Refurbishment and commissioning of a dual-band 23/31 GHz tipping radiometer at potential radio astronomical sites Open Access

Radio astronomy observations, especially in microwaves, are heavily affected by tropospheric water vapor,¹ limiting both source brightness measurements and phase coherence in radio interferometry observations. To address these issues, water vapor radiometers (WVRs)2–4 are frequently deployed at radio astronomy observatories to measure precipitable water vapor (PWV) as an aid for data calibration and telescope management. WVRs are also often deployed to characterize the long-term atmospheric opacity at candidate telescope sites prior to the construction of new radio astronomy facilities, such as the proposed African Millimeter Telescope (AMT)⁵ or the Next-Generation Very Large Array (ngVLA).⁶ 

The main atmospheric water vapor emission lines in the microwave range occur at 22.235 and 183.31 GHz.⁷ Both lines can be used to trace the atmospheric water vapor content, but generally, the observation band and the geographical location of the observatory determine the use of a cm-wave or a mm-wave WVR.⁸ Several WVR architectures can be found in the literature, e.g., single-band,9,10 multiple-band,11,12 system-on-chip-based,13–15 and spectrometer-based,16,17 whose designs depend not only on the scientific applications of the instrument but also on the water vapor emission line to be characterized.

The Hartebeesthoek Radio Astronomy Observatory (HartRAO) of the South African Radio Astronomy Observatory acquired the Generation III tipping radiometer WVR-III¹⁸ in the early 2000s to measure the atmospheric water vapor. This device was developed by the Federal Institute of Technology (ETH Zurich) of Switzerland in cooperation with CAPTEC GmbH: the former was in charge of mechanism control, communication interfaces, data acquisition, and instrument calibration, and the latter was in charge of mechanical design, construction, and documentation. The HartRAO application of the WVR-III was for path delay correction in satellite geodesy: because atmospheric path delay is related to the water vapor content, the measurements of PWV may be used to effect a delay correction.

The WVR-III aided several observational campaigns from 2000 to 2007 at HartRAO prior to suffering numerous failures in power supplies, control systems, data acquisition electronics, and various other circuits. Under the joint South Africa–Mexico Bilateral Programme on Water Vapor Radiometry¹⁹ (SA-MX), this instrument was refurbished over the course of 2019–2021 and renamed as WVR-III+, with preliminary efforts in this regard previously reported in 2020.²⁰ The WVR-III+ was returned to the University of Pretoria in mid-2021 (see Fig. 1) and recommissioned at HartRAO in early 2022. After several months, the radiometer was moved to the High Energy Spectroscopic System (H.E.S.S.) in Namibia as part of the site characterization tasks for the AMT project. We report here on the detailed refurbishment of this instrument, including requalification data taken at the H.E.S.S. site²¹ in the second half of 2022.

In the following, we describe the basis of atmospheric measurements using a ground-based microwave radiometer. We rely on the techniques described here and implement them on the refurbished WVR-III+.

The objective of a radiometer is to measure the intensity of incident microwave radiation, which is then expressed as a brightness temperature. This is accomplished by measuring the atmospheric scene intensity and then retrieving the brightness temperature by performing a Y-factor measurement with two targets at known (hot and cold) temperatures. Figure 2 presents a simplified radiometer block diagram in which the microwave radiation coming from the atmosphere is received by the feed horn and preamplified and then down-converted in frequency for acquisition. The signal is amplified again and passed to a bandpass filter to limit the frequency response. Then, a square-law detector provides a DC signal proportional to the noise power of the incoming signal, which in turn is proportional to the atmospheric brightness temperature. Finally, a computer acquires the radiometer output for signal processing and data retrieval.

Due to the phenomenon of refraction, electromagnetic signals traversing the Earth’s atmosphere experience a modification in their path length, commonly referred to as path delay. This refraction is caused by several physical mechanisms, but the interaction with atmospheric water vapor is responsible for a considerable part of the change in the path length.²² Measurements in radio interferometry and satellite geodesy are disturbed by this increase in path length, and hence, a means for measuring and correcting for the path delay is needed.

The simultaneous measurements of the atmospheric emission at the two frequencies of the WVR-III+ may also be used as a check of the accuracy of the model used for the atmospheric emission by noting if the results from the two frequencies are consistent with the derived PWV value. Measurement of the sky brightness temperature over a range of elevation angles provides a means to calculate the zenith opacity of the atmosphere,²⁶ which is a fundamental parameter for the characterization of astronomical sites.

