2. The receiver system

The receiver system consists of the feed and LNAs, both of which are cryogenically cooled, and the RF amplifier chain. (Dunning et al. Reference Dunning, Bowen, Bourne, Hayman and Smith2015) described the basic properties of the feed. The design is shown in the left-hand panel of Figure 2 and is based on a quad-ridged horn. The middle and right-hand panels of this Figure show different visualisations of the complete receiver package. This design provides a wide bandwidth with comparable performance to narrower band systems by making use of a central dielectric spear to improve the beam properties at high frequencies, and a corrugated skirt to improve the beam properties at low frequencies.

The dielectric spear consists of three layers giving a graded dielectric constant: a central quartz section, a solid polytetrafluoroethylene (PTFE; more commonly known as $\text{Teflon}^{\text{TM}}$ ) section and a slotted PTFE outer section. The feed horn operates over the frequency range ${\sim}700$ to ${\sim}4.5\,\text{GHz}$ and provides two nominally orthogonal linear polarisations, each with symmetric outputs. The exceptionally wide frequency coverage of the feed required the development of cryogenically cooled LNAs with large dynamic range that cover the entire band. The three-stage LNAs are cooled to 20 K, the feed horn and the dielectric spear are cooled to 70 K, and the outer feed rings are at ambient temperature (because of thermal losses, there is a temperature gradient across the dielectric spear; see Smart et al. Reference Smart, Dunning, Smith, Carter, Bourne, Doherty, Castillo and Dong2019 for more detail). Following the cryogenically cooled LNAs, the signals from the two polarisations are further amplified by ambient temperature amplifiers before digitisation. To assist with the mitigation of RFI, the UWL band for each polarisation is split into three sub-octave RF bands using a triplexer (created using a power splitter and a diplexer) and band-limiting filters. These bands are separately amplified and digitised. As shown in the bottom panel of Figure 3, the three RF bands (labelled Bands 1 to 3) have a relatively flat response over the corresponding digital bands, which cover from 704–1 344, 1 344–2 368, and 2 368–4 032 MHz (see Section 3 for more detail on the digital bands). The separate sub-octave RF bands provide protection against intermodulation products from strong RFI signals affecting other parts of the UWL band. Switchable attenuators allow the choice of signal level for the three RF bands, with RF Band 1 requiring significantly more attenuation than the other two because of the strong RFI in this band. Default attenuator settings ensure that each band normally remains in a linear regime in the presence of RFI^{Footnote f}.

Figure 2.

Representation of the feed system: (left) the internal structure of the feed, (centre) the feed, platform, and RF shielded cabinets, and (right) the completed system being installed on the telescope.

Figure 3.

The sampling bands for the UWL. The bottom panel shows the filters determining the three RF bands (RF bands 1, 2, and 3). The digital bands are formed using two sampling frequencies ( $f_{1S} = 4\,096\,\text{MHz}$ and $f_{2S} = 2\,560\,\text{MHz}$ ). These sampling frequencies permit the 26 sub-bands (denoted SBx) each 128 MHz wide. Those shown in red are obtained from the first and second Nyquist zones of $f_{1S}$ . Those in blue are from the second Nyquist zone of $f_{2S}$ . See text for further details.

A key aspect of the design was to digitise the signals at RF in the focus cabin and then to transmit digital signals to the telescope tower that contains the signal processor instrumentation. The alternative of using analogue RF over fibre for the long ( ${\sim}150\,\text{m}$ ) path to the tower was rejected because of potential linearity issues related to the strong RFI, particularly for RF Band 1. Also, analogue transmission over such a long path would be subject to differential delays in the two polarisations with varying ambient conditions, leading to instabilities in derived polarisation parameters.

The UWL cannot be rotated (both software and hardware limits are in place to ensure this) and the linear polarisations have been aligned at $\pm45^\circ$ to the elevation axis of the antenna. The UWL feed design precludes the injection of noise into the feed horn. The artificial noise signal is therefore coupled directly to the inputs of the LNAs. Note that the coupling is non-directional; see Section 5. We also have a capability to radiate noise directly into the feed from a transmitter on the surface of the dish. However, this is currently not used in any standard observing mode. It is possible to vary the amplitude of the injected signal; for the results presented here, the amplitude was between 1 and 2 K across the band.

The noise temperature of the LNAs, $T_{\text{LNA}}$ , was determined using laboratory measurements in a cryogenic test dewar. The results are shown as the solid, cyan line in the upper panel of Figure 4 and is close to 5 K across the entire band. The system temperature (note that, as we are not attempting a rigorous analysis here, we ignore efficiencies and beam pattern weightings) is the sum of several components, which are individually defined below:

(1) \begin{eqnarray} T_{\text{sys}} &=& T_{\text{rcvr}} + T_{\text{atm}} + T_{\text{Gal}} + T_{\text{CMB}}\nonumber\\ &&+ T_{\text{spill}} + T_{\text{scatt}} + T_{\text{RFI}}. \end{eqnarray}

The $T_{\text{rcvr}}$ measurements (which include $T_{\text{LNA}}$ ), also shown in the top panel of Figure 4, were made at the Parkes observatory site with the receiver on the ground facing up, using an ambient temperature absorber as a hot load and the sky as a cold load. $T_{\text{rcvr}}$ , which includes contributions from losses in the feed, is ${\sim}9\,\text{K}$ across most of the band, rising from around 3 GHz to ${\sim}18\,\text{K}$ at ${\sim}4\,\text{GHz}$ .

