MeerTRAP: Real-time Transient Search
- MeerTRAP is a commensal, real-time search program on MeerKAT that detects fast radio bursts, pulsars, and short-duration transients.
- It integrates coherent and incoherent beamforming with GPU-based processing and adaptive RFI mitigation to rapidly process high time-resolution data.
- The system enables sub-arcsecond FRB localisation and efficient multi-messenger follow-up through simultaneous radio and optical imaging.
MeerTRAP is a commensal, real-time fast-transient search programme and backend on the MeerKAT radio telescope that performs fully commensal, high time resolution searches for fast radio bursts (FRBs), pulsars, and other short-duration radio transients while other MeerKAT science is ongoing. Across the literature it is expanded as “Meer(more) TRAnsients and Pulsars”, “More TRAnsients And Pulsars”, and the MeerKAT Radio Transients and Pulsars real-time transient-search backend, reflecting a common instrumental role: beamformed discovery, rapid triggering, voltage capture, interferometric localisation, and follow-up, including simultaneous optical coverage with MeerLICHT in some observing modes (Rajwade et al., 2021, Jankowski et al., 2020).
1. Development and survey role
MeerTRAP was commissioned on MeerKAT in 2019 and was framed from the outset as a commensal instrument for the multi-messenger era. Commissioning started in April 2019, first commensal science observations with the real-time system started in August 2019, and the early science emphasis included new FRBs, Galactic transients, and targeted follow-up of FRB 121102 and SGR J1935+2154 (Rajwade et al., 2021). A complementary systems paper then formalised the real-time triggering problem for apparently one-off FRBs, emphasizing that repeating FRBs can be localised by repeated targeted observations whereas one-off FRBs require real-time discovery and triggering (Jankowski et al., 2020).
The subsequent literature established MeerTRAP as both a survey and a backend family rather than a single-purpose FRB detector. Early results reported the first twelve Galactic fast transients with DMs in the range of $33$– (Bezuidenhout et al., 2022) and the first three MeerTRAP FRBs, discovered with a hybrid coherent-plus-incoherent strategy (Rajwade et al., 2022). Later work extended this to sub-arcsecond FRB localisations, host-galaxy identification, persistent-radio-source searches, repeating-FRB burst phenomenology, and larger Galactic transient samples (Driessen et al., 2023, Caleb et al., 2023, Jankowski et al., 2023, Tian et al., 13 May 2025, Pastor-Marazuela et al., 8 Jul 2025, Turner et al., 14 Jan 2025, Tian et al., 20 Oct 2025, Mfulwane et al., 7 Feb 2026).
| Period | Representative result | Paper |
|---|---|---|
| 2019–2021 | Commissioning, real-time triggering infrastructure, first science highlights | (Rajwade et al., 2021, Jankowski et al., 2020) |
| 2022 | First 12 Galactic fast transients; first 3 MeerTRAP FRBs | (Bezuidenhout et al., 2022, Rajwade et al., 2022) |
| 2023 | Sub-arcsecond FRB localisations and survey-rate analysis | (Driessen et al., 2023, Caleb et al., 2023, Jankowski et al., 2023) |
| 2025–2026 | Hyperactive repeater, large FRB host sample, PRS searches, expanded Galactic transient census | (Tian et al., 13 May 2025, Pastor-Marazuela et al., 8 Jul 2025, Turner et al., 14 Jan 2025, Tian et al., 20 Oct 2025, Mfulwane et al., 7 Feb 2026) |
2. Instrument architecture and commensal operation
MeerTRAP operates on the 64-dish MeerKAT array through a beamforming-and-search chain centred on FBFUSE and TUSE. FBFUSE forms up to 768 coherent beams on the sky and also forms an incoherent beam from up to 64 antennas; TUSE is the GPU-enabled search cluster that ingests these beamformed data in real time (Rajwade et al., 2021, Smirnov et al., 2023). In production mode, data arrive from the 32 beamformer nodes of FBFUSE as SPEAD2 packets, are read with mkrecv, ordered into POSIX shared-memory ring buffers, transposed per beam, and passed into the Cheetah pipeline, which sends them to Astro-Accelerate for GPU-based single-pulse searching (Rajwade et al., 2021).
