The CHIME Pulsar Project: System Overview

The CHIME telescope consists of four 5 m focal length parabolic cylinders that are 20 m wide and 100 m long. The cylinders are oriented north–south, each mounted with 256 dual-polarization “cloverleaf” antennas (Deng & Campbell-Wilson 2014) placed 30 cm apart below a 68 cm wide ground plane at the parabolic focus. Two low-noise amplifiers are coupled directly to each antenna. These are coupled via 60 m of low-loss coaxial cables to custom band-defining filter amplifiers located in a radio frequency–shielded room containing analog-to-digital converters and the frequency channelization portion of the CHIME correlator. The basic telescope properties are given in Table 1.

Table 1. Telescope Specifications

Field of view	120° (N–S)
	1.3°–2.5° (E–W)
Receiver noise temperature	∼50 K (nominal)
Frequency range	400–800 MHz
Polarization basis	Orthogonal linear
Telescope longitude	$-119^\circ 37^{\prime} 25\buildrel{\prime\prime}\over{.} 238$
Telescope latitude	$49^\circ 19^{\prime} 14\buildrel{\prime\prime}\over{.} 553$
Telescope elevation a	547.918 m

Note.

^a Above the GRS 80 ellipsoid.

Download table as:  ASCIITypeset image

CHIME employs a hybrid FX correlator that consists of custom-built electronics and adapted commodity processing units. The first stage, known as the F-Engine, is responsible for signal digitization and frequency separation, while the second stage, the X-Engine, is responsible for spatial correlation and beamforming.

2.2.1. F-Engine: Digitization and Channelization

2.2.2. X-Engine: GPU Correlator and Beamformer

The CHIME/Pulsar back end consists of 10 independent compute nodes that are connected directly to the CHIME correlator through a series of network switches. Each compute node consists of a Supermicro motherboard, an Intel Xeon E5-1650 central processing unit (CPU), 128 GB of RAM, and a solid-state drive. Data are received through a 10 Gbps Intel Network Interface Controller and temporarily buffered in RAM before being processed through a fast Fourier transform (FFT)–based coherent dedispersion algorithm on a single liquid-cooled NVIDIA Titan X GPU. The resulting data are temporarily stored and further processed on a local 60 TB data archiver.

The CHIME/Pulsar back end ultimately writes a variety of data products with variable file sizes several hundred times every sidereal day at a time-averaged rate of 67 Mbps. In order to handle the output data rate, we built a local data-archive server using components similar to the hardware described above. The archiving server employs two redundant arrays of independent disks for secure, short-term storage of CHIME/Pulsar data products. This short-term data archive can accommodate several months of continuous typical data acquisition and is sufficient in the event that higher data-rate modes (see Section 3.3) become emphasized for a moderate time span. These data are eventually transferred offsite via ground shipment of physical hard drives to CHIME/Pulsar institutions for uploading to multipurpose computing facilities for offline processing and long-term preservation. The primary long-term data archiving occurs on the Cedar cluster of Compute Canada. ²¹

We digitally form 10 dual-polarization tied-array beams within the X-Engine by summing all 1024 inputs phased to specified celestial coordinates. The 10 independent beams allow for tracking of 10 different sky positions simultaneously within the primary beam of CHIME. The repointing of each beam is performed over the network using the Representational State Transfer application programming interface and typically occurs within milliseconds of command execution.

The complex gain calibration (amplitude and phase) of the input data is achieved via point-source calibration with bright continuum sources such as Cas A, Cyg A, and Tau A (CHIME Collaboration et al., in preparation). During a bright continuum source transit, the visibility matrix for CHIME is treated as approximately the product of the complex gain and sky signal (as viewed through the beam response function). In this limit, computing an eigendecomposition of the visibility matrix allows us to determine the complex gain for each antenna at each frequency channel in the direction of the calibrator source. The eigenvectors corresponding to the two largest eigenvalues represent the complex gains for the X and Y polarizations of each input. After determining the eigenvectors, we also remove the known interferometric phase based on the physical separation of the antennas by a process known as fringe stopping (e.g., Thompson et al. 2001).

