SMART Pulsar Survey with MWA

• SMART Pulsar Survey is an all-southern-sky pulsar survey using the MWA to detect steep-spectrum pulsars and millisecond pulsars with long 80-minute dwell times.
• It employs coherent tied-array beamforming, high time resolution, and AI-assisted snapshot folding to efficiently process vast low-frequency data.
• The survey has already discovered PSR J0125−5854 and established a 154-MHz MSP census, setting the stage for further low-frequency pulsar studies in the SKA-Low era.

The SMART Pulsar Survey—the Southern-sky MWA Rapid Two-metre survey—is an all-southern-sky pulsar survey conducted with the Murchison Widefield Array (MWA) at low radio frequency. In its published form, SMART combines very large instantaneous sky coverage, 80-minute integrations, coherent tied-array beamforming from voltage data, and survey-scale computational triage to search for pulsars over the sky south of declination $+30^\circ$. Its scientific niche is the detection and characterization of steep-spectrum pulsars and millisecond pulsars (MSPs), particularly at high Galactic latitude and in long-period binaries, while its methodological importance lies in showing how low-frequency survey speed can be coupled to modern candidate filtering and, more recently, AI-assisted reduction of folding cost (Tan et al., 17 Jun 2026, Lee et al., 14 Aug 2025, Fu et al., 10 Nov 2025).

1. Survey architecture and observing system

SMART is an all-southern-sky pulsar survey with the MWA. As described in the discovery paper for PSR J0125−5854, it covers the sky south of declination $+30^\circ$ using 71 pointings, each of 80 minutes duration, at a central frequency of 154.24 MHz with 30.72 MHz bandwidth. The survey records tile-level voltages with the MWA Voltage Capture System (VCS), beamforms those voltages, and writes pulsar-search data in PSRFITS format with 100 $\mu$s time resolution and 3072 channels across the band (Tan et al., 17 Jun 2026).

A major design feature is the MWA’s very large field of view. The survey paper states a survey speed of about $450~{\rm deg^2\,h^{-1}}$ and notes that the whole southern sky can be surveyed in <100 h of telescope time. SMART’s long dwell time is central to its observational strategy: at high Galactic latitudes, earlier Parkes surveys used much shorter integrations—about 157 s in the 70-cm survey and about 270 s in HTRU—whereas SMART uses 80-minute integrations, roughly 20–40 times longer (Tan et al., 17 Jun 2026).

Parameter	Value	Source
Sky coverage	South of declination $+30^\circ$	(Tan et al., 17 Jun 2026)
Number of pointings	71	(Tan et al., 17 Jun 2026)
Dwell time per pointing	80 min	(Tan et al., 17 Jun 2026)
Central frequency	154.24 MHz	(Tan et al., 17 Jun 2026)
Bandwidth	30.72 MHz	(Tan et al., 17 Jun 2026)
Time resolution	100 $\mu$s	(Tan et al., 17 Jun 2026)
Channels	3072	(Tan et al., 17 Jun 2026)

The published SMART program distinguishes between an earlier pilot search of several 10-minute observations and the later deep-pass survey, begun in 2024, which searches the full 80-minute observations. The first deep-pass discovery is PSR J0125−5854, discussed below. This suggests that SMART’s present scientific identity is tied less to short exploratory pointings than to full-observation, low-frequency, long-dwell processing of the southern sky (Tan et al., 17 Jun 2026).

2. Search strategy and baseline processing

SMART uses a conventional pulsar-search workflow at the survey level. The J0125−5854 paper states that candidate searching used a Fourier-domain periodicity search with PRESTO; folded candidates were then inspected and refined. For that discovery, localization and confirmation additionally involved re-beamforming the VCS voltages on a dense grid, folding with DSPSR, measuring signal-to-noise with PSRCHIVE tools, and later refining position with MWA and MeerKAT tied-array localization (Tan et al., 17 Jun 2026).

A crucial present limitation is explicit: no acceleration searches are included in the current pipeline. The survey is therefore currently biased toward isolated pulsars and binaries with long orbital periods, rather than compact binaries whose orbital motion would smear the signal during an 80-minute integration. This is not merely an implementation detail; it defines the present SMART selection function and explains why wide, slowly accelerated binaries are especially visible in early results (Tan et al., 17 Jun 2026).

The published MSP census extends the survey’s processing beyond discovery mode. That work targeted known southern MSPs using SMART voltage data and applied a two-stage analysis. A broad first pass used 10 kHz / 100 $\mu$s search-mode products with PRESTO prepfold and incoherent dedispersion; detected targets were then re-beamformed into voltage beams and processed with dspsr using coherent dedispersion, followed by RFI cleaning with clfd and DM refinement with pdmp. For polarimetry and rotation-measure work, the census used full-Stokes data products and RM synthesis (Lee et al., 14 Aug 2025).