The control stack is correspondingly heterogeneous. The high-level software uses Python, the main processing pipeline uses C++11, and GPU acceleration is implemented with CUDA. The head-node MasterController interfaces with MeerKAT’s CAM control system through katcp, dispatches load, and forwards metadata changes such as beam positions; per-node NodeController processes configure and monitor the executing pipelines, while a web server exposes node health, bandpass summaries, and recent-candidate thumbnails (Rajwade et al., 2021). The compute deployment described for commissioning comprised 66 compute servers in total, with 1 head node and 65 compute nodes, each compute node equipped with 2 Intel Xeon 8C/16T processors, 2 Nvidia GTX 1080 Ti GPUs, 256 GB RAM, and 10-GbE NICs (Rajwade et al., 2021).
Survey geometry is deliberately hybrid. At L-band, MeerTRAP combines a coherent search with an average field of view of and an incoherent search utilizing a field of view of (Rajwade et al., 2022). In the tied-array-beam literature, coherent beams are typically tiled at the 25 per cent sensitivity level of adjacent beam maxima during survey mode, while denser Nyquist-sampled tilings are used for follow-up and localisation (Bezuidenhout et al., 2022).
3. Real-time search, triggering, and localisation workflow
The core search problem is to detect dispersed single pulses in real time while preserving enough raw information to permit localisation and detailed burst analysis. In the FRB-triggering system description, MeerTRAP uses a GPU-based single-pulse detection pipeline built with AstroAccelerate, followed by a candidate-processing stack that performs multi-beam clustering, known-source matching against catalogues, and filtering of candidates associated with RFI and known emitters such as pulsars. These tools typically reduce the candidate rate by 70 to 90 per cent, and an image-based machine-learning classifier was expected to reduce it further to once fully deployed (Jankowski et al., 2020).
The search parameter space is broad. Multiple papers report incoherent dedispersion over $0$–, maximum boxcar widths of , and L-band time resolutions of ; UHF operation is reported at (Rajwade et al., 2021, Jankowski et al., 2023, Bezuidenhout et al., 2022). Selected candidates are written with 0 padding on either side (Caleb et al., 2020, Bezuidenhout et al., 2022).
A central design constraint is the transient buffer. MeerTRAP retains typically about 1 of channelised complex voltage data in POSIX shared-memory psrdada buffers, so candidate handling must complete fast enough that the event is not overwritten. The quoted timing requirements are 2 seconds after the single-pulse pipeline and 3 seconds in total (Jankowski et al., 2020). When a candidate passes threshold, VOEvent packets are used to trigger writeout of the voltage buffers and to alert internal facilities such as MeerLICHT, with comet used on the MeerTRAP head node and the MeerKAT telescope-internal broker (Jankowski et al., 2020).
RFI mitigation is an explicit subsystem rather than an afterthought. IQRM, the Inter-Quartile Range Mitigation algorithm, was developed to infer a time-variable frequency-channel mask directly from short-duration data blocks and was integrated into the MeerTRAP real-time transient search pipeline. It is non-parametric, trend-robust, training-free, and reported to run about 100 times faster than real time on a single CPU core for MeerTRAP/Lovell-type setups (Morello et al., 2021). This suggests that MeerTRAP’s real-time viability depends not only on beamforming and GPU dedispersion but also on aggressive adaptive masking of narrow-band RFI.
Localisation is mode-dependent rather than uniform. The literature reports at least four distinct routes: short-timescale commensal imaging, transient-buffer voltage imaging, multibeam tied-array-beam inference with SeeKAT, and beam-centre localisation when only a brightest beam is available. In a 15-FRB localisation study, two events were localised with 8-second commensal imaging, eight with transient-buffer imaging of 300 ms voltage dumps, and five with SeeKAT, but only one of those five SeeKAT cases achieved sub-arcsecond localisation while four remained at arcminute scale (Pastor-Marazuela et al., 8 Jul 2025). A common misconception is therefore that any MeerTRAP detection is automatically sub-arcsecond; the published record shows that localisation precision depends strongly on trigger mode, beam occupancy, and whether voltage or commensal imaging products exist.
4. Fast radio bursts, localisation precision, and host-galaxy science
The first dedicated FRB discovery paper from the programme reported FRB 20200413A with 4, FRB 20200915A with 5, and FRB 20201123A with 6. The first two were discovered only in the incoherent beam, while FRB 20201123A was found in a single coherent beam, showed a faint post-cursor burst about 200 ms after the main burst, and was associated within an approximately 7 localisation region with a luminous galaxy that dominates the posterior host probability (Rajwade et al., 2022).
