3.1. Standing waves

Standing waves are introduced as a result of broadband signals entering the telescope along multiple paths and creating an interference pattern. The principal (on-source) standing wave at Parkes is created by radiation reflecting from the feed towards the apex of the telescope. The frequency interval of this standing wave is c/2F, where F is the focal length (Parkes F/D = 0.41), and corresponds to 5.6 MHz at Parkes. Reflections off other parts of the dish and feed support legs result in standing waves at other frequencies. Standing waves from off-source interference can be even more complex. As with other baseline artefacts, the effect of standing waves can be mitigated by careful calibration. However, this becomes more difficult if the phase or amplitude of the wave shifts over time (Briggs et al. Reference Briggs, Sorar, Kraan-Korteweg and van Driel1997). Both Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013) and Kleiner et al. (Reference Kleiner, Pimbblet, Heath Jones, Koribalski and Serra2016) noted the presence of standing waves in Parkes multibeam data and further corrected by fitting and subtracting high-order polynomials from their spectra. Standing waves can even be problematic for radio interferometers (Popping & Braun Reference Popping and Braun2008).

These results can be understood by noting that the amplitude, a, of the standing wave relative to the power in the direct signal, A, is a/A = 2γ, where γ is the voltage ratio of the delayed and undelayed signals. So, a scattered power of only 0.01% will give rise to a standing wave amplitude ratio of 2% for the multibeam, and a scattered power of only 0.0001% will result in an amplitude ratio of 0.2% for the MPIPAF. The large apparent difference in the reflection coefficient of the two receivers (~100) is partly a result of the increased efficiency of the MPIPAF. The higher efficiency of the MPIPAF leaves less energy available for multipath reflections from the feed, but this can explain at most a 1.4 times reduction in reflected power compared to the multibeam, as suggested from a comparison of the measured multibeam (Staveley-Smith et al. Reference Staveley-Smith1996) and MPIPAF (Chippendale et al. Reference Chippendale, Beresford, Deng, Leach, Reynolds, Kramer and Tzioumis2016) feed efficiencies of 50–64% and 64–75%.

However, this cannot be the only factor. We must also consider that the MPIPAF fully samples the focal plane, such that neighbouring beams, which are separated by 15 arcmin on the sky, also have the same high efficiency. On the other hand, the multibeam feed array undersamples the aperture plane, and neighbouring beams are separated by 28 arcmin, or ~2 beamwidths (Staveley-Smith et al. Reference Staveley-Smith1996). Therefore, the efficiency of this receiver for hypothetical beams separated by 15 arcmin is effectively zero – i.e. most incident radiation is reflected. The low MPIPAF standing wave amplitude must therefore be a result of the low amounts of power reflected from the focal area around a given beam, and not just the power reflected from the beam itself. A more accurate analysis of the excellent standing wave performance of the MPIPAF relative to the traditional multibeam array requires a full electromagnetic simulation and diffraction analysis, and is outside the scope of this paper.

Modulation of the primary beam pattern by standing waves on interferometers is a related problem (Popping & Braun Reference Popping and Braun2008). Using APERTIF on the Westerbork Synthesis Radio Telescope, Oosterloo, Verheijen, & van Cappellen (Reference Oosterloo, Verheijen and van Cappellen2010) have shown that this can also been suppressed using PAFs. Aside from the lower standing wave, additional suppression can be obtained by utilising the frequency flexibility inherent in beamforming. For the MPIPAF, the beamformer weights can be adjusted in 1-MHz sub-bands. As the primary reflection standing wave period of 5.6 MHz is well sampled by the 1-MHz resolution of the beamformer and the 90-mm spatial sampling period of the MPIPAF chequerboard at the focal plane (spacing between PAF element feeds) it may be possible to suppress the standing wave even further via more advanced beamforming techniques in the future.

