Tianlai Cylinder Pathfinder Array

• Tianlai Cylinder Pathfinder Array is a fixed transit-style radio interferometer designed for 21 cm intensity mapping to probe cosmic large-scale structure and BAO.
• It employs three adjacent cylindrical reflectors with intentionally unequal feed spacing to suppress grating lobes and achieve a narrow east–west beam with a broad north–south view.
• The system integrates advanced digital processing, calibration strategies, and beam characterization to overcome challenges like mutual coupling, standing waves, and regularization in map-making.

Searching arXiv for recent and core Tianlai Cylinder Pathfinder papers to ground the article. The Tianlai Cylinder Pathfinder Array is a fixed, transit-style radio interferometer built to test the instrumental and analysis methods required for post-reionization 21 cm intensity mapping. As deployed at Hongliuxia in Xinjiang, it consists of three adjacent north–south cylindrical reflectors, each $40\,\mathrm{m}\times 15\,\mathrm{m}$, instrumented with 31, 32, and 33 dual-polarization feeds, for a total of 96 feeds or 192 polarized signal channels (Li et al., 2020). Its scientific purpose is to measure the aggregate redshifted H I emission from large sky voxels rather than individual galaxies, thereby probing large-scale structure and the baryon acoustic oscillation signal over cosmological volumes (Zuo et al., 2018). The array’s architecture—a narrow east–west synthesized response, broad north–south field of view, and passive drift-scan observing mode—makes it a natural platform for wide-field survey cosmology, but also makes beam characterization, gain calibration, mutual coupling control, and map-making regularization central technical problems (Cianciara et al., 2017).

1. Array configuration and observational concept

The pathfinder is part of the broader Tianlai program, which was organized in stages as Pathfinder, Pathfinder+, and Full Array (Das et al., 2018). Within the pathfinder stage, the cylinder array is one of two co-located instruments, the other being a 16-dish array (Zuo et al., 2018). The cylinder pathfinder itself is described as three adjacent parabolic cylinders with long axes oriented north–south, operated as a drift-scan or transit interferometer (Zuo et al., 2018).

Across the three reflectors, the installed feed counts are 31, 32, and 33, with slightly different feed spacings of 41.33 cm, 40.00 cm, and 38.75 cm respectively (Zuo et al., 2018). This unequal spacing is intentional and is used to reduce grating lobes or, in related map-making language, to suppress grating-lobe degeneracies that become severe for regular feed spacing larger than half a wavelength (Zhang et al., 2016). The same basic geometry is reported consistently across instrumentation, calibration, and beam papers, with the cylinders labeled A, B, C from east to west (Wang et al., 2024).

The array is located at the Hongliuxia site in Balikun County, Xinjiang, described as radio quiet (Das et al., 2018). A cylinder telescope focuses in only one dimension, so the Tianlai reflectors provide a narrow beam in the east–west focusing direction while leaving the north–south response broad and feed-limited (Cianciara et al., 2017). This geometry yields a large instantaneous field of view and makes the system well suited to survey operation, in which Earth rotation supplies sky coverage (Li et al., 2020). A plausible implication is that the instrument’s sky response is intrinsically anisotropic: angular resolution and calibration demands differ strongly between the focused and unfocused axes.

The pathfinder’s initial observing emphasis is in the 700–800 MHz band, corresponding to the redshifted 21 cm line over $1.03 > z > 0.78$ (Das et al., 2018). More broadly, Tianlai forecasting work treated the pathfinder as a technology demonstrator for 21 cm intensity mapping over this redshift range, with the longer-term aim of BAO and dark-energy measurements in larger future arrays (Xu et al., 2014).

2. Feed system, optical geometry, and electromagnetic design

The feed subsystem is a defining component of the cylinder pathfinder. The installed feed is a wideband, dual-polarized modified four-square antenna, called a “four-hex” in the feed-design paper because two corners of each square radiator are folded toward the ground plane (Cianciara et al., 2017). The design goal is operation across 650–1420 MHz, corresponding to the H I 21 cm line from approximately $z\approx 1.2$ to $z=0$, although Tianlai’s initial observing emphasis is around 700–800 MHz (Cianciara et al., 2017).

Several construction features are explicitly tied to pathfinder requirements. The dipole tip-to-tip length is $0.62\lambda_c$ with $\lambda_c \approx 29$ cm, and the feed includes a circular disk ground plane surrounded by a cylindrical rim, described as a cylindrical “can”, plus a subreflector that forms a resonant structure analogous to a short-backfire antenna (Cianciara et al., 2017). A quarter-wave balun is used between radiators and ground plane to suppress imbalance current and preserve the intended radiation pattern (Cianciara et al., 2017). The paper states that the folded radiator corners reduce cross-coupling between neighboring feeds, reduce coupling between the two orthogonal polarizations within a feed, and improve impedance match (Cianciara et al., 2017).

