3.1 Phase and amplitude calibration

For each of the SMART observations, there are a handful of standard calibration observations taken before and after the nominal voltage (VCS) recording. These are required in order to derive the relative complex amplitudes and phases of each MWA tile (as a function of frequency channel), which are a critical part of the beamforming process. This collection of tile-based amplitudes and phases across the observing bandwidth are the ‘calibration solution.’ We transfer these calibration solutions from the standard calibrator field to the SMART observation field. The MWA is generally phase-stable on reasonably long-time scales, and our experience has been that the use of a single calibration solution works well over $\sim$ 4 800-s observing durations of SMART pointings.

To extract the calibration solution, we used the purpose-built birli Footnote ^e and mwa_hyperdrive Footnote ^f packages on the dedicated calibration observation. First, birli is used to flag known misbehaving tiles (based on metadata collected by the observatory during observing sessions), apply passband, digital gain, and geometric corrections, perform basic radio frequency interference (RFI) mitigation, and average the visibilities to 2-s/40-kHz resolution. These pre-processed visibilities are then passed to mwa_hyperdrive, along with nominal sky and primary beam models, which attempts to solve for the amplitude and phase of each tile at each frequency channel and incorporates ionospheric refractive source shifting effects (Jordan et al. Reference Jordan2025), i.e. bulk source position offset correction across the field-of-view. This amounts to a direction-independent calibration procedure that is equivalent to an iterative application of the classic antsol approach (e.g. Bhatnagar & Nityananda Reference Bhatnagar and Nityananda2001; Mitchell et al. Reference Mitchell, Greenhill, Wayth, Sault, Lonsdale, Cappallo, Morales and Ord2008). The calibration solutions obtained in this manner are collated at a central repository and can be used to expedite future re-processing. In general, there are two or more dedicated calibrator observations for each SMART observation, and the one that maximises the detection sensitivity of known bright pulsars in the field is adopted as the preferred calibration solution for a given observation.

3.2 Tied-array beamforming with VCSBeam

The algorithm used for the beamforming operation in vcsbeam Footnote ^g is identical to that described by Ord et al. (Reference Ord, Tremblay, McSweeney, Bhat, Sobey, Mitchell, Hancock and Kirsten2019), with the addition of the multi-pixel functionality (Swainston et al. Reference Swainston, Bhat, Morrison, McSweeney, Ord, Tremblay and Sokolowski2022) and the ability to invert the polyphase filterbank (PFB) to produce coarse-channelised VDIF (VLBI data interchange format) data (McSweeney et al. Reference McSweeney2020). To summarise, the basic beamforming operation involves:

1. 1. Applying predetermined calibration solutions to complex-valued voltages, to account for differences between the responses and signal paths of each MWA tile and polarisation (Section 3.1);

2. 2. Adding a term to the phases of the voltages to account for the geometric delay of a signal coming from the target look-direction;

3. 3. Summing the voltages from each tile/polarisation together; and

4. 4. (Optionally) converting the dual polarisation beamformed voltages to Stokes parameters.

The advent of the MWAX backend removed the fine-channelisation (i.e. the second PFB stage) step that was previously performed by dedicated field programmable gate array (FPGA) boards on site before the signal was passed to either the correlator or the VCS backend (cf. Tremblay et al. Reference Tremblay2015). Consequently, MWAX voltage data are written to disk as coarse-channelised, Nyquist-sampled voltages (1.28 MHz). We therefore re-implemented the PFB in software and added it as a new module to vcsbeam (see Figure 2). In doing this, backwards compatibility is maintained: i.e. the PFB is only applied if the VCS observation is of the MWAX VCS type. In the case of legacy VCS data, the sampled voltage time series is already fine channelised (i.e. 10-kHz/100- $\unicode{x03BC}$ s resolutions).

The PFB implementation in vcsbeam can be run in two additional modes that were not available in the FPGA implementations. Firstly, the output voltages can be either typecast into 4-bit complex integers (as in the FPGA implementation), or retain the full 64-bit floating point (‘double’) precision used during the calculation. In terms of computing resources (run-time, memory), there is no disadvantage to using the latter, since the voltages stay in memory, and are promoted to 64-bit floats during beamforming in any case. We have found, however, that the retention of the full precision makes negligible difference to the signal-to-noise ratio ( $\mathrm{S}/\mathrm{N}$ ) of the final beamformed products. Secondly, the asymmetric rounding scheme of the FPGA PFB, described in the appendix of McSweeney et al. (Reference McSweeney2020), has been replaced with symmetric rounding; however, the original asymmetric scheme is still available as an option. To keep the data products from SMART as homogeneous as possible, we have included a specialised ‘SMART’ option which reproduces the FPGA implementation of the PFB exactly. The deleterious effects of demoting to 4-bit integers and using an asymmetric rounding scheme are found to be marginal.

Some aspects of vcsbeam have been improved to make better use of the available compute resources, especially when executed on a supercomputing infrastructure. For instance, vcsbeam makes use of Message-Passing Interface (MPI) to handle the embarrassingly parallel processing of multiple coarse channels, as well as gathering the channels together to write out a single psrfits file. Multiple seconds of data, corresponding to several files for each coarse channel, are read in parallel using CPU threads and buffered into computer memory. The approach leverages the high data-transfer bandwidth of a distributed filesystem while limiting the impact of latency and resource contention. The core beamforming GPU kernel itself has also evolved to employ modern GPU programming techniques that make more efficient use of modern GPUs. Benchmarked on an AMD MI250X GPU of Pawsey’s Setonix supercomputer, the execution time of the compute kernel has improved by a factor of 8 compared to the original GPU implementation on the same hardware. Further, vcsbeam can be compiled for both NVIDIA and AMD GPUs, allowing the software to run on the majority of modern supercomputing infrastructure.