Figure 4.

The measured system temperature (T $_{\text{sys}}$ ) for the UWL receiver system on the Parkes radio telescope and several of its components are shown in the upper panel. The red and dark blue traces indicate the A and B polarisations, respectively, for $T_{\text{sys}}$ (upper pair) and $T_{\text{rcvr}}$ (lower pair). The solid, cyan line ( ${\sim}5\,\text{K}$ ) is the measured $T_{\text{LNA}}$ , the green line is the estimated $T_{\text{spill}}$ , and the orange curve ( ${<}5\,\text{K}$ ) is the estimated sky background temperature, $T_{\text{Gal}} + T_{\text{CMB}}$ . The upper, black line is the sum of the individual estimated noise components. The horizontal dotted line indicates 20 K. The horizontal lines delineated by circles give an approximation to the frequency coverage and system temperature for other receivers available at Parkes. The 10/40cm receiver is represented in red, the 20-cm multibeam receiver in pink, and the H-OH receiver in blue. The lower panel provides measures of the aperture efficiency from laboratory-based measurements and simulation (black line). The higher, blue line in this panel provides an estimate of the main beam efficiency.

To determine the expected $T_{\text{sys}}$ , we assumed an atmospheric contribution ( $T_{\text{atm}}$ ) of around 2 K at these observing frequencies (Smith Reference Smith1982). The contribution of ‘spillover’ radiation, $T_{\text{spill}}$ , from the ground entering the feed past the main reflector was estimated by integrating the feed response falling outside the reflector, assuming a ground temperature of 290 K. This is shown in Figure 4 as the green line. The relatively large variations are caused by frequency-dependent fluctuations in outer parts of the feed beam pattern. $T_{\text{scatt}}$ is radiation from any source scattered into the feed from the telescope structure (our modelling does not account for scattering). As far as possible, RFI is eliminated from the signal (see Section 6) and so we do not consider it further in this section.

The sky temperature depends upon the Galactic location being observed. An estimate of $T_{\text{Gal}}$ was obtained from the reprocessed Haslam et al. (Reference Haslam, Salter, Stoffel and Wilson1982) 408-MHz image of Remazeilles et al. (Reference Remazeilles, Dickinson, Banday and Ghosh2015) for the approximate sky position relevant to the measurements and assuming a spectral index of −2.6 (see range of values in Lawson et al. Reference Lawson, Mayer, Osborne and Parkinson1987 for more details). Note that both the Haslam et al. (Reference Haslam, Salter, Stoffel and Wilson1982) and Remazeilles et al. (Reference Remazeilles, Dickinson, Banday and Ghosh2015) image scales include the cosmic microwave background contribution, $T_{\text{CMB}}$ , and so this was subtracted from the image value before scaling to the required frequency. All these contributions have been plotted in the upper panel of Figure 4 with the summation being shown as the upper, solid, black line.

The measured on-telescope system temperature ( $T_{\text{sys}}$ ) as a function of frequency is shown in the top panel of Figure 4 and tabulated in Table 1. These measurements were obtained by comparing spectra obtained with an ambient temperature absorber over the feed with spectra obtained when observing the sky. The $T_{\text{sys}}$ measurements were carried out with the telescope pointed at the zenith when both the Sun and the Galactic centre were below the horizon (between UTC 12:00 and 20:00 each day over the period 2018 October 29 to 2018 October 31). The $T_{\text{sys}}$ values match, or significantly improve on, all current and previous Parkes receivers covering the same observing bands (see horizontal bars in Figure 4). The predicted $T_{\text{sys}}$ (from Equation (A1) is within a few Kelvin of the measured value over most of the band, with slightly larger discrepancies at the low and high ends. At the low end, the discrepancy could easily be accounted by a small underestimation of the spillover. At the high end, it may result from scattered power from various sources, including the telescope structure itself, entering the feed and/or leakage of ground radiation through the outer parts of the reflector. We note that these $T_{\text{sys}}$ measurements were determined from the output of the RF chain and do not include any extra noise contributions from the digitiser and signal processor systems.