As part of the calibration process, data from poorly behaving antennas and contaminated frequencies are masked. A static list of inputs that are known to be malfunctioning are removed, and this list is updated on a daily basis. Calibration solutions with more than 5% of inputs deemed bad are flagged as poor quality, and the most recent preceding solutions deemed adequate are instead used until the next calibrator transit occurs. While radio-frequency interference (RFI) excision in CHIME data can be a serious challenge, some channels are known to always be contaminated, and these channels are masked in the calibration solutions.

During the “precommissioning” period of CHIME operations (before 2018 August), phase-only calibration was computed every few days. Since 2018 September, daily amplitude and phase-calibration solutions have been calculated.

For each X-Engine GPU node, the resulting beamformed data from each of the four processed frequency channels are scaled down, encoded into (4+4) bit complex numbers, and packetized in VLBI Data Interchange Format (VDIF). ²² Each individual packet consists of 625 samples of one of the two polarizations for each of the four frequency channels and a VDIF header of 32 bytes, resulting in a total packet size of 5032 bytes. The corresponding pulsar data rate on each FX correlator node for all 10 beams is 0.2516 Gbps, with a total network load of ∼64 Gbps for all pulsar beams. Each individual pulsar back-end node receives ∼6.4 Gbps of data from the correlator in the form of User Datagram Protocol (UDP) packets. The incoming data rate to each pulsar node corresponds to 1.6 × 10⁵ packets s^–1. We have optimized our network-capture algorithms in order to properly receive packets with negligible packet loss. ²³ The received packets are assembled and written to memory in preparation for further processing.

The main purpose of the CHIME/Pulsar system is to autonomously monitor radio pulsars with regular cadence. The beamformed dual-polarization time series are coherently dedispersed and processed to form the data product defined by the scheduled observation configuration. Most observations occur in “fold mode,” where the dedispersed baseband data are manipulated to yield time-averaged pulse profiles for all 1024 frequency channels over a specified integration time using a set of pulsar rotation parameters for coherent averaging of the signal. In most cases, the number of profile phase bins for a given fold-mode observation is 256, although this number is nominally defined by the science case and, therefore, dictated by the scheduling process. By default, 10 s integrations are formed for all fold-mode observations, although this could, in principle, be adjusted on a per-pulsar basis. When desired, observations may also be conducted in “filterbank mode,” where the dedispersed baseband time series are converted into a Stokes-I data stream that is then downsampled in time. In special circumstances, it is scientifically worthwhile to directly record beamformed baseband data for specific pulsars and FRBs of interest. We have developed the appropriate software capabilities to enable all three types of observation with the CHIME/Pulsar system. The basic output data specifications of the CHIME/Pulsar system are given in Table 2.

3.3.1. Fold Mode

For fold-mode observations, the CHIME/Pulsar back end uses the Digital Signal Processing for Pulsars (DSPSR) suite, ²⁴ an open-source GPU-based library (van Straten & Bailes 2011). Written in C++ and CUDA, ²⁵ DSPSR is a high-performance, general-purpose tool for high-time-resolution radio pulsar studies using accelerated computing. It is used in many astrophysical applications (e.g., Karuppusamy et al. 2012; Martinez et al. 2015; Price et al. 2016) and extensively used as a part of real-time processing instrumentation for various telescopes around the world.

The DSPSR suite has the ability to read from and write to data buffers created using the PSRDADA library. ²⁶ The PSRDADA is an open-source software that allows the creation of flexible and well-managed ring buffers with a variety of applications for piping data from process to ring buffer, and vice versa. In addition to enabling the transfer of data between processes, PSRDADA provides various utilities to manage data within the ring buffer itself.

For fold-mode observations, the incoming data streams from the CHIME X-Engine are sorted and arranged as a polarization-frequency-time series prior to entering the real-time processing stage. The assembled data are then passed on to the DSPSR process using PSRDADA buffers. Even though it has the ability to accept data with (4+4) bit encoding, DSPSR runs significantly faster when instead receiving data with (8+8) bit encoding. We have thus modified the PSRDADA software to convert data to (8+8) bit encoding before passing them to DSPSR in order to circumvent this issue. Once the data are passed into the DSPSR GPU buffers, DSPSR executes several kernels that perform coherent dedispersion and timing-model folding of the data before writing to disk. A schematic of the processing chain for fold-mode observations is presented in Figure 2.