4.1 Dedispersion and folding

For the standard data analysis of non-recycled pulsars presented in this paper, we incoherently dedisperse and fold psrfits files using dspsr at the native frequency channel resolution ( $10\, \mathrm{kHz}$ ) with 10-s subintegrations and 256 phase bins. The folded data are stored in the psrfits archive format, enabling flexible data analysis with psrchive. In contrast with the non-recycled pulsar detections, the MSP detections in paper III were phase-coherently dedispersed and folded with up to 1 024 phase bins. However, the flux densities, dispersion measures (DMs), and rotation measures (RMs) have been measured using the same methods for both the non-recycled pulsars and the MSPs. These methods will be briefly summarised here for completeness.

4.2 RFI mitigation

The exceptionally radio-quiet nature of the Observatory site (in comparison to most other observatories) and our preferential strategy for SMART survey observations to be scheduled during nightly hours make the mitigation of RFI non-essential for the majority of routine pulsar data processing. Indeed, sensitive pulsar searches will benefit from a careful assessment of the periodic RFI environment (and devising suitable steps to mitigate it), and this is part of the SMART processing plans under active development. Typically, much of the SMART data are remarkably clear of RFI, and even extreme cases (due to either impulsive or narrowband type RFI) are limited to a small fraction ( $\lesssim$ 10%) of the data. Importantly, having access to full-resolution data allows presto tools (such as rfifind) or psrchive tools (such as paz, pazi, or psrzap) to be effectively used to excise any RFI present in our data. Oftentimes, this tends to be science-dependent, and so we prefer to retain that flexibility for users. However, for folded data products, we have opted to perform automated RFI mitigation using clfd (Morello et al. Reference Morello2019). In our testing, this is sufficient to remove the majority of obvious RFI.

Table 3.

Summary of the 205 non-recycled pulsars re-detected in SMART observations. For each pulsar, we list the J-name, the B-name (where available), the Galactic longitude and latitude (l, b), the spin period, and the DM from pdmp. For the best detection of each pulsar, we list the $\mathrm{S}/\mathrm{N}$ reported by psrchive, the mean and peak flux densities at 154 MHz ( $S_\mathrm{mean}$ and $S_{p}$ ), the observed RM and the estimated (ionosphere-corrected) RM from the ISM, the estimated pulse widths at 10% and 50% of the peak flux density, the SMART pointing label (see Table 2), the integration time ( $T_\mathrm{obs}$ ), and the mean offset from the pointing centre.

Table 3. Long description

Table with multiple columns listing details of 205 non-recycled pulsars. Columns include J-name, B-name, Galactic longitude and latitude (l, b), spin period, DM from pdmp, mean and peak flux densities at 154 MHz, observed RM, estimated RM from the IISM, pulse widths at 10% and 50% of the peak flux density, SMART pointing label, integration time, and mean offset from the pointing centre.

Notes: $^{*}$ More precise than the measurement in the ATNF pulsar catalogue (v2.7.0).

$^{{\#}}$ No previous RM measurement available in the ATNF pulsar catalogue (v2.7.0).

$^{\#\#}$ Main and interpulse measurements.

$^{\dagger}$ Since the main and interpulse overlap for the Crab pulsar, the $W_{10}$ is made across both pulses.

4.3 Improved dispersion measures

The $\sim$ 20% fractional bandwidth of SMART data ( $30.72\,\mathrm{MHz}$ centred at $154.24\, \mathrm{MHz}$ ) is quite comparable to previous pulsar surveys in the southern skies. For instance, the Parkes multibeam surveys employed 288 and $380\, \mathrm{MHz}$ of bandwidth for the legacy and HTRU surveys, respectively (Manchester et al. Reference Manchester2001; Keith et al. Reference Keith2010). The inverse-frequency-square law dependence of the dispersion delay ( $t _\mathrm{dm} \propto \nu^{-2}$ , where $\nu$ is radio frequency) still means a much longer frequency lever arm in the SMART frequency band, which allows us to make precise measurements of DMs, even from a single detection with a modest $\mathrm{S}/\mathrm{N}$ . Indeed, higher precisions (down to $\sim 10^{-6}\mathrm{cm^{-3}\,pc}$ ) can be achieved when observations are made in a distributed channel set up, and with phase-coherent dedispersion techniques, as we have demonstrated in observations of MSPs (Kaur et al. Reference Kaur2019, Reference Kaur, Ramesh Bhat, Dai, McSweeney, Shannon, Kudale and van Straten2022). For non-recycled pulsars in the SMART data, we typically limit our processing to incoherent dedispersion within the $30.72\, \mathrm{MHz}$ contiguous bandwidth.

The improved DMs obtainable from the SMART data are especially important for two main reasons. Firstly, in cases where the catalogued DMs are inaccurate or imprecise, these low-frequency measurements may provide better measurements. Secondly, in the case of frequency-dependent DMs (cf. Cordes et al. Reference Cordes, Shannon and Stinebring2016) that arise from multipath propagation effects, DM measurements at low frequencies may differ in measurable ways to those from higher radio frequency observations (e.g. Kaur et al. Reference Kaur, Ramesh Bhat, Dai, McSweeney, Shannon, Kudale and van Straten2022). Aside from this subtlety, it is conceivable that profile evolution in frequency can potentially result in some systematic bias between the measured DMs. With observations using wide-band instrumentation becoming a common practice, this seems especially relevant (e.g. Hassall et al. Reference Hassall2012). We therefore measure and tabulate DM measurements from our SMART detections.

