 06, respectively, in Galactic longitude direction, and 0. 03 ± 0. 03 and 0. 11, respectively, in Galactic latitude direction. On the other hand, the mean CORNISH−VLBI position offset and its standard deviation is −0. 02 ± 0. 02 and 0. 07, respectively, in the Galactic longitude direction, and −0. 02 ± 0. 03 and 0. 10, respectively, in the Galactic latitude direction. The conclusion from this analysis is that the astrometry of the B-configuration GLOSTAR images presented in this paper is accurate to better than 0. 1.

Comparison with the D-configuration catalog
In the first GLOSTAR catalog, we have reported a total of 1575 discrete sources detected in the same region presented in this paper, but based on data obtained with the VLA in its most com-S. A. Dzib et al.: A global view on star formation: The GLOSTAR Galactic plane survey. VI. Radio Source Catalog II: 28 • < < 36 • and |b| < 1 • , VLA B-configuration. pact (D) configuration. The GLOSTAR images from the VLA D-configuration observations have an angular resolution of 18 , i.e., 18 times larger than the VLA B-configuration images presented in this paper, which also excluded the shortest baselines to further filter out extended emission, resulting in some expected differences. First, given the higher angular resolution observations of the B-configuration, the extended radio sources reported by Medina et al. (2019) are resolved out and, in some cases, only the brightest peaks of extended sources are detected. It is worth noting that some compact sources detected in the Bconfiguration images can be seen projected on the area of the ex-tended sources, although they do not represent their direct counterpart. Their possible relation must be studied further in future (e.g., upper-panel of Figure 3). Second, some multiple component radio sources that are unresolved or slightly resolved in the D-configuration images will be resolved in the B-configuration images (middle-panel of Figure 3) and the integrated flux densities of the individual components can be estimated. Third, some individual radio sources detected in the D-configuration images can be resolved and appear as fragmented radio sources in the B-configuration images (lower-panel of Figure 3). In total, we have found that 95 sources in the D-configuration images are resolved into 224 B-configuration sources (see column (12) of Table 1). The components of fragmented radio sources are grouped and treated as a single source. However, the information of each single component is given in Table 1 and the fragmented sources are listed in Table 2. The integrated flux density reported in Table 2 is obtained by adding the integrated flux densities of the individual fragments. In total, 72 sources recovered from BLOBCAT are grouped into 22 fragmented sources.
Other sources were detected as single compact sources in both D-and B-configuration images and can be considered as direct counterparts. The mean position error of D-configuration sources is 1. 2 (Medina et al. 2019) and the beam size of the B-configuration images is 1. 0. Thus, by adding these values in quadrature we used a maximum angular separation of 2 between sources in both catalogs to consider them as direct counterparts. The number of matching sources between both catalogs with this criterion is 372. A further 312 matching sources are found using an angular separation of 9 , half of the angular resolution of the D-configuration images, for which the association must be investigated further. Given the differences described above, the matching of sources between both catalogs is not expected to be one-to-one.
In column (12) of Table 1 we list the GLOSTAR Dconfiguration name to which the B-configuration source is related. With the D-configuration name we have also labeled those sources that are related to two or more B-configuration sources and if they are considered as individual (I) or fragmented (F) sources. In total, 908 B-configuration sources were related to 780 D-configuration sources. The remaining 551 Bconfiguration sources have no counterpart in the D-configuration catalog. Most of these sources are located in the inner parts of the Galactic plane (see Figure 4) where the noise level is higher in the D-configuration images because of the bright and extended radio sources (see the lower panel of Figure 1 in Medina et al. 2019). As the noise levels could be as high as 500 µJy beam −1 , it explains why most of the sources were not detected in the Dconfiguration images, but are detected in the B-configuration images where the noise level is about 10 times lower.

Source sizes
The source sizes are obtained following Medina et al. (2019), who determined the source effective radius. BLOBCAT determined the number of pixels comprising each source, which can be used to estimate the area (A) covered by the source using the pixel size of 0. 25 × 0. 25. Then the effective radius can be determined using The effective radius distribution is shown in Figure 5, and the value for each source is listed in column (10) of Table 1.