The MSP census also makes explicit a general low-frequency constraint within SMART: incoherent dedispersion in the first-stage search can suppress detectability when intrachannel smearing approaches the pulsar period. The recovery of PSR B1937+21 after coherent dedispersion is the clearest published example. A plausible implication is that SMART’s raw detectability of short-period, moderate-DM MSPs is shaped as much by dedispersion mode as by nominal telescope sensitivity (Lee et al., 14 Aug 2025).

3. Scientific niche: low-frequency MSPs, steep spectra, and wide binaries

SMART’s most distinctive scientific niche is the intersection of low radio frequency, very long dwell time, and very large sky coverage. The J0125−5854 paper frames this niche explicitly in four points: low frequencies favor steep-spectrum pulsars; long dwell times offset the MWA’s modest collecting area; large sky coverage is especially valuable at high latitude; and, even without acceleration searches, long-period binary MSPs remain accessible because line-of-sight acceleration changes little over an 80-minute observation (Tan et al., 17 Jun 2026).

This niche is now supported by the dedicated MSP census. That study reports 40 MSPs detected in SMART at 154 MHz, with 11 being the first published detections below 300 MHz. It further reports significant rotation measures for 25 MSPs and apparent phase-dependent RM variations for three. Comparison with published profiles at other frequencies is said to support earlier work suggesting that MSP pulse-component separations vary negligibly over a wide frequency range, consistent with compact magnetospheres, while integrated profiles tend to be more polarized at low frequencies (Lee et al., 14 Aug 2025).

A recurrent misconception is that low-frequency, acceleration-free surveys are intrinsically unsuitable for MSP work. SMART’s published results qualify that assumption rather than support it. The current pipeline is indeed incomplete for compact binaries, but the combination of low frequency and long dwell time is demonstrably effective for steep-spectrum MSPs and for binaries with sufficiently long orbital periods. The first deep-pass discovery, PSR J0125−5854, is an existence proof of that selection space, and the 154-MHz census shows that the survey is already capable of systematic southern MSP characterization at low frequency (Tan et al., 17 Jun 2026, Lee et al., 14 Aug 2025).

The same paper that reports J0125−5854 cites earlier SMART simulations indicating that full search processing of the 80-minute observations is expected to yield up to 55 MSPs with DMs up to roughly 100 pc cm$^{-3}$, including up to 15 new discoveries. It immediately adds the necessary qualification that, because the current SMART pipeline does not perform acceleration searches, the realized yield under the present search mode will be less than 55 (Tan et al., 17 Jun 2026).

4. Major observational results

The first major discovery from the deep-pass survey is PSR J0125−5854, reported as the first MSP discovered with the MWA and the first pulsar discovery from SMART deep-pass searches. It was found in a SMART observation taken on 2018 October 22 with a folded-candidate significance of $17\sigma$, a period of 24.590998(2) ms, and DM $= 11.663(4)\ {\rm pc\,cm^{-3}}$. Follow-up with the MWA and MeerKAT showed that it is a steep-spectrum pulsar, with $+30^\circ$0, at high Galactic latitude $+30^\circ$1 and distance roughly 0.5–1 kpc (Tan et al., 17 Jun 2026).

The astrophysically distinctive result is its orbit. The current timing-based solution gives a binary period of $+30^\circ$2, projected semi-major axis $+30^\circ$3, eccentricity $+30^\circ$4, and minimum companion mass $+30^\circ$5. The favored interpretation is a wide pulsar–helium-white-dwarf system, although the paper stresses that a fully phase-coherent timing solution has not yet been obtained and further data are required (Tan et al., 17 Jun 2026).

The discovery is important not only as a new pulsar but as a demonstration of SMART’s effective discovery space. The paper argues that J0125−5854 was detectable in SMART because it combines the survey’s preferred properties: a steep spectrum, long dwell-time detectability, high Galactic latitude, and an orbit so wide that acceleration smearing is negligible over 80 minutes. The authors add that the pulsar was detected in the first of the 71 observations taken for SMART, which they interpret as an encouraging early benchmark for the deep-pass search (Tan et al., 17 Jun 2026).

The second major published product is the dedicated low-frequency MSP census. That paper extends SMART from discovery to population characterization by providing coherently-dedispersed full-polarimetric integrated pulse profiles and mean flux densities for all detected MSPs, together with 25 significant RMs. It also identifies apparent phase-dependent RM variations for three MSPs and argues that low-frequency integrated profiles tend to be more polarized than higher-frequency profiles, consistent with depolarization by superposed orthogonal polarization modes at higher frequencies. These products are framed as a resource for future low-frequency MSP monitoring and for improving survey simulations for SKA-Low (Lee et al., 14 Aug 2025).