The conversion from the measured atmospheric emission by the WVR-III+ to linearized brightness temperatures is calculated following the literature.²³ Here, we summarize the salient points.

For a linear detector, two reference points are sufficient to define a linear temperature scale such as kelvins. In the case of a radiometer, time and temperature-dependent drifts of front-end gain and reference loads, system voltages, or a myriad of other parameters result in a continually changing instrumental response. To effect a real-time calibration of the radiometer, perhaps the most commonly used approach is tipping scans.²⁷ In this procedure, multiple measurements are made over a range of elevation angles, and a subsequent linear fit to these values is used to determine the calibration factors to be applied to the data. In the case of WVR-III+, which has fixed feed horns for both frequencies, a pair of tipping mirrors rotates synchronously to provide for observations at various elevation angles.

In particular, although the hot calibration load has a constant noise temperature, its precise value is not accurately known. Thus, the hot load temperature has to be calibrated using an independent method. A well-known means for performing this calibration is to use the isotropic nature of the cosmic microwave background radiation (CMB), which has a brightness temperature of 2.7 K. In particular, T_B measurements at all elevation angles will have the same contribution from the CMB, therefore, providing an independent means of calibrating the temperature scale.²³ 

The radiometer tipping curves are calculated using the assumption that the measured brightness temperature can be represented as a linear function of the air mass m. Initially, a linear fit can be calculated and then extrapolated to m = 0. This fit will register a T_B > 2.7 K for m = 0, and the data points are initially shifted downward by the offset from T_B(m = 0) = 2.7 K. This correction factor is referred to as the “hot load correction,” ΔT_H, and is added to the hot load temperature. The linear regression of the brightness temperature is then calculated iteratively with new hot load corrections until the y-intercept reaches T_B(m = 0) = 2.7 K. The determination of the hot load correction is achieved by measuring three distinct elevations for each tipping curve. Figure 3 shows the tipping curve as an example of a full-sky tip observation. The dashed line in this figure corresponds to the initial linear function that best fits the observations, in a weighted least-squares sense. The solid line is the result after several hot load corrections have been applied.

Direct calibration by means of physical blackbody targets at known temperatures is a viable alternative for the radiometer calibration. Compared to direct physical calibration, the tip curve procedure that we adopt offers the primary advantage that it avoids the complications of including mechanical means for inserting blackbody absorbers in the signal path and of maintaining these absorbers at carefully controlled temperatures, both of which would add significant complexity to the instrument.

The original architecture of the WVR-III was based on a decentralized embedded system, as shown in Fig. 4. An external host computer controls and communicates with all the subsystems of the radiometer via an RS-485 Serial Interface Hub. The signal acquisition of the front-ends and the switching control were done by a microcontroller-based data acquisition module (DAQ), whereas the tipping (allowing elevation and azimuth rotation) was done by two independent complex programmable logic devices (CPLDs). The host computer was provided with a LabVIEW-based graphical interface, which allowed the acquisition and processing of the radiometric data, retrieval of PWV, atmospheric opacity, phase delay, recording of weather variables, and general instrument management. This architecture performed automated tip scans²⁶ for real-time determination of precipitable water vapor and path delay retrieval using the supplied software. In addition, the graphical interface provided flexible scan scheduling and convenient data accessibility.

The original radio frequency (RF) electronics of the WVR-III are shown in Fig. 5. The receivers are centered at 23.8 and 31.65 GHz, respectively, and are based on a load-switching architecture, similar to the one discussed in Sec. II A. They each consist of a single front-end amplification and downconversion stage, but the 23.8 GHz receiver has a Gunn oscillator tuned at 23.69 GHz, whereas the 31.65 GHz receiver’s is tuned at 31.54 GHz. The RF amplification and downconversion in each band is carried out by a low-noise block converter (LNB in Fig. 5) provided by Spacek Labs, model P23-5 for the 23 GHz band and model P31-5 for the 31 GHz band, each featuring ∼25 dB downconversion gain and a noise figure of 3.5 dB. The detection blocks (DC Amp in Fig. 5) consist of tunnel diode-based square-law detectors in a series with OP07 operational amplifiers. Switches A (MRI-SPSTK in receiver 1, MRI-SPSTR in receiver 2) and B (QuinStar QSD K-band in receiver 1 and QuinStar QSD Ka-band in receiver 2) were controlled by the DAQ module, which was also responsible for acquiring and transmitting the signals from the receivers to the host computer via the RS-485 interface.

