Radio telescopes now produce so much data that scientists can no longer afford to write everything to disk and analyze it later. This paper tests a move to real‑time processing for searches of fast radio bursts (FRBs), short, bright radio flashes whose origins are still not fully known. The authors set up a compact pipeline around Heimdall, a widely used open‑source FRB search tool, and compared its traditional offline mode with a real‑time mode that keeps data in memory.

Heimdall looks for single pulses by performing incoherent de‑dispersion, a step that corrects the frequency‑dependent delay caused by the interstellar medium. In the offline case Heimdall reads a SIGPROC filterbank file (a stored data format). In the real‑time case it reads from a PSRDADA ring buffer, which is a circular block of memory designed to handle continuous high‑rate data with low delay. The team ran tests on real data from 16 cylinders of the North‑South arm of the Northern Cross radio telescope and varied input files and search parameters to probe robustness.

The main result is that the two modes give very similar answers. For the key measured quantities that describe an FRB — the time of arrival (TOA) and the dispersion measure (DM, a number related to how much the pulse is delayed by intervening plasma) — the PSRDADA and SIGPROC runs agreed to about 3%. Signal‑to‑noise ratios (SNR) were typically within about 10%. In timing and detection speed, the real‑time (PSRDADA) mode stayed inside the required data flow for most runs that contained real FRBs, although in this particular testbed the execution times were often comparable to the offline SIGPROC mode.

Why this matters: future instruments such as the Square Kilometre Array will produce far larger data volumes, making real‑time pipelines essential. Keeping data in RAM with ring buffers lets a system keep up with the incoming stream and save only scientifically interesting pieces, saving disk space and avoiding slow input/output steps. Demonstrating that a real‑time Heimdall setup returns results consistent with the established offline mode is a useful step toward practical deployment.