In this paper
AbstractIntroductionMethodExperimental configurationsFeed Array and signal chainBackend and raw data preprocessingData reduction proceduresNoise temperature budgetModified Y-factor algorithm for mutual coupling extractionResultsEquivalent receiver noiseInfluences of beamforming weightsMutual couplingEquivalent system noise temperatureDiscussionConclusionReferencesWorksPhysics and Astronomy
System noise temperature measurements of a 19-element phased array feed
Abstract
We present multi-epoch system noise temperature measurements of a 19 element hexagonal room-temperature phased array feed (PAF) using a modified Y-factor method. A Python-based pipeline with iterative channel calibration and maxSNR beam forming is developed to do data reduction, and a covariance-based decomposition scheme is proposed to extract mutual coupling noise temperature Tmc. Long-term observations from March to June 2026 capture temporal receiver noise drift mainly induced by LNA temperature variations. Within the 1.37-1.47 GHz operating band, the measured beam equivalent system noise temperature of the PAF is about 73-89 K. The measured Tmc baseline ranges from about 2 K to 8 K, while under thermal loading, the resonant peak of Tmc can increase to more than 20 K. The baseline of Tmc is consistent with the S-parameter mea surements, which show the Tmc of the full array ranges less than 10.7 K. The baseline fluctuation of Tmc most likely arises from the superposition of thermal and non-thermal components of the spatial electromagnetic fields. A detachable stainless steel shield is deployed to quantify sidelobe spillover noise, and comparative tests show unshielded en vironmental interference increases receiver noise temperature by over 60 K. The proposed measurements and data processing framework provides a practical reference for noise per formance characterization of ground-based radio telescope PAF prototypes.
Introduction
Phased Array Feeds (PAFs) represent the transformative focal-plane receiver technology for contemporary and next-generation radio telescopes, enabling simultaneous electronically steerable multi-beam synthesis via digital beamforming and substantially accelerating sky survey speed with a wide instantaneous field of view [21, 17, 18]. Radio telescopes equipped with PAFs, such as the Australian Square Kilometre Array Pathfinder (ASKAP), Parkes Telescope, Westerbork APERture Tile In Focus (Apertif), Green Bank Telescope (GBT), have been deployed for observations targeting fast radio bursts, pulsars, HI survey, magnetic fields in galaxy clusters, continuum sources, and large-area radio sky surveys [10, 1, 6, 19, 16, 9, 14] (Li et al. 2021). In addition to these observations, low-noise amplifiers and cryogenic PAF hardware have also been designed and manufactured to meet the ultra-low noise requirements for astronomical observations [5, 25, 8, 13, 3].
Cryogenic and room-temperature L and C band PAF prototypes with various antenna types including Vivaldi, dipoles, ’rocket’ have been developed to achieve wideband astronomical observations over the past two decades [22, 19, 13, 20, 5] (Wu et al. 2013). For instance, the beam-formed system temperature normalized by aperture efficiency of cryogenic PAF for the GBT has been measured as -- K near 1350 MHz for the boresight beam, achieving comparable noise temperature performance of the 1400 MHz cryogenic single-feed receiver on the GBT [15, 14]. The system temperature of L band room-temperature Apertif PAF is about 70K [19].
System noise temperature acts as the core sensitivity metric determining the detection threshold of faint cosmic radio sources, and its budget consists of multiple coupled noise constituents: LNA intrinsic thermal noise, antenna/cable resistive loss, cosmic sky background radiation, antenna sidelobes and backlobes noise, and correlated noise induced by element mutual coupling [23, 11]. Mutual coupling, arising from near-field electromagnetic interaction between closely spaced radiators and beam-dependent active impedance mismatches at LNA input ports, introduces off-diagonal correlated terms in array covariance matrices and creates an unavoidable noise penalty for dense PAF designs [26, 24]. Fundamental theoretical frameworks, established by Warnick & co-workers systematically model coupling-induced receiver noise degradation, derive active impedance matching criteria, and contrast three mainstream amplifier matching schemes [26, 23]. S-parameter full-wave simulation can predict inter-port transmission coefficients and convert coupling strength into equivalent noise temperature contributions, yet purely numerical predictions require on-telescope hot/cold load calibration validation [11, 4] (Wu et al. 2013).
We in this paper conduct a quantitative diagnosis on how the overall noise performance including mutual coupling contributions, of the 19-element hexagonal PAF designed in (Wu et al. 2013) and recently upgraded, varies over time and under different ambient environmental conditions. Moreover, we present a modified Y-factor method to decompose the mutual coupling components from accumulated overall system noise. This paper is structured as follows. Section 2 details the PAF hardware layout, hot/cold loads and shielding configuration for Y-factor method, full signal chain and end-to-end data processing pipeline. Section 3 presents measured receiver noise, and total system noise temperature against frequency and time, with quantitative contributions of inter-element coupling noise and spillover. Section 4 makes discussion of the physical mechanisms governing time-variant noise fluctuations including coupling, spillover and RFI effects, and comparison between our multi-epoch measurement scheme and others. Section 5 summarises main experimental conclusions and outlines prospective optimisations.
Method
Experimental configurations
We measured the equivalent system noise temperature () of the 19-element PAF receiver using the Y-factor method, with observations conducted on four days between March and June 2026 (March 04, April 12, May 08, and June 10). This method uses power measurement data obtained with a hot load and a cold load. The power ratio between the two states yields the ratio of their equivalent noise temperatures, from which we calculate the system noise budget.