Flux densities
To estimate the reliability of the radio source flux densities determined from the GLOSTAR images, we compare the results from the BLOBCAT extraction with those in the CORNISH catalog. As most of the sources are expected to be extragalactic sources whose variability is expected to be low (typically only of a few percent over timescales of several years) these provide a good point of comparison. We compare sources that are point like, and thus used the peak flux density. Moreover, we have only compared sources whose peak flux densities are above 2.7 mJy in both catalogs, as this is the 7σ base point in the high reliability source catalog of CORNISH (Purcell et al. 2013). We found 207 sources that meet these criteria. A difference between both catalogs is the mean observed frequency. CORNISH observed at 5 GHz, and GLOSTAR observed a wider bandwidth centered at 5.8 GHz. The measured flux density thus is expected to differ slightly for this observational mismatch due to the spectral index. Assuming that the extragalactic objects have a mean spectral index of −0.7 (Condon 1984), the CORNISH flux values will be on average 10% higher than in GLOSTAR.
In the upper panel of Figure 6 we plot the peak flux densities measured by the CORNISH survey as a function of the peak flux densities measured in this work. The dashed line indicates the equality line, and most of the sources are around this line. The lower panel of this figure shows the distribution of the peak flux density ratio between the results from both catalogs. A Gaussian fit to the distribution indicates that the mean value is 1.11 ± 0.03 with a standard deviation of 1.28. Considering the expected higher values in the CORNISH catalog, we conclude that the integrated and peak flux densities of the GLOSTAR-B radio sources are accurate to within 10%.

Spectral indices
The spectral index of a radio source gives us information on the dominant emission mechanism. Using the wide frequency coverage of our observations we can estimate the spectral index within the observed band. We measure the peak flux density in each of the imaged frequency bins for compact and point like sources that have S/N> 10 in our main source extraction, a total of 988 sources. The constraint of the S/N was chosen to consider that the noise level in each of the imaged frequency bins will be ∼ 3× higher, and may thus not be detected in the individual frequency bin images or their determined values will be affected by noise. On the other hand, over the area covered by an extended source, flux density variations that will depend on its structure may be observed at each frequency, and are hence not considered for this analysis. To obtain the spectral index we assume that in the observed frequency range the flux density is described by the linear equation: log S ν = α · log ν + C.
A weighted least-square fitting is made to the measured flux densities. Independent spectral index values for each source are listed in column (11) in Table 1. The distribution of the determined spectral indices is shown in Figure 7. The distribution of spectral indices has a mean value of −0.66 ± 0.02. Spectral indices at radio frequencies in the same area of the presented images have been measured previously from the GLOSTAR D-configuration (Medina et al. 2019) and in the THOR survey described Figure 8 shows the histogram distributions of the spectral index differences of the three sets. We found that the measured spectral indices are consistent among the three data sets, given that the mean of the differences is consistent with zero. We thus conclude that the spectral indices measured on GLOSTAR B-configuration images are reliable.
In star forming regions, three main mechanisms are known to produce compact radio continuum emission, and are related to different astrophysical phenomena (Rodríguez et al. 2012). The majority of radio sources show thermal free-free radio from ionized gas (e.g., HII regions, externally ionized globules, proplyds, jets) that has a spectral index ranging from 2.0 (optically thick) at low frequencies to −0.1 (optically thin) at high frequencies.
Magnetically active low-mass stars, may show nonthermal gyrosynchrotron with spectral indices ranging from −2.0 to +2.0.
Nonthermal synchrotron emission arising from colliding winds in high mass binaries as well from jets ejected by high mass stars interacting with the ambient interstellar medium (ISM) have a typical spectral index of −0.7. However, other phenomena not related to star formation also can produce thermal radio emission, namely gas ionized in planetary nebulae (PNe), while synchrotron emission is observed from extragalactic sources. Background active galactic nuclei (AGN) will mostly emit optically thin synchrotron emission with spectral indices ∼ −0.7. On the other hand, a fraction (up to 20%) of extragalactic background radio sources show a flat or positive spectral indices (e.g., Callingham et al. 2017). These represent a population of star forming galaxies, and progenitors of AGNs (i.e., high frequency peakers Dallacasa et al. 2000;Dallacasa 2003). Given the diversity of the radio sources, to better understand their nature, information at other wavelengths is required. This will be discussed in the following subsection.