In the cylinder coordinate system, the two orthogonal dipoles correspond to longitudinal and transverse polarizations in one notation (Cianciara et al., 2017), while later system papers denote them X and Y, with X = N–S polarization and Y = E–W polarization (Li et al., 2020). This suggests a shift in operational labeling rather than a change in the underlying dual-linear architecture.

The reflector geometry is also standardized across the instrument literature. The cylinders are 15 m wide in the focusing direction and 40 m long, with $f/D = 0.32$ and focal length 4.8 m (Cianciara et al., 2017, Sun et al., 2022). In the feed-design stage, the focal-line implementation considered a 32-element line feed with center spacing $0.71\lambda_c$, described as “nearly touching,” and full-system simulations included all 32 feeds, the cylinder, and support structures (Cianciara et al., 2017). In the deployed pathfinder, the unequal 31/32/33 population on the three cylinders modifies that idealized uniform case while retaining the same dense focal-line philosophy (Zuo et al., 2018).

Electromagnetic simulations show that the single-feed-plus-cylinder system has the expected cylinder response: at 1050 MHz, full width at half maximum is about $1.2^\circ$ in the focusing cut and about $61.3^\circ$–$1.03 > z > 0.78$0 in the non-focusing cut (Cianciara et al., 2017). For the full feed-array-plus-reflector model, the central feed at 1050 MHz has focusing-plane widths of about $1.03 > z > 0.78$1–$1.03 > z > 0.78$2 and non-focusing widths of about $1.03 > z > 0.78$3–$1.03 > z > 0.78$4 depending on polarization (Cianciara et al., 2017). These values define the basic pathfinder beam concept: fine east–west angular response from the cylinder aperture and wide north–south survey coverage from the unfocused axis.

3. Signal chain, correlators, and backend evolution

The array exposes 192 analog signal channels from its 96 dual-polarization feeds (Wang et al., 2024). In the system-performance paper, the analog chain is described as follows: feed-mounted LNAs amplify the signals; the RF is carried by 15 m coaxial cable to optical transmitters under the cylinders; the signal is sent over 8 km single-mode fiber to the station house; there it is converted back to RF, downconverted to an IF band, and digitized for correlation (Li et al., 2020). The IF band is given as 125–235 MHz in the system paper and 135–235 MHz in the construction overview, reflecting slightly different descriptions of the same nominal 100 MHz backend band (Li et al., 2020, Das et al., 2018).

An earlier custom cylinder correlator is summarized in the commissioning overview as an FX correlator with 192 input signals, sampled at 250 MSPS, using 14-bit ADCs, 2048-point FFT channelization, 1008 output channels of 122 kHz width, and 4 s averaging (Das et al., 2018). A later backend implementation replaced this with a fielded ROACH2+GPU correlator described as the production digital signal processor for the Hongliuxia-deployed cylinder pathfinder (Wang et al., 2024). That system consists of six CASPER ROACH2 boards and seven GPU servers, handling all 192 inputs, sampling at 250 Msps, channelizing into 1024 channels, and processing the central 896 channels, corresponding to 109.375 MHz from 692.8125–802.1875 MHz (Wang et al., 2024).

The ROACH2+GPU correlator is a packetized FX design. The F-engine performs digitization, polyphase-filter-bank channelization, equalization, requantization to 4+4-bit complex samples, and packetization over 24 total 10 GbE links (Wang et al., 2024). The X-engine is implemented on GPUs using xGPU, with each server processing 128 frequency channels via four hashpipe instances (Wang et al., 2024). The unique-baseline count for 192 inputs is explicitly given as 18,528, and the correlator retains the four correlation products $1.03 > z > 0.78$5, providing full-polarization outputs (Wang et al., 2024).

The backend is synchronized by a shared 250 MHz sample clock and 1-PPS timing, and integrated visibility data are centralized on a master/storage server and written in HDF5 format (Wang et al., 2024). The normal integration time is about 4 s, and the system is reported to operate in real time over the 896-channel processed band (Wang et al., 2024). Commissioning tests verified ADC correctness, correlator phase linearity, and on-sky response, while long-duration operation demonstrated continuous stability over a month (Wang et al., 2024).