The black line in the bottom panel of Figure 4 shows the expected aperture efficiency for the UWL as installed on the Parkes 64-m telescope. As described by (Smart et al. Reference Smart, Dunning, Smith, Carter, Bourne, Doherty, Castillo and Dong2019), these were determined using laboratory-based measurements of the feed radiation patterns with a model of the telescope. The calculated efficiency is around 65% over the observing band. We compare these results with the astronomically measured efficiencies described below. The upper, blue line in the bottom panel of Figure 4 provides an estimate of the main beam efficiency. These values were also determined from the measured feed patterns using Equation (A7). Across most of the band, the mean main beam efficiency is ${\sim}0.96$ , reducing to ${\sim}0.92$ at the low-frequency end of the band.

3. The signal processing systems

The two polarisation channels for each of the three bands are digitised in the focus cabin^{Footnote g}. The analogue-to-digital converters (ADCs) provide 12-bit resolution samples that are formatted into serial data streams complying with the JESD204B (Joint Electron Device Engineering Council Standard Document 204 Revision B) protocol for serially connected data converters. These signals are transported to the signal processors in the telescope tower via single-mode optical fibres, with two fibres required per digitiser^{Footnote h}. The ADCs operate at one of two separate sampling frequencies, 2 560 and 4 096 MHz, to ensure unbroken frequency coverage of the observing band from 704 to 4 032 MHz, as is shown in Figure 3.

Table 1.

Parameters for the 26 UWL sub-bands: the corresponding RF band that feeds them (column 2); sub-band frequency ranges (column 3); the central observing frequency (column 4); the median system temperature (column 5) and system equivalent flux densities as determined using PKS B0407–658 (column 6) summed over both polarisations; an estimate of the antenna (column 7) and main beam (column 8) efficiencies, the full-width half-maximum (FWHM) of the measured beam shape (column 9), and the fraction of the sub-band containing useful data (column 10, see main text for definition).

The six digitiser data streams (nine when the focus cabin RFI reference signal is implemented, see Section 7) are fed to a pre-processor unit in the telescope tower which utilises Field Programmable Gate Arrays (FPGAs) to implement polyphase frequency channelisers that produce 26 contiguous sub-bands each 128 MHz wide for each polarisation (see Table 1). The pre-processor subsystem is built around a commercial platform with a Xilinx^® Kintex^® Ultrascale^TM FPGA and eight Quad Small Form-factor Pluggable cages for data exchange. Each FPGA board receives data from two ADCs and eight such boards are housed in custom-made enclosures totalling eight rack units. The total designed capacity is 32 GHz (256 channels of 128 MHz) of real-time processing bandwidth and 1.28 Tb $\text{s}^{-1}$ of Ethernet data output.

Currently, each sub-band delivered by the pre-processor is produced using a critically sampled polyphase filterbank. To eliminate the effects of aliasing at the sub-band edges, an oversampled polyphase filterbank is planned, but has not yet been implemented. There are two separate versions of the FPGA signal processing firmware to match the two ADC sampling frequencies. The primary differences between the two firmware versions are the number of output sub-bands and the length of the finite impulse response (FIR) filter for the polyphase channelisers. For the 4 096-MHz sample clock firmware, there are 16 output sub-bands and 1 024 coefficients in the channeliser filter. For the 2 560-MHz sample clock, there are 10 output sub-bands and 640 coefficients in the channeliser filter. The channeliser FIR filters are carefully designed to achieve the best balance between passband flatness and attenuation of out-of-band components while also maintaining steep transition bands to minimise gain loss at the sub-band edges and aliasing between sub-bands. The FIR filters are designed using a least squares algorithm with a Kaiser window function to smooth the resulting impulse response and have a passband ripple of ${\sim}0.02\,\text{dB}$ with a transition at the band edges crossing the adjacent sub-band at −6 dB. The relationship of the sub-band frequencies to ADC sampling frequencies is illustrated in the top panel of Figure 3 and tabulated in Table 1.

The polyphase filterbank is designed to preserve full numerical precision throughout both the FIR and Fast Fourier Transform (FFT) stages. Accordingly, there is ‘bit-growth’ from the input 12 bits from the ADCs to 23 bits at the filterbank sub-band outputs. A number representation of 23 bits is not compatible with a CPU/GPU computational environment and would also require significant additional network infrastructure to transport. For these reasons, a scaling module is implemented for each sub-band output to reduce the number representation down to 16 bits (signed two’s-complement). This scaling operation is an area where dynamic range issues of numerical overflow, underflow, or saturation can occur, so the location of the 16 bits that are selected from the input 23 can be controlled through software-adjustable fixed gain that has eight settings ranging exponentially from unity gain through to a gain of 128.