Zoom In Zoom Out Reset image size

Output fold-mode data products are written to structured files (“archives”) that are readable by the PSRCHIVE software suite ²⁷ (Hotan et al. 2004; van Straten et al. 2012) for offline processing and analysis. An example of the CHIME/Pulsar system fold-mode data from an observation of PSR B1937+21 is given in Figure 3.

Zoom In Zoom Out Reset image size

3.3.2. Filterbank Mode

The CHIME/FRB system is discovering many repeating FRBs (CHIME/FRB Collaboration et al. 2019a, 2019b; Fonseca et al. 2020), as well as single pulses from unknown radio pulsars and “rotating radio transients” (RRATs; McLaughlin et al. 2006). Some fraction of the CHIME/Pulsar observing time is thus expected to be used for follow-up observations of these sources. To observe repeating FRB sources, the pulsar back end must be able to record coherently dedispersed high-time-resolution filterbank data. Since these extragalactic sources are known to have dispersion measures (DMs) upward of 1000 pc cm⁻³, it is also essential that the back end can support such DM ranges.

The readily available DSPSR filterbank module digifil cannot support real-time performance at such large DMs at the CHIME operating frequency range. To overcome this, we have developed a new codebase for the filterbank mode with the ability to dedisperse incoming data to a maximum DM of 3300 pc cm⁻³ in real time based on work by Naidu et al. (2015). The significant change in our implementation is the input data format. Based on benchmark tests on FFT performance, the efficiency of the GPUs can be improved by a factor of 5 if the input data format is changed from polarization-frequency-time series to polarization-time-frequency series, as shown in Figure 4.

Zoom In Zoom Out Reset image size

The received UDP packets are arranged into a polarization-time-frequency series and passed on to the GPU for coherent dedispersion at a user-specified DM. These dedispersed data are then passed on to the CPU for downsampling and then manipulated into a community-established filterbank format before writing to disk. All of the code for the CPU processes after GPU kernel execution is written in Intel AVX2 intrinsics.

An automated scheduler is necessary to make maximal use of the continuous operation of CHIME for pulsar observations. The transit nature of the telescope means that sources are set to be observed only when they are within 1.75° of the meridian. This results in certain parts of the sky having more sources than the number of available beams. To address this, we have developed a probabilistic scheduling algorithm that employs user-defined weights to prioritize certain sources, or positions of interest, over others. This priority system is used to resolve scheduling conflicts. There are five base weight values, where the highest weight is assigned to MSPs currently used in a PTA for GW detection, as well as other interesting pulsars (e.g., relativistic pulsar–binary systems) such that they will be observed every day. The remaining four weighting values are defined such that each additional tier has twice the weight of the previous tier. There is also a feedback mechanism that reduces the priority associated with a source (as long as it is not assigned the highest priority level) based on the number of days it has been observed over the previous 14 days. The weight of a previously scheduled source will be reduced by a factor equivalent to the number of days observed over the period.

The scheduler adds the next source to the schedule through an iterative process. It looks for sources whose transit time began within 7 minutes of the end of the transit time of the previous source for each one of the 10 beams. In cases where two or more sources fall within the time frame, the scheduler will automatically select the source with a weight that corresponds to being observed daily and the starting time of transit being closest to the end of the transit time of the previous source. If none of the sources have such weighting, each source will be given a probability equivalent to its weighting divided by the weighting of all sources in conflict, and a source will be randomly selected with the determined probabilities.

In order to evaluate the effectiveness of the scheduler, we examined the number of times each source is scheduled over a 60 day period. Figure 5 shows the histograms of sources and how often they are scheduled. We found that when the telescope is not transiting the Galactic plane, the scheduler is highly effective in cycling through all sources, with more than half of the sources being observed on a near-daily cadence (i.e., scheduled >30 times over 60 days), with only about 10 of the 559 sources having a cadence of 2 weeks or more (i.e., scheduled <4 times over 60 days). On the more crowded field of R.A.s between 18 and 20 hr, sources on average will have a longer cadence, with less than 20 sources scheduled with a near-daily cadence and about 290 out of the 620 sources having an observing cadence of 2 weeks or longer, among them 14 sources that were not scheduled at all. While this circumstance is less than ideal, it nonetheless demonstrates that the scheduler is able to cycle through most of the sources, even with a large number of conflicts.