During the system noise temperature measurements, a custom-built stainless steel screen was erected around the array. The screen was tilted at an angle of relative to the ground with a height of approximately 1 m, to provide partial shielding of the array from noise radiated by the ground, horizon, mountain and nearby buildings. When the array equipped with the screen observes the background radio sky, the backend records the voltage of electromagnetic waves from the radio sky as cold load, which corresponds to the cold temperature in the Y-factor method.
For the cold-load observations, we conducted additional measurements without the steel shielding screen, so that we could compare the results with those obtained under the shielded condition.
As for the hot load, we assembled and fixed sixteen pyramidal foam microwave absorbers on a plywood base plate to form a hot load with dimensions of and a height of 0.34 m. At 1.5 GHz, the absorbers exhibit a minimum reflection of dB for normal incidence, according to the manufacturer’s datasheet. During operation, the hot load just covers the array equipped with the screen. The tips of the microwave absorbers are mm above the ground and mm above the dipoles of the array.
The photos and a sketch (upper left) in Figure 1 show the sizes of the PAF system, the steel shielding screen, the hot-load etc., and the configurations during the measurements of the four days.

Figure 1. A sketch in upper left of this figure and five photos show the sizes of the PAF system, the steel shielding screen, the hot-load etc., and the configurations during the measurements of the four days. The date is noted on the upper right of each photo.
Feed Array and signal chain
The PAF under test consists of 19 hexagonal arranged back-cavity dipole elements with element spacing ( mm), and the element number is from 0 to 18. The PAF bandwidth is 1.05-1.45 GHz with dual polarization (Wu et al. 2013). The maximum radial distance from the centre of the array is 350 mm. The 19-element array is deployed within a radio quiet zone − the mountain saddle of the YinXiao Cave depression in LongShan, HuNan province.
The radio-frequency signal from each PAF antenna element first enters a low-noise amplifier (LNA, ZX60-P162LN+). The amplified signal passes sequentially through a bandpass filter (BPF, 1370–1470 MHz), a second amplifier (ZRL-2150+), and another bandpass filter. After further amplification by the ZFL-2500VHX+ amplifier, the signal is transmitted via optical fiber (OF) using an optical transmitter module (OTM) and an optical receiver module (ORM).The received electrical signal goes through a 1370–1470 MHz bandpass filter and then enters a mixer (Mix). A tunable local oscillator (LO) generator provides the reference signals for frequency down-conversion. The intermediate-frequency output is filtered by a DC–40 MHz low-pass filter (LPF), then amplified by a broadband amplifier with 40 dB gain, 5 dB noise figure. A second DC–40 MHz low-pass filter suppresses high-frequency spurious components before the signal is fed into the digitizing data acquisition unit.
Figure 2 illustrates the electromagnetic wave receiving chain, from antenna input to voltage data logging on the server.

Figure 2. Signal flow block diagram illustrating the electromagnetic wave receiving chain, from antenna input to voltage data logging on the server.
Backend and raw data preprocessing
We adopt the ROACH2 platform equipped with the HMCAD1511 analog-to-digital converters. This is an 8-bit ADC operating at a sampling rate () of 77 MSps ( MHz) and analog full-scale input V. The sampled data are transmitted to the server for offline processing.
We record the baseband data, and calculate a spectrum using fast Fourier transform (FFT) with sampling points and the Hanning window. The highest frequency of the spectrum, given by the Nyquist frequency, is . The time required to record 2048 sampling points is microseconds. The minimum frequency and frequency resolution is MHz, and the channel number is 1024. In order to reduce the data size, the 1024 initial frequency channels are rebinned into 64 channels via power summation of every 16 channels, and the rebinned frequency interval is MHz. We in this paper calculate the system noise temperature as a function of frequency with the rebinned frequency interval . The total observation time of each measurement is about 10.5 s with time resolution microseconds.
We carried out one system noise temperature measurement together with one backup observation on each of the four observing days. Within each observing run, we collected three datasets: hot load, cold load with the metal screen, and cold load without the screen. Each dataset occupies approximately 15.77 GB, summing to 47.31 GB per observing run. To extend the frequency coverage, we varied the local oscillator (LO) frequency across different measurement sessions. The LO configurations are as follows: 1.35 GHz and 1.41 GHz on March 04; 1.35 GHz on April 12; 1.35 GHz and 1.49 GHz on May 08; and 1.35 GHz, 1.40 GHz, and 1.45 GHz on June 10. The total size of all raw observational data is approximately 757 GB, and the storage volume of preprocessed data reaches 1.152 TB.
Data reduction procedures
We developed a python-based calibration pipeline to perform consistent phase and amplitude correction for the 19-element phased array antenna system. The pipeline processes segmented time-domain voltage datasets acquired from individual antenna channels. The data reduction procedures are as follows:
Raw data files are systematically screened to discard invalid samples, such as incomplete matrices, to guarantee high-quality inputs for downstream processing;
Before calibration, segment-wise zeroing is applied to preprocessed voltage amplitudes via mean subtraction to remove DC baseline offsets;
To minimize the system noise temperature, iterative mutual calibration is implemented across the array: each antenna element is sequentially selected as the temporary reference to correct the complex responses of all remaining receiving array channels;
Fully calibrated time-series data are forwarded to the beamforming stage, where the maximum signal-to-noise ratio (maxSNR) criterion is utilized to solve for the optimal complex amplitude weighting coefficients of every antenna element.