3.2. System temperature

We determined the system temperature, T _sys using the calibrator observation of PKS 1934-638 directly preceding the target observations. Chippendale et al. (Reference Chippendale, Beresford, Deng, Leach, Reynolds, Kramer and Tzioumis2016) determined T _sys/η for the MPIPAF to be in the range 45–60 K and to be relatively stable between the two measured polarisations from observations with the dish pointing near zenith. In the three months of observing (August–October), however, we find T _sys/η to be significantly higher (increasing with decreasing frequency) and less stable, ranging from ~70–140 K (mostly between ~70–110 K), with significant variation between the two polarisations and the observation dates and beams (see Figures 4 and 5). We also note that adjacent beams are separated by 0.25°, or ~1 beamwidth, so there is some overlap. We measure a mean correlation in the spectral noise between adjacent beams, in the same linear polarisation, of ~10%. As expected, this is slightly less than the level of correlation of 13–20% previously measured for adjacent ASKAP BETA beams, which are separated by 0.78°, or ~0.7 beamwidths (Serra et al. Reference Serra2015).

Figure 4. Comparison of polarisation A and B T _sys/η variation with frequency for the central beam (beam 8) from October 4 observations with the test values upon installation on Parkes (Chippendale et al. Reference Chippendale, Beresford, Deng, Leach, Reynolds, Kramer and Tzioumis2016). The higher values measured by us at low frequencies are a result of the beamformer delay slips noted in the main text.

Figure 5. Polarisation A and B T _sys/η average values over the frequency range ν = 1400–1420 MHz in each of the 16 beams over observations from August, September, and October (panels (a) and (b), respectively). August and September are dashed lines and October are solid lines.

Some variation in T _sys/η is to be expected as new beam weights were not made for each spectral line observation and the state of the hardware was not carefully controlled between observations. The probable cause of the T _sys/η variation are delay slips (mostly single-sample) between different ports of the MPIPAF digitiser when the digital receiver is power cycled (Bannister & Hotan Reference Bannister and Hotan2015). In a production system, such as ASKAP or Bonn, this can be calibrated using an on-dish noise source.

The central beam (beam 8) appears to be one of the most stable beams, remaining roughly constant during the August and September observations and at a different value for the October observations (dashed and solid lines Figure 5, respectively). The T _sys values for polarisation A were slightly more constant than those for polarisation B. If a PAF is permanently installed on the Parkes telescope, the T _sys/η noise can be lowered below the Chippendale et al. (Reference Chippendale, Beresford, Deng, Leach, Reynolds, Kramer and Tzioumis2016) values by cooling the receiver systems. Simulations suggest that a next-generation ‘rocket’ design can achieve T _sys/η = 20–25 K (Dunning et al. Reference Dunning, Bowen, Hayman, Kanapathippillai, Kanoniuk, Shaw and Severs2016), which is a 40 K improvement on the MPIPAF and a 15–20 K improvement on the Parkes multibeam receiver.

3.3. Radio frequency interference

There is significant RFI present in the observations, as the Parkes telescope is located in New South Wales and suffers interference from radio, television, mobile phones, as well as satellites. In the frequency range of our observations, the main contributors are the navigational satellite signals (e.g., GPS) at ν < 1290 MHz (see Figure 6). This is in contrast to the excellent RFI situation at the Murchison Radio-astronomy Observatory (MRO) where, with the exception of satellite RFI and tropospheric ducting events (Indermuehle et al. Reference Indermuehle, Harvey-Smith, Wilson and Chow2016), low-frequency observations are largely free of terrestrial contamination.

Figure 6. The power spectrum (in arbitrary units) for a sample observation before (blue) and after (brown) applying a realtime projection algorithm for RFI mitigation. Substantial reduction in RFI levels is achieved for the strong RFI signals near 1 150, 1 210, 1 230, 1 310 and 1 440 MHz. We show zoomed-in spectral regions near 1 150, 1 230 and 1 440 MHz in the lower panels. The 1 152 MHz signal is suspected internal RFI and has also been noted in ASKAP spectra. The mitigated signal from 1 159–1219.5 MHz is most likely satellite RFI. The 1232.6 MHz signal is an aliased fundamental of the coarse filter bank readout clock, also noted by Chippendale & Hellbourg (Reference Chippendale and Hellbourg2017). Based on the ACMA Register of Radiocommunications Licenses, the signal at 1439.5 MHz is a telecommunications signal.