The main technical details of the receivers are listed in Table I. As can be seen, the output noise power is integrated over a 200 MHz bandwidth in each channel. The WVR-III observation sequence followed the scheme discussed in Sec. II A and is now implemented in WVR-III+, as described in Sec. IV B 2.

The WVR-III was sent to Mexico and arrived in early 2019, where the Mexican team started the refurbishment. At that point, several technical tests were done to evaluate the physical status of the radiometer. As a result, we decided to replace many of the radiometer components, primarily due to the lack of technical details of the hardware and the absence of the software source code. The exceptions were the receivers, optics, and associated mechanical controllers, which were fully functional, but some of their components also lacked part or serial numbers, making it difficult for us to directly ascertain the features mentioned in the user manual provided by HartRAO. After testing the front-ends, measurements were obtained within the range specified by the manufacturers. We take this fact as a decisive point to not modify this analog front-end from the original design. In contrast, most of the control electronics were either non-functional or outdated and were completely replaced. In this regard, we proposed a new architecture based on a single-board computer (SBC), which allows a single processing unit to execute all the radiometer tasks that were originally done by distinct modules. Moreover, an SBC provides powerful computational capabilities for data processing and analysis in a compact design, in contrast to the previous external personal computer.

The refurbishment of the former WVR-III presented several technological and logistical challenges, as the instrument had been disabled for more than a decade due to technical failures. In addition, given the changes in HartRAO management, technical documentation was lost during this period. Despite this, we managed to continue with the repair of the WVR-III and we detail the work performed in this section.

The refurbished architecture of the WVR-III+ (see Fig. 6) is based on a centralized processing core: the Raspberry Pi 3 B+ computer²⁸ (hereafter, RPi) and its peripherals to manage the main radiometer tasks.²⁹ In addition to replicating the original WVR-III working scheme, this new system also provides autonomy to the tipping process, software customization, and real-time data processing through the implementation of the SBC. Furthermore, the user is able to remotely communicate with the instrument for debugging, data consulting, and system management.

The old WVR-III mechanical system was based on a DC-motor control loop for the elevation angles of tipping mirrors. This control system used an inclinometer as the feedback element, a CPLD as the control unit, and a DC motor as the actuator element. Due to challenges in modifying the mechanical structure, we chose to keep the servomechanical elements but replace the control unit and the feedback element in the chain. Thus, we implement new modules in the WVR-III+; the LSM9DS1 Inertial Measurement Unit as the feedback element, the RPi as the control unit, and a pulse width modulation interface plus an H-bridge based power stage as part of the actuator element, as shown in Fig. 7.

This scheme provides a closed-loop control over the elevation mechanism. The control law was designed to achieve an over-damped response with a rise time $≤2.5s$⁠, a settling time $≤4s$⁠, and an allowable tolerance of ±0.5°. An example of the step-response is shown in Fig. 8.

Despite the $∼5°$ HPBW of the radiometer feed horns being 50 times larger than the uncertainty required by the control system, a high-precision pointing is required for the control system. Elevation angle deviations of >1° lead to errors of ≳2.5 K in the calibration of a K-band radiometer.²⁶ Thus, a pointing precision of $<1°$ is suitable for this type of radiometric system.

To evaluate the actual pointing precision of the system, we analyze the mean (μ) and the standard deviation (σ) of the angle achieved for a given setpoint angle using a set of eight elevations. The data consist of a single day with 3780 elevations taken at 90°, 60°, 50°, 45°, 40°, 35°, 30°, and 28°, which were continuously repeated during the day. Following the data in Table II, the control system works within the estimated allowable tolerance, achieving the setpoint with a typical uncertainty of ±0.1°.

Despite the mechanical capability of WVR-III+ to azimuth rotation, we did not implement a control system for it. For the present, we will operate the WVR-III+ at a single fixed azimuth; an extension of the elevation control concept to the azimuth may be made in the future.

The WVR+III consists of two 200 MHz channels centered at 23.8 and 31.65 GHz, with noise power originally measured over an integration time of 100 ms in each channel (see Fig. 5 and Table I). As mentioned throughout Sec. III, the various RF components were all in working condition, so we kept the front-end architecture and components in the new design. However, a new signal conditioning stage was required to process the receiver output by using the RPi. Furthermore, the switches of the receivers (also shown in Fig. 5) required a new control interface because the original was removed from the instrument.