As with our earlier partial census published in Paper II, the vast majority of our pulsar detections are in the signal-to-noise ratio ( $\mathrm{S}/\mathrm{N}$ ) range $\sim$ 10–100 in 30–80 min observations over the full SMART frequency band. For the purpose of DM determination, we make use of the pdmp routine from psrchive to perform a fine search in period and DM. We perform the search with 10-s subintegrations and $320\,\mathrm{kHz}$ subbands with a DM step sizeFootnote ⁱ of $0.005\,\mathrm{cm^{-3}\,pc}$ (for non-recycled pulsars) or $0.001\,\mathrm{cm^{-3}\,pc}$ (for MSPs). When a DM correction of 2 $\sigma$ or greater is measured, we correct the residual dispersion delays using the pam routine. A drift in period is seldom seen in our data as we make use of most up-to-date pulsar ephemerides available, either from MeerTimeFootnote ^j (Keith et al. Reference Keith2024a,Reference Keithb) or from the ATNF Pulsar Catalogue (Manchester et al. Reference Manchester, Hobbs, Teoh and Hobbs2005). In the rare cases where a period correction is needed, we use the $\mathrm{S}/\mathrm{N}$ -optimised value reported by pdmp.

Figure 3.

Example spline fits to the integrated pulse profiles of PSRs B0818 $-$ 41 (top left), B1659 $-$ 60 (top right), B2045 $-$ 16 (bottom left), and B2327 $-$ 20 (bottom right). The top panel of each subfigure shows the measured profile (black connected dots) and the spline fit (red lines), whilst the bottom panel shows the fit residuals. The insets show the heuristic ‘whiteness’ score of the residuals for a range of noise trials, $\hat{\sigma}_i$ , which determine the smoothing factor $s=N_{b}\hat{\sigma}_i^2$ (see Section 4.4 for details). The spline fit to B2327 $-$ 20 exhibits oscillations at the leading and trailing edges due to the sharp transitions.

4.4 Integrated pulse profiles and pulse widths

For each pulsar detection, we average the subintegrations and frequency channels to produce an integrated pulse profile. The detected sample of pulsars includes a diverse range of profile shapes, including a number of profiles with broadening due to multipath propagation in the ISM. The complexity of the profiles, along with the range of $\mathrm{S}/\mathrm{N}$ ratios, makes it challenging to reliably measure pulse widths without component fitting, which is difficult to reliably automate. Nevertheless, measurements of pulse widths are important for studies of the intrinsic frequency-dependence of pulsar emission and scattering in the ISM.

Considering that we primarily want to model the profiles to estimate the pulse widths and onpulse phase regions, we have chosen to model each profile by fitting a smoothing spline; i.e. a smooth curve constructed from multiple polynomial segments that are joined seamlessly at boundary points called ‘knots’. We use the splrep functionFootnote ^k from scipy to find the number and location of the knots, and the coefficients of the basis polynomials. This approach enables the straight-forward computation of profile derivatives, which can be used to estimate the onpulse/offpulse. We fit splines of degree 5 (so that the second derivative is a cubic spline) with periodic boundary conditions at the edges of the pulse profile.

The splrep fitting algorithm takes a smoothing condition, s, which determines the ‘rigidity’ of the spline; it should be chosen such that the noise in the profile is removed whilst retaining as much pulse structure as possible (i.e. to avoid over- or under-fitting). In general, a good choice of smoothing condition is $s\sim N_{b}\hat\sigma^2$ , where $N_{b}$ is the number of phase bins and $\hat\sigma$ is the standard deviation of the noise in the profile. To find the best-fit spline, we performed a log-spaced grid search in the range $\hat\sigma \in [0.2\hat\sigma_\mathrm{est}, 5\hat\sigma_\mathrm{est}]$ , where $\hat\sigma_\mathrm{est}$ is an initial estimate of $\hat\sigma$ . We determined $\hat\sigma_\mathrm{est}$ by finding the phase window of width $N_{b}/8$ bins with the minimum integrated flux density, and measuring the standard deviation of the sample values in that window. For each trial i, we fit a spline with $s_i = N_{b}\hat\sigma_i^2$ and performed the following tests:

• • a Ljung-Box test, where the null hypothesis is that all of the autocorrelations of the residuals up to a lag of 20 samples are jointly zero; and

• • a Wald-Wolfowitz runs test, where the null hypothesis is that the sequence of signs of the residuals above or below the mean is random.

We then multiplied the p-values of the two tests to determine a heuristic score for the ‘whiteness’ of the residuals.Footnote ^l The spline with the maximum score was selected as the best fit. Our testing indicates that this method yields reasonable fits for a wide range of profile shapes. In Figure 3 we show the best-fit smoothing splines for four pulsars with a variety of profile features. PSRs B0818 $-$ 41 and B1659 $-$ 60 have broad features that are well modelled by a more rigid spline, whereas PSRs B2045 $-$ 16 and B2327 $-$ 20 have sharper features (relative to the bin width). The best-fit spline for B2327 $-$ 20 exhibits ‘Runge’s phenomenon’, where sharp transitions induce oscillations in high-degree polynomials (in this case degree 5) fitted to equispaced samples. Increasing the number of phase bins would likely improve the fits; however, we have opted to uniformly process with 256 phase bins for this work. On the other hand, for pulsars with structure at a similar timescale to the bin width, the smoothing spline interpolates peaks and troughs in the profile (see B2045 $-$ 16 and B2327 $-$ 20 in Figure 3). In some cases, this may improve the estimate of the peak amplitude. However, for the peak flux densities listed in Table 3 and Figure 4, we have used the peak of the data rather than the spline, as we are not confident that the spline fits consistently give more accurate estimates of the peak amplitude.

