VLBI Ecliptic Plane Survey: VEPS-1

We began observations in the search mode in 2015 February. The participating stations included the three core stations of the Chinese VLBI Network (CVN): seshan25, kunming, and urumqi. However, sometimes these stations were not available at the same time, or occasionally one of them had a risk of failure. In that case, one or two international stations joined. Depending on the participating stations, the longest baseline length in each session can be varied from 3200 km to 9800 km.

Figure 1 shows the geographical distribution of all participating stations. kashima34, sejong, and hobart26 have contributed to past VEPS observations. They are relatively large antennas, and have good common visibility of the ecliptic zone. Before joining in the VEPS survey, we performed fringe tests to sejong, hobart26, and kashim34 in 2014 December, 2015 July, and 2016 January, respectively.

Zoom In Zoom Out Reset image size

Our observations were performed at a 2048 Mbps data rate, with 16 Intermediate Frequency (IF) channels and 2-bit sampling. The first eight IFs of 32 MHz bandwidth were distributed in the range of [8.188, 8.444] GHz, and the remaining eight IFs of 32 MHz bandwidth were in the range of [8.700, 8.956] GHz. The data volume was around 16 TB for each station, in one 24 hr session.

We used the Chinese VLBI Data Acquisition System for data acquisition at the Chinese stations, ADS3000+ at kashim34, and K5/VSSP32 for sejong. The maximum data rate of the K5/VSSP32 is limited to 1024 Mbps, so the data were sampled with 1 bit. For hobart26, the Digital Base-band Converter (DBBC) board was used. At the time, the firmware supported only 16 MHz-wide IFs, so we observed at 16 MHz-wide IFs instead, and correlated them against the low parts of 32 MHz-wide IFs recorded at other stations.

Each target source was observed in two scans of 90 s length, with a gap between consecutive observations of the same source of at least 3 hr. The sequence of observations was optimized to minimize slewing time with the use of the astrometry VLBI scheduling software package sur_sked. Every hour, four calibrators from the pool of 190 bright sources were inserted in the schedule. Two of them were observed at elevations below 30°, and two at elevations of 50° or more above the horizon. They were also used for bandpass calibration, antenna gain calibration, as ties for astrometric global solutions, and to improve the estimation of the atmosphere path delay in the zenithal direction.

By 2016 September, 13 sessions had been observed. The summary of these observations is presented in Table 1. In general, the observations were successful, despite a number of failures. urumqi had a receiver problem in the first 12 hr in VEPS01. seshan25 and kunming made use of incorrect B1950 source positions for antenna control in VEPS03. hobart26 had a timing problem after 9 hr in VEPS09. kunming data in VEPS012 were lost due to a hard disk failure. If a target source was not observed due to station failure, we reobserved it in the next VEPS sessions.

We ran two absolute astrometry dual-band VLBA programs that targeted ecliptic plane compact radio sources: the dedicated survey of weak ecliptic plane calibrators with the VLBA BS250 program in 2016 March–May, and the VLBA Calibrator Survey 9 (VCS-9) in 2015 August–2016 September.

The BS250 program had 111 target sources within $7\buildrel{\circ}\over{.} 5$ of the ecliptic plane, which were observed in four 8 hr segments. The targets were the weakest calibrators, with correlated flux densities at baseline projection lengths greater than 5000 km in the range [30, 50] mJy. Each target was observed in three scans of 180 s length. The target sources were scheduled in such a way that the minimum gap between their consecutive observations was at least 2 hr. Every 1.5 hr, a block of four calibrators from the pool of bright sources was inserted in the schedule in such a way that two of them were observed at elevations below 30°, and two at elevations 50° or more above the horizon, similar to the VLBI observations in the search mode. We used the same frequency setup for the BS250 as in the VCS-II program (Gordon et al. 2016): four IFs of 32 MHz bandwidth spread in the range of [2.22, 2.38] GHz, and 12 IFs spread in the range of [8.43, 8.91] GHz, recorded simultaneously at a 2 Gbps data rate. Of the 111 target sources, 37 were detected in the search mode of the VEPS program, and the remainder were detected in other surveys.