The output voltage from the power detectors at the end of the receivers were originally acquired by the DAQ module and sent to the host computer through the RS-485 interface (see Fig. 4). These outputs have a dynamic range of 10 V (0–10 VDC). Since the RPi cannot directly acquire analog signals, it was necessary to implement an external analog-to-digital converter (ADC) for signal acquisition. The newly implemented ADC is the ADS1115 board, which provides 16-bit conversion, four-channels, programmable gain amplifiers, and I2C serial communication interface for data transmission.

The ADS1115 has been configured to provide a full-scale range of 4 V (0–4 VDC), with a resolution of 61 μV for a 16-bit acquisition mode, and a default sampling rate of 128 Hz (modifiable up to 860 SPS). It became evident that we could not directly connect the receivers’ outputs to any of the ADC channels because of a mismatch in dynamic ranges. Therefore, we designed an analog signal conditioning circuit between the receivers and the ADC board, converting the 10 V range of the receivers to the 4 V range of the ADC prior to digitization. Figure 10 shows the implemented signal conditioning stage, which primarily consists of operational amplifiers (providing the gain) and optical isolation of the primary stages of the chain. TL081 operational amplifiers were used in the Amplifier stages, and the HCNR201 chips were used in the Optical Coupling stages. Analog channels are sequentially acquired by the RPi, following the switching scheme discussed in Sec. IV B 2.

The frequency response of the conditioning stage was characterized to estimate the operable bandwidth and the practical gain. For this purpose, we made Bode plots of the overall conditioning stage; the frequency response is shown in Fig. 11. Following the plots, we obtain a gain of −7.33 dB for a 10 kHz bandwidth without a phase shift. The gain of the signal conditioning stage is digitally compensated for in the ADC readings made by the RPi.

The radiometric observations of the WVR-III+ follow the scheme addressed in Sec. II A, which means that the receivers sequentially observe three different noise power sources: the water vapor emission collected by the feed horn as the sky, the internal noise source power as the hot load, and the power of an attenuator as the cold load. The measurements obtained by observing these sources provide V_A, V_H, and V_C, respectively.

Originally, the DAQ module controlled switches A and B inside the receivers, represented by the external lines connected to such switches in Fig. 5. For the new design, we implemented a relay-based interface to control the switches. We use the scheme shown in Fig. 12, where four of the RPi’s general purpose input/output pins (GPIO) are connected to transistor-driven relays: using the 2N2222 transistors and the Axicom V23079 relays.

The observation sequence of the radiometric measurements is shown in Table III (also see Fig. 5). Note that state 2 provides the noise power due to the attenuator as a termination load at room temperature (usually ≳ 300 K), i.e., the Cold Load—the accurate temperature of the cold load is measured by transducers B and A, as described in Sec. IV C 2. State 4 provides the Hot Load noise power as a cascade combination of the noise source and the attenuator $(≈1000K)$⁠.

The acquisition cycle of the radiometer proceeds with the three measurements of States 1, 2, and 4 in both receivers up to a maximum 5 s per state and another 5 s allocated for processing. This results in a maximum measurement time of 20 s per elevation angle. With six elevation measurements used in the tip curve, the total time required to estimate zenith opacity is less than 2 min, which is fast enough assuming that the time variability of the atmosphere is on the order of several minutes.²² 

The Allan variances of the 23 and 31 GHz channels (see Fig. 13) were measured to determine the receiver and noise source stabilities. The measurements were made over a 3-min interval with 100 ms sampling. The Allan times for the 23 GHz channel are 1.7 and 1.8 s for the cold and hot loads, respectively. For the 31 GHz channels, the corresponding times are 4.8 and 5.1 s. The cold load values can be considered to be the Allan variance of the receiver, while the hot load values are indicative of the stability of the noise source. These Allan times allow for a measurement sensitivity of ≲0.2 K.

To determine whether the RPi generates any harmful radio frequency interference (RFI), we measured RF levels close to the RPi board using a near-field probe connected to a Rohde & Schwarz FSW50 spectrum analyzer. The spectra shown in Fig. 14 represent five different tested locations using the probe: near the RPi power supply components (turquoise), above the CPU chip (orange), near the RJ-45 Ethernet connector (violet), and two arbitrary locations over the board (magenta, lime).