To estimate the onpulse regions, we identified the maxima of the spline which exceed $3\hat\sigma$ , and then the corresponding minima of the spline which flank these maxima. The intervals between the flanking minima were considered part of the onpulse. If none of the samples between two of the initially identified onpulse regions fell below $1\hat\sigma$ , the two regions were considered to be part of the same onpulse region. Lastly, if multiple pulse components were identified nearby in phase (e.g. PSRs B0148 $-$ 06 and B0957 $-$ 47 in Figure 4), they were considered part of the same onpulse region – i.e. interpulses were treated separately where they could be identified, but otherwise the components were considered part of the same onpulse region.

Spline fitting enables estimation of the profile widths by solving for the roots at a given amplitude. As such, for all detections with peak amplitudes greater than $2\hat\sigma$ and $10\hat\sigma$ , respectively, we measured $W_{50}$ and $W_{10}$ (i.e. the pulse width at 50% and 10% of the peak amplitude). For consistency, the peak amplitudes used for estimating the 50% and 10% amplitudes were measured from the spline fits. Each profile fit was visually inspected to verify that the pulse width measurements are reliable.

Figure 4.

Integrated pulse profiles for the best detection of each of the 205 non-recycled pulsars in SMART observations. For each pulsar, we list the B-name (or if unavailable, we use the J-name instead), the spin period in ms, the DM in cm^{− 3} pc, and the peak flux density $S_{p}$ in Jy (the uncertainty in the last digit is listed in parenthesis, rounded to 1 significant figure).

Figure 4. Long description

Panel A: A line graph shows the integrated pulse profile for the pulsar B1238+2152. The horizontal axis ranges from 0 to 1, representing the pulse phase, and the vertical axis represents the flux density. The graph shows a peak around 0.5. Panel B: A line graph for the pulsar B1237+25 shows a peak around 0.3 on the horizontal axis, which ranges from 0 to 1, and the vertical axis represents the flux density. Panel C: A line graph for the pulsar B1236-68 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel D: A line graph for the pulsar B1237-41 shows a peak around 0.4 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel E: A line graph for the pulsar B1254-10 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel F: A line graph for the pulsar B1309-12 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel G: A line graph for the pulsar B1309-53 displays a peak around 0.3 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel H: A line graph for the pulsar J1313+0931 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel I: A line graph for the pulsar B1317-53 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel J: A line graph for the pulsar B1325-43 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel K: A line graph for the pulsar J1335-3642 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel L: A line graph for the pulsar B1336-64 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel M: A line graph for the pulsar B1352-51 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel N: A line graph for the pulsar B1358-63 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel O: A line graph for the pulsar J1418-3921 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel P: A line graph for the pulsar B1424-55 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel Q: A line graph for the pulsar B1426-66 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel R: A line graph for the pulsar B1436-63 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel S: A line graph for the pulsar B1449-64 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel T: A line graph for the pulsar B1451-68 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel U: A line graph for the pulsar B1504-43 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel V: A line graph for the pulsar B1507-44 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel W: A line graph for the pulsar B1510-48 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel X: A line graph for the pulsar B1524-39 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel Y: A line graph for the pulsar B1552-53 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel Z: A line graph for the pulsar B1554-48 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel AA: A line graph for the pulsar B1556-44 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel AB: A line graph for the pulsar B1600-27 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel AC: A line graph for the pulsar B1600-49 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel AD: A line graph for the pulsar B1604-00 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel AE: A line graph for the pulsar J1612-2408 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel AF: A line graph for the pulsar B1612+07 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel AG: A line graph for the pulsar B1611-55 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel AH: A line graph for the pulsar J1627+1419 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel AI: A line graph for the pulsar B1630-59 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel AJ: A line graph for the pulsar B1642-03 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel AK: A line graph for the pulsar B1641-68 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel AL: A line graph for the pulsar J1648-3256 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density. Panel AM: A line graph for the pulsar J1654-2713 displays a peak around 0.5 on the horizontal axis, ranging from 0 to 1, with the vertical axis representing the flux density. Panel AN: A line graph for the pulsar B1700-32 shows a peak around 0.5 on the horizontal axis, ranging from 0 to 1, and the vertical axis represents the flux density.

4.5 Mean flux densities

The system sensitivity was simulated following the method described by Meyers et al. (Reference Meyers2017) and Paper III, which involves simulating the tied-array beam over the whole sky. To account for the direction and frequency dependence of the beam, the sensitivity was estimated from the average of four uniformly spaced time and frequency steps, for a total of 16 simulations per observation. It should be noted that this does not fully correct the bias introduced by beam fading (i.e. the change in on-source power over time), so we have limited our analysis to periods when the target is above 20% beam power at 154 MHz. The simulations are used to estimate the noise expected in each phase bin, $\hat{\sigma}_\mathrm{sim}$ . The mean flux density $S_\mathrm{mean}$ was then calculated as follows:

(1) \begin{equation} S_\mathrm{mean} = \frac{1}{N_{b}} \frac {\hat{\sigma}_\mathrm{sim}}{\hat{\sigma}_\mathrm{data}} \sum_{i=1}^{N_{b}} (p_i - \bar{p}_\mathrm{off}),\end{equation}

where $N_{b}$ is the number of phase bins, $\hat{\sigma}_\mathrm{data}$ is the standard deviation of the noise in the profile estimated from the spline-fit residuals (see Section 4.4), $p_i$ is the uncalibrated flux density of the ith phase bin, and $\bar{p}_\mathrm{off}$ is an estimate of the profile baseline calculated from the mean of the offpulse samples. By estimating $\hat{\sigma}_\mathrm{data}$ from the spline-fit residuals, we are able to estimate the noise level reliably, even for pulsars with small or non-existent offpulse regions (e.g. due to scatter broadening). For the pulsars with no offpulse samples (e.g. scattered pulsars such as B0531+21, B0833 $-$ 45, and B1844 $-$ 04), $\bar{p}_\mathrm{off}$ was measured from a short phase window that minimised the integrated flux density. For each mean flux density, we assign a statistical uncertainty $\hat{\sigma}_\mathrm{sim}/\sqrt{N_{b}}$ and a systematic uncertainty of 30% for sources away from the Galactic plane ( $|b|\gt10^\circ$ ) and 40% for sources near or on the Galactic plane (for a discussion on the sources of systematic uncertainty, see Section 3.2.1 of Paper III). For the peak flux density, the statistical uncertainty is simply $\hat{\sigma}_\mathrm{sim}$ . The statistical and systematic uncertainties were summed in quadrature to calculate the total uncertainty.

4.6 Faraday rotation measures

For each detection, we measured the profile-averaged RM following the method described in Paper III. To summarise, we measured the Faraday dispersion function (FDF) for all of the onpulse phase bins, then added the amplitudes of the FDFs together to make a profile-averaged FDF. The RM was determined as the Faraday depth for which the amplitude of the FDF peaked. This procedure was bootstrapped by adding white noise to the data to create a sample of RM point estimates. For low- $\mathrm{S}/\mathrm{N}$ cases with multiple peaks in the bootstrap distribution due to noise, we discarded samples outside of the RM peak, which was typically easy to identify by manually inspecting a histogram of the samples. We then measured the mean and standard deviation of the bootstrap samples to estimate the RM and the 1 $\sigma$ statistical uncertainty.

For pulsars with catalogued RMs below $300\,\mathrm{rad}\,\mathrm{m^{-2}}$ , we performed the search in the range $[{-}300, 300]\,\mathrm{rad}\,\mathrm{m^{-2}}$ with $80\, \mathrm{kHz}$ subbands. For pulsars with unknown RMs or catalogued RMs above $300\,\mathrm{rad}\,\mathrm{m^{-2}}$ , we performed the search in the range $[{-}600, 600]\,\mathrm{rad}\,\mathrm{m^{-2}}$ with $40\, \mathrm{kHz}$ subbands. The chosen frequency resolutions ensure that sensitivity is maintained within the search range (Brentjens & de Bruyn Reference Brentjens and de Bruyn2005). All searches were done with a Faraday depth resolution of $0.2 \,\mathrm{rad}\,\mathrm{m^{-2}}$ , which oversamples the FDF by a factor of $\sim$ 12.5.

The estimated ionospheric contribution $\mathrm{RM}_\mathrm{iono}$ was subtracted from the observed value $\mathrm{RM}_\mathrm{obs}$ to estimate the RM due to the ISM, i.e. $\mathrm{RM}_\mathrm{ISM} = \mathrm{RM}_\mathrm{obs} - \mathrm{RM}_\mathrm{iono}$ . Details on the estimation of $\mathrm{RM}_\mathrm{iono}$ are discussed in Paper III. In short, we use spinifex (Mevius et al. Reference Mevius, Thomson and Dijkema2025) for spatial and temporal interpolation of the total electron content (TEC) using a single-layer ionospheric model with global ionospheric measurements (from the JPL). The uncertainties in $\mathrm{RM}_\mathrm{iono}$ calculated by spinifex are doubled to better reflect the accuracy observed previously in pulsar observations at the telescope site (Lee et al. Reference Lee, Bhat, Sokolowski, Meyers and Magro2024).

5.1 Pulse profiles

The integrated pulse profiles for the best detections (by $\mathrm{S}/\mathrm{N}$ ) of the 205 pulsars re-detected in the SMART data are presented in Figure 4. Since the previous census reported in Bhat et al. (Reference Bhat2023b), many of the pulsar detections have been reprocessed with longer integration times and additional RFI mitigation, and hence are presented here with a higher $\mathrm{S}/\mathrm{N}$ . Figure 5 shows the $W_{10}$ – $W_{50}$ distribution for the non-recycled pulsars for which both width measurements could be made. For two pulsars (J0514 $-$ 4408 and J1057 $-$ 5226), we show the measurements for both the main- and inter-pulse. The remaining measurements are all main pulses. Pulsars with large differences in $W_{10}$ and $W_{50}$ generally fall into two categories: those with significant broadening due to scattering (e.g. B0853 $-$ 33, B1818 $-$ 04, B1920+21); and those with multiple profile components with different amplitudes (e.g. B1237+25, B1451 $-$ 68, B1659 $-$ 60). The marker colours indicate the DM of each pulsar, which helps to identify which measurements fall into these categories. Since pulsars with higher DMs are generally more scattered, we also observe a bias of higher-DM pulsars generally having larger pulse widths.