The output data streams from the FPGA pre-processor are sent as multicast Ethernet packets as 10 Gb $\text{s}^{-1}$ Ethernet (10 GbE) to a 64-port Cisco^® 3164Q network switch (they are aggregated as $4\,\times\,10\,\text{GbE}$ on a 40-GbE port on the switch). During observations, the GPU-based Medusa signal processor subscribes to the multicast streams and runs further processing on them to form science data products, as described below. For the UWL RF bands, the FPGA signal processor outputs data as User Datagram Protocol (UDP) packets of size 8 272 B, each containing a single ‘Data Frame’, as defined in the VLBI Data Interchange Format (Whitney et al. Reference Whitney, Kettenis, Phillips and Sekido2009) specification (VDIF; http://www.vlbi.org/vdif/). The FPGA pre-processor outputs a total of 52 VDIF UDP streams, one for each polarisation of each sub-band, with 128 MHz bandwidth using complex sampling and 16-bit resolution per complex component; the total ingest data rate at the switch is 215 Gb $\text{s}^{-1}$ .

The use of multicast Ethernet allows copies of the data to be sent not only to Medusa, but also to the Breakthrough Listen (BL) data recorder (Price et al. Reference Price2018). Data are transported from the UWL switch to the BL switch via eight 40 GbE links, which are configured as an aggregated group using Link Aggregration Control Protocol.

4. Timing and synchronisation

An observatory distributed clock system, which is referenced to the observatory’s hydrogen maser frequency standard, is used to derive precise 128-MHz and 1-Hz reference signals for the data acquistion systems. Phase-locked synchronisation signals at 2.56 MHz for the 4 096 MHz system and 1.6 MHz for the 2 560 MHz system, known as ‘SYSREF’ signals, are required by the ADCs. These are derived from the 128-MHz reference signals and distributed with it to the focus cabin over optical fibres. A local synthesiser in the cabin generates the sampling clocks from the 128-MHz reference and distributes them to all ADCs along with a copy of the appropriate SYSREF clock. The combination of the sampling clock, SYSREF and a 1-Hz gated synchronisation from the FPGA receivers, along with local copies of these signals supplied directly to the FPGA, allows for precise, repeatable synchronisation with deterministic latency. The ADCs are synchronised to a chosen SYSREF edge, and therefore the synchronisation error is expected to be sub-nanosecond, relative to the maser standard. Time synchronisation to terrestrial time standards at a level of a few nanoseconds is obtained by measuring the offset between the observatory clock 1-s pulses and those from a Global Positioning System (GPS) receiver. These are subsequently referenced to International Atomic Time (TAI) or other time standards using time offsets published by the Bureau International des Poids et Mesures (BIPM)^{Footnote i}.

To illustrate the timing stability of the UWL system, Figure 5 shows the pulsar timing residuals for PSR J1909–3744 in the 10-cm observing band. This is the most stable pulsar observed by the PPTA project (Manchester et al. Reference Manchester2013). The residuals in black were obtained using the 10-cm receiver and the PDFB4 processor. Those in red are for an identical band, but obtained through the UWL system. The rms of the timing residuals (that span around 6 months) obtained with the Medusa system (over this relatively small band) is 145 ns and the residuals are dominated by the uncertainty on their measurement. The UWL is clearly adequate for high-precision pulsar timing, but we will continue to test the timing precision and accuracy of the system and to improve our knowledge of the delays between this system and earlier receiver and signal processor combinations. Our intention is to determine the absolute delays required to convert any measured pulse arrival times to the intersection of the axes of the telescope (see Manchester et al. Reference Manchester2013) for details on how this was carried out for the PDFB4 signal processor).

Figure 5.

Timing residuals from PSR J1909–3744 obtained by the PPTA project team. Points in black are from observations with the 10-cm receiver and PDFB4 signal processor. Those in red are observations covering the same band, but with the UWL system.

5. System performance

Here, we describe the system performance as measured through the entire system (from the feed to the final data products). Figure 6 shows spectra with a frequency resolution of 488 Hz across the three RF bands. During this 2-min observation, the telescope was pointed towards the zenith. The sub-band spectral shapes are significantly affected by RFI (see Section 6) and quasi-periodic oscillations. For instance, the spectrum contains the characteristic small-scale ripple with a periodicity of ${\sim}5.7\,\text{MHz}$ (see zoomed region in the figure), which arises from reflections in the 26-m space between the vertex and the underside of the focus cabin. The narrowband spikes in the spectra are not yet fully understood; many will be externally generated RFI, although those at frequencies related to 1 024 and 2 560 MHz are linked to the timing system and sampling frequencies. Band 3 has a currently uncompensated ${\sim}10\,\text{dB}$ slope across the band which originates in an RF amplifier within the digitiser module. This slope and the sub-band edges are stable and can be calibrated through observations by means of the injected calibration signal.

Figure 6.

Average spectra across three RF bands obtained on 2019 February 24 at 22:48:23 UTC. The red and blue traces represent the A and B polarisations, respectively. The vertical, dotted lines indicate the sub-band boundaries and the principal RFI transmissions are labelled. The narrow spikes at 1 024, 1 920, and 3 072 MHz are digital processing artefacts related to the clock signals. NBN stands for the National Broadband Network and BT for Bluetooth. The zoomed region in RF Band 2 highlights a characteristic ripple that is described in the text. The main text also provides an explanation of the slope across RF Band 3.