Counterparts at other wavelengths
To gain more insight on the nature of the radio sources, we have searched for counterparts at shorter wavelengths. The search was focused on catalogs that could give us evidence for ionized gas, dense cold gas and dust, which are indicators of massive young stars and the regions they form in. We now briefly describe the catalogs used in our search.
The APEX Telescope Large Area Survey of the Galaxy (ATLASGAL) observed the galactic plane at a wavelength of 870 µm (345 GHz), and angular resolution ∼ 20 (Schuller et al. 2009). The emission at this wavelength is dominated by dense cool gas and dust. Several ATLASGAL source catalogs have been released by Contreras et al. (2013), Csengeri et al. (2014), and Urquhart et al. (2014), who list > 10, 000 dense clumps. We will compare our results with the compact source catalog presented by Urquhart et al. (2014). The differences between the angular resolution of ATLASGAL and the observations presented here have to be taken into account. Thus, we have used an offset of 18 (roughly the beam size of ATLASGAL images) to consider an ATLASGAL source to be a potential counterpart for the compact radio source. We found 143 radio sources matching the position of 83 submillimeter sources.
The Herschel infrared Galactic plane survey (Hi-GAL) observed the inner Galaxy in 5 bands distributed in the wavelength range between 70 µm and 500 µm (Molinari et al. 2010(Molinari et al. , 2016.  is ∼ 1. 2. We have used a 2 offset between a Hi-GAL source and a GLOSTAR source to consider them as counterparts, and found 98 sources that matched this criterion. The Wide-field Infrared Survey Explorer (WISE) mapped the entire sky in four infrared bands centered at 3.4, 4.6, 12.0, and 22.0 µm. The WISE ALL-sky Release Source Catalog contains astrometry and photometry for over half a billion objects (Wright et al. 2010). The angular resolution of the observations is 6 at the shortest wavelength and the position errors are around 0. 2. To consider a WISE source as the counterpart to a GLOSTAR radio source we have used a maximum offset of 2 . A total of 125 sources match this criterion.
The Galactic Legacy Infrared Mid-Plane Survey Extraordinaire (GLIMPSE; Benjamin et al. 2003;Churchwell et al. 2009) mapped a large fraction of the Galactic plane with the Spitzer space telescope. It observed in four near-infrared sub-bands covering the range from 3.6 to 8.0 µm, and with angular resolutions ∼ 2 . While the filter widths of the 3.6 and 4.5 µm bands are similar to those of WISE, the GLIMPSE survey has focused on Galactic plane observations and is more sensitive than the former. The interstellar medium emission at these wavelengths comes mainly from warm and dusty embedded sources. Considering the angular resolution of our observations and the position uncertainties in the GLIMPSE survey we have used an offset of 2 for the counterpart searching, leading to 251 matching sources.
The United Kingdom Infrared Deep Sky Survey (UKIDSS) is a suite of five public surveys at near infrared wavelengths (NIR) of varying depth and area coverage (Lucas et al. 2008)  H (1.65 µm) and K (2.20 µm) bands. The spatial resolution of UKIDSS-GPS observation is typically better than 0. 8. For the counterpart search of GLOSTAR radio sources in the UKIDSS-GPS catalog we have used this value of 0. 8, finding 389 matches. After our careful search of counterparts at other wavelengths, it is noticeable that 640 radio sources have no counterparts at any other wavelength. The spectral index distribution of the sources in this category shows that they have preferably negative spectral indices (see Fig 7). We will further discuss these sources in the next section.

Classification of radio sources
Using the counterparts of the radio sources in the catalogues described in the previous section, a robust classification can be carried out. A single classification was performed for fragmented sources listed in Table 2 instead of separate classifications of their fragments, and thus the classification was done to 1409 sources. The classification criteria are based on our findings of the emission properties and/or counterparts of near-infrared (NIR; UKIDSS), mid-infrared (MIR; GLIMPSE and WISE), farinfrared (FIR; Hi-Gal), and submillimeter (SMM; ATLASGAL). Images from the above mentioned infrared surveys are plotted at the position of the radio sources, some examples can be seen in Appendix B. For SMM and FIR, we show the emission properties of each GLOSTAR source. For NIR and MIR, we did a visual inspection of the three-color images for the UKIDSS (red K-band, green H-band and blue J-band), the GLIMPSE (red 8.0 µm, green 4.5 µm, and blue 3.6 µm), and the WISE (red 22.0 µm, green 12.0 µm and blue 4.6 µm) surveys. The sources have been classified into 5 groups, using the following criteria:  (17) in Table 1 and in column (5) of Table 2 for fragmented sources.