Measurements of pulse widths at multiple frequencies can also allow the investigation of the frequency evolution of pulse profiles in attempts to explore the emission geometries or more broadly the frequency dependence of pulsar emission. Pilia et al. (Reference Pilia2016) and Kumar et al. (Reference Kumar, Taylor, Stovall, Dowell and White2025) present the related analysis for large pulsar samples at low frequencies with LOFAR and the LWA, spanning 15–167 MHz and 26–88 MHz, respectively. While both studies reveal a general widening of the pulse profiles at lower frequencies, as expected from radius-to-frequency mapping or the birefringence phenomenon, there are many exceptions, and overall there does not appear to be a specific common trend. Further, as acknowledged by Pilia et al. (Reference Pilia2016), it is also important to deconvolve scattering tails before meaningful width comparisons are possible at low frequencies. A detailed analysis along these lines, including the comparison with the width estimates at higher frequencies, is therefore deferred to a future publication where we will report the pulse broadening analysis of our MWA sample. Nonetheless, our width estimates are robust and therefore can serve as useful indicators of profile complexity or the degree of pulse broadening, with larger values of the pulse width ratio $W_{10}/W_{50}$ implying either a multi-component pulse profile or dominant scattering tails.

5.2 Flux Densities

Measurements of mean and peak flux densities ( $S_\mathrm{mean}$ and $S_{p}$ ) obtained using the method described in Section 4.5 are listed in Table 3. We emphasise that these are only first order estimates, as they do not account for any variability due to (refractive) scintillation, and moreover their estimation relies on simulating the beam patterns for telescope performance parameters. Nonetheless, our large sample enables us to infer broader trends in the population, particularly spectral indices and the presence of spectral flattening and turnovers.

In Figure 6, we present a comparison of our flux densities with 22 published measurements from LOFAR (Bilous et al. Reference Bilous2016) and 149 from MeerKAT (Keith et al. Reference Keith2024a). Note that the small overlapping sample with LOFAR is due to their census targeting only northern pulsars. On average, the MWA flux densities are consistent with those from LOFAR, with the outliers likely a result of the aforementioned sources of systematic error (scintillation and calibration). For the detections in common with MeerKAT, we measured two-point power-law spectral indices between the geometric centre frequenciesFootnote ^m of the MWA and MeerKAT bands (154.02 MHz and 1 223.61 MHz, respectively). The spectral indices range from $-0.2$ to $-2.5$ , with a mean of $-1.56 \pm 0.13$ (standard error) and a standard deviation $0.48$ . This is slightly shallower than the mean spectral index based on the study of a larger pulsar population, as measured by Jankowski et al. (Reference Jankowski, van Straten, Keane, Bailes, Barr, Johnston and Kerr2018) to be $-1.60 \pm 0.03$ (standard error) with a standard deviation of $0.54$ when low and high frequency turnovers are modelled out. This may be due to the lack of commonly-detected pulsars with steep spectral indices (e.g. $\lt$ $-2.0$ ), that may be below the detectable flux density of the MeerKAT observations at L-band. The lack of common steep-spectrum pulsars is also reflected in the slightly smaller standard deviation of our sample. It is also possible that the bias is due to some pulsars turning over in the SMART frequency band. However, even if this is the case, the observed bias is small. This may indicate that most pulsars are turning over below 154 MHz, or the pulsars exhibiting significant turnovers in the SMART frequency band were not detected.

Figure 6.

Comparison of the mean flux densities for non-recycled pulsars in SMART data sets (139–170 MHz) with LOFAR mean flux densities (110–188 MHz; Bilous et al. Reference Bilous2016) and long-term averaged mean flux densities from the MeerKAT Thousand-Pulsar-Array (MTPA) monitoring program (896–1 671 MHz; Keith et al. Reference Keith2024a). Top: Scatter plot of flux density measurements compared between telescopes. When comparing with LOFAR, we show the line of equality and the factor-of-two error band. When comparing with MeerKAT, we show lines indicating the two-point spectral index. The colour scale shows the dispersion measure of each pulsar. Bottom: Histogram of two-point spectral indices estimated from MWA and MeerKAT flux densities. We also show the mean spectral index distribution of the known pulsar population measured by Jankowski et al. (Reference Jankowski, van Straten, Keane, Bailes, Barr, Johnston and Kerr2018).

Figure 6. Long description

The image contains three graphs. Panel A: A scatter plot compares flux density measurements between LOFAR and MWA telescopes. The x-axis represents MWA S_mean in millijanskys, ranging from 10^1 to 10^4. The y-axis represents LOFAR S_mean in millijanskys, also ranging from 10^1 to 10^4. The plot includes a line of equality and a factor-of-two error band. The color scale indicates the dispersion measure of each pulsar. Panel B: Another scatter plot compares flux density measurements between MeerKAT and MWA telescopes. The x-axis represents MWA S_mean in millijanskys, ranging from 10^1 to 10^4. The y-axis represents MeerKAT S_mean in millijanskys, ranging from 10^0 to 10^2. The plot includes lines indicating the two-point spectral index. The color scale again shows the dispersion measure of each pulsar. Panel C: A histogram displays the distribution of two-point spectral indices estimated from MWA and MeerKAT flux densities. The x-axis represents the spectral index, ranging from -3 to 0. The y-axis represents the probability density, ranging from 0 to 1.25. The histogram includes a comparison with the mean spectral index distribution of the known pulsar population measured by Jankowski et al. (2018).