We adopt a narrowband complex baseband signal model analogous to the formalism presented in Boonstra & van der Veen [2]. The signal bandwidth is sufficiently narrow such that inter-element propagation delays only introduce constant phase offsets rather than waveform distortion. Considering an M-element PAF targeted at a source signal during calibration observations, denote the narrow band complex baseband signal at frequency from the -th antenna channel at time . The single-channel signal model can be expressed as , where represents the lumped amplitude gain of the -th receiver channel, accounting for cable resistive losses, amplifier gain variations and environmental attenuation; denotes the cumulative instrumental phase offset originating from hardware delay mismatches; represents the geometric phase shift induced by the propagation time delay of the source wavefront relative to the reference antenna element; and is the uncorrelated system noise contribution from passive loss noise, LNA noise and thermal background spillover.
The signals from all M elements can form a column vector , which equals to
where is array gain vector combined with all channel-dependent complex response terms. The core goal of iterative mutual calibration is to solve for the unknown complex entries of by eliminating the inter-channel amplitude and phase inconsistencies.
The calibration procedure consists of two steps: phase alignment and amplitude equalization. First, FFT-based cross-correlation is employed to achieve precise time-domain synchronization between each target antenna and the reference channel. After removing DC offsets from both the reference and target signals, the optimal temporal shift is determined from the correlation peak, and each waveform is circularly realigned to match the reference phase.
An amplitude scaling factor is then calculated according to the variance ratio between the reference signal and each individual channel, which normalizes the fluctuation power of all antenna outputs to a consistent level. The relative variance error between each calibrated channel and the reference is quantitatively recorded to evaluate calibration performance. This two-dimensional joint calibration effectively eliminates inter-channel phase delays and amplitude inconsistencies caused by instrumental differences and transmission variations across the array.
In the maxSNR technique, we first solve the generalized eigenvalue equation where and is the signal-plus-noise covariances and noise covariances obtained from the observations, the dominant eigenvector associated with the maximum eigenvalue to the eigenvalue equation. We in this study consider the hot-load is the signal source, and the cold sky is the external noise source apart from the PAF receiver. In this step, we obtain the signal-of-interest covariance matrix , where represents the array response, and the superscript denotes the Hermitian conjugate (conjugate transpose) of a matrix. We then solve the equation to obtain the beamformer weight [11].
Noise temperature budget
The beam equivalent system noise temperature can be expressed as [23, 11]
where is the radiation efficiency of PAF and here we adopt ; the sum of noise temperature in the main lobe caused by the cosmic microwave background (CMB) radiation [7], galactic background , and atmospheric noise ; We assume that the equivalent spillover noise temperature due to electromagnetic waves leak from sidelobes and backlobes into the receiving system is ; is the beam equivalent receiver noise temperature. We have , where is the noise temperature contribution of discrete electronic devices from the first LNA to the ADC, the loss noise including resistive losses in the antenna, cables, and connectors ahead of the LNA, dominated by the coaxial cable loss, the noise temperature due to the mutual coupling which will be dealt with in Eq. (4).
Assuming that there is perfect impedance match between the element antenna and the immediately following LNA, and given that noise temperature scales linearly with the power received by the PAF system, the equivalent PAF receiver noise temperature can be solved by
where and are the known voltage noise covariance matrices calculated from the measurement data when the PAF pointed to the cold load and hot load, the weighting factor matric of array dipole elements calculated from the maxSNR technique. We assume that the temperature of hot load equals to the isotropic ambient temperature. The measured ambient temperature was near 288 K on March 04, 292 K on April 12, 289 K on May 08, and 301 K on June 10. In order to easily compare the measured on the different dates, we set the ambient temperature to be 288 K unless otherwise specified.
Modified Y-factor algorithm for mutual coupling extraction
Electromagnetic waves reflected from neighboring antenna elements, as well as reflected signals caused by impedance mismatches along the RF feed chains, can re-excite the array apertures and couple into the PAF receiver channels, leading to fluctuations in the system noise temperature. This phenomenon is referred to as the inter-element mutual coupling effect. The noise temperature contribution originating from such mutual coupling is incorporated into the equivalent receiver noise temperature during measurement.
We estimate the noise contribution induced by inter-element mutual coupling by computing the proportion of sum of the off-diagonal entries relative to the total sum of all entries within the cold-load covariance matrix and hot-load covariance matrix . To the 19-element array, the mutual coupling noise here is defined as
where the proportion is
with being the matrix entry in , and subscript * denoting cold or hot. Note that . This definition allows us to investigate how the mutual coupling noise varies with the cold and hot loads.
We can calculate the and from Eqs. (3) and (4), and the result of the modified Y-method are
and
where we let . When the mutual coupling can be neglected, i.e. , the Eq. (6) becomes for the Y-factor method.