A separate digital backend has also been developed for transient work. The FRB system forms 96 digital beams from the cylinder array, currently using one polarization because of processing limits, and covers approximately 40 square degrees at 3 dB (Yu et al., 2024). This establishes that the pathfinder has become a dual-purpose platform: a visibility-producing cosmology instrument and a beamforming transient instrument.

4. Calibration architecture and gain stability

Calibration in the Tianlai cylinder pathfinder is driven by the fact that receiver gains vary in time and the analog signal path includes long fiber transport whose effective electrical length changes with temperature (Zuo et al., 2018). The array therefore uses a two-step calibration strategy. First, a periodically broadcast artificial noise source provides relative phase calibration to remove time-varying cable-delay phases; second, bright celestial point sources provide absolute calibration of phase and amplitude (Zuo et al., 2018).

The fundamental feed-voltage model is written as

$1.03 > z > 0.78$6

with the corresponding visibility

$1.03 > z > 0.78$7

(Zuo et al., 2018). In the presence of a single dominant point source, the sky contribution simplifies to a rank-1 form,

$1.03 > z > 0.78$8

where

$1.03 > z > 0.78$9

Thus the dominant eigenvector of the visibility matrix yields the effective complex response $z\approx 1.2$0 toward the calibrator (Zuo et al., 2018).

The calibration paper uses stable principal component analysis (SPCA) to decompose the measured visibility matrix into a low-rank calibrator term, sparse outliers, and dense residual noise,

$z\approx 1.2$1

with $z\approx 1.2$2 (Zuo et al., 2018). This improves robustness against RFI, bad feeds, and missing data. Applied to first-light drift-scan observations on 27 September 2016, the method successfully recovered gains, identified malfunctioning feeds, and extracted the east–west beam profile from source transits (Zuo et al., 2018).

The relative calibration source is important because the RF-over-fiber transport is about 7 km in that paper’s description, and temperature-dependent cable or fiber delays induce substantial phase drift (Zuo et al., 2018). Nighttime phase drifts are typically only a few degrees over a night, while daytime variations are larger and faster (Zuo et al., 2018). The 2020 system-performance analysis similarly showed that gain phase drifts correlate strongly with environmental temperature and that daytime phase fluctuations are much larger than nighttime ones (Li et al., 2020).

A major conclusion of simulation-based error propagation is that the current calibration is limited primarily by absolute calibration, not relative calibration (Yu et al., 2023). The main error source is the approximation of a single dominating point source during bright-source calibration; residual phase error is typically at the $z\approx 1.2$3–$z\approx 1.2$4 rad level, while the relative calibration floor from the artificial calibrator is about $z\approx 1.2$5 rad (Yu et al., 2023). This suggests that future improvement depends more on richer sky modeling and calibration formalism than on tighter control of the relative phase reference alone.

5. Beam characterization, polarization structure, and instrumental chromaticity

Beam knowledge is a central problem for the cylinder pathfinder because the instrument is fixed, highly anisotropic, and intended for precision foreground subtraction. Several complementary approaches have been used: electromagnetic simulation, bright-source transit analysis, and unmanned-aerial-vehicle measurements.

Transit measurements constrain the east–west beam naturally, because sources drift through the narrow focusing direction. Using bright-source transits, the eigenvector/SPCA calibration paper measured east–west beam FWHM at 750 MHz of $z\approx 1.2$6 for $z\approx 1.2$7 and $z\approx 1.2$8 for $z\approx 1.2$9 (Zuo et al., 2018). The 2018 construction overview reports the same values as representative main-lobe widths at 750 MHz (Das et al., 2018). By contrast, a later electromagnetic study comparing simulation with observations found Gaussian-fit HPBW at 750 MHz of $z=0$0 for Y polarization and $z=0$1 for X polarization, with the measured beam widths somewhat wider than simulation but following the same frequency trends (Sun et al., 2022). This difference reflects distinct beam definitions and analysis pipelines rather than a single settled value.

Direct measurement of the north–south beam is much harder. The 2025 UAV-beam paper was motivated precisely by the fact that the large cylinders cannot be placed in an anechoic chamber and source transits do not adequately constrain the broad north–south response (Li et al., 2 Aug 2025). In that study, a DJI M300 RTK carrying a calibrated broadband noise source flew at 1220 m relative height, satisfying the far-field condition for the 15 m east–west aperture over 700–800 MHz (Li et al., 2 Aug 2025). The beam was assumed factorizable,

$z=0$2

and the cylinder beam was recovered from sky-subtracted auto-correlations after correcting for the transmitter pattern (Li et al., 2 Aug 2025).