There have been several past studies that used low-frequency flux density measurements for exploring the broadband spectra of pulsars. Bilous et al. (Reference Bilous2016) measured flux densities in the LOFAR high band for 158 pulsars and used them along with those from other published work for performing their spectral fits. LWA measurements (10–88 MHz) have also been used for spectral analysis, though that was for modest samples (Stovall et al. Reference Stovall2015; Kumar et al. Reference Kumar, Taylor, Stovall, Dowell and White2025). Bilous et al. (Reference Bilous2016) used single or broken power-law models for their analysis, which yielded a mean spectral index of $-1.4$ . This is somewhat shallower than the mean value of $-1.56$ from our analysis and $-1.6$ from the more comprehensive analysis by Jankowski et al. (Reference Jankowski, van Straten, Keane, Bailes, Barr, Johnston and Kerr2018). Although the geometric centre of the LOFAR frequency band is only $\sim$ 10 MHz lower than the SMART band, turnovers in the lower half of the band (i.e. between 110–150 MHz) could significantly lower the band-averaged flux densities. As such, the intriguing discrepancy in mean spectral index may be interpreted as an indication of spectral flattening or turnover below the SMART band. However, as noted by several authors (e.g. Bilous et al. Reference Bilous2016, Reference Bilous2020; Jankowski et al. Reference Jankowski, van Straten, Keane, Bailes, Barr, Johnston and Kerr2018), the quality and results from spectral fits generally depend on the number and extent (in frequency coverage) of measurements and their sparsity. The frequency coverage of the data, along with the complexity and diversity of pulsar spectra and the uncertainties inherent to estimating low-frequency flux density measurements from beamformed observations, make the conclusions drawn from our analysis tentative. A more detailed analysis of the spectral characteristics of our southern pulsar sample will hence be instructive, and will be reported in a future publication using the pulsar_spectra modelling software and catalogue (Swainston et al. Reference Swainston, Bhat, Morrison, McSweeney, Ord, Tremblay and Sokolowski2022).

5.3 Dispersion and Rotation Measures

Measurements of DM and RM obtained using the methods described in Sections 4.3 and 4.6 are listed in Table 3. For the sample of non-recycled pulsars, we were able to estimate RMs for 117 pulsars. Thus, including RMs for 25 MSPs reported in Paper III, we have RMs measured for 142 pulsars detected in SMART data sets. For five pulsars (PSRs J0026 $-$ 1955, J0038 $-$ 2501, J0418 $-$ 4154, J0600 $-$ 5756, and J1612 $-$ 2408), there are no RM measurements available in the latest pulsar catalogue (v2.7.0; Manchester et al. Reference Manchester, Hobbs, Teoh and Hobbs2005); for these, we present measurements here for the first time (the locations of these pulsars are shown in Figure 10). For the 112 non-recycled pulsars with RM measurements listed in the pulsar catalogue, Figure 7 shows a plot of our measurements against those from the pulsar catalogue (for a similar comparison of the MSP RMs, see Paper III).

Figure 7.

Comparison of RM measurements for non-recycled pulsars in SMART data sets with the RMs reported in the ATNF pulsar catalogue (i.e. PSRCAT; v2.7.0). The top plot shows all pulsars with both MWA and PSRCAT RMs available; the bottom plot shows just those with an RM less than $100\,\mathrm{rad\,m^{-2}}$ , with the corresponding region indicated in the top plot. The highest RMs correspond to PSRs B1727 $-$ 47 and B1907+10 (annotated), which are both located on the Galactic plane and have DMs of $122.9 \,\mathrm{cm^{-3}\,pc}$ and $149.8 \,\mathrm{cm^{-3}\,pc}$ , respectively. Several outliers are annotated: PSRs J0133 $-$ 6957, B0833 $-$ 45, B1352 $-$ 51, and B1524 $-$ 39 (cf. Han et al. Reference Han, Manchester and Qiao1999, Reference Han, Manchester, Lyne, Qiao and van Straten2006; Johnston et al. Reference Johnston, Hobbs, Vigeland, Kramer, Weisberg and Lyne2005).

Low-frequency observations provide the advantage of obtaining high-precision estimates of DM and RM. For example, Figure 8 shows our RM uncertainties compared with the catalogued measurements for the same pulsars. The majority of the catalogue measurements (86 out of 112) come from long-term monitoring with MeerKAT (Keith et al. Reference Keith2024a), which has enabled better precisions than can currently be obtained from single-epoch measurements due to the limited accuracy of global ionospheric models (e.g. a median uncertainty in $\mathrm{RM}_\mathrm{ISM}$ of $0.06\,\mathrm{rad\,m^{-2}}$ for MeerKAT compared with $0.15\,\mathrm{rad\,m^{-2}}$ for SMART). However, as Figure 8 illustrates, even without long-term monitoring, the precision that we can obtain in the SMART data is often comparable or even better than the best available measurements. For the 29 pulsars that we have obtained more precise RM measurements than reported in the catalogue, we note this in Table 3.

Figure 8.

Histograms comparing the uncertainties in RM from this work (before and after ionospheric correction) and the ATNF pulsar catalogue (v2.7.0) for the 112 pulsars with measurements available from both. The subset of catalogue measurements from long-term monitoring with the MeerKAT Thousand Pulsar Array (MTPA; Keith et al. Reference Keith2024a) are hatched.