Sources previously classified
Classification of the radio sources is a main part of the catalog construction. However, some of the sources could have been previously classified. A search for previous classifications of the radio sources has been performed using the SIMBAD astronomical database (Wenger et al. 2000), within a radius of 2 . We found that 269 radio sources have counterparts in the SIMBAD database. Most of these sources, 138, are only classified as radio sources, and are thus of an unknown nature. The classification of the remaining 107 sources suggests that these are Galactic sources. In Table 3 we have separated these sources in seven types of sources, and have listed the source class in SIMBAD considered for these types and the number of sources of each type.
Article number, page 9 of 25 A&A proofs: manuscript no. main

Overview of the final classification
The final radio source classification is a combination of the method described in Section 3.8, and the use of known information of individual sources recovered from the SIMBAD database. For our final catalog we additionally consider the SIMBAD classification for sources for which we find no counterpart in the searched IR and sub-mm catalogs. This was, for example, the case for pulsars that are usually not detected at infrared wavelength and are non-thermal radio emitters. Based on the criteria described in the previous section they were first classified as EgCs and after consulting the SIMBAD database they were finally classified as pulsars. A similar situation occurred with a Wolf-Rayet star and a cataclysmic variable star, wherein they were initially classified as a PN and as a Radio-star, respectively. We use the classification recovered from SIMBAD in these cases.
In Table 4 we give the number of sources found in each of the classifications. Interesting sources are the HII region candidates, as they are related to massive star formation, that we will discuss later.

Comparison of classifications
Classifications of some of the detected radio sources have been performed in previous radio surveys and here we will compare the results of these classifications. We will focus on a comparison based on the low resolution GLOSTAR D-configuration image presented by Medina et al. (2019), and with the classification based on the CORNISH survey which has a similar angular resolution observation as our B-configuration data (Purcell et al. 2013).

GLOSTAR D-configuration
Some differences between the present catalog and that derived from the D-configuration data are expected, since for the latter, the search for counterparts was done using larger angular separations than in this work on account of the lower resolution. 54 of the HII region candidates identified by us were also detected in D-configuration images, and were also identified as HII regions. Four other HII region candidates had a D-configuration counterpart, but were unclassified. The remaining 32 HII region candidates were not detected in the D-configuration images. Additionally, six sources classified as HII region candidates in the Dconfiguration catalog have been classified as PNe in the present work as their positions are no longer coincident with ATLAS-GAL sources.
We have detected 32 radio sources that were classified as PNe in the D-configuration catalog, of which 30 are classified as PNe in this work, one is classified as a WR star and another as an EgC because it is no longer positionally coincident with IR emission. We also detect 13 sources identified as radio stars in the D-configuration, of which 10 were related to a radio source also classified as radio stars. The remaining source, G028.098-00.781, was resolved into three different sources, two of them were classified as extragalactic sources and the other as PN. A further 749 sources detected in D-configuration images were also detected in the B-configuration images. Among these, 688 were classified as EgCs, consistent with the suggestion by Medina et al. (2019) that most of their unclassified radio sources are background extragalactic radio sources. The remaining 61 unclassified sources in D-configuration have now been classified, 43 as radio-stars, 14 as PNe, and 4 as HII region candidates. A comparison between the method used in the D-configuration data and in the B-configuration data shows a consistency better than 90% in the resulting classes.