Results
Equivalent receiver noise
The results are presented in Figure 3. For the local oscillator (LO) frequency of 1.35 GHz, the lowest (#1 in Figure 3) averaged values are 67.14±3.13 K, 75.36±3.88 K, 70.33±3.87 K, and 78.62±4.81 K for the observations on March 04, April 12, May 08, and June 10, respectively. At an LO frequency of 1.41 GHz, the measured average on March 04 is 65.12±3.63 K. For the 1.40 GHz LO configuration on June 10, the average is 75.15 ± 3.28 K. In addition, the derived values at 1.42 GHz are approximately 73 K and 75 K for the March 04 and June 10 observations, respectively. For comparison, we also plot the top3 averaged in Figure 3. The averaged receiver temperature and standard deviation are listed in Table 1. Note that when calculating the average values, we discarded anomalous data points with extreme high or low values, as well as samples falling outside the operating bandpass.
| Measurement Date | LO Frequency | Rank | Ref Index | Mean (K) | Std. Deviation (K) |
| 2026-03-04 | 1.35 GHz | 1 | 18 | 67.14 | 3.13 |
| 2026-03-04 | 1.35 GHz | 2 | 14 | 67.87 | 3.09 |
| 2026-03-04 | 1.35 GHz | 3 | 6 | 68.00 | 3.44 |
| 2026-03-04 | 1.41 GHz | 1 | 7 | 65.12 | 3.63 |
| 2026-03-04 | 1.41 GHz | 2 | 14 | 65.68 | 3.41 |
| 2026-03-04 | 1.41 GHz | 3 | 2 | 66.88 | 2.56 |
| 2026-04-12 | 1.35 GHz | 1 | 14 | 75.36 | 3.88 |
| 2026-04-12 | 1.35 GHz | 2 | 5 | 75.96 | 5.39 |
| 2026-04-12 | 1.35 GHz | 3 | 1 | 79.06 | 8.20 |
| 2026-05-08 | 1.35 GHz | 1 | 1 | 70.33 | 3.87 |
| 2026-05-08 | 1.35 GHz | 2 | 14 | 71.89 | 4.06 |
| 2026-05-08 | 1.35 GHz | 3 | 18 | 72.02 | 4.25 |
| 2026-05-08 | 1.49 GHz | 1 | 14 | 78.50 | 6.60 |
| 2026-05-08 | 1.49 GHz | 2 | 18 | 78.54 | 6.93 |
| 2026-05-08 | 1.49 GHz | 3 | 1 | 79.01 | 7.21 |
| 2026-06-10 | 1.35 GHz | 1 | 14 | 78.62 | 4.81 |
| 2026-06-10 | 1.35 GHz | 2 | 6 | 79.69 | 6.90 |
| 2026-06-10 | 1.35 GHz | 3 | 18 | 79.95 | 4.96 |
| 2026-06-10 | 1.40 GHz | 1 | 18 | 75.15 | 3.28 |
| 2026-06-10 | 1.40 GHz | 2 | 14 | 75.78 | 3.85 |
| 2026-06-10 | 1.40 GHz | 3 | 6 | 79.29 | 4.54 |
| 2026-06-10 | 1.45 GHz | 1 | 8 | 81.64 | 5.40 |
| 2026-06-10 | 1.45 GHz | 2 | 6 | 82.26 | 4.75 |
| 2026-06-10 | 1.45 GHz | 3 | 12 | 82.29 | 5.91 |
Table 1. Top-3 references with the minimum average receiver temperature under different LO frequencies and measurement dates.

Figure 3. The equivalent receiver noise temperature () as a function of radio frequency. The panels from top to bottom correspond to measurements conducted on 2026 March 04, April 12, May 08, and June 10, respectively. The labels in the lower-right corner of each panel indicate the reference antenna index and local oscillator (LO) frequency used for array calibration. For example, ref18 LO1.35GHz denotes that the measurement adopts reference antenna 18 with an LO frequency of 1.35 GHz. (#1), (#2), (#3) denotes the top3 averaged in each measurement. Each measurement covers a frequency bandwidth of 38.5 MHz spanning 1350–1490 MHz, with a frequency interval of approximately 0.60 MHz. The total observation time of each measurement is about 10.5s with time resolution .
As shown in Table 1, the standard deviations of measured are distributed between approximately 2 K and 8 K for all local oscillator (LO) frequencies. However, we can see that in Figure 3 the anomalous data points deviate substantially from the average. Part of this discrepancy arises from RFI contamination, while another component can be attributed to aliasing, which will be discussed in § 4. The severe signal distortion observed above 1.478 GHz originates from the intrinsic bandpass response of element antenna (1.05–1.45 GHz), combined with the bandpass filter covering 1.37–1.47 GHz.
An interesting trend can be observed for the LO frequency of 1.35 GHz: the average rose from 67 K on March 4 to 75 K on April 12, then decreased to 70 K on May 8, and finally climbed back to 79 K on June 10. For the 1.40 GHz LO channel, measured in June is approximately 10 K higher than that obtained in March. This temporal drift in may be partially attributed to temperature variations of discrete electronic components. We measured the physical temperature of the LNA with a thermometer on May 8 and June 10, which were 37.5 °C and 37.3 °C, respectively. When the measurement was conducted on March 4, there were no barriers surrounding the PAF system. Afterwards, enclosures made of wood-fiber panels were installed around the PAF to shield the array from wind and rain. On April 12, we observed elevated equipment temperature caused by poor air circulation inside the enclosure; ventilation improvements including drilled holes were subsequently implemented, which yielded a noticeable reduction in the PAF operating temperature within the barriers. The ambient temperature reached approximately 16 °C on May 8, considerably lower than the 28 °C ambient temperature recorded on June 10.