In order to compare and quantify the offsets relative to the published measurements (from the ATNF pulsar catalogue or other published work), we use the z-score of residuals, which is defined as

(2) \begin{equation} z = \frac{X_\mathrm{ref} - X_\mathrm{meas}}{\sqrt{\sigma_\mathrm{ref}^2 + \sigma_\mathrm{meas}^2}},\end{equation}

where $X_\mathrm{ref}\pm \sigma_\mathrm{ref}$ is a reference measurement (e.g. from the pulsar catalogue or published work) and $X_\mathrm{meas}\pm \sigma_\mathrm{meas}$ is the measured value from SMART data sets presented in this work. A summary of this is shown in Figure 9 that shows a plot of ${z_\mathrm{DM}}$ vs. ${z_\mathrm{RM}}$ , where ${z_\mathrm{DM}}$ and ${z_\mathrm{RM}}$ are the z-scores for DM and RM, respectively. This can be used to identify the outliers. For instance, there are handful of cases with $ | {{z_\mathrm{RM} }} | \gt 20 $ ; e.g. for PSR B1352 $-$ 51 (J1355 $-$ 5153), our measured RM is $-72.6\pm 0.5\,{\mathrm{rad\,m^{-2}}}$ , compared with $55\pm 6\,{\mathrm{rad\,m^{-2}}}$ from Han et al. (Reference Han, Manchester, Lyne, Qiao and van Straten2006). Our RM measurements generally show a good agreement with those from published work (see Figure 7), and we note that a similar agreement was also noted in the earlier work of Riseley et al. (Reference Riseley2020) using image-based measurements (albeit for a much smaller sample). Given that, we are inclined to believe that there may be a sign error in the published RM for B1352 $-$ 51. A particularly striking case is the Vela pulsar (B0833 $-$ 45) with $ | {z_\mathrm{RM} } | \gt 100 $ , though this is not surprising in light of recent observational evidence of the presence of small-scale magnetised plasma structure as reported in Lee et al. (Reference Lee, Bhat, Sokolowski, Meyers and Magro2024). These scores can be hence used to identify anomalous sight lines where either discrete and/or magnetised plasma structures are present. In most cases, the outliers in ${z_\mathrm{DM} }$ (for which we note those with $ | {z_\mathrm{DM}} | \gt 6 $ ) are likely a result of inaccurate DM measurements from past observations (either due to limited S/N or small fractional bandwidths).

Figure 9.

Distribution of residual z-scores (see Equation 2) between the RM and DM estimates from SMART data sets and from the ATNF pulsar catalogue (v2.7.0). Some significant outliers have been annotated.

Figure 10.

Galactic distribution of the line-of-sight component of the mean magnetic field $\langle B_{\parallel} \rangle$ , estimated for a sample of 117 non-recycled pulsars (circles; from this work) and 25 MSPs (squares; from Paper III) using the RM and DM measurements deduced from SMART data sets. The pulsars with RMs measured for the first time in this work are emphasised. The grey dots show all pulsars from the ATNF pulsar catalogue (v2.7.0). Declinations above $+30^\circ$ , which are out of reach of the MWA, are shaded in grey.

Figure 10. Long description

A scatter plot representing the galactic distribution of the line-of-sight component of the mean magnetic field estimated for a sample of 117 non-recycled pulsars and 25 millisecond pulsars using RM and DM measurements. The horizontal axis represents the galactic longitude in degrees, ranging from -90 to +90. The vertical axis represents the galactic latitude in degrees, ranging from -90 to +90. The plot includes circles for non-recycled pulsars and squares for millisecond pulsars. The color of the markers indicates the mean magnetic field strength in microgauss, with a color bar on the right side showing the scale from -4 to +4 microgauss. Grey dots represent all pulsars from the ATNF pulsar catalogue. Declinations above 30 degrees, which are out of reach of the MWA, are shaded in grey. The plot shows a variety of data points spread across the galactic coordinates, with some clustering visible in certain regions.

A summary of our RM measurements is shown in Figure 10 that shows the Galactic distribution of estimated values for the mean magnetic field strength $ \langle B_{\parallel} \rangle $ towards the sight lines. The values are within $ \pm 4\, \unicode{x03BC}$ G, as we may expect for pulsars within the local ISM. A near isotropic distribution of the sample, with a sizeable fraction at high Galactic latitudes, makes them particularly useful in constraining the models for the Galactic magnetic field, both within the halo and the local ISM (e.g. Oswald et al. Reference Oswald, Weltevrede, Posselt, Johnston, Karastergiou and Lower2025).

5.4 Pulse Broadening Times

As seen from Figure 4, many of our pulsar profiles show a significant broadening due to scattering in the ISM. A sizeable fraction of these is located at fairly high Galactic latitudes (see Figure 1), where there is currently a paucity of available measurements. These will be invaluable for constraining the model parameters towards the high-latitude regions (Ocker et al. Reference Ocker, Cordes and Chatterjee2020, Reference Ocker, Cordes and Chatterjee2021), which are particularly important for interpreting the measurements of fast radio bursts (FRBs). As well known, at low radio frequencies, many pulsar profiles tend to evolve to complex multi-component shapes. The deconvolution-based approaches (Bhat et al. Reference Bhat, Cordes and Chatterjee2003; Young & Lam Reference Young and Lam2024) are likely to be more amenable in such cases, rather than traditional forward model approach (Geyer et al. Reference Geyer2017; Krishnakumar et al. Reference Krishnakumar, Mitra, Naidu, Joshi and Manoharan2015). As the deconvolution-based methods make no assumption of the intrinsic pulse shapes, they are likely to yield more reliable estimates of pulse broadening times (Bhat et al. Reference Bhat, Cordes, Camilo, Nice and Lorimer2004), however the peak signal-to-noise ratio needs to be sufficiently high ( $\mathrm{S}/\mathrm{N} \gtrsim 20$ ) for them to converge. They also enable a robust estimation of the pulse broadening function (PBF), and having the PBF shapes accurately determined at multiple well-separated frequencies may also allow for the investigation of the frequency dependence of the PBF – a subtle effect expected when inner scale effects are dominant (cf. Geiger et al. Reference Geiger2025, Cordes et al. submitted). Measuring pulse broadening times and related interpretations are deferred to a future publication, where more data could be combined to increase the $\mathrm{S}/\mathrm{N}$ .