The system equivalent flux density for the telescope with this receiver system ( $S_{\text{sys}}$ ) and the calibrator source ( $S_{\text{cal}}$ ) were determined by observing PKS B0407–658 and its surroundings (we note that we obtain almost identical results with the calibrator PKS B1934–638). Each observation lasted 2 min at positions $1^\circ$ North of PKS B0407–658, pointed directly at the source, $1^\circ$ to the South, again pointing at the source, and then again to the North. During each observation, the noise injection is switched at a frequency of 11.123 Hz. The data were processed using the Psrchive program fluxcal to produce flux calibration files for each band. This process assumes that PKS B0407–658 has a flux density of 15.4 Jy at 1 400 MHz and a spectral index of −1.20 over the entire band (obtained from the Australia Telescope Compact Array calibrator database^{Footnote j}). The measured $S_{\text{sys}}$ and $S_{\text{cal}}$ values as a function of frequency are shown in the upper and lower panels of Figure 7, respectively. The lower red and blue lines in each panel of the figure show the determinations independently for each polarisation. The upper, black line in the top panel is the more commonly quoted summation of both polarisations. The median $S_{\text{sys}}$ values for each sub-band are listed in Table 1.

Figure 7.

The measured system equivalent flux density ( $\text{S}_{\text{sys}}$ ) for the UWL receiver system on the Parkes radio telescope is shown in the upper panel, where the red and blue traces indicate the A and B polarisations, respectively, and the black line is the sum of the two polarisations. These measurements were obtained using observations of PKS 0407–658. The step between RF bands 1 and 2 is discussed in the text. The equivalent flux density of the injected calibration signal ( $\text{S}_{\text{cal}}$ ) for each polarisation is shown in the lower panel.

Whereas the $T_{\text{sys}}$ measurements, described in Section 2, were obtained from the output of the receiver system, while the telescope was pointing at zenith, the $S_{\text{sys}}$ measurements have been determined using an astronomical source through the entire UWL system (including the digitisers, FPGAs, and GPU processors). The discontinuous break in the $S_{\text{sys}}$ measurements seen between RF bands 1 and 2 in Figure 7 results from the extra attenuation required in Band 1 to avoid saturation effects in the analogue or digital systems which could otherwise occur because of the very strong mobile phone transmitters in this band^{Footnote k}. As Figure 6 shows, these transmitters can be up to 50 dB above the receiver noise floor. However, this extra attenuation results in a significant digitiser noise contribution, adding an extra ${\sim}3\,\text{Jy}$ to the measured system equivalent flux density.

The $S_{\text{cal}}$ values clearly contain an oscillation with a ${\sim}100\,\text{MHz}$ periodicity. The noise source is injected into the LNA system, but the coupler is not directional. This oscillation is believed to result from reflection off the tip of the dielectric spear. The oscillation period and structure are stable and, to date, have not affected the calibration of astronomical data sets.

Using Equation (A6) in Appendix A, we can obtain estimates for the aperture efficiency, $\epsilon_{\text{ap}}$ , from the system equivalent flux density values and the system temperature values (Figure 7). We note that the $\text{T}_{\text{sys}}$ measurements were obtained at nighttime, with the telescope pointing at zenith using an independent spectrum analyser and hence do not include extra noise from the digitisers. However, the $S_{\text{sys}}$ measurements were obtained with a telescope zenith angle of $40^\circ$ (elevation angle of $50^\circ$ ). We, therefore, have extra contributions to $\text{T}_{\text{sys}}$ from the atmosphere, spillover, and from the digitiser system. For a typical sub-band (we have chosen sub-band 5 spanning from 1 344 to 1 472 MHz), we estimate an extra contribution to $\text{T}_{\text{sys}}$ during the PKS 0407–658 observations of ${\sim}4\,\text{K}$ giving $\epsilon_{\text{ap}} \approx 0.7$ . From Equation (A5) in Appendix A, this converts to a gain of $G_{\text{DPFU}}\sim 0.8\,\text{K \,Jy}^{-1}$ .