According to the manufacturer’s datasheet of Mini-Circuits ZX60-P162LN+ which was used as the first-stage LNA in the PAF receiver, the noise figure is 0.39 dB at −45°C, 0.65 dB at 25°C, and 0.84 dB at 85°C. Using piece-wise linear interpolation, we obtain noise-figure values of 0.61 dB and 0.69 dB for this LNA at 15°C and 38°C, respectively. The corresponding equivalent noise temperatures are 43.4 K and 49.6 K, yielding an increase of approximately 6 K. The overall receiver noise temperature reaches 67 K and 79 K at 15°C and 38°C. This may suggest that roughly 50 % of the receiver-noise variation over the temperature range from 15°C to 38°C can be attributed to this first-stage LNA.
We compare the measurements on June 10 in Figure 4. The panels from top to bottom in this figure correspond to measurements conducted for the cold load with the steel shielding screen (measurements 1), backup observation 1-6 mins after measurements 1 (measurements 2), and the cold load without the screen (measurements n1). Minor subtle fluctuations of against different reference antenna can be observed within several minutes from the figure. Notably, after removing the metal shielding screen, the measured significantly increases from approximately 75 K (with the screen) to over 142 K. This phenomenon is likely caused by electromagnetic interference coupled through the antenna sidelobes, where the noise contributors may be mountainous terrain, vegetation and surrounding buildings nearby.

Figure 4. The equivalent system noise temperature () as a function of radio frequency on June 10. The panels from top to bottom correspond to measurements conducted for the cold load with the steel shielding screen (measurements 1), backup observation 1-6 mins after measurements 1 (measurements 2), and the cold load without the screen (measurements n1). The labels in the lower-right corner of each panel indicate the reference antenna index and local oscillator (LO) frequency used for array calibration. For example, ref14 LO1.35GHz denotes that the measurement adopts reference antenna 14 with an LO frequency of 1.35 GHz. (#1), (#2), (#3) denotes the top3 averaged in each measurement. Each measurement covers a frequency bandwidth of 38.5 MHz spanning 1350–1490 MHz, with a frequency interval of approximately 0.60 MHz. The total observation time of each measurement is about 10.5s with time resolution .
The noise received by the antenna array can be decomposed into contributions originating from the main beam, sidelobes and backlobes. The noise temperature of the antenna array can be estimated as , where , and are the main beam, sidelobe beam and backlobe beam efficiencies respectively. The beam efficiency (: mb, side, back) is defined as the ratio of the gain-weighted beam solid angle integrated over the corresponding angular region to the total array beam solid angle , i.e. , where and , where is the antenna gain. These efficiencies satisfy the normalization condition . In the case of stainless-steel screen installed, we can estimate the sidelobe noise temperature via , where is the L-band emissivity of stainless steel (), is the environmental temperature on June 10, and represents the cold-sky radiation reflected by the inner surface of the steel screen. We here take (Wu et al. 2013) and . When the steel screen was removed, the emissivity becomes , and due to the surrounding mountains. If we assume that the increase of sidelobe-induced noise from the screened and unscreened case results in 67 K rise in the receiver noise, this would suggest a beam efficiency of 0.76, which is lower than the mean value () of beam efficiency in Chippendale et al. [4] by 15.8%. In the next work, we will pay attention to the radiation-pattern measurement to understand the beam efficiency and antenna noise budget.
Influences of beamforming weights
We compare the influences of different reference antennas and beamforming weights on receiver noise in Figure 5. We find that compared with equal-weight beamforming, the maxSNR beamforming combined with the noise-whitening method can suppress receiver noise and yields estimates with lower variance and reduced bias. This is in line with previous studies [e.g. 11]. Furthermore, our results indicate that the performance gain of the maxSNR beamforming method may be diminished when the antenna-calibrated measurement data inherently exhibits low variance. Future work will include further comparisons of additional beamforming strategies and their impacts on measurements.

Figure 5. The equivalent receiver noise temperature () as a function of radio frequency on June 10 under LO 1.35GHz. The top panel is for the reference antenna 14 with equal weight and maxSNR weight, while the bottom panel is for the reference antenna 18. Each measurement covers a frequency bandwidth of 38.5 MHz, with a frequency interval of approximately 0.60 MHz. The total observation time of each measurement is about 10.5s with time resolution . The solid circles, squares, triangles and diamonds denote the measured data points at the corresponding frequencies.
Mutual coupling
We plot mutual coupling temperature and coupling proportion (sum of the off-diagonal entries relative to the total sum of all entries within the cold-load covariance matrix and hot-load covariance matrix ) in Figure 6 (2026 March 04) and 7 (2026 June 10).

Figure 6. The mutual coupling noise () (top panel) and the coupling proportion (bottom panel) as a function of radio frequency on 2026 March 04. Labels in this figure, namely LO* and ref* denote the LO frequency and reference antenna used in the measurements, e.g., LO1.35GHz and ref18 means the LO frequency is 1.35GHz, and the number of reference antenna is 18. Each measurement covers a frequency bandwidth of 38.5 MHz, with a frequency interval of approximately 0.60 MHz. The solid and dashed lines are for the hot and cold respectively. The solid circles, squares, triangles and diamonds denote the measured data points at the corresponding frequencies. The total observation time of each measurement is about 10.5s with time resolution .
Figure 6 illustrates and across two LO frequencies spanning 1.35–1.45 GHz, with paired hot (solid lines, circle markers) and cold (dashed lines, square markers) measurements for reference antenna ref18 (LO1.35GHz: LO 1.35GHz) and ref7(LO1.41GHz: LO 1.41GHz). The measurements on Figure 7 uses three LO frequencies 1.35GHz, 1.40GHz, and 1.45GHz, and corresponding references antennas are ref14, ref 18 and ref8.