Zoom In Zoom Out Reset image size

We also allow for flexibility and autonomous schedule interruption in the case of special target-of-opportunity events. For example, if there is a candidate alert from the CHIME/FRB system, or if a radio magnetar is reported to burst, we can override the standard schedule to point at the location as soon as it is visible. It is also possible to grid around a given source position simultaneously by employing multiple tracking beams.

Multiple web-based tools are in place to actively monitor the continuous operation, including a live data-product viewer and a Grafana ²⁸ visualization platform for displaying system, transmission, and data-storage metrics collected using the Prometheus toolkit. ²⁹

During the “precommissioning” period of CHIME operations, from 2018 July to 2019 February, we gradually increased the number of pulsar back-end nodes to the current 10 available and employed a monthly operation cycle with a week of software deployment for new features between cycles. The CHIME/Pulsar system typically had an uptime of ∼70% during the precommissioning phase of operation (including downtime due to software-deployment weeks). As expected, the on-sky observing time has increased to ∼100% as the commissioning period continues and transitions to the “full-science” mode of CHIME operation. As of publication, the CHIME/Pulsar back end is continuously acquiring data alongside the various back ends currently deployed for the CHIME telescope.

Figure 6 shows a selection of MSPs and their daily signal-to-noise ratio (S/N) variations as observed by the CHIME/Pulsar back end over the 2019 commissioning period. These S/N variations indicate that the system is stable during the commissioning period. Some degree of S/N variation remains, though we believe this is due to imperfect excision of the time-varying RFI environment, imperfections in the calibration solutions, and potentially scintillation effects toward low-DM pulsars.

Zoom In Zoom Out Reset image size

Frequency channels affected by bright RFI (notably the LTE cellular network band between ∼730 and 755 MHz and several digital television bands between 500 and 600 MHz) are identified from the amplitude and phase-calibration solutions provided for each antenna and frequency channel (see Section 3.1). Since these solutions are calculated and applied on a daily basis, the particular frequency channels that are masked can vary on similar timescales, but the RFI environment is generally stable.

The bulk of RFI excision occurs during postprocessing (e.g., while analyzing the folded or filterbank data products). A large fraction of the corrupted data can be removed by masking an empirically constructed list of frequency channels where RFI is persistent (approximately 15% of the band). This corrupted channel mask can, in principle, also be combined with the calibration-based mask applied in the beamforming stage. More refined RFI excision can be attained by using techniques available in standard pulsar-processing software packages, e.g., PSRCHIVE or PRESTO (Ransom 2011), but is left to user discretion. For example, in Figure 7, we have excised RFI using both the persistent corrupted channel mask and a method based on RFI cleaning utilities from the CoastGuard software suite (Lazarus et al. 2016), where several robust statistical quantities on a per-channel, per-subintegration basis are evaluated to determine whether the samples are corrupted. ³⁰ Typically, we mask ∼25% of the 1024 frequency channels, yielding ∼300 MHz of usable bandwidth spread noncontiguously across the full 400 MHz observing band. As our understanding of the system and RFI environment evolves, these will be tuned to improve sensitivity.

Zoom In Zoom Out Reset image size

CHIME is nominally capable of observing all known pulsars down to a decl. of ∼−20°. To date, we have redetected over ∼500 known pulsars with precommissioning CHIME/Pulsar observations. Out of these ∼500 observable sources, 209 pulsars have published flux densities at 600 MHz in the ATNF Pulsar Catalogue ³¹ (Manchester et al. 2005). For these pulsars, we can compare their detected average S/N with the expected value to assess the sensitivity of the CHIME/Pulsar system.

We calculate the expected S/N for each pulsar using the radiometer equation. Many pulsars have steep spectral indices, and the