The goal of the VCS-9 program is the densification of the grid of VLBI calibrators; it observed over 11,000 sources spread approximately uniformly over the sky at declinations above −45°, in one scan of 60 s each. There is some overlap between the source lists of VCS-9 and VEPS. The positional accuracy of the sources observed in VCS-9 is worse than those observed in the BS250 campaign, because its integration time was lower by a factor of nine, but is still significantly higher than in the VEPS search mode. Therefore, we consider the VCS-9 program to provide observations in the refining mode, for the purposes of this study. By 2017 February, about half of the VCS-9 observations had been processed, so we report here only the VCS-9 results available so far out of the sources detected in the VEPS search mode. VCS-9 observed with the wide-band C-band receiver, and simultaneously recorded eight IFs, 32 MHz-wide, spread within [4.13, 4.61] GHz, and eight IFs spread within [7.39, 7.87] GHz. Scheduling VCS-9 observations was done in a similar way to those of BS250, except that only one scan per source was observed, and the interval of time between calibrators was reduced to 1 hr. VCS-9 was observed in segments of 3.5–10 hr.

The IVS runs a number of VLBI observing programs, primarily for geodesy, with occasional use for astrometry (e.g., Le Bail et al. 2016). As a subgroup of the IVS, the Asia-Oceania VLBI network (AOV), which includes facilities from five countries (Australia, China, Japan, New Zealand and South Korea) in the Asia-Oceania region (Lovell et al. 2015), ran a number of experiments beginning in 2015. We made an attempt to improve the coordinates of some VEPS sources detected in the search mode and provide additional measurements of telescope position with the same experiments in two such 24 hr sessions, AOV010 in July and AUA012 in 2016 August.

Both sessions included sensitive AOV antennas: parkes and hobart26 in Australia; seshan25, kunming, and urumqi in China. In addition, tianma65 (Tianma 65 m Radio Telescope, or TMRT) in China and tsukuba in Japan participated in AOV010. We conducted observations at a 1 Gbps data rate, with 16 IFs of 16 MHz bandwidth and 2-bit sampling, which is the highest data rate that all of the participating stations were able to provide in 2016.

Because two geodetic stations, kashim11 and koganei, could observe only within a narrow X-band (8.1–8.6 GHz), we adjusted the frequency sequence by balancing the uncertainty of the group delay and the amplitude of the highest side lobe in the delay resolution function. The following frequency sequence was used in the session AOV010: [8.19799, 8.21399, 8.23399, 8.25399, 8.33399, 8.41399, 8.51399, 8.53399, 8.55799, 8.57399] GHz. The amplitude of the highest side lobe is 0.52.

We used the geodetic software sked (Gipson 2016), which is widely used in the preparation of many other IVS sessions, to schedule these experiments. We set a group of weak sources as targets. Among them, 32 sources were previously detected in the VEPS search mode observations, with declinations in the range [−31°, −15°] and correlated flux densities in the range [30, 80] mJy. Another group of strong compact sources were selected automatically to provide a uniform sky coverage. We used the astrometric mode of sked for automatic scheduling; it is described in more details in Le Bail et al. (2016). After taking a few iterations by adjusting some control parameters of the sked, we were able to schedule 60% of the total number of observations for target sources, while keeping enough scans for small antennas to achieve the geodesy goals.

We correlated the VEPS search mode observations and the IVS sessions with the DiFX software correlator (Deller et al. 2011), which was installed on a powerful hardware platform in 2014 (Shu et al. 2015) at the Sheshan Campus of the Shanghai Astronomical Observatory. The data from the Chinese domestic stations were recorded on disk packs and then shipped to Shanghai, while the data from international stations were transferred to Shanghai via a high-speed network. The data volume for each station is approximately 16 TB, eight times bigger than that recorded in regular IVS geodetic sessions, so the data correlation of one VEPS session usually took more than 24 hr.