Figure 7. Same with Figure 6, but the date of measurements took place was 2026 June 10.
In Figure 6, when LO frequency is 1.35GHz, hot starts at K, gradually climbs to 14–18 K between 1.37–1.38 GHz, and features tall, narrow resonant spikes above 22 K. Cold remains close to 0–2 K with smaller matching resonant peaks. The curve reflects this behavior: a slowly rising baseline around 0.02–0.05 for hot calibration, plus sharp resonant peaks exceeding 0.10, while cold signals stay weak with narrow spikes. For LO frequency 1.41GHz, hot forms a flat, stable baseline near 8 K with only a few isolated narrow resonances, while cold is constrained to –2 K. The baseline is uniformly low at for both hot and cold, with only minor narrow resonant spikes at select frequency points. Similar to the LO1.35GHz dataset, hot calibration produces systematically stronger mutual coupling signatures than cold calibration across both bands, and resonant coupling appears as discrete narrow peaks superimposed on smooth, slowly varying baselines.
In Figure 7, for LO1.35GHz, hot maintains a stable baseline around 8–10 K, with sharp narrow resonant peaks up to 16 K; cold stays much lower at roughly 0–4 K. The corresponding follows the same trend: the hot curve hovers near 0.02 with resonant peaks reaching 0.20, while the cold signal exhibits weaker but consistent resonant features. The LO1.40GHz setup delivers distinctly elevated coupling signals. Its hot rises to a broad plateau of 16–22 K, far higher than the other two bands, with a smooth, wide hump rather than narrow discrete resonances. Cold remains suppressed at 2–4 K. shows a broad, mild peak around 0.06 for hot calibration and 0.04 for cold, without sharp narrow spikes. When using LO1.45GHz, the overall coupling returns to levels comparable to LO1.35GHz.
Across all LO frequency setup, hot calibration consistently yields significantly larger and values than cold calibration, indicating that hot-load measurement amplifies observable mutual coupling effects. However, we find that for the measurement of LO1.35GHz on June 10, at many frequency bins, cold values are larger than hot . This phenomenon could be caused by RFI, and further measurements are necessary to understand such phenomenon.
The measurements on both dates demonstrate that: 1. hot-load measurements produce larger and than cold-load measurements. This validates that hot calibration provides higher signal contrast to characterize receiver mutual coupling; 2. We can see that a smooth, frequency-dependent baseline coupling modulated by sharp, narrow resonant spikes probably originating from standing waves or array element crosstalk; 3. low-frequency bands (1.35–1.39 GHz) consistently show stronger frequency evolution of the baseline coupling than higher-frequency bands (1.40+ GHz) in the measurements.
The consistent gap between hot and cold traces across both datasets confirms that the measured and can reflect the receiver’s intrinsic mutual coupling. The dramatic variation among measurements with LO bands and different reference channels demonstrates that mutual coupling is highly sensitive to reference hardware and front-end assembly. The hot and cold measurements capture distinct hardware configurations, enabling us to disentangle the contributions of mutual coupling to the overall performance of the PAF receiver system.
Equivalent system noise temperature
The measured system noise temperature is shown in Figure 8. For the local oscillator (LO) frequency of 1.35 GHz, the averaged values are 74.87 ± 3.13 K, 83.09 ± 3.88 K, 78.06 ± 3.87 K, and 86.35 ± 4.81 K for the observations on March 04, April 12, May 08, and June 10, respectively. At an LO frequency of 1.41 GHz, the measured average on March 04 is 72.85 ± 3.63 K. For the 1.40 GHz LO configuration on June 10, the average is 82.88 ± 3.28 K. In addition, the derived values at 1.42 GHz are approximately 81 K and 83 K for the March 04 and June 10 observations, respectively.

Figure 8. The equivalent system noise temperature () is plotted as a function of radio frequency. The panels from top to bottom correspond to measurements conducted on 2026 March 04, April 12, May 08, and June 10, respectively. The labels in the lower-right corner of each panel indicate the reference antenna index and local oscillator (LO) frequency used for array calibration. For example, ref18 LO1.35GHz denotes that the measurement adopts reference antenna 18 with an LO frequency of 1.35 GHz. Each measurement covers a frequency bandwidth of 38.5 MHz spanning 1350–1490 MHz, with a frequency interval of approximately 0.60 MHz. The total observation time of each measurement is about 10.5s with time interval .
The averaged equivalent system noise temperature and the main components over the frequency band 1350-1390MHz for the measurements on 2026 March 04 and June 10 are shown in Table 2.
| 2026 March 04 LO1.35GHz ref18 | 2026 June 10 LO1.35GHz ref14 | |
| 55 ± 3 K (measured) | 64 ± 5 K (measured) | |
| 2 ± 1 K (measured) | 4 ± 2 K (measured) | |
| 5.73 K (estimated) | 5.73 K (estimated) | |
| 10 K (estimated) | 10 K (estimated) | |
| 2 K (estimated) | 2 K (estimated) | |
| 75 ± 3 K | 86 ± 5 K |
Table 2. Averaged equivalent system noise temperature and the main components over the frequency band 1350-1390MHz for the measurements on 2026 March 04 and June 10 in our observations, adopting ambient temperature of 288K and maxSNR beamformer.