As seen in the spectrum, there is unlikely to be any emission in the RF bands; however, emission in the 110 MHz IF band is possible. Given a 200 MHz integration bandwidth, a typical T_A of 290 K would result in power levels of around −63 dBm at the input of the IF amplifier, where the system would be more susceptible to RFI. Although this is comparable with the levels measured above the CPU of the RPi, a minimum of 70 dB of shielding is easily achieved by the receiver’s metal enclosure and braided ground coaxial cable, resulting in a negligible contribution of RFI to measurement error.

In addition to the modules discussed so far, there is a sensor suite in the WVR-III+ that monitors weather variables and receiver temperatures. Here, we describe the purpose of these complementary sensors.

Several PWV or path delay retrieval algorithms depend upon the meteorological conditions at the radiometer site. Hence, apart from the radiometric measurements, a radiometer is often provided with a weather station. One of these systems was included in the original WVR-III, but was an external system, as shown in Fig. 4. This weather station was consulted via a serial protocol by the host computer, where the data were processed simultaneously with the radiometric data.

We replaced the weather station with a compact weather sensor incorporated into the radiometer structure, providing for greater compactness. A suitable solution for the WVR-III+ was the BME280 module, used to measure the same variables as the weather station of the original WVR-III: the air temperature, relative humidity, and barometric pressure. This modular sensor is connected as a slave to the RPi and consulted via the I2C protocol (see Fig. 15).

Radio frequency receivers are susceptible to gain fluctuations. These fluctuations directly affect the overall sensitivity of the radiometer: gain fluctuations on timescales longer than intrinsic calibration timescales (for example, via a Dicke switch) will decrease the radiometer sensitivity. Among the various origins of these phenomena is the operating temperature of the receiver; a good radiometer design must include some means for stabilizing the operating temperature.

The WVR-III was originally provided with a heating system that kept the receiver boxes at a controlled operating temperature with the intention of minimizing receiver gain fluctuations. This system consists of a temperature controller module, in which a setpoint value was introduced by the user. Then, the controller turned on two resistive heaters to change the temperature of the baseplates of each receiver. Finally, the control loop was fed back through a PT100 temperature sensor located on the surface of the 23 GHz receiver baseplate. This system was independent of the radiometer architecture and its tasks, so the user could adjust the control system before and during radiometric operation. Apart from the controller’s sensor, three other PT100 sensors were used to monitor temperatures inside the receiver enclosures: two on each of the baseplates of both receivers and one more only in the 23 GHz receiver noise source case. Temperature data were collected by the DAQ module and reported to the host computer.

For the WVR-III+, we kept the heating system scheme, but replaced both the temperature control unit (due to out-of-date technology) and the temperature sensors inside the receiver boxes. In addition, a fifth temperature sensor was added in the 31 GHz noise source case, as shown in Fig. 16. The current controller is a Shinko BCS2R10, and all five sensors are PT100 type. The new temperature transducers are the MAX31865 modules, which read the PT100 signals and send the data to the RPi via the SPI protocol.

The heating control continues to be a separated process from the radiometric tasks. The detector layout now allows for monitoring and comparison of the temperature changes in the baseplates and noise sources of each receiver. In this way, we have uniform monitoring of the environment inside the receivers for possible corrections in data retrieval and to prevent damage to the hardware due to extreme temperatures and avoiding condensation at dew point temperatures. If any of the five temperature transducers measures below 0 °C or above 50 °C, a report is sent to the system logger and the radiometric tasks are stopped (also see Sec. IV E).

Since the temperature controller only allows heating and not cooling, the Setpoint Value shown in Fig. 16 is adjusted depending on the weather of the site and should be changed during the year. This setpoint should be approximately equal to the maximum temperature expected at a given time of the year to avoid large fluctuations in the temperature to be controlled. For instance, a setpoint T_sp lower than the outdoor temperature T_amb causes the control action to shut down until the setpoint is reached by natural cooling. This implies a significantly longer settling time than would be achieved with a controlled chiller. Depending on the time of day when the above-mentioned process occurs, it could be affected by sudden temperature changes between the sunrise and sunset. In contrast, T_sp ≈ T_amb means lower temperature differences in which there will be less settling time both active (for heating) and natural (for cooling).

As a review of the stability of noise sources in the WVR-III+, we analyze a week of data taken from April 21–28, 2024. As shown in Fig. 17, the voltages of the noise sources remain constant within one tenth of a volt around the average voltage. Note that despite the outdoor temperature gradients (from $∼15°C$ to $∼35°C$⁠) during the day, the overall system response is stable within the specified range in both receivers.