CORNISH Survey
The CORNISH survey has a similar angular resolution, and it also used IR information to classify detected radio sources. In their classification effort, they consider UCHII regions and dark HII regions; we detected radio emission from 40 CORNISH sources classified as UCHII and from one dark HII region, which are classified as HII region candidates by us. We also have detected 25 CORNISH radio sources classified as PNe, that are also classified as PNe in our work. Seventeen radio-stars in the COR-NISH survey have counterparts in our catalog; we classified fifteen of them also as radio-stars and two of them as EgCs because we did not find IR counterparts. We have detected 171 COR-NISH radio sources classified as IR-Quiet (not detected at IR wavelengths); we classify 167 of them as EgCs, 4 as radio-stars as IR emission is now reported in the position of these sources, and one as a pulsar based on the SIMBAD database. Finally, out of the 29 radio-Galaxy sources in CORNISH, 28 are classified as EgCs in our work, and the remaining source is classified as radio-star given its IR properties. The consistency in classification between CORNISH and GLOSTAR B-configuration is S. A. Dzib et al.: A global view on star formation: The GLOSTAR Galactic plane survey. VI. Radio Source Catalog II: 28 • < < 36 • and |b| < 1 • , VLA B-configuration.
high, especially for HII regions and PNe where we found a 100% agreement.

THOR survey
The HI/OH/Recombination line survey of the Milky Way (THOR; Bihr et al. 2015;Beuther et al. 2016;Wang et al. 2020) observed the northern hemisphere of the Galactic plane with the VLA in C-configuration and using its L-band receivers (1 to 2 GHz). It covers Galactic longitudes from 14. • 5 to 67. • 4 and Galactic latitudes from −1. • 25 to 1. • 25, i.e., including the area of the maps presented in this paper. The angular resolution of the THOR radio continuum images is 25 and they have noise levels from 0.3 to 1.0 mJy beam −1 (Wang et al. 2018). While a detailed comparison of the source classification by THOR and GLOSTAR B-configuration is limited given the different image angular scales, it is still, however, useful.
To compare GLOSTAR B-configuration sources with THOR sources, we used a maximum separation of 2. 5, which is the position accuracy of the THOR survey (Wang et al. 2018). We found 554 sources matching in position between these surveys. From these sources, the THOR survey identified 34 HII regions (26 of which are identified as HII region candidates) and 9 as PNe (also identified as PNe with our classification criteria). Our classification for the remaining eight sources identified as HII regions by the THOR survey is four EgCs and four PNe. Differences between the classification of these sources can be caused by the different angular scales used for comparison with IR surveys. In fact, the median match radius of IR HII regions (identified with the WISE survey; Anderson et al. 2014) with the THOR survey is ∼ 60 (Wang et al. 2018), almost two orders of magnitude larger than our angular resolution.

HII region candidates
We have identified 93 HII region candidates 7 . Out of these sources, 71 were previously related to HII regions from our analysis of the D-configuration images or from CORNISH or from THOR or from the SIMBAD database and 22 are new detections from this work. A characteristic of the radio emission from HII regions is that the spectral index values ranges from −0.1 (optically thin free-free radio emission) to 2.0 (optically thick freefree radio emission). Observationally, the spectral index distribution of hundreds of young HII regions at similar frequencies show a mean value of 0.6 (Yang et al. 2019(Yang et al. , 2021. We have determined the in-band spectral index (see Section 3.6) for 57 HII region candidates, 13 of which are new. In Figure 9, we plot the spectral index of these HII regions as a function of the R eff. the Y-factor and the S/N. Surprisingly, some of the determined spectral indices are negative and well below the lowest value expected for free-free radio emission of α = −0.1, indicating that the radio emission nature is non-thermal. The results are not affected by the size of the source, as most of them are slightly resolved sources (R eff θ resolution /2 = 0. 