Discussion
Aliasing affects the measurements with LO frequencies of 1.40–1.45 GHz. Taking LO = 1.40 GHz as an example, signals from 1.36–1.40 GHz alias into the plotted 1.40–1.44 GHz band. Owing to such spectral overlap, the corresponding noise data can only act as a reference for the overall system noise performance across the bandwidth of 1.36–1.45 GHz. In our case, the receiver output power under aliased conditions can be written as , where is the Boltzmann constant, denotes the receiver gain, and , are the bandwidths corresponding to the negative-frequency and positive-frequency components, respectively. We find that the Y-factor remains unchanged. The measured results are consistent with this simplified theoretical analysis. Note that the presence of strong RFI may influence the Y-factor measurement. We will upgrade and reconfigure hardware in subsequent measurement campaigns to suppress this spectral folding effect.
We can alternatively calculate the receiver noise temperature due to the first stage LNAs and subsequent circuits by the cascaded noise factor formula
where , , denotes the noise factor of the first stage amplifier, second stage amplifier, third stage amplifier, and , , the gain of each stage amplifier respectively. From the formula, we see that if the noise factor is sufficiently small or the gain is large, the noise temperature of a cascaded system is mainly dominated by the first stage LNA.
| name | gain | noise figure | contribute to NF | contribute to |
| ZX60-P162LN+ | 20dB | 0.6dB | 0.6dB | 43K |
| ZRL-2150+ | 25dB | 1.5dB | 0.053dB | 4K |
| ZFL-2500VHX+ | 20dB | 5.5dB | 0.0013dB | 0.15K |
Table 3. The parameters of main amplifiers used in the PAF and their contributes to total NF and .
We used the ZX60-P162LN+ manufactured by Mini-circuits as the first stage LNA. It has a noise figure of 0.6 dB in room temperature and gain of 20 dB at about 1.4GHz. The measured total equivalent receiver noise figure using a noise source in front of the first level LNA is about 0.67 dB, which lead to a noise temperature of about 48 K. The gain, noise figure and noise temperature contribution of each amplifier (typical) are listed in table 3. This result is approximately consistent with our noise temperature measurements.
Simulations and S-parameter measurements of the six inner-ring antenna elements in the PAF (Wu et al. (2013)) show that the maximum transmission coefficient between the central element and an adjacent inner-ring element reaches -18 dB and -24 dB. Given an ambient temperature of 288 K, the corresponding mutual coupling noise temperatures are estimated to be approximately 4.6 K and 1.2 K respectively. S-parameter simulations and measurements for the central element and 12 outer-ring elements show , yielding a mutual coupling noise contribution in the range of 0.03-0.3 K. The overall mutual coupling noise temperature of the full array ranges approximately from 7.6 K to 10.7 K. Our measurements agree well with S-parameter data, and further indicate that mutual coupling’s contribution to system noise temperature depends on ambient radio sky temperature and RFI levels.
Chippendale et al. [4] presented system noise temperature measurements for phased-array antennas operating in receiving band from 0.7 GHz to 1.8 GHz. Similar to our experiments, they also used a large microwave absorber at ambient temperature and the unobstructed radio sky as the hot and cold load. The main differences between their work and ours are: 1) they used a broadband reference antenna as noise source centred over and pointed at the array under test, which can make noise measurements with reproducible, well-defined beams directed at the zenith and thereby at the centre of the absorber target; 2) our multi-band, multi-epoch measurements utilise deployable shielding screens to isolate sidelobe spillover noise; 3) we propose a covariance-based method to extract mutual coupling temperature . Validated against S-parameter simulations and tests, our work further identifies the dependence of coupling noise on radio frequency interference (RFI) and radio sky temperature, a finding unreported in the CSIRO research. Additionally, our pipeline incorporates iterative channel calibration and long-term multi-epoch experiments over months, whereas the CSIRO workflow focuses on single experiment.
Conclusion
In this paper, we present comprehensive system noise temperature measurements of a 19-element hexagonal room-temperature PAF via the modified Y-factor method across multiple frequency bands and multi-month observation campaigns. We developed a complete Python-based data reduction pipeline containing DC offset removal, iterative inter-channel mutual calibration and maxSNR beamforming to do the data reduction. A detachable stainless steel shielding screen was deployed to quantitatively isolate sidelobe spillover noise originating from terrain and surrounding environments. Comparative measurements demonstrate that unshielded sidelobe pickup elevates the receiver noise temperature by more than 60 K, clearly verifying the contamination of spillover radiation to noise characterization.
Based on the covariance matrix off-diagonal terms extracted from hot/cold calibration datasets, a quantitative decomposition scheme for mutual coupling noise temperature was proposed as shown in Sec. 2.6. The baseline of the measured mutual coupling noise temperature ranges from approximately 2 K to 8 K, and can reach up to 20 K under certain observational conditions. Cross-validation with element-to-element S-parameter results shows that inner adjacent elements with dB generate coupling noise of 1.2–4.6 K, while distant outer-ring elements ( dB) only contribute approximately 0.03–0.3 K. The integrated mutual coupling noise of the full array falls within 7.6–10.7 K, which also matches well with electromagnetic simulation predictions.
Consistent offsets of between hot and cold calibration states confirm that hot-load observations provide higher signal levels of inter-element crosstalk by contrast with cold-load. We further reveal that mutual coupling noise is jointly induced by the thermal radiation associated with the ambient radio-sky temperature and non-thermal radiation most likely originating from RFI, an effect unreported in previous PAF noise measurement studies including Chippendale et al. [4]. Long-term monitoring over March–June 2026 captures obvious temporal drift of receiver noise, which may be mainly attributed to the temperature fluctuation of front-end LNAs and heat dissipation conditions inside the protective enclosure.