The feed has been designed to have a constant beam width across the band (Smart et al. Reference Smart, Dunning, Smith, Carter, Bourne, Doherty, Castillo and Dong2019). However, the telescope beam width as measured on the sky remains proportional to the observing wavelength. Theoretical full-width half-power beam widths calculated from $1.02\lambda/D$ (where $\lambda$ is the observing wavelength and D, the telescope diameter) give 23, 12, and 4 arcmin for 700, 1 400, and 4 032 MHz, respectively. A high dynamic range measurement of the actual beam shape over the wide band can be obtained by determining the flux density (Stokes I) of the Vela pulsar (PSR J0835–4510) as a function of the angular separation between the known pulsar position and the telescope pointing direction (we define the pointing in azimuth, A, and elevation, E). We have carried out multiple observations of Vela with the telescope offset in a grid of ( $A \cos E$ , E) positions. Figure 8 shows representative beam shapes (for sub-band 5 and centred on 1 408 MHz). The complete set of beam shape data files is available for download (see Appendix B). The results from the grid pointings in elevation and in azimuth are not identical. This is primarily caused by the presence of a feed leg in the elevation plane. We overlay predictions of the beam shape obtained from electromagnetic simulations. The blue dashed lines are an ideal case with no attempt to model the effect of the prime focus cabin. The red, dashed line includes a simplified model of the cabin (assumed to be a circular blockage), but does not include the effect of the feed cabin support legs. We do not expect a perfect match between the observing beam shape and the prediction, but note that they match remarkably well and that the sidelobe structure is dominated by the effect of the focus cabin. Data files containing the predictions are also available as part of our public data collection.

Figure 8.

Beam pattern obtained through scans in azimuth (left) and elevation (right) for the 20-cm observing band (sub-band 5, Table 1). The measured response is shown using the black line and error bars. The red and blue dashed lines are modelled beam shapes that are described in the text.

The on-axis polarimetric response of the receiver system is modelled using an updated version of the Measurement Equation Modeling (MEM) technique originally described in (van Straten Reference van Straten2004). The required updates, motivated by (Liao et al. Reference Liao, Chang, Kuo, Masui, Oppermann, Pen and Peterson2016), remove the assumption that the system noise has insignificant circular polarisation and enable modelling of an artificial noise source that is coupled after the feed. We have used this updated MEM algorithm to measure the properties of the receiver (including the differential gain, differential phase, and cross-coupling of the receptors; and the Stokes parameters of the artificial calibration signal). As inputs, we observed the bright millisecond pulsar, PSR J0437–4715, over a wide range of parallactic angles; we also included both on-source and off-source observations of Hydra A (3C218). The measured receptor ellipticities are close to $0^\circ$ across the whole band, indicating that the degree of mixing between linear and circular polarisation is low. The non-orthogonality of the receptors is also very low, as characterised by the intrinsic cross-polarisation ratio (IXR; Carozzi & Woan Reference Carozzi and Woan2011), which varies between 40 and 60 dB across the band. This is much greater (noting that larger values of the intrinsic cross-polarisation ratio correspond to lower non-orthonormality) than the minimum recommended by (Foster et al. Reference Foster, Karastergiou, Paulin, Carozzi, Johnston and van Straten2015) of 30 dB for high-precision pulsar timing. We also found that, at higher frequencies, the reference signal produced by the artificial noise source deviates from 100% linearly polarised. Up to ${\sim}30$ % circular polarisation has been observed at ${\sim}4\,\text{GHz}$ ; therefore, the polarisation of astronomical signals must be calibrated using the technique described in Section 2.1 of (Ord et al. Reference Ord, van Straten, Hotan and Bailes2004).

We provide the set of calibrator parameters as determined using the updated MEM algorithm as part of our publicly available data collection (see Appendix B).

We have not yet studied the off-axis polarisation properties of the system in detail, but provide theoretical models for the off-axis polarisation purity in our data collection.

5.1. Demonstration of astronomical observation modes

To demonstrate the end-to-end system, we show in Figure 9 a wide bandwidth observation of PSR J1559–4438 (B1556–44). The observation has been divided into eight frequency channels. The observation has been both flux density and polarisation calibrated. Details of the pulse profile variations with frequency will be published elsewhere, but we note that these results agree, in common observing bands, with previous measurements (e.g., Johnston et al. Reference Johnston, Karastergiou, Mitra and Gupta2008 and Johnston and Kerr Reference Johnston and Kerr2018).

Figure 9.

Pulse profiles for PSR J1559–4438 (B1556–44) in eight different frequency bands. The top section of each panel shows the position angle of the linear polarisation. The profiles are shown as total intensity (black), linear polarisation (red) and circular polarisation (blue). The position angles (P.A.) have been corrected in infinite frequency assuming a rotation measure of $-5\,\text{rad \,m}^{-2}$ .

An example of the spectral line observing mode is shown in Figure 10. We observed both NGC 45 (HI Parkes All Sky Survey (HIPASS) J0014–23) and an off-source position for 5 min. The off-source position was $1.2^\circ$ offset in both azimuth and elevation. We recorded 4 096 channels per sub-band and a spectrum was written to disc every second. During both observations, the calibration signal was switched at 100 Hz and the output data product (available as part of our public data collection) contains lower resolution spectra representing the on and off states of the calibrator as well as the astronomy spectrum. The calibration process for the astronomy spectrum was carried out offline, using the on/off calibration spectra and the on-/off-source observations. The resulting Hi spectrum (black) is overlaid on the spectrum obtained from the HIPASS survey (Barnes et al. Reference Barnes2001) (red line). We note that this source is slightly extended and so we also overlay in the figure (blue, dotted line) the integrated spectrum obtained after identifying the source in the original HIPASS data cube (H236) using the Duchamp source finder (Whiting Reference Whiting2012). The spectral shape matches well with the HIPASS result and our flux scale lies between the HIPASS spectrum and Duchamp results.

