PRATUSH Radiometer: 21-cm Cosmic Dawn Probe

• PRATUSH radiometer is a space-based instrument that detects the global 21-cm signal from Cosmic Dawn/EoR using a calibrated wideband antenna and digital spectrometer.
• The design employs analog bandpass calibration, in-situ antenna impedance monitoring, and an FPGA-based digital backend to achieve mK-level sensitivity.
• Its mission leverages lunar-farside operations to mitigate terrestrial RFI, enabling high-fidelity measurements of subtle spectral inflections critical for cosmological studies.

PRATUSH, short for Probing ReionizATion of the Universe using Signal from Hydrogen, is a proposed space-based radiometer for detecting the global, sky-averaged redshifted 21-cm signal from the Cosmic Dawn (CD) and Epoch of Reionization (EoR). The experiment concept is centered on precision low-frequency spectral radiometry in a radio-quiet orbital environment, with a broader science envelope of 40–200 MHz and a baseline design optimized for the 55–110 MHz Cosmic Dawn band. In its baseline form, PRATUSH combines a smooth wideband antenna, calibrated analog bandpass measurements, in-situ antenna impedance monitoring, and an FPGA-based digital correlation spectrometer, with lunar-farside operation providing the principal mitigation of terrestrial radio-frequency interference (RFI) (Rao et al., 8 Jul 2025).

1. Scientific target and observational regime

The primary scientific objective of PRATUSH is the detection of the global 21-cm signal from neutral hydrogen during CD/EoR. In this context, the 21-cm hyperfine line at rest frequency 1420.406 MHz, when redshifted to meter wavelengths, traces the average thermal and ionization history of the intergalactic medium. The targeted cosmological signature is a weak spectral distortion with a maximum amplitude below a few hundred mK, whereas astrophysical foregrounds are hundreds to thousands of kelvin, so spectral smoothness and calibration fidelity are fundamental rather than auxiliary requirements (S. et al., 8 Jul 2025).

The PRATUSH literature treats the science observable as the sky-averaged spectrum rather than an image-domain quantity. The experiment is intended to detect the global signal’s turning points / inflections, which encode the underlying astrophysics and cosmology. The mission overview also uses the controversial EDGES low-band claim as a benchmark, while the laboratory-model paper emphasizes that the instrument class is aimed at a high-confidence detection of the global 21-cm signal from Cosmic Dawn and the Epoch of Reionization (Rao et al., 8 Jul 2025).

Two mission phases are described in the laboratory-model digital-receiver paper: Phase I: Low Earth Orbit (LEO), intended as a technology demonstration and pathfinder, and Phase II: Lunar orbit / lunar farside operation, identified as the main science phase because the lunar farside offers exceptional shielding from terrestrial and solar radio noise. The experiment-concept paper presents the lunar-orbiter baseline optimized for farside science (S. et al., 8 Jul 2025).

2. Mission concept and baseline payload

PRATUSH was proposed to the Indian Space Research Organization (ISRO) during the announcement of opportunity for science payloads in 2018 and is in the pre-project studies phase. The concept model is under development and is expected to lead to the engineering model followed by the flight model, subject to mission approval. Because of the measurement’s sensitivity, PRATUSH is designed to operate as a dedicated solo spacecraft in order to avoid payload-to-payload interference (Rao et al., 8 Jul 2025).

The rationale for the lunar-farside configuration is both observational and instrumental. The farside strongly suppresses terrestrial RFI, including the terrestrial FM band (88–108 MHz) that overlaps the science band; it also avoids the terrestrial ionosphere, ground coupling, and horizon/topography-induced contamination that affect ground-based instruments. The concept of operation is to observe in the radio-quiet region on the lunar farside and downlink the data on the nearside (Rao et al., 8 Jul 2025).