Comparison of the Allan time (see Sec. IV B 3) with the temperature stability indicates that temperature variations are substantially slower than any reasonable integration time, thus ensuring that temperature instabilities will not degrade the system sensitivity.

The efficient operation of various internal stages of the WVR-III+ depends on a stable bias supply, provided by a dedicated power supply card (custom-engineered) to deliver the required voltage and current to each subsystem. Currently, the WVR-III+ incorporates a primary and a secondary power supply: the former provides AC–DC conversion and the latter consists of DC–DC converters.

The primary power supply (PPS) converts the AC power line to DC-power required by the secondary power supply (SPS), the elevation mechanism and the heating system. The PPS is assembled in a separate metal box from the radiometer housekeeping and is connected using a 5 m circular connector power cable. The features of the PPS are shown in Table IV.

The custom SPS board draws a +12 V bias from the WVR-III+ PPS. The SPS consists of three DC–DC COSEL converters, two solid-state regulators, and two LM2596 modules, together with a Cincon DC–DC converter. Each radiometer channel requires its own voltage distribution, so the SPS board offers distinct voltage outputs for each channel. The SPS outputs and the modules where they are implemented are shown in Table V and are distributed as follows:

• The two +12 V outputs supply the LM2596 modules, which are special, high-stability DC–DC converters used to provide +6.7 V to the Gunn oscillators. They are needed due to the extremely low DC input impedance of the oscillators—around 4 Ω for the 23 GHz channel and 2.5 Ω for the 31 GHz channel. The converters thus provide 1.67 and 2.68 A for the two channels, respectively. The LM2596 modules can support these current levels without disturbing the low voltage values.

• The +15 V is delivered by an ECLB60-12S15 Cincon module, which takes the input +12 V and converts it into a stable voltage level of +15 V.

• The −15 V output is generated by the ZUS101215 COSEL module in reverse configuration. In addition, the same −15 V are step-down to −10 V (required for the source switching board) by using an LM337 voltage regulator.

• The +5 V output is generated using the ZUS101205 COSEL module in a standard configuration.

• The +28 V output is obtained using the ZUW31215 COSEL module, which generates a symmetrical output of ±15 V. Using these values in a sum of voltage sources, a single output of +30 V is obtained, and latter adjusted to +28 V by the LM317 voltage regulator.

The Cincon and COSEL modules are supplied by the +12 V input and are excellent DC–DC converters that can maintain a stable output voltage signal despite possible disruptions at the voltage input. The LM337 and LM317 are stable voltage regulators, which can hold stable voltage levels with constant current.

Under normal operating conditions (considering the heating system on), the total power requirement of the radiometer is 390 W at maximum load.

The WVR-III+ is comprised of three main components to provide autonomy and high performance through the use of an SBC: the hardware, described in the previous sections and elsewhere;²⁹ the firmware, which allows communication between the SBC and peripherals; and the software interface, which performs the tip scans (sky surveys).

The firmware consists of Python-coded drivers that interact with the WVR-III+’s electronic components. The drivers are used to create a module that performs tasks, e.g., temperature system readouts or servo-mechanism control. The module is implemented as the script wvrsa and is executed on the RPi as the main control program. This program performs a tip scan using the new core and firmware to communicate with each subsystem for request and control data. The flow chart of wvrsa is shown in Fig. 18. At the beginning, the main program imports libraries and Python modules to control the WVR-III+ and the user defines a list with the desired elevation angles to scan, and the switching mode. Then, the program executes a set of hardware tests, including the following.

1. Test servomechanism control: sequentially runs two different elevation angles to check the correct functioning of the servomechanism.

2. Test receiver temperatures: confirms connection with the four temperature transducers inside the receivers by taking a single temperature reading and restricts the code execution within the operable range.

3. Test RF acquisition: checks if the receivers’ outputs are within the linear operating regime (0–10 VDC) by taking a single ADC reading.

If these tests are successful, then the survey module begins; see Fig. 19(a). In case of errors, processing is passed to the logger module [see Fig. 19(b)] and the main program is stopped. The survey module performs the task of data acquisition, which includes setting up the servo-mechanisms, reading the weather variables, reading the receiver temperatures, and acquiring the RF signals for the switching mode defined by the user. Finally, the measurements are saved into a CSV (comma-separated values) file. The internal functions of the modules shown in Fig. 19 are described in the following.