6. Radio frequency interference

All radio astronomy receiver systems are affected by RFI. Ground-based, aircraft, and satellite transmissions are especially problematic for wideband centimetre-wavelength systems such as the UWL. Figure 6 shows the entire UWL spectrum, with some known RFI sources labelled.

Figure 10.

The UWL spectrum of the Hi emission from NGC 45 (black) overlaid on the expected HIPASS spectrum (red). The UWL and HIPASS spectra have spectral resolutions of 31.25 and 62.5 kHz and integration times of 300 and 450 s, respectively. The, higher, blue-dotted line is a reprocessing of the HIPASS data cube using the Duchamp source finder and integrating over the entire extended structure of the source.

In Australia, licensed ground-based transmitters can be identified through the Australian Communications and Media Authority website^{Footnote l}. These include numerous mobile phone signals including the 3rd generation (3G) networks between 850 and 900 MHz and the 4th generation (4G) networks around 760, 780, 1 800, 2 100, and 2 680 MHz. National Broadband Network (NBN) transmissions exist around 2.37 and 3.56 GHz.

As Figure 6 shows, the strongest persistent RFI bands are mobile phone transmissions in RF Band 1. These are primarily from transmission towers located 10 to 20 km away, both north and south of the telescope. These RFI signals enter the system through far-out sidelobes of the antenna response and hence are quite variable depending on antenna pointing. Despite the low gain of the antenna sidelobes, these signals can be 50 dB or more above the local noise floor. Over the full RF Band 1 bandwidth, they typically add up to 35 dB of power to the digitised signal. Additional attenuation is required in the RF Band 1 chain to avoid saturation effects and the consequent formation of intermodulation products across the RF Band 1 spectrum. RF bands 2 and 3 do not suffer from these problems since the RFI signals are much weaker in these bands.

There are also numerous satellites emitting in the UWL band, including the various global navigational satellite systems (such as GPS, Beidou, GLONASS, and Galileo) as well as communication systems such as the Thuraya (1 525–1 559 MHz) and Iridium (1 617–1 627 MHz) satellites. The positions of satellites for which public ephemerides are available can be monitored by the observer during an observation. For post-observation analysis of suspect or distorted bandpasses, an online resource exists^{Footnote m} permitting the observer to find RFI generating satellites that might have been near the main beam of the telescope at a given time in the past.

Short duration impulsive or transient RFI signals are also common. Because of their typically low duty cycle, they are not prominent in spectral plots such as Figure 6, but can affect astronomical observations. There are two main classes for such signals: (a) broadband signals, mainly from switching transients in local electrical systems or from lightning, and (b) highly modulated relatively narrowband (typically a few MHz) transmissions, from a variety of sources, for example, aircraft navigation systems, mobile phone handsets, and many satellite transmissions.

On 2019 June 5, we observed using a modified version of the spectral line/continuum mode that recorded maximum signal powers in $64\,\mu\text{s}$ samples of each 125 kHz channel, as well as providing the average power in each channel. The observation lasted for 15 min.

In Figure 11, we have selected three regions that exhibit significant transient RFI. The black line in this figure gives the mean spectrum and the blue line indicates the peak sample values recorded for each channel in the 15-min observation.

Figure 11.

Mean (black line) and peak-hold (blue line) spectra for parts of the band that contain significant impulsive interference. The vertical, green lines indicate sub-band boundaries. Labels in the upper panel indicate DME signals from the following towns and cities: Parkes (1 and 8), Nowra (4 and g), Canberra (5 and h), Sydney (6 and d), Richmond (7), Cowra (a), Williamtown (b), Cooma (c), Wagga Wagga (e), and Albury (f). ADS-B is indicated by 9 and air traffic control ground interrogation by 3. An internal clock system produces signal 2 at 1 024 MHz. The broad signature around 1730 MHz is an alias of strong RFI at the 1 860 MHz and the broad signal around 1 085 MHz is an alias from a strong mobile phone band just below 960 MHz. Several of the strong peaks, especially for the peak-hold data in the top panel, have wide sidebands; these originate from the short FFTs used to provide the high time resolution data.

Aircraft navigation signals cover a wideband from 962 to 1 150 MHz and typically consist of sequences of microsecond duration pulses with a low duty cycle. Consequently, they contain very little power on average but are prominent in the peak-hold spectra. The airport in Parkes has a Distance Measuring Equipment (DME) beacon transmitting at 1 018 MHz with corresponding airborne beacons interrogating on 1 081 MHz; both are strong signals in the UWL spectrum. Numerous other signals from other DME systems are seen between 1 020 and 1 150 MHz.