50). Kalcheva et al. (2018) obtained the spectral index to known ultra-compact HII regions and found that about 18% of their sample were consistent with negative spectral indices. They discuss their results and conclude that different interferometer array configurations and time variability could explain these negative values. These effects, however, are not present in the GLOSTAR observations presented here. Hence, these radio sources with negative spectral indices could be related to HII regions but they are not the HII regions themselves (e.g. Purser et al. 2016).
Compact radio sources have been found related to HII regions. The most known and nearby case is the Orion Nebula Cluster (ONC) where around 600 compact radio sources are found in the HII region ionized by the Trapezium (Forbrich et al. 2016;Vargas-González et al. 2021). As first suggested by Garay et al. (1987); Garay (1987), most of these fall into two broad categories: first, sources with thermal radio emission from circumstellar matter, often protoplanetary disks ("Proplyds"), that are photo-evaporated in the intense UV field of the brightest Trapezium star θ 1 Ori C (O7Vp). Second, nonthermal sources associated with the coronal activity of magnetically active low mass members of the ONC many of which also show X-ray emission and are (highly) variable Dzib et al. 2021). Other example regions are NGC 6334 ) and the M17 (Rodríguez et al. 2012) star forming regions (SFRs) where several radio sources are found close to prominent HII regions. Also, in these SFRs a significant fraction of the compact radio sources have been found to produce non-thermal emission. That high number of such radio sources can be detected on the ONC is a result of the cluster's close distance, D, of just ≈ 400 pc (Menten et al. 2007;Kounkel et al. 2017). Extrapolating the Orion case to a distance of a few kilo-parsecs, most of these magnetically active stars would not be detected, except for the two strongest sources with measured integrated flux densities of a few tens of mJy. We note that, NGC 6334 and M17, mentioned above, are also relatively nearby, at D = 1.34 +0.15 −0.12 and 1.98 +0.14 −0.12 kpc, respectively Wu et al. 2014). Time variable weak radio sources around the well-studied ultracompact HII region W3(OH) (D = 2.0 kpc) are other examples (Wilner et al. 1999).
For more distant SFRs, thermal radio emission is only detected from the HII region itself that is excited by the central high mass star(s) of a cluster.
It should be noted that other phenomena can produce compact non-thermal radio emission in massive star forming regions that are at work in Orion and can be observed in regions at distances of a few kpc, such as non-thermal radio jets (e.g. Carrasco-González et al. 2010;Purser et al. 2016) and wind collision regions in massive binary stars (e.g. Dzib et al. 2013). Non-thermal radio jets from YSOs, however, are rare and only a handful of cases are known. Similarly, non-thermal radio emission from wind collision regions are not common and are usually associated with evolved massive stars that have powerful winds.
On the other hand, the radio sources with positive spectral index could be true HII regions. More extensive multi-wavelength radio analysis is needed to characterize their emission and determine their turnover frequency, emission measure, etc. In general, all the radio sources that are associated with HII regions discussed in this work deserve further attention and more detailed studies.