The cascaded noise factor calculation indicates that the first-stage LNA dominates the receiver noise budget, consistent with our measured ranging from 65 K to 82 K under different LO configurations. Combined with sky background, spillover, loss and mutual coupling contributions, the overall beam-equivalent system noise temperature of the PAF is measured to be 73–89 K within the 1.37–1.47 GHz operating band. The measurement framework, channel calibration workflow and mutual coupling decomposition method established in this work offer a practical reference for noise performance evaluation of room-temperature PAF prototypes for ground-based radio telescopes. In future work, we will compute adaptive beamforming weights from long-term covariance matrix measurements and measure the radiation pattern of the PAF receiver to jointly characterize the temporal variation of system noise.
Acknowledgements We thank the reviewer for the helpful comments. This work was supported by the Science and Technology Innovation Program of Hunan Province (No. 2024JC0001 No. 2025RC4002), and the Investigation of Technological Infrastructure Resources (No. 2023FY101101). This work was also supported by the CAS Project for Young Scientists in Basic Research, Grant No. YSBR-063 and the National Natural Science Foundation of China (Grant No. 11073027, 11673031, 12373104). We are grateful to Mr. Shengyi Zhou (People’s Government, Longshan County) and Mr. Jie Zhu (People’s Government of Guitang Town) for their strong local supports.
References
- [1]Bannister, K. W., Shannon, R. M., Macquart, J.-P., et al. 2017, ApJ, 841, L12 2
- [2]Boonstra, A., & van der Veen, A. 2003, IEEE Transactions on Signal Processing, 51, 25 5
- [3]Burnett, M. C., Kunzler, J., Nygaard, E., et al. 2020, in 2020 XXXIIIrd General Assembly and Scientific Symposium of the International Union of Radio Science (IEEE), 111 2
- [4]Chippendale, A. P., Hayman, D. B., & Hay, S. G. 2014, PASA, 31, e019 2, 9, 15, 16
- [5]Cortes-Medellin, G., Vishwas, A., Parshley, S. C., et al. 2015, IEEE Transactions on Antennas and Propagation, 63, 2471 2
- [6]Deng, X., Chippendale, A. P., Hobbs, G., et al. 2017, PASA, 34, e026 2
- [7]Fixsen, D. J. 2009, ApJ, 707, 916 6
- [8]Groves, III, W. M., & Morgan, M. A. 2017, PASP, 129, 085001 2
- [9]Heywood, I., Bannister, K. W., Marvel, J., et al. 2016, MNRAS, 457, 4160 2
- [10]Hotan, A. W., Bunton, J. D., Chippendale, A. P., et al. 2021, PASA, 38, e009 2
- [11]Landon, J., Elmer, M., Waldron, J., et al. 2010, AJ, 139, 1154 2, 6, 10
- [13]Melis, A., Chiello, R., Comoretto, G., et al. 2020, Journal of Astronomical Instrumentation, 9, 2050013 2
- [14]Pingel, N. M., Pisano, D. J., Ruzindana, M., et al. 2021, AJ, 161, 163 2
- [15]Roshi, D. A., Shillue, W., Simon, B., et al. 2018, AJ, 155, 202 2
- [16]Röttgering, H., Afonso, J., Barthel, P., et al. 2011, Journal of Astrophysics and Astronomy, 32, 557 2
- [17]van Ardenne, A., Bregman, J. D., van Cappellen, W. A., Kant, G. W., & de Vaate, J. G. B. 2009, IEEE Proceedings, 97, 1531 2
- [18]van Cappellen, W. A., & Bakker, L. 2010, in 2010 IEEE International Symposium on Phased Array Systems and Technology (IEEE), 109 2
- [19]van Cappellen, W. A., Oosterloo, T. A., Verheijen, M. A. W., et al. 2022, A&A, 658, A146 2
- [20]Veidt, B., Burgess, T., Messing, R., Hovey, G., & Smegal, R. 2009, in 2009 13th International Symposium on Antenna Technology and Applied Electromagnetics and the Canadian Radio Science Meeting, Vol. 13 (IEEE), 87 2
- [21]Veidt, B., & Dewdney, P. 2005, in 11th International Symposium on Antenna Technology and Applied Electromagnetics [ANTEM 2005] (IEEE), 177 2
- [22]Verheijen, M. A. W., Oosterloo, T. A., van Cappellen, W. A., et al. 2008, in American Institute of Physics Conference Series, Vol. 1035, The Evolution of Galaxies Through the Neutral Hydrogen Window, ed. R. Minchin & E. Momjian (AIP), 265 2
- [23]Warnick, K. F., & Jeffs, B. D. 2008, IEEE Antennas and Wireless Propagation Letters, 7, 565 2, 6
- [24]Warnick, K. F., & Jensen, M. A. 2005, IEEE Transactions on Antennas and Propagation, 53, 2490 2
- [25]Warnick, K. F., Maaskant, R., Ivashina, M. V., Davidson, D. B., & Jeffs, B. D. 2016, IEEE Proceedings, 104, 607 2
- [26]Warnick, K. F., Woestenburg, B., Belostotski, L., & Russer, P. 2009, IEEE Transactions on Antennas and Propagation, 57, 1634 2