5. Computational bottlenecks and AI-assisted acceleration of SMART processing

A separate SMART-related development addresses a different bottleneck: not candidate ranking after full processing, but the cost of the folding stage itself. The 2025 AI pipeline paper argues that, in FFT-based pulsar searches relevant to SMART, folding dominates compute time, particularly for long observations. It therefore inserts a lightweight classification stage before full-data folding: candidate DM–period combinations are first folded only in the corresponding de-dispersed 1-D time series, creating “snapshot” candidates, and only those passing a deep-learning classifier are sent on to expensive full-data folding (Fu et al., 10 Nov 2025).

The classifier uses two time-domain features from the snapshot folds: the average pulse profile and the time-phase diagram. It is a hybrid multi-input network built around multi-scale resizing, a 1-D ResNet branch with squeeze-and-excitation attention for the profile, and a denoised 2-D CNN branch with CBAM attention for the time-phase image. In internal validation on real multi-telescope data from FAST, Parkes, and Arecibo, the model achieved accuracy 98.30%, precision 98.25%, recall 98.44%, and F1 98.34% (Fu et al., 10 Nov 2025).

Its direct SMART relevance comes from blind testing on SMART candidates and from a survey-like folding benchmark. In one 80-minute MWA observation containing 43 known pulsars, the standard search produced 25,691 potential DM–period combinations. Applying the classifier at threshold 0.5 selected only 436 for full-data folding, while manual inspection confirmed that these retained the fundamental candidates of all 43 known pulsars plus some harmonics. The paper therefore reports a roughly 59-fold reduction in full folds and presents this as a conservatively estimated speed-up factor of 60 in the folding step over a large parameter space (Fu et al., 10 Nov 2025).

The same study emphasizes that this is earlier in the workflow than traditional pulsar-candidate classifiers used in PRESTO-style pipelines, including those used in LOTAAS and first-pass SMART processing. In restricted-parameter-space tests, such as the NGC 5904 benchmark and simulated FAST data, the reduction in full folds was about tenfold while retaining all known detectable pulsars in the searched restricted parameter space. This suggests that, for SMART, AI can function not only as a candidate-ranking tool but as a compute-saving gate before the most expensive search stage (Fu et al., 10 Nov 2025).

6. Position in the survey landscape, limitations, and prospects

SMART occupies a distinctive position among pulsar surveys. Compared with earlier Parkes high-latitude surveys, it trades raw dish sensitivity for very large field of view and much longer dwell times. Compared with highly sensitive but narrower-field systems such as the MPIfR-MeerKAT Galactic Plane Survey, it operates in a different part of parameter space: low frequency, all-southern-sky coverage, and strong sensitivity to steep-spectrum high-latitude objects. The J0125−5854 paper also draws an explicit analogy with LOTAAS, another low-frequency, long-dwell survey that found many high-latitude pulsars and long-period-binary MSPs, reinforcing the idea that acceleration-free low-frequency surveys are naturally biased toward such systems (Tan et al., 17 Jun 2026).

Its principal present limitation remains the absence of acceleration searches in the main deep-pass pipeline. That limitation is operational, not merely theoretical: the survey paper explicitly states that the current pipeline is biased against compact binaries and that the realized MSP yield will therefore fall below earlier predictions that assumed fuller search capability. The AI pipeline also identifies a second limitation, namely that time-domain “snapshot” features alone can preserve high recall while leaving more false positives in harsher RFI environments than in the relatively clean MWA case (Tan et al., 17 Jun 2026, Fu et al., 10 Nov 2025).

The published MSP census adds further low-frequency caveats. Flux calibration carries substantial uncertainties because of beam and sky-model systematics, and apparent phase-dependent RM variations can be produced not only by intrinsic propagation or magnetospheric effects but also by scattering and instrumental zero-Faraday-depth contamination. The census is also not fully complete, because full coherent reprocessing of all non-detections was not feasible at current data volumes (Lee et al., 14 Aug 2025).

Even with those caveats, the early SMART record is scientifically consequential. It has already produced the first MWA-discovered MSP, established a southern 154-MHz MSP census with 40 detections, and motivated a survey-specific AI strategy that can reduce the dominant folding cost by about 10× in restricted searches and by a conservatively estimated 60× in a large-parameter-space SMART test. Taken together, these results suggest that SMART is not only a survey of the southern sky at 154.24 MHz, but also a pathfinder for how low-frequency, long-dwell pulsar surveys may be designed and processed in the SKA-Low era (Tan et al., 17 Jun 2026, Lee et al., 14 Aug 2025, Fu et al., 10 Nov 2025).