• Servo-mechanism control: manages a discrete-time controller for the elevation mechanism (see Fig. 7) and set the mirrors to the desired angle.

• Weather station monitor: requests ambient temperature, relative humidity, and barometric pressure measurements from the weather transducer (see Fig. 15).

• Temperature system monitor: obtains n samples specified by the user from each temperature transducer and computes the median (see Fig. 16). The result represents the reported temperature of each sensor.

• RF signal acquisition: used to digitize the output signals from both RF receivers (see Fig. 10) and manages the observation sequence shown in Table III.

• Save data to file: saves all measurements into a CSV file and stores them in the RPi.

• Logger: registers all events (errors, exceptions, and messages) that happen during wvrsa execution. The file gives information about these events, including when and where in the execution they occurred.

When a cycle of observations through the specified set of elevation angles is complete, i.e., to obtain a single-tip curve, the survey module is continuously executed. For a typical acquisition time of 20 s per elevation, a daily dataset is expected to contain about 4320 rows of data, provided there are no significant software delays. The infinite loop can be broken only if the user interrupts the program (usually with a control-c keystroke) or a critical error arises during data acquisition. At the end of a 24 hour period, the program executes an automated procedure to upload a copy of the CSV file to a Secure File Transfer Protocol server and Google Drive storage account to backup the measurements.

The WVR-III+ system is mainly controlled by the RPi, as described in Sec. IV. An SBC-based solution offers several hardware and software benefits for communication, which we took advantage of for the new WVR-III+ design. In this sense, the Raspbian operating system (version 9, Stretch) of the RPi allows communication with other computers for both internal and external networking by using Ethernet, WiFi, or 4G-LTE modules. Currently, the Internet connection is provided via the Ethernet port, which is used to upload all data files to cloud services for storage and access by the research team. Figure 20 shows the communication diagram of the system, including the serial interfaces and the network connection.

The renewed WVR-III+ provides the radiometric characteristics listed in Table VI. Note that RF features are preserved from the original WVR-III front-ends.

The Allan variance measurements yielded maximum integration times of 1.8 and 5.1 s for the 23 and 31 GHz channels, respectively. Comparing these results with the integration times of 2 and 5 s applied in practice (see Table III) suggests that the WVR-III+ can be considered stable over the integration periods, thus synchronous calibration schemes are superfluous. Furthermore, since the WVR-III+ can achieve an ideal temperature sensitivity of $∼0.2K$⁠, and that a typical K-band PWV sensitivity8,12 lies $∼0.5K/mm$⁠, the radiometer would be able to detect PWV changes of 1 mm for a signal-to-noise ratio of 10.

Generic system specifications of the WVR-III+ are shown in Table VII.

As mentioned in the literature,²⁶ a deviation of ±1° in the desired elevation angle leads to calibration errors (rms) up to 3.43 K at 23.8 GHz and up to 0.1 K at 31.65 GHz. In comparison, the new WVR-III+ system can perform tip scans with an angular tolerance of ±0.1°, which is accurate enough for the calibration process.

The temperature stability of the heating system achieves around 1 °C around the setpoint, but can increase for larger outdoor temperature gradients. To ensure this stability, the WVR-III+ must operate in an ambient temperature range of 0–60 °C.

A 390 W dissipation at a maximum load suggests that the WVR-III+ will probably require a high-power converter for remote sites where a power line is not provided.

The African Millimeter Telescope project proposes to place a millimeter-wave observatory on the Gamsberg Plateau, which is located 40 km from the H.E.S.S. Observatory (see Fig. 21) in Khomas, Namibia. This makes the H.E.S.S. facilities an ideal deployment site for the initial logistics of the AMT. As a starting point, we placed the WVR-III+ at this site to evaluate the atmospheric conditions in the cm-band and also the behavior of the instrument under remote site conditions.

The WVR-III+ was installed at H.E.S.S. (1859 m elevation) by the Observatory’s technical staff in mid-2022. Currently, this radiometer is located on a small rise, just over a half kilometer from the Observatory buildings (a map is shown in Fig. 22). The radiometer is installed along with other remote sensing instruments, which are powered by a solar panel station and connected to the Internet via a 4G network of a Namibian commercial supplier.

The WVR-III+ is currently monitoring the water vapor concentration at the H.E.S.S. site (see Fig. 23), alongside the Survey 3 radiometer, which was previously used for site characterization in Mexico.10,30 The Survey 3 is also a water vapor radiometer, but with a single channel at 210 GHz (in the wing of the 183 GHz water vapor emission line). Data from both radiometers are uploaded to the cloud for data retrieval and analysis by team members in Mexico, South Africa, and Namibia, providing a broad overview of the atmospheric variables at H.E.S.S.