The baseline payload is organized into three main subsystems:

Subsystem	Function	Baseline implementation
Antenna	Smooth beam and smooth return loss	Monocone over a shaped reflector
RF system	Bandpass calibration, gain conditioning, in-situ antenna return-loss measurement	Analog receiver plus one-port VNA section
Digital receiver	Digitization, correlation spectroscopy, flagging, averaging, storage/downlink of reduced products	FPGA-based digital backend

The baseline antenna is a monocone with a shaped reflector, optimized over 55–110 MHz and fitted to a 2 × 12U bus. The reflector shape is based on an optimized log-spiral curve to keep the return loss smooth. The electronics are placed in EMI-shielded compartments, and the antenna is mounted above the spacecraft bus. This configuration reflects the project’s central design requirement: broadband response is insufficient unless the antenna and receiver remain spectrally smooth enough that residual structure after calibration is nearly Gaussian and at or below the mK scale (Rao et al., 8 Jul 2025).

3. Radiometric measurement principle and statistical sensitivity

PRATUSH is designed around the idea that the measured low-frequency sky spectrum should be as smooth as possible except for the cosmological contribution. Foregrounds are expected to be spectrally smooth, whereas the 21-cm signal has turning points; accordingly, the instrument is required to deliver a bandpass-calibrated, antenna-corrected spectrum whose residuals after smooth foreground modeling are at the mK level or below. The mission concept uses Maximally Smooth (MS) functions as the diagnostic of spectral smoothness (Rao et al., 8 Jul 2025).

The antenna response enters directly through the return loss. The experiment concept defines the antenna temperature after return loss as

$T_A = T_{sky}(1-|\Gamma|^2),$

where $\Gamma$ is the antenna reflection coefficient. In the validation pipeline, simulated beam and return-loss curves are combined with GMOSS, lunar emission contributions, and candidate global 21-cm signal models. A specific result reported for the baseline concept is that return-loss correction is essential: without it, a smooth foreground-only spectrum can have residuals of order 7.1 mK, whereas with in-situ return-loss correction the Cosmic Dawn signal becomes distinguishable (Rao et al., 8 Jul 2025).

A statistical foundation directly relevant to a PRATUSH-type radiometer is provided by the derivation of the radiometer equation in "Statistical Topics Concerning Radiometer Theory" (Hunter et al., 2015). In that treatment, the instantaneous amplitude is modeled as a zero-mean normal random variable,

$A = \sigma z,$

with measured power proportional to the square,

$A^2 = \sigma^2 z^2.$

Using the Nyquist/Johnson relation,

$P = kT\beta,$

and the time-bandwidth product,

$N = 2\beta\tau,$

the temperature sensitivity becomes

$\sigma_T = \frac{T}{\sqrt{\beta\tau}}.$

For a PRATUSH-like radiometer, this gives the ideal thermal-noise floor: sensitivity improves as $1/\sqrt{\tau}$ provided that the signal is Gaussian and white over the relevant band, the variance is finite and stable, and the samples entering the average are statistically independent (Hunter et al., 2015).

The same statistical analysis also identifies the principal caveat for long integrations. Simulations of $1/f$ and $1/f^2$ noise show that these processes are non-stationary, their variance grows with observing time, and they produce long-range correlations. When small gain fluctuations of this form are added to an otherwise white-noise radiometer signal, the measured mean-power distribution broadens beyond the radiometer-equation prediction and the improvement with integration ceases to follow the ideal $\Gamma$0 law. This is the formal statistical basis for treating gain stability, calibration cadence, and low-frequency systematics as first-order performance drivers for PRATUSH rather than secondary engineering details (Hunter et al., 2015).

4. Calibration architecture and suppression of instrumental structure

The baseline PRATUSH calibration scheme is inspired by SARAS-3 and combines a Dicke switch, a noise-source / attenuator calibration unit, a phase switch, and a double-differencing method. The receiver cycles through six states, OBS00, OBS11, CAL00, CAL01, CAL10, and CAL11. The calibrated bandpass spectrum for a cycle is written as

$\Gamma$1

where $\Gamma$2 is an absolute temperature scale from hot/cold load calibration. In this formulation, differencing suppresses receiver offsets and internally generated noise while isolating the sky contribution from calibration artifacts (Rao et al., 8 Jul 2025).