The UAV measurements showed that the north–south beam is broad, repeatable across flights and feeds, and exhibits a shoulder near $z=0$3 with steepening near $z=0$4 (Li et al., 2 Aug 2025). The discrepancy from simulation at the shoulder reaches about 3 dB, and the profile is slightly asymmetric, with the shoulder more pronounced on the northern side (Li et al., 2 Aug 2025). This is a noteworthy result because earlier simulation-based beam models typically assumed a smoother, more symmetric unfocused-axis response.

The east–west UAV beam agrees qualitatively with source-transit measurements and electromagnetic simulation, but is described as slightly broader, asymmetric, and with higher sidelobes than the idealized model (Li et al., 2 Aug 2025). The paper also notes polarization-dependent anomalies, including a small central dip in the YY H-plane profile for one example feed, possibly arising from cross-coupling (Li et al., 2 Aug 2025).

Beam separability has also been tested with sky-based methods. A 2026 study used the seasonal north–south motion of the Sun to argue that the primary beam can be decomposed into independent east–west and north–south components, measuring the E–W beam from solar transits and fitting the N–S beam with a sky model (Geng et al., 22 Apr 2026). The E–W beam width remained largely invariant across different solar elevations, supporting the factorization assumption (Geng et al., 22 Apr 2026). This suggests that separable beam models, while approximate, may be adequate for some calibration and map-making tasks.

Polarization behavior is scientifically important because foreground leakage into Stokes $z=0$5 can be driven by mismatch between the two orthogonal beams rather than simply by poor port isolation (Cianciara et al., 2017). The feed-design paper gives the explicit leakage expressions

$z=0$6

$z=0$7

with Stokes $z=0$8 formed as

$z=0$9

to show how unequal, frequency-dependent co- and cross-polar responses leak polarized foreground structure into total intensity (Cianciara et al., 2017). In the full-system simulation, the cross-polar response is typically 20 dB or more below the co-polar response over most angles, but the paper stresses that this does not eliminate leakage risk because beam mismatch remains (Cianciara et al., 2017).

6. Mutual coupling, standing waves, system temperature, and map-making limitations

The cylinder pathfinder’s instrumental complexity is dominated not only by beam shape but also by spectral and baseline-dependent nonidealities. Mutual coupling between feeds, standing waves between feed and reflector, and calibration/model mismatches all enter directly into sensitivity and foreground behavior.

In the feed-array-plus-reflector simulations, feed-to-feed coupling remains below $0.62\lambda_c$0 dB at all frequencies in the 650–1420 MHz study, even for neighboring feeds with parallel polarizations (Cianciara et al., 2017). The later full-cylinder electromagnetic study finds stronger reflector-mediated same-polarization coupling, with adjacent YY pairs reaching nearly $0.62\lambda_c$1 dB peak coupling and adjacent XX pairs about $0.62\lambda_c$2 dB, while cross-polar $0.62\lambda_c$3 and $0.62\lambda_c$4 couplings are generally below $0.62\lambda_c$5 dB (Sun et al., 2022). The same paper shows that the coupling decreases with feed separation and that simulated coupling can explain correlated-noise structures seen in real data, including Kelvin-level offsets on short baselines (Sun et al., 2022).

The visibility contamination from coupling is modeled perturbatively as

$0.62\lambda_c$6

which reduces under symmetry assumptions to

$0.62\lambda_c$7

(Sun et al., 2022). This identifies the coupling-induced term as a correlated-noise floor rather than ordinary thermal noise. Observationally, baseline 15Y–16Y shows correlated noise up to about 4 K, while 2Y–16Y shows about 1 K, in good agreement with estimates from the simulated $0.62\lambda_c$8 values (Sun et al., 2022).

Standing waves and spectral ripple are another major systematic. Delay-transform analysis of auto-correlations identified a strong peak at about 142 ns, traced directly to the 15 m feed cable, and a lower-delay feature below 50 ns, interpreted as a mixture of feed–reflector standing wave and standing waves in the 4 m IF cable (Li et al., 2020). The reflector-feed path length of 9.6 m corresponds to a standing-wave period of about 31.1 MHz, which is reproduced in electromagnetic simulation and observed band-pass structure (Sun et al., 2022). These spectral ripples are scientifically important because they impose non-smooth frequency structure on otherwise smooth foregrounds (Li et al., 2020).