Correlation of mixed observing modes with different bandwidths or sampling bits can be challenging. Fortunately, the DiFX can fully support the correlation of 1-bit sampled data from sejong against 2-bit sampled data from the other stations. For the correlation of 16 MHz bandwidth data from hobart26 to the 32 MHz bandwidth data from the other stations, the zoom mode was selected to pick up the overlapped frequency band. Moreover, it was optional to make correlations only on the 16 MHz bandwidth on the baselines to hobart26, while the other stations with 32 MHz bandwidth went through an independent correlation pass, the same as the usual correlation procedures. We selected an accumulation period of 0.125 s, and 512 spectral channels per IF. This setup provides a wide field of view that allowed us to detect a source within several arcminutes of their a priori position. In fact, we observed fields around the pointing direction rather than individual sources.

Correlation of the VLBA experiments was performed at the Socorro array control center using the same DiFX software correlator. The DiFX correlator output was converted into FITS-IDI format using the difx2fits program for further analysis.

The visibility data produced by the correlator were then processed using the VLBI data analysis software ${ \mathcal P }\,{ \mathcal I }\,{ \mathcal M }\,{ \mathcal A }\,$.¹¹ A detailed description of the analysis strategy and a comparison between the methods adopted in the past and those used for processing our data can be found in Petrov et al. (2011a). We ran the pipeline, which includes a coarse fringe search, manual adjustment of phase calibration, generation of the auto-correlation and cross-correlation bandpass masks, computation of the complex bandpass calibration, a fine fringe search, amplitude calibration, amplitude normalization, and computation of the total group delay. The group delays, fringe amplitudes, and related information were written in databases for further analysis.

The databases were loaded into the VTD/Post-Solve software¹² for preliminary astrometric data analysis. At the beginning of the astrometric analysis, we discarded observations with a signal-to-noise ratio (S/N) < 6, which corresponds to the probability of false detection around 0.001. The S/N is defined here as the ratio of the peak amplitude to the mean amplitude of the noise. In the beginning, we used a coarse parameter estimation model; we estimated only the positions of the sources not detected before with VLBI and a clock function that is represented by a linear spline with knots placed at equal intervals of 1 hr. After the elimination of outliers, we refined the model by adding an estimation of source positions and residual atmosphere path delay in the zenithal direction. We then reduced the S/N limit gradually to 5.2. We discarded all new sources with less than two detections.

Next, we updated the source positions, re-ran fringe fitting and repeated the procedure of astrometric data analysis. For those sources that were marked as outliers after the second step of iterations, we computed the expected group delay based on results of parameter estimation. We then reran fringe fitting for these observations with a narrow fringe search window, and repeated the astrometric analysis.

Databases cleaned for outliers were saved for the final astrometric analysis.

We ran an astrometric analysis in the global mode, which is the usual approach for processing absolute astrometry VLBI surveys (Petrov et al. 2007). This means that we used all VLBI experiments in geodesy and absolute astrometry acquired so far, as well as the new data. We ran two solutions. In solution A we used all geodesy data acquired until 2016 December 1, absolute astrometry VLBI data acquired until 2015 January 1 (before the start of the VEPS program), and the VEPS-1 data in the search mode. In solution B, we used all dual-band VLBI data acquired until 2017 February, including the BS250 and VCS-9 campaigns, but excluding single-band group delays acquired in the search mode of the VEPS program.

We estimated the following: source coordinates, station positions, and velocities as global parameters; pole coordinates, UT1 angle, their rates, and daily nutation offsets for every observing session; clock function and residual zenith path delay in the atmosphere, modeled with a B-spline of the first degree with a step of 60 and 20 minutes, respectively. No-net-rotation constraints were applied to the estimates of the source coordinates in such a way that the net rotation of 212 sources marked as "defining" in the ICRF catalog (Ma et al. 1998) was zero, with respect to their catalog positions. The data-analysis procedure is very close to that used for deriving other VLBI catalogs, e.g., in Petrov (2013).