A distinctive feature of PRATUSH is the inclusion of a dedicated one-port VNA section for in-orbit measurement of the antenna’s complex impedance or $\Gamma$3. The VNA uses a DAC-generated swept tone, open, short, and 50 $\Gamma$4 standards, and an RF switch network, with calibration to correct cable, switch, and fixture effects. The motivation is explicit: the antenna may change because of thermal cycling across lunar day/night transitions and because of the plasma environment in lunar orbit, so the impedance model must be monitored in situ if residual systematics are to remain below mK-level contamination (Rao et al., 8 Jul 2025).

The concept study also treats internal spectral contamination as a design constraint. For standing waves in cables and internal reflections, the quoted periodicity estimate is

$\Gamma$5

The stated design intent is to keep cables only centimeters long, so the standing-wave period is much broader than the cosmological spectral structure and therefore less likely to masquerade as the signal. EMI from spacecraft electronics is identified as perhaps the largest technical risk, and the shielding requirement is described as likely needing effectiveness of $\Gamma$6 dB, supported by multi-layer shielding, linear power supplies, careful clock selection, twisted pairs/coax, PCB shielding practices, strong on-ground EMI testing, and maximal physical separation where possible (Rao et al., 8 Jul 2025).

A common misconception is that lunar-farside radio quietness alone would make global-signal detection straightforward. The PRATUSH design literature instead treats radio quietness, antenna smoothness, calibration accuracy, standing-wave control, and EMI suppression as coupled requirements. This suggests that the instrument’s scientific performance depends as much on suppressing internally generated spectral structure as on external RFI avoidance.

5. Digital correlation spectrometer and laboratory implementation

The Digital Correlation Spectrometer (DCS) is a central subsystem of PRATUSH. Its functional roles include analog-to-digital conversion, phase switching, spectral channelization, self- and cross-power spectra generation, RFI mitigation, calibration control, data packetization and transfer, and onboard recording / pre-processing. In the laboratory model, the two analog receiver outputs are split into 0° and 180° phase-shifted paths; this phase-switching arrangement is used to help cancel additive system contributions from the analog chain and samplers. The digital pipeline then performs digitization, windowing, FFT-based channelization in an FX correlator architecture, spectral integration on the FPGA, and Ethernet streaming of the integrated spectra to the controller (S. et al., 8 Jul 2025).

The flight-baseline concept and the laboratory model are related but not identical:

Aspect	Baseline concept	Laboratory model
ADCs	Two 12-bit ADCs, 250 MSps	Two quad 10-bit ADCs (EV10AQ190CTPY), 250 Msps
FPGA / channelization	Virtex-5QV FPGA, 2048-point FFT, 244 kHz resolution	Virtex-6 FPGA (XC6VSX315T-FF1516-2), 16384-channel spectrum, 30.516 kHz effective resolution
Control / integration	16384 spectra averaged onboard over about 134 ms	On-chip integration time ≈ 134 ms; Raspberry Pi 4 Model B as master controller

In the laboratory implementation, the digital receiver is built around the pSPEC platform. The ADC input full-scale range is about 500 mVpp (approximately -2 dBm), and the ADC inputs are maintained at about -27 dBm total power to leave roughly 4 bits of headroom against strong RFI and clipping. A 4-term Nuttall window is applied before the FFT; the paper reports ideal sidelobe suppression of about 98 dB and practical suppression of about 80 dB with 10-bit ADCs and finite FPGA precision. The firmware uses a 16384-point FFT implemented as a split-FFT $\Gamma$7 architecture, with two parallel 8192-point streaming FFT IP cores plus a custom 2-point parallel FFT stage (S. et al., 8 Jul 2025).

The controller for the laboratory model is a Raspberry Pi 4 Model B SBC, selected for its compromise among compute capability, memory, I/O, Linux support, documentation, low power, and portability. The paper also emphasizes the engineering cost of this choice: lower CPU performance, limited memory and storage bandwidth, slower SD-card I/O, greater sensitivity to timing jitter, more difficult real-time coordination between acquisition and control, higher risk of data corruption during long runs, and the need to manage RFI from the SBC electronics. These issues are not incidental; the reported output data rate is about 8 MBps (approximately 64 Mbps), and long runs can show 3–10% data corruption from packet drops (S. et al., 8 Jul 2025).