Variable radio sources
By comparing the peak flux densities measured in the GLOSTAR images with those reported in CORNISH (Purcell et al. 2013), we have looked for sources with variable radio emission. The search was restricted to sources with Y < 2.0, i.e., with compact radio emission. Variability is identified when the source is detected in both catalogs, GLOSTAR and CORNISH, but the flux ratio is larger than 2.0. We also identify variability when we detect a source in the GLOSTAR catalog with a peak flux den- sity level > 2.7 mJy beam −1 (the CORNISH detection limit) and it is not reported in the CORNISH catalog. Finally, a source detected as unresolved in the CORNISH catalog that is not in the GLOSTAR high reliability catalog or in the catalog of sources with 5 to 7σ, is also considered as a variable radio source.
With the above criteria we have identified 49 variable sources. They are listed in Table 5 together with the peak flux densities in the compared catalogs. The GLOSTAR and COR-NISH classification are listed when available. Most of the identified variable radio sources are EgC or IR-quiet sources, and have only been detected at radio wavelengths. Since extragalactic background sources are not expected in general to show pronounced variability, thus some of these sources could be interesting Galactic radio sources whose nature has to be explored.

Extragalactic Objects
As has been shown in our previous works, most of the detected radio sources are expected to be background extragalactic objects (Medina et al. 2019;Chakraborty et al. 2020). Following the formulation by Fomalont et al. (1991), and using an observed area of 57,600 arcmin 2 , assuming a nominal noise level of σ = 65 µJy beam −1 and a threshold 7× the noise level, we estimate that the number of expected background extragalactic sources in our image is 1138 ± 664. This number suggests that most of the detected radio sources above 455 µJy beam −1 (7σ) are of extragalactic origin.
From our classification criteria we have compiled a list of 1159 sources that have most probably an extragalactic origin. They are labeled as EgC in Table 1. Their extragalactic nature is also supported with the negative spectral index of most of them. The spectral index is determined for 777 of these sources and 157 (20%) have a flat or positive spectral index, consistent with the expected number of extragalactic radio objects that are expected to have positive spectral indices at our frequencies.
Similar to our previous work (Chakraborty et al. 2020), we have also studied the Euclidean-normalized differential source counts of the point sources characterized as being extragalactic in origin. We have binned the source integrated flux densities in logarithmic space and divided the raw counts in each bin by the fraction of image area over which a source with a given integrated flux density value can be detected, known as the visibility area (Windhorst et al. 1985). The differential source counts have been calculated by dividing the visibility area weighted source counts in each bin, by the total image area (Ω in steradians) and bin width (∆S in Jy). These differential source counts have been normalized by multiplying with S 2.5 , where S is the mean integrated flux density of sources in each bin (Windhorst et al. 1985). The normalized differential source counts is shown in Figure  10, where the error bars are Poissonian. We have compared our findings with two simulated catalogues, the SKA Design Study simulations (SKADS, Wilman et al. 2008) and the Tiered Radio Extragalactic Continuum Simulations (T-RECS, Bonaldi et al. 2019). We have also compared our findings with the observed extragalactic source populations at low-frequency as well as high-frequency, which include: the TIFR GMRT Sky Survey at 150 MHz (TGSS-ADR1; Intema et al. (2017)), BOOTES field at 150 MHz using LOFAR (Williams et al. 2016), Lockman Hole field at 1.4 GHz with the LOFAR (Prandoni et al. 2018), COS-MOS field at 3 GHz with VLA (Smolčić et al. 2017), ECDFS field at 5 GHz (Huynh et al. 2015) and 9 GHz (Huynh et al. 2020) with ATCA and the 1.4 GHz source counts based on observations with VLA by Condon (1984). In all cases we have scaled the source counts to 5.8 GHz using a spectral index, α = −0.7. We have found that the source count of these sources classified as extragalactic in the GLOSTAR survey is statistically similar and consistent with the previously observed extragalactic source population as well as with the simulated catalogs. This shows that the majority of these sources are indeed of extragalactic origin. Figure 7 shows that the peak of the spectral index distribution is −0.66 ± 0.02. As discussed in the previous subsection most of the unidentified radio sources will be of extragalactic origin, however some of these will be interesting Galactic non-thermal radio sources. These sources are only detected at radio frequencies and it will be hard to distinguish them. Very Long Baseline Interferometry (VLBI) observations to all unidentified compact radio sources could help to distinguish the Galactic from the extragalactic radio sources. Position measurements on the scale of several months to years could measure the proper motions and trigonometric parallaxes of these sources. As the extragalactic background sources are not ex-S. A. Dzib et al.: A global view on star formation: The GLOSTAR Galactic plane survey. VI. Radio Source Catalog II: 28 • < < 36 • and |b| < 1 • , VLA B-configuration.  (Purcell et al. 2013). pected to exhibit proper motions or trigonometric parallaxes, these observations can help to distinguish between the two classes of radio sources. A wide field VLBI survey of the Galactic Plane is now feasible thanks to the DiFX software correlator (Deller et al. 2011) which allows multiple-phase center correlation inside primary beam of the interferometer.  Fig. 10. The Euclidean-normalized differential source counts of the point sources classified as of extragalatic origin in the GLOSTAR survey. We have compared the source counts with the simulated radio sky and previously observed source populations. For details of simulated catalogues and different observed source populations see text.

Perspective on the search of non-thermal galactic sources
they require less computational power. The observed frequency by the GLOSTAR survey also has the advantage that the radio emission is less affected by scattering than the lower frequencies used for conventional pulsar searches. As an interesting example, the radio emission from PSR J1813-1749, one of the most energetic known pulsars, was first detected in radio continuum images at 5 GHz (Dzib et al. 2010 while searches for the pulsed radio emission at lower frequencies failed (Helfand et al. 2007;Halpern et al. 2012;Dzib et al. 2018). It turns out that interstellar scattering in the direction of this pulsar is very high, and PSR J1813-1749 is the most heavily scattered known pulsar (Camilo et al. 2021). Pulsar radio continuum emission is characterized by point-like structure and steep radio spectrum (α = −1.4 ± 1.0; Bates et al. 2013). These are criteria that are shared with EgCs in our classification scheme, making them hard to differentiate. Dedicated observations to search for pulsed emission can distinguish them, with the advantage that the observations will be target intended instead of a blind survey.