A later sample paper analysed 11 MeerTRAP FRBs discovered by the end of 2021 and used the coherent-versus-incoherent survey comparison as a population diagnostic. It estimated FRB all-sky rates at 1.28 GHz of 8 sky9 d0 above 1 for the coherent survey and 2 sky3 d4 above 5 for the incoherent survey, and argued that there seems to be a deficit of low-fluence FRBs, suggesting a break or turnover in the rate-versus-fluence relation below 6 (Jankowski et al., 2023). Because the same platform generated both surveys, this comparison is unusually clean instrumentally.
MeerTRAP’s first sub-arcsecond-localised FRB with MeerKAT was FRB 20210405I, detected in the incoherent beam with 7 and 8 while commensal with ThunderKAT. The burst was sufficiently bright that ThunderKAT 8-second images could localise it to sub-arcsecond precision, enabling association with the disk/spiral galaxy 2MASS J17012499 at 0 (Driessen et al., 2023). FRB 20210410D then provided a complementary case: 1, sub-arcsecond localisation in 2-second images, host-galaxy redshift 2, and an inferred host-galaxy DM contribution that places the source above the 3 scatter of the Macquart relation after subtraction of Milky Way and IGM terms (Caleb et al., 2023).
By 2025, the host-localisation programme had scaled substantially. A localisation and host-galaxy study reported 15 apparently non-repeating FRBs detected with MeerKAT, of which 11 were localised to arcsecond precision and 4 only to arcminute precision; 9 host galaxies were associated with 4 confidence, 2 FRBs had ambiguous two-galaxy associations, and spectroscopic redshifts were measured for 6 hosts spanning 5 to 6 (Pastor-Marazuela et al., 8 Jul 2025). This suggests that MeerTRAP had moved from demonstration cases to an internally coherent host-sample builder.
5. Repeaters, image-plane burst recovery, and counterpart searches
MeerTRAP has also been used as a burst-phenomenology instrument for repeating FRBs. In simultaneous multi-telescope observations of FRB 121102 on 10 September 2019, the newly deployed system detected 11 bursts in 7 hours, with 4 simultaneous detections at Nançay during the last hour of overlap. The data extended over MeerKAT’s 8–9 usable L-band, two bursts exhibited faint precursors separated by about 28 ms and 34 ms, and the measured drift-rate evolution was consistent with published behaviour between 600 and 6500 MHz, with slope 0 (Caleb et al., 2020). The paper explicitly states that the coincident Nançay detections helped verify and characterise the MeerTRAP transient detection system.
The repeating source FRB 20121102A provided a different use case in which MeerTRAP was the burst trigger and MeerKAT imaging was the localisation channel. During the MJD 58736 epoch, the MeerTRAP backend detected 11 bursts over 2.5 hours, and 7 of those bursts were recovered in 2-second image-plane imaging. The same study found that the persistent radio source associated with FRB 20121102A was detected in all five MeerKAT epochs, with the first four 2019 epochs stable at roughly 260–290 1Jy beam2 and the final 2022 epoch at 3Jy beam4, a drop of more than one-third. With an average rms noise of 5 in a 2-second image and the TraP 6 threshold, the authors inferred a detectability threshold of 7 at 1.3 GHz (Rhodes et al., 2023). A plausible implication is that MeerTRAP’s scientific role includes not only beamformed detection but also validation of image-plane transient-search strategies for commensal interferometers.
The discovery of FRB 20240619D showed the same architecture at higher burst yield. MeerTRAP first detected three bursts within about two minutes on 2024 June 19 in MeerKAT L-band, then follow-up with UHF, L-band, and S-band recorded 249 bursts in total, of which 46 were detected in UHF, 177 in L-band, and 26 in S-band; 12 were simultaneous between the UHF and L-band sub-arrays, leaving 237 unique bursts. The localisation, refined against RACS-mid tied to the Radio Fundamental Catalogue, gave 8, 9, with about 0.9 arcsec uncertainty. Above a fluence completeness limit of about $0$0, the cumulative burst rate followed $0$1 with $0$2 in UHF and $0$3 in L-band; the source was about three times more active in L-band than in either UHF or S-band, and the bursts were mostly $0$4 linearly polarised with $0$5–$0$6 circular polarisation fractions (Tian et al., 13 May 2025). Simultaneous MeerLICHT observations detected no optical counterpart and set a q-band fluence upper limit of $0$7 and an optical-to-radio fluence ratio limit of 0.034 (Tian et al., 13 May 2025).