We added, in quadrature, a floor of 0.5 nrad to the position uncertainties computed with the law of error propagation, in order to accommodate the contribution of unaccounted systematic errors. The value of the floor was found empirically from the set of trial solutions, and comparisons of the source coordinate estimates from independent subsets of observations. The uncertainties inflated with the given floor are in a closer agreement with the differences in source coordinate estimates from VLBA observations.

Because VEPS search mode experiments were observed in the single band, we computed the ionosphere contribution from the total electron content (TEC) maps provided by the CODE analysis center for processing Global Navigation Satellite System data (Schaer 1998), using the technique described in detail in Petrov et al. (2011a). Our previous extensive analysis of the residual contribution of the ionosphere at 8 GHz to the source coordinate estimates, after applying the a priori path delays computed from TEC maps to the data-reduction model, does not exceed 7.5 nrad in quadrature in the worst case (Petrov et al. 2011b). Because uncertainties of source positions derived from VEPS search mode observations are greater than that, we simply ignored the contribution due to errors in the TEC model.

Using the results of fringe fitting, we apply the ${ \mathcal P }\,{ \mathcal I }\,{ \mathcal M }\,{ \mathcal A }\,$ task split to perform coherent averaging over time and frequency, after phase rotation according to group delays and phase delay rates. It also applies calibration for system temperature, gain curves, bandpass renormalization, combines all visibilities of a given source, and writes averaged visibilities and their weights into output binary files in the FITS format. The data were then suitable for imaging with the NRAO Astronomical Image Processing System (AIPS) and the Caltech difmap package (Shepherd et al. 1994).

We ran this procedure through all segments of the BS250 campaign. Some sources were observed in more than one segment; because the observations took place within two months, we ignored possible source variability and merged the calibrated visibilities.

The visibilities coherently averaged over all spectral channels within an individual IF, and over time with an integration time of 8 s. The resulting data were exported to difmap for imaging and to calibrate the residual phase errors. We performed a traditional hybrid mapping procedure consisting of several iterations of cleaning (Högbom 1974), as well as phase and amplitude self-calibration. We first self-calibrated the phases against a model of point-like sources using the startmod task. We then performed phase self-calibration and mapping under uniform weighting, followed by natural weighting. Every source was first imaged with an automatic pipeline. The results were examined, and those sources for which the automatic pipeline did not provide satisfactory results were reimaged manually. The main reason for manual reimaging was the necessity to remove outliers due to inaccurate on–off source flagging, and due to spikes in system temperature readings caused by radio frequency interference (RFI). Thus, the data were edited and those parameters including clean boxes were set by hand, to avoid large sidelobes and get more correct morphology. Figure 2 shows some sample images. The final images in FITS format, as well as self-calibrated visibilities, are accessible from the project web site http://astrogeo.org/veps.

Zoom In Zoom Out Reset image size

Images are needed to predict the necessary integration time when planning phase-referencing observations. In the future, we are going to reanalyze the data with the source structure contribution applied, in order to improve position accuracy and tie the reported position to a particular source-image feature.

Because the VEPS search mode sessions had too few observations per source, typically 2–6, imaging was not feasible. Instead, we used the non-imaging procedure developed in Petrov et al. (2011b) for computing the mean correlated flux density. First, we calibrated the raw visibilities data for the a priori antenna gain G(e) and system temperature T_sys: ${F}_{\mathrm{corr}}=v\cdot {T}_{\mathrm{sys}}(t,e)/G(e)$. Second, we adjusted antenna gains using publicly available brightness distributions of the calibrator sources, made with observations under other programs that can be found in the Astrogeo VLBI FITS Image Database.¹³ Using images in the form of CLEAN components, we computed the predicted flux densities of the calibrator sources for every observation. Using logarithms of the flux densities of the calibrator sources derived from the calibrated visibilities and known images, we computed the multiplicative gain corrections using least squares. These gain corrections were applied to the target sources and the correlated flux density estimates were corrected for errors in a priori gain calibration. This procedure is known to have errors of around 15% (Petrov et al. 2011b, 2012).