Since PAFs can fully sample the focal plane and have flexible beamforming capability, they are perfect for the application of advanced RFI mitigation and suppression techniques (see Fridman & Baan Reference Fridman and Baan2001, for an overview). Indeed, the ASKAP Boolardy Engineering Test Array (BETA) was used to test one of these RFI mitigation methodologies, a spatial filtering technique based on projecting out the interferer signature (Hellbourg et al. Reference Hellbourg, Weber, Capdessus and Boonstra2012). This test demonstrated the effectiveness of the projection algorithm in suppressing RFI contamination in ASKAP PAF data (Hellbourg, Bannister, & Hotarn Reference Hellbourg, Bannister and HotarP2016). This success encouraged us to attempt a similar technique for the MPIPAF data taken at the Parkes site. Full mitigation typically requires much higher temporal and spectral resolution than we have in the MPIPAF data (~ microseconds and ~ kHz vs. 4.5 seconds and 18.5 kHz, respectively). However, we were able to demonstrate again the power of the projection method in some of our observations. (Figure 6 shows spectra taken before (brown) and after (blue) applying RFI mitigation. See Chippendale & Hellbourg Reference Chippendale and Hellbourg2017, for further details)

For most of our observations, RFI contamination was removed from the spectra using more conventional threshold flagging techniques, by excluding individual spectral channels with flux density, S _νobs > 5σ _i , where σ _i is the channel RMS noise. We excluded all optical sources with redshifts placing them within the frequency range of GNSS satellites (1240–1252 MHz). We also excluded, by manual inspection, all spectra containing occasional GPS L3 emissions at 1376–1384 MHz. We did not exclude sources falling within the receiver breakthrough at ~1350 MHz as this appeared relatively stable and, unlike other RFI regions had a reasonably low channel RMS (0.06–0.08 Jy, c.f. clean spectral channels RMS ~0.04 Jy).

3.4. The Large Magellanic Cloud

The LMC was observed as a check of the accuracy of the flux density and frequency calibration of the MPIPAF, as well as a basic check of the reduction pipeline. We compare our results with accurate archival observations from Staveley-Smith et al. (Reference Staveley-Smith, Kim, Calabretta, Haynes and Kesteven2003), which used the Parkes multibeam receiver. As the observed LMC emission is over the range ν ~ 1418–1419.5 MHz, the LMC observations are in the RFI free section of the band and we can use all channels containing LMC emission.

We gridded the MPIPAF data without any further calibration or adjustment except to apply a Gaussian smooth of FWHM = 7 arcmin to remove small residual scanning lines which can still be seen faintly in Figure 7 at −68° < δ < −67°. Unlike the previous multibeam observations, no cross-scans were taken to mitigate against such artefacts.

Figure 7. Column density map of the MPIPAF observations of the Large Magellanic Cloud (LMC) overlaid with the yellow contours taken from the archival map from Staveley-Smith et al. (Reference Staveley-Smith, Kim, Calabretta, Haynes and Kesteven2003). The contours are (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9) × 5.58 × 10²¹ atoms/cm⁻². The beam size is shown with the black ellipse in the top left corner.

In Figure 7, we compare the MPIPAF column density map with contours from the archival multibeam observations from Staveley-Smith et al. (Reference Staveley-Smith, Kim, Calabretta, Haynes and Kesteven2003). The multibeam contours match well with the MPIPAF image. One thing of note in our map with the MPIPAF data is the edge effects at the top and bottom of the map, which are due to the gridding process under/over-estimating the flux along the edges.

Figure 8 shows a comparison of the brightness temperature values in the individual pixels of the MPIPAF and multibeam image cubes, excluding any boundary regions. There is excellent agreement with the MPIPAF temperature having a small (~1.25 K) zero-point offset. The zero-point offset is mainly due to the in-scan bandpass calibration procedure adopted. This has the effect of removing uniform background emission.