Aircraft-borne Automatic Dependent Surveillance - Broadcast (ADS-B) systems transmit at 1 090 MHz. In addition, ground-based air traffic control radars send an interrogation signal at 1 030 MHz. The 1 090 MHz signals will be visible at ranges of up to hundreds of kilometres for aircraft at cruising altitude and so at any given time signals from many aircraft will be seen. ADS-B transmissions consist of short sequences (lasting ${\sim}120\,\mu\text{s}$ ) of digitally encoded bursts with low duty cycle, where each individual burst is only $0.5\,\mu\text{s}$ in duration.

We know that much of the broadband impulsive interference that we detect is locally generated. There were initial concerns that our focus cabin digitisation system would generate an unacceptable level of RFI. However, we estimate that the digitiser packaging and the shielded cabinets together provide at least 100 dB of RFI attenuation and no such leakage has been detected in the recorded signals. Even though active mobile, Wi-Fi, and/or Bluetooth devices are not permitted on site, we do observe signals across these bands. We are still investigating the source of these signals, but it is likely that some originate in electronic devices used by the general public at the Telescope Visitors Centre, which is only ${\sim}100\,\text{m}$ from the telescope.

In Table 1, we list the fraction of each sub-band that was removed by automatic pulsar data processing scripts in more than 50% of 236 pulsar observations taken on 2019 February 17. In total, such automated processing removes ${\sim}814\,\text{MHz}$ of the band. To reject aliasing effects resulting from the use of critically sampled digital filterbanks, we remove $\pm 5\,\text{MHz}$ at each sub-band boundary, a total of 260 MHz of bandwidth. The planned implementation of oversampled filterbanks will remove the need to reject these sub-band edges.

7. Future developments and conclusion

Iterative improvements will continue to enhance the UWL receiver system. Planned upgrades include a novel, single-dish calibration scheme in which the noise source is modulated in a pseudo-random sequence continuously throughout an observation. The signal processor system can then use the known pseudo-random sequence to extract the time-dependent calibration solutions.

A further planned upgrade will be to develop oversampling (replacing critical sampling) of the input data stream, preventing signal degradation at the sub-band edges, and effectively eliminating aliasing effects. This will require updates to the FPGA firmware.

The current UWL system provides a platform to research RFI mitigation techniques. Transmissions from mobile phone and NBN towers are confined in bandwidth and are quasi-continuous. In principle, such signals can be removed using adaptive filtering techniques based on having an RFI reference signal that can be correlated with the astronomy signal (Kesteven et al. Reference Kesteven, Manchester, Brown and Hampson2010). In practice, this is difficult because of the great strength of the interfering signals, requiring very high S/N in the reference signal, and the fact that multiple sources contribute to the signal in some RFI bands. For example, the 4G band containing 760 MHz is transmitted from both Alectown, about 11 km north of the telescope, and from a tower near Parkes, about 15 km south-west of the telescope. In principle, some satellite signals could also be removed using adaptive filter techniques, but obtaining a reference signal with sufficient S/N is a challenge. Signals that are easily detectable in the peak-hold spectra, but only marginally visible in the average spectra, are best removed in the time domain. Signal processing algorithms to achieve this are currently being evaluated but are not yet implemented in the standard pipeline processing (for instance, Nita & Gary Reference Nita and Gary2010). Localisation and removal of local sources of transient RFI remains an ongoing effort.

The telescope control system is currently being upgraded to a web-based system. When complete this will enable a range of observing modes, from almost hands-off scheduled observing, to hands-on control of the telescope and observing system.

The astronomical signal processing system will also be upgraded to provide (1) commensal observing modes, (2) transient detection capability, (3) the ability to fold multiple pulsars simultaneously, (4) to phase resolve spectra within a pulsar period, and much more. We will continue to characterise the system including using holographic measurements to study beam patterns in detail (Iheanetu et al. Reference Iheanetu, Girard, Smirnov, Asad, de Villiers, Thorat, Makhathini and Perley2019).

The observing system described here is provided to the astronomical community. Teams can propose to observe with the system twice per year using the standard Australia Telescope National Facility proposal system (https://opal.atnf.csiro.au/). One of the primary challenges facing users of the system is the massive data volumes. Observation data files can relatively easily become TBs in size. Not only such files are difficult to transfer to local computer systems, our current astronomical software suites for single-dish telescopes are ill-equipped to deal with these data volumes. In addition, traditional calibration procedures such as feed rotation and frequency switching cannot be applied and new methods will be required to extract the required scientific results from the wideband data (such as new wideband timing methods, e.g., Pennucci Reference Pennucci2019).

In the longer term, we plan to upgrade the Parkes receiver suite with higher frequency, single-pixel wideband receivers and a cryogenically cooled, phased array feed. With this suite of receiver systems and versatile signal processor and control systems, the Parkes telescope will continue to provide cutting-edge observations of astronomical sources in the Southern Hemisphere for many years to come.