Figure 24 shows the time series of atmospheric thermal emission, generated with data taken with the WVR-III+ installed on the surroundings of the H.E.S.S. site. Various atmospheric emission models31,32 suggest that water vapor densities of $<2g/cm2$ (or PWV = 20 mm) lead to atmospheric brightness temperatures of $<30K$ in the K-band. As can be seen in Fig. 24, the brightness temperatures at the two frequencies are within those ranges, with higher values at 23.8 GHz than at 31.65 GHz. Both frequencies have the same temporal pattern, showing larger values in the middle of the day and lower values in the morning and evening. Such measurements are also consistent with the PWV values reported in the literature33–35 for the H.E.S.S site, where PWV values of $∼5mm$ and $∼35mm$ are expected during the driest (May to September) and the wettest (October to April) seasons, respectively. A relationship between the brightness temperature measured with the WVR-III+ and the PWV, specific to this site, has yet to be determined.

The current WVR-III+ system presents several improvements to the WVR-III from which it is derived. We have refurbished the acquisition, sensors, and control systems of the radiometer along with the software for data acquisition, control, and data reduction. In the process, we implement technology based on Python and Raspberry Pi, providing better autonomy, accessibility, and customization. Given the computational capabilities of this SBC, the WVR-III+ is capable of performing retrieval algorithms immediately at the end of the day. At present, our team is developing a new process to run natively on the RPi, but for the moment, the data analysis is done offline. The WVR-III+ refurbishment provides an extension of the useful life of this instrument for at least 5 more years. However, it is important to keep the front-end system under constant monitoring as this part of the radiometer is about two and a half decades old.

The preliminary results of field measurements are at the H.E.S.S facility. An observatory in Namibia validates the proper functioning of the refurbished system and the data reduction techniques. The WVR-III+ will take atmospheric data for the statistical analysis of the water vapor content at the H.E.S.S. site and other potential radio astronomical sites, such as Mt. Gamsberg, over the next few years. A complete development and discussion on this point will be presented in a future paper.

Finally, this work is a demonstration of the success of the South Africa–Mexico Bilateral Programme on Water Vapor Radiometry with the involvement of the University of Pretoria and the South African Radio Astronomy Observatory from South Africa, the National Institute of Astrophysics Optics and Electronics, the Institute of Astronomy-UNAM, and the Institute of Radio Astronomy and Astrophysics-UNAM from Mexico.

This work was carried out with the financial support of the National Research Foundation (NRF) of South Africa and the Consejo Nacional de Humanidades, Ciencias y Tecnologías (CONAHCyT) of México under the South Africa–Mexico Joint Science & Technology Research Collaboration with Project Nos. UID114676 (NRF) and 291778 (CONACyT) granted from 2018 to 2024. The project was also supported by the DGAPA-UNAM under Project Nos. IT100318 and IT101622 and by the Institute for Radio Astronomy and Astrophysics. We acknowledge the invaluable staff support offered by the Hartebeesthoek Observatory and the High Energy Spectroscopic System. The authors thank colleagues at the University of Namibia and Radboud University for logistical support. Finally, we thank the referees whose suggestions improved the quality of this work.

The authors have no conflicts to disclose.

J. Cuazoson: Formal analysis (equal); Methodology (equal); Software (equal); Visualization (equal); Writing – original draft (equal). D. Hiriart: Conceptualization (equal); Resources (equal); Writing – original draft (equal); Writing – review & editing (equal). T. Stander: Conceptualization (lead); Funding acquisition (equal); Writing – original draft (supporting). R. Botha: Resources (supporting); Validation (supporting). J. Contreras: Data curation (lead); Software (equal); Validation (supporting); Writing – original draft (supporting). D. Ferrusca: Methodology (supporting); Supervision (lead); Validation (lead). E. Ibarra-Medel: Methodology (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). S. Kurtz: Funding acquisition (equal); Project administration (lead); Writing – review & editing (equal). R. Neate: Data curation (supporting); Methodology (supporting); Validation (supporting). D. Rojas: Validation (supporting); Visualization (equal); Writing – review & editing (equal). M. Velázquez: Formal analysis (supporting); Methodology (supporting); Supervision (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.