The system temperature budget is consistently treated as a key performance metric. In the feed-design study,

$0.62\lambda_c$9

with representative values $\lambda_c \approx 29$0 K, $\lambda_c \approx 29$1 K in the initial 700–800 MHz band, and a design goal that spillover contribute no more than about 15 K (Cianciara et al., 2017). In the later pathfinder system analysis, the measured mean system temperature is about 90 K, with cylinder averages of 99.7 K, 85.8 K, and 82.0 K for A, B, and C respectively (Li et al., 2020). The electromagnetic study likewise adopts $\lambda_c \approx 29$2 K, with estimated receiver temperature about 48 K and antenna temperature about 42 K (Sun et al., 2022). This convergence suggests that $\lambda_c \approx 29$3 K is the practical working system temperature of the deployed pathfinder in its initial band.

Map-making studies identify an additional class of limitation: incomplete harmonic coverage and the need for regularization. In $\lambda_c \approx 29$4-mode form, the drift-scan measurement equation is

$\lambda_c \approx 29$5

or abstractly

$\lambda_c \approx 29$6

(Yu et al., 2023). Because the matrix $\lambda_c \approx 29$7 is rank-deficient or ill-conditioned, direct inversion is unstable. The unregularized least-squares estimator,

$\lambda_c \approx 29$8

must therefore be replaced by regularized methods such as truncated SVD or Tikhonov regularization (Yu et al., 2023). In Tianlai-specific simulations, Tikhonov regularization yields smoother maps and avoids some of the comb-like artifacts produced by hard singular-value truncation (Yu et al., 2023). A separate end-to-end error study shows that discarding singular modes below a threshold can itself generate visible comb-like artifacts, especially from lost low-$\lambda_c \approx 29$9 information (Yu et al., 2023).

The present feed layout also limits angular reconstruction. Because the populated north–south baseline extent is only about 12.4 m, the current array measures modes only up to approximately

$f/D = 0.32$0

well, and loses a significant fraction of higher angular modes once noise is present (Yu et al., 2023). Reconfiguring the outer feed positions to extend the effective north–south baseline can improve transfer-function behavior substantially, though with tradeoffs in short-baseline coverage (Yu et al., 2023). This suggests that the current pathfinder geometry is not only a commissioning choice but also a substantive limit on recoverable sky information.

7. Scientific uses, commissioning outcomes, and broader significance

The cylinder pathfinder has already served as a platform for several distinct classes of scientific and technical work. In its original cosmological role, it was designed to test whether wide-field cylinder interferometers can support post-reionization 21 cm intensity mapping, with the long-term aim of BAO and dark-energy measurements (Das et al., 2018). Forecasting treated the pathfinder itself as able to detect the H I power spectrum but not as a competitive cosmology machine; the full Tianlai array was the configuration expected to deliver strong dark-energy and primordial non-Gaussianity constraints (Xu et al., 2014).

Commissioning established that the array could acquire stable drift-scan visibilities, detect bright astronomical source transits, recover beam profiles, and support continuous backend operation (Li et al., 2020, Wang et al., 2024). The eigenvector/SPCA calibration method validated the use of bright-source transits for initial gain calibration even with imperfect beam knowledge, and identified beam-center offsets with maximum deviation 108 s, corresponding to $f/D = 0.32$1, and median 28 s, corresponding to $f/D = 0.32$2 (Zuo et al., 2018). These results showed that the real instrument includes measurable feed misalignment and beam-to-beam nonuniformity.

The pathfinder has also become an FRB-search instrument. A dedicated transient backend now forms 96 beams, detects candidates in quasi-real time, and achieved an end-to-end recall rate of 88% on valid injected mock FRBs (Yu et al., 2024). During commissioning, it detected signals from four known pulsars and discovered FRB 20220414A (Yu et al., 2024). This does not alter the array’s original cosmological purpose, but it demonstrates that the hardware and digital architecture support additional science modes.

Across all these uses, a common conclusion emerges. The Tianlai Cylinder Pathfinder Array is a functioning, observatory-scale transit interferometer whose main limitations are not raw operability but instrumental complexity: embedded beams are not simple, standing waves make the band-pass non-smooth, short-baseline correlated noise is non-negligible, and map-making requires regularization that can itself imprint artifacts (Cianciara et al., 2017, Li et al., 2020, Yu et al., 2023). At the same time, the combination of electromagnetic modeling, transit calibration, UAV beam measurements, scalable digital correlation, and $f/D = 0.32$3-mode analysis has made those limitations measurable rather than opaque. That is the pathfinder’s central significance within the Tianlai program: it has turned the cylinder-array concept from a forecasting instrument into a quantitatively characterized platform for 21 cm survey methodology.