Conclusions
As part of the GLOSTAR survey, the VLA in its B-and Dconfiguration was used to observe a large portion of the Galactic plane in the C-band. In this paper we present the B-array observations covering the area within 28 • ≤ < 36 • and |b| < 1 • , which we previously investigated using data obtained in the D-configuration (Medina et al. 2019). Using a combination of automatic source extraction with BLOBCAT (Hales et al. 2012) and visual inspection we have identified 3325 radio sources.
The catalog of these radio sources is divided in two parts. The catalog of highly reliable radio sources contains 1457 entries. Detailed properties of these sources are given, such as the positions, the signal-to-noise-ratio, integrated and peak fluxes, the ratio between these two values (also known as the Y-factor), the effective radii, and the spectral indices. The weak source catalog lists 1866 sources with a signal-to-noise ratio between 5 to 7. Only their basic properties such as positions, the signal-tonoise-ratio, integrated and peak fluxes are given.
The highly reliable radio sources were further investigated. The positions of these sources were compared with the positions from the CORNISH catalog (Purcell et al. 2013) and radio sources detected with the VLBI from the Radio Fundamental catalog. We found that the positions of GLOSTAR sources are in agreement with those in these catalogs to better than 0. 1. We have also compared our integrated flux densities with those in the CORNISH catalog and conclude that the GLOSTAR integrated flux densities are accurate to within 10%, apart from clearly variable sources. From a comparison with the GLOSTAR D-configuration catalog, we find that 908 of them are related to 780 sources detected in the D-configuration images. In particular, 22 D-configuration sources are partially resolved and appear as fragmented sources in the new high resolution images. A total S. A. Dzib et al.: A global view on star formation: The GLOSTAR Galactic plane survey. VI. Radio Source Catalog II: 28 • < < 36 • and |b| < 1 • , VLA B-configuration.
of 72 highly reliable B-configuration sources comprise these 22 fragmented sources.
To further investigate the nature of the highly reliable radio sources, we have used information from surveys at infrared and sub-millimeter wavelengths as well as consulting the SIM-BAD database. The classification of the radio sources resulted in 93 HII region candidates, 64 PNe, 81 radio stars, and most of the remaining sources as EgCs. We compared our classification with the classification done to the D-configuration radio sources, and to the sources from the CORNISH survey. We find that the classification from the catalogs agree in more than 90% of the sources. An interesting result is that many sources classified as HII region candidates, however, have a negative in-band spectral indices suggesting that the radio emission is predominantly non-thermal and, thus, they are not the HII regions themselves, but likely related to nearby YSOs. These sources could be other radio emitter objects related to star formation. They deserve a further study using deeper and radio multi-wavelength observations to better characterize their radio emission nature, and the nature of the objects themselves. Finally, by comparing the integrated flux densities from GLOSTAR and CORNISH, whose observations are separated by 7 years, we have identified 49 variable sources, whose nature has to be explored for most of these sources.

Acknowledgments
This research was partially funded by the ERC Advanced Investigator Grant GLOSTAR (247078). AY would like to thank the help of Philip Lucas and Read Mike when using the data of the UKIDSS survey. RD and HN are members of the International Max Planck Research School (IMPRS) for Astronomy and Astrophysics at the Universities of Bonn and Cologne. HB acknowledges support from the European Research Council under the Horizon 2020 Framework Program via the ERC Consolidator Grant CSF-648505. HB also acknowledges support from the DFG in the Collaborative Research Center SFB 881 -Project-ID 138713538 -"The Milky Way System" (subproject B1). VY acknowledge the financial support of CONACyT, México. This work (partially) uses information from the GLOSTAR database at http://glostar.mpifr-bonn.mpg.de supported by the MPIfR, Bonn. It also made use of information from the ATLAS-GAL database at http://atlasgal.mpifr-bonn.mpg.de/ cgi-bin/ATLASGAL_DATABASE.cgi supported by the MPIfR, Bonn, as well as information from the CORNISH database at http://cornish.leeds.ac.uk/public/index.php which was constructed with support from the Science and Technology Facilities Council of the UK. This work is based in part on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. This publication also makes use of data products from the Wide-field Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration. Herschel is an ESA space observatory with science instruments provided by European-led Principal Investigator consortia and with important participation from NASA. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France. We have used the collaborative tool Overleaf available at: https: //www.overleaf.com/.