6. Galactic transients, RRAT-like sources, and pulsar discovery space
From the beginning, MeerTRAP’s Galactic yield made clear that the project was not restricted to FRBs. The first twelve Galactic fast transients discovered in the real-time commensal survey had DMs in the range $0$8–$0$9, repeat pulses were detected from seven of the twelve sources, pulse periods were determined for four, and four sources were localised to the arcsecond level with a tied-array-beam localisation method. PSR J1843−0757 yielded a coherent timing solution with 0 and 1, while its waiting-time distribution was better described by a clustered Weibull model than by a Poisson process (Bezuidenhout et al., 2022).
The sample later grew substantially. A 2025 paper reported 26 new Galactic radio transients, mostly RRATs, plus one independently discovered RRAT and two independently discovered pulsars. Using Gaussian fits to individual pulses and arcsecond image-based localisations from transient-buffer voltage data, the authors derived timing solutions spanning multiple years for five sources and measured spin periods for eight more, including one source which appears to rotate every 17.5 seconds. The timing parameters implied ages of several Myr and low surface magnetic field strengths characteristic of RRATs, while some sources exhibited component switching and periodic microstructure (Turner et al., 14 Jan 2025).
A second 2025 Galactic survey paper pushed this further, reporting 30 new Galactic sources identified via single-pulse search, with 9 localised to arcsecond precision in the image domain, periods constrained for 14 sources ranging from 121 ms to 7.623 s, and a phase-coherent timing solution for MTP0063 / PSR J1817−1932 with 2, 3, and RMS timing residual 4. Effelsberg follow-up detected regular but faint emission from three sources and confirmed PSR J2218+2902 at 17.5 s, described as the fourth slowest in the radio pulsar population at the time of writing; PSR J1243−0435 showed periodic microstructure with a peak at 118.6 Hz, corresponding to 5, and possible nulling was noted in PSR J1911−2020 and PSR J1243−0435 (Tian et al., 20 Oct 2025).
MeerTRAP also became part of a distinct imaging-plus-time-domain discovery mode through the source designated PSR J2009−2026. In dynamic imaging observations of the Great Saturn–Jupiter Conjunction, a bright transient lasting about 45 minutes and peaking at about 5.6 mJy was first identified in the image plane; MeerTRAP then detected five single pulses in an L-band observation and twelve in a later UHF observation, establishing the object as a radio-emitting neutron star. PTUSE follow-up refined the timing to 6 and 7, and the paper described this as the first direct discovery of a pulsar via single-epoch imaging (Smirnov et al., 2023). This suggests that MeerTRAP’s discovery space includes not only classical beamformed single-pulse searches but also identification of objects whose transient phenomenology first appears in interferometric image products.
7. Scientific context, capabilities, and limitations
The scientific framing of MeerTRAP has consistently been broader than “FRB finder.” The overview paper positioned it within multi-messenger astrophysics, explicitly coupling real-time radio transient detection and rapid localisation to simultaneous optical follow-up with the 0.65 m MeerLICHT telescope, which covers the entire MeerKAT field of view with 60-second snapshots to about 20th magnitude in SDSS-8 (Rajwade et al., 2021). The later FRB and Galactic-transient papers demonstrate that this architecture supports host-galaxy identification, repetition-rate constraints, polarimetry, survey-rate measurements, and neutron-star timing.
At the same time, the literature is cautious about several unresolved points. The low-fluence FRB turnover inferred from the coherent-versus-incoherent survey comparison is only at the 1.4-9 level (Jankowski et al., 2023). Persistent-radio-source searches toward 25 localised FRBs, including several localised by MeerTRAP, detected 14 radio sources and 12 non-detections, but the authors explicitly state that they cannot definitively classify the detected sources as PRSs and recommend higher-resolution follow-up with e-MERLIN (Mfulwane et al., 7 Feb 2026). For nearby or low-latitude events such as FRB 20210405I, the Galactic-versus-extragalactic interpretation can depend sensitively on Milky Way ISM and halo DM models (Driessen et al., 2023). For some Galactic single-pulse transients, DMs marginally exceed the Galactic contribution depending on the electron-density model assumed, raising the possibility of halo placement without compelling an extragalactic classification (Turner et al., 14 Jan 2025).
A further limitation is bibliographic rather than astrophysical. The arXiv record “Timing analysis of rotating radio transients discovered with MeerKAT” (Letsele et al., 2024) is described in the supplied details as an IOP/JPCS LaTeX conference-paper style guide rather than a MeerTRAP science paper, and therefore does not contribute genuine MeerTRAP results. Within the actual science literature, however, MeerTRAP emerges as a commensal discovery-and-localisation system whose published output spans FRB population studies, host-galaxy association, repeating-burst phenomenology, and the expanding census of RRAT-like and long-period Galactic neutron stars.