Figure 8. Pixel-by-pixel comparison of temperatures in the MPIPAF and multibeam image cubes. The number of pixels compared is 96,000. The line of best fit is shown in solid red. The shading indicates the data point density, with lighter shading indicating increasing density.

3.5. GAMA G23 field

We extracted the spectrum for each optically identified galaxy listed in NED for the G23 field in the redshift range 0.003 ⩽ z ⩽ 0.23, covering the available bandpass of the MPIPAF band 2. In each channel, we averaged the flux from a 9 pixel box (3 × 3 pixel − 12 × 12 arcmin) centred on the galaxy to ensure we did not lose any flux. We then extracted 600 channels (~11 MHz) around the central redshifted frequency for each galaxy.

3.5.1. H I stacking

We perform H i stacking using the H i mass spectra rather than the originally extracted flux density spectra. We computed the observed-frame H i mass spectrum following Equation 1 from Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013),

(2) $$\begin{equation} \begin{array}{l}\displaystyle \frac{M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}},\nu _{\mathrm{obs}}}}}{M_{\odot }\,\mathrm{MHz}^{-1}}=4.98\times 10^7\left(\frac{S_{\nu _{\mathrm{obs}}}}{\mathrm{Jy}}\right)\left(\frac{D_L}{\mathrm{Mpc}}\right)^2, \end{array} \end{equation}$$

where S _νobs is the observed-frame flux density and D _L is the luminosity distance.

To stack spectra, all spectra must be shifted and aligned at the rest frequency, 1420.406 MHz. We do this by shifting the spectral axis from observed to rest frame [i.e. ν_rest = ν_obs(1 + z)] and to conserve total mass,

(3) $$\begin{equation} \begin{array}{l}\displaystyle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}},\nu _{\mathrm{rest}}}}=\frac{M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}},\nu _{\mathrm{obs}}}}{1+z}. \end{array} \end{equation}$$

We compute the stacked H i mass spectrum as done by Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013),

(4) $$\begin{equation} \begin{array}{l}\displaystyle M_{\mathrm{stacked,{\it i}}}=\frac{\sum _i w_i M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}},\nu _{\mathrm{rest}},{\it i}}}}{\sum _i w_{\it i}}, \end{array} \end{equation}$$

where $M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}},\nu _{\mathrm{rest}},i}}$ is an individual galaxy’s H i mass in channel i, M _{stacked, i} is the final stacked mass in channel i, and w _i is the weight given by

(5) $$\begin{equation} \begin{array}{l}\displaystyle w_i=\frac{1}{\sigma _i^2 D_L^4}, \end{array} \end{equation}$$

where σ _i is the RMS noise in channel i, which we calculated using the miriad task imstat. We removed RFI contamination from the extracted spectra as described in Section 3.3 (excluded channels with S _νobs > 5σ _i and excluded spectra containing GPS satellite RFI at 1376–1384 MHz). We then fit and subtracted a fourth-order polynomial to the stacked spectra to leave a flat baseline in the final spectrum.

We stacked the $M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}}$ spectra in redshift bins of 0.05 up to z = 0.20, with the final bin, 0.20 ⩽ z ⩽ 0.23. We determined the RMS noise in the stacked spectra by randomising and reassigning each redshift to a different pair of coordinates from the input NED catalogue, ensuring that no redshift was assigned to its original coordinates. We then extracted and stacked these mock spectra identically to the galaxy spectra. The stacked mock spectra should not result in a positive detection, as our mock spectra are not centred on galaxies, and should give an approximation of the RMS noise. We performed the randomised stacking 10 times in each redshift bin and inspected each mock stacked spectrum for a possible signal mimicking a detection.

There was no detection in the 0.00 ⩽ z ⩽ 0.05 bin after excluding two direct detections at z = 0.0043 and z = 0.0055 (see Section 3.6.2). Due to the increasing RFI levels at z > 0.10, there were no direct or indirect H i detections in these redshift bins.