A major contribution of the laboratory-model study is dynamic flagging for corrupted spectra. For channel $\Gamma$8, the median absolute deviation is defined as

$\Gamma$9

where $A = \sigma z,$0 is the value of channel $A = \sigma z,$1 in spectrum $A = \sigma z,$2 and $A = \sigma z,$3 is the median value of channel $A = \sigma z,$4 across the dataset. This is converted to an approximate standard deviation through

$A = \sigma z,$5

for Gaussian-distributed data. Channels exceeding a threshold in units of $A = \sigma z,$6 are flagged, and if 16 consecutive channels are flagged the entire spectrum is treated as corrupted and dropped. A moving window can also be used so that slow drifts do not cause over-flagging (S. et al., 8 Jul 2025).

6. Demonstrated performance and related calibration developments

The laboratory digital receiver has been integrated with the PRATUSH laboratory-model analog receiver and tested end-to-end using open, short, 50 $A = \sigma z,$7 load, and antenna simulator / shaped termination states. The reported demonstrations include successful bandpass calibration, successful state switching over the six calibration states, stable acquisition and packet transfer from FPGA to SBC, generation of calibrated spectra after flagging and averaging, and validation that the receiver output is dominated by thermal noise at the expected level (S. et al., 8 Jul 2025).

For an 8-hour run with a 50 $A = \sigma z,$8 termination, the residuals after a maximally smooth fit had rms ≈ 72 mK. When the data were combined over 44 hours of effective integration, the residuals reached 12.5 mK rms at the native 30.51 kHz resolution, and after boxcar averaging to 610 kHz the residuals improved to 3.4 mK rms. The residuals were reported to be Gaussian distributed, indicating no strong unmodeled spectral structure in the receiver response at the achieved level (S. et al., 8 Jul 2025).

At mission scale, the concept study assumes a 2-year lifetime, about 200 hours total of usable prime-cone observing time, roughly 15% of orbital time as useful science time, and an orbital period of about 2 hours. The observing sequence is to perform a VNA measurement of antenna return loss, make the bandpass-calibrated sky observation, perform another VNA measurement before exiting the prime cone, and downlink the data when Earth is in view. Under these assumptions, the final noise level is stated to reach roughly mK thermal-noise rms at 244 kHz resolution. For sensitivity tests, the injected fiducial 21-cm absorption is modeled as a Gaussian with amplitude $A = \sigma z,$9 mK, center frequency 78 MHz, and FWHM 23.5 MHz; the posterior on the Gaussian parameters is reported as consistent with the injected signal (Rao et al., 8 Jul 2025).

A related calibration development of direct relevance to PRATUSH-style experiments is the physics-informed neural calibration framework demonstrated on the REACH receiver in "Radiometer Calibration using Machine Learning" (Leeney et al., 23 Apr 2025). That work addresses the same generic problem faced by global 21-cm radiometers: impedance mismatch, standing waves, spectral ripples, and receiver-chain systematics in a wide-beam, non-steerable instrument. The framework uses a multi-layer perceptron to infer source-independent Noise Parameters and gain from thermocouple data, PSD measurements, and VNA reflection-coefficient measurements, then propagates those parameters through an explicit receiver model. Reported performance includes RMSE = 0.05 K on relative residuals at 1 MHz channel width over 60–130 MHz on real receiver data, and RMSE = 0.14 K at 12 kHz and 0.04 K when binned to 1 MHz in full-chain simulations. Although this is not presented as the baseline PRATUSH calibration chain, it is explicitly relevant to PRATUSH-style sky-averaged 21-cm radiometry, particularly where time-varying receiver behavior and mismatch-driven spectral structure limit conventional calibration assumptions (Leeney et al., 23 Apr 2025).