We found a detection for ~1100 stacked galaxies at 0.050 ⩽ z ⩽ 0.075. This signal was not found to be mimicked in the mock spectra from random lines of sight. In Figure 9, we plot the stacked galaxy spectrum (blue) and the random mock spectrum (brown). Both spectra exhibit similar residual baseline curvature.

Figure 9. Stacked $M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}}$ spectrum (blue) for 1 094 galaxies at 0.05 ⩽ z ⩽ 0.075. The average mock spectrum from randomising the redshifts of the NED catalogue and stacking the spectra (shown in brown). The noise level in the mock spectrum is lower than that of the data as it is the mean of 10 simulations. The dashed green and grey lines indicate the left and right edges of the stacked H i emission determined by visual inspection and the rest frame H i line, respectively.

Our stacked H i detection spans the range from 1419.8–1420.8 MHz (shown in Figure 9 by the dashed green lines), which is significantly narrower than the Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013) South Galactic Pole stacked spectra (i.e., ~1 MHz vs. ~3.6 MHz for z = 0.05–0.075 and z = 0.04–0.13, respectively). Our stacked detection is most likely narrower than the results of Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013) because of our smaller sample size (i.e., ~1/3 that of the South Galactic Pole region from Delhaize et al. Reference Delhaize, Meyer, Staveley-Smith and Boyle2013) and lower redshift range, hence less confused sources entering our sample.

We integrated over the emission region to calculate the average H i mass of our stacked galaxies using

(6) $$\begin{equation} \begin{array}{l}\displaystyle \langle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}}\rangle =\int ^{\nu _2}_{\nu _1} \langle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}},\nu }\rangle d\nu , \end{array} \end{equation}$$

where ν₁ and ν₂ are the edges of the emission region. We find an average integrated H i mass of $\langle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}} \rangle = 1.24 \pm 0.18 \times 10^9 h^{-2} M_{\odot }$ . Similar to Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013), we computed the error in $\langle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}} \rangle$ by integrating the $\langle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}} \rangle$ random mock stack. We find our $\langle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}} \rangle$ value to be lower than the South Galactic Pole value of $\langle M_{\mathrm{H{\fontsize{4}{5}\selectfont{{\I}}}}} \rangle = 6.93 \pm 0.17 \times 10^9 h^{-2} M_{\odot }$ from Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013), indicating we have detected lower mass galaxies.

We investigated the noise behaviour of the stacked MPIPAF data by stacking the mock random line of sight flux density spectra, as described previously. We calculated the RMS of a stack of N randomly chosen spectra to determine the noise behaviour with increasing number of stacked spectra. To estimate the noise behaviour, we fix the weighting of each spectra using the mean RMS calculated from 1 000 individual mock spectra (σ = 0.142 Jy) over the frequency range 1321–1352 MHz (corresponding to the redshift range of the stacked H i detection). Figure 10 shows the change in RMS noise in the stacked spectrum with number of spectra included in the stack with the error bars calculated as the 1σ standard deviation of the RMS for 100 random stacks of N spectra. We find the noise decreases as expected for Gaussian noise with a gradient of −0.49 ± 0.01 for the MPIPAF data. This is similar to the noise behaviour for the Parkes multibeam data (gradient ~− 0.5) from Delhaize et al. (Reference Delhaize, Meyer, Staveley-Smith and Boyle2013).

Figure 10. The RMS noise in the stacked flux density signal vs. the number of stacked spectra. The error bars denote 1σ errors on the RMS. The dashed line shows the expected trend, assuming Gaussian noise, in decrease in noise with number of spectra with a gradient of −0.5.

3.6. Comparison with HIPASS detections

We present results of targeted observations of two HIPASS sources, J1413-65 (Circinus) and J1909-63A (NGC 6744) from August 3 and September 4, respectively, observed with the MPIPAF, in addition to two HIPASS sources we detected in the GAMA G23 field, J2242-30 (NGC 7361) and J2309-30 (ESO 469-G015). We obtain HIPASS spectra from the archival HIPASS data cubes from the first HIPASS data release (Meyer et al. Reference Meyer2004).

3.6.1. Targeted HIPASS observations

The Circinus galaxy was only observed with the central MPIPAF beam providing a single spectrum. We therefore compare the MPIPAF observation with the pencil beam spectrum along the same line of sight through the galaxy from the archival HIPASS data cube, as this galaxy is resolved and not contained within a single beam. The MPIPAF and HIPASS spectra agree well in shape and flux density with the only difference in that we detect some additional flux on the high frequency end of the spectrum ( $\nu \rlap{>}{\lower1.0ex\hbox{$\sim $}}1418.5$ MHz, top panel of Figure 11). This is most likely due to a slight position offset between the nearest HIPASS data cube pixel to the MPIPAF spectrum position (HIPASS pixel position: α, δ = 213.41°, − 65.35°, MPIPAF line of sight position: α, δ = 213.36°, − 65.31°).

Figure 11. Targeted HIPASS galaxy line of sight spectra of Circinus and NGC 6744, panels (a) and (b), respectively. The Circinus spectrum is from a single line of sight, while the NGC 6744 spectrum is the integrated line of sight spectrum from the 16 MPIPAF beams. The MPIPAF and HIPASS spectra are shown in blue and brown, respectively. The dashed green lines indicate the edges of the galaxy emission determine by visual inspection.

Although the NGC 6744 observation utilized all 16 MPIPAF beams, not all 16 beams lie upon the galaxy. NGC 6744 has an angular radial size in H i of ~15 arcmin (~30 arcmin in diameter, Ryder, Walsh, & Malin Reference Ryder, Walsh and Malin1999), while 13 of the MPIPAF beams have angular separations >30 arcmin from the centre of NGC 6744. We integrated the flux from the three beams with angular separations <24 arcmin. We also calculated the total integrated flux density from the archival HIPASS cube within a 13 × 13 pixel box centred on NGC 6744 (maximum angular separation 24 arcmin, matching the separation of the MPIPAF beams). The spectral shape and flux density of the integrated MPIPAF spectrum agrees with the HIPASS spectrum (lower panel of Figure 11).

3.6.2. G23 H I detections

We have two direct H i detections of HIPASS detected galaxies, NGC 7361 and ESO 469-G015, at z = 0.0043 and z = 0.0055 (Figure 12 panels (a) and (b), respectively). We found these direct detections through visual inspection of the H i spectra extracted from the G23 field based on optical identifications from NED. Unlike the targeted HIPASS galaxy observations, we can compare the total integrated flux for these two galaxies as they are completed covered by the drift scan. We compare our MPIPAF spectra with the HIPASS spectra integrated over the same area from the archival data cubes (20 × 20 arcmin) for these two galaxies in Figure 12 (blue and brown lines, respectively). The integrated MPIPAF spectrum of NGC 7361 shows very good agreement with the HIPASS data both in spectral shape and flux density. While the integrated MPIPAF spectrum of ESO 469-G015 has a similar spectral shape, it has a slightly lower flux density. Nevertheless, both spectra agree within the combined uncertainties (Table 3).

Figure 12. Direct H i detection of HIPASS galaxies NGC 7361 and ESO 469-G015 at z = 0.0043 and z = 0.0055, panels (a) and (b), respectively. The MPIPAF and HIPASS integrated spectra are shown in blue and brown, respectively. The dashed green lines indicate the edges of the galaxy emission determine by visual inspection.

Table 3. Integrated fluxes for MPIPAF and HIPASS galaxy spectra shown in Figure 12.

3.7. Galactic centre hydrogen and positronium recombination lines

We inspected the Galactic Centre spectra from each beam at the 14 Hydrogen and 11 positronium RRL frequencies predicted from the Rydberg equation within the band (Table 4). We were unable to use large sections of the spectra due to RFI contamination. Also present in the spectra are 1-MHz beamformer ‘jumps’, some of which were not removed during bandpass calibration. These 1-MHz ‘jumps’ have also been seen in early ASKAP data and are not caused by RFI or the Parkes dish, but are due to the discretization of the beamformer weights. We were able to model and remove the spectral shape of the ‘jumps’ with a first-order polynomial fit to each 1-MHz spectral interval as the ‘jumps’ occur at exactly 1-MHz intervals (e.g., 1380.5 MHz, 1381.5 MHz, 1382.5 MHz, etc.). We find detections for the H165α, H166α, H167α, H168α, and H170α hydrogen recombination lines in individual MPIPAF beams (Figure 13(a) shows the stacked signal for these five lines in all 16 MPIPAF beams). In Table 5, we list calculated line parameters from Gaussian fits to the spectra in Figure 13(a). Our calculated peak intensity line temperatures agree with previous single dish Galactic Centre hydrogen RRL studies (Table 5). All other hydrogen recombination lines lie within regions with high RFI contamination and are undetected.

Figure 13. Stacked Galactic Centre hydrogen and positronium spectra, panels (a) and (b), respectively. For hydrogen, we stacked the spectra from all 16 MPIPAF beams, for individual recombination lines and recovered a detection for the 165α, H166α, H167α, H168α, and H170α lines. The positronium spectrum is the combined stack of the Ps131α, Ps132α, Ps133α, and Ps135α lines in all 16 MPIPAF beams and does not show a detection.

Table 4. Rydberg numbers (n) and corresponding hydrogen (H) and positronium (Ps) frequencies.

Table 5. Galactic Centre FWHM line widths and line temperatures (T _L) for hydrogen radio recombination line (RRL) detections from (a) this work and T _L from previous studies, (b) Roberts & Lockman (Reference Roberts and Lockman1970), (c) Riegel & Kilston (Reference Riegel and Kilston1970), (d) Kesteven & Pedlar (Reference Kesteven and Pedlar1977), (e) Hart & Pedlar (Reference Hart and Pedlar1980).

For positronium, however, we did not have any clear direct detections above the noise. In an attempt to improve the signal to noise, we stacked the Galactic Centre spectra from each beam centred on the predicted frequencies of positronium recombination lines to look for a stacked detection. We were unable to stack spectra at all predicted frequencies due to the presence of RFI, as mentioned above. We excluded the RFI contaminated sections of the spectra. This reduced the number of stacked positronium spectra to 64, centred on the predicted Ps131α, Ps132α, Ps133α, and Ps135α lines. For each spectrum, we extracted 450 channels centred on the predicted line frequency and fit a second-order polynomial to remove the baseline from the stacked spectrum. We find no detection in the stacked positronium spectra in either emission or absorption as shown in Figure 13(b) and set a 3σ upper limit on the stacked recombination line signal of <0.09 K. Using this upper limit, we calculate the recombination rate to be <3.0 × 10⁴⁵ s⁻¹, assuming the positronium line would have a FWHM of 4.2 MHz, as the positronium line width is thermally broadened to 30 times the hydrogen line width at 1 400 MHz. The positronium upper limit improves upon the results of Anantharamaiah et al. (Reference Anantharamaiah, Radhakrishnan, Morris, Vivekanand, Downes and Shukre1989), who placed a 3σ upper limit of <29.3 K (recombination rate <1.1 × 10⁴⁴ s⁻¹, assuming a line width FWMH of 4.2 MHz) on the detection of the positronium Ps133α line from the Galactic Centre from VLA observations. It should be noted that the recombination rate upper limit from Anantharamaiah et al. (Reference Anantharamaiah, Radhakrishnan, Morris, Vivekanand, Downes and Shukre1989) is lower than our value due to the differing flux limits (<3.4 mJy vs. <84 mJy from the VLA and Parkes, respectively), which is a result of the arcsecond vs. arcminute resolution of the VLA (~12 × 6 arcsec) and Parkes (~15 × 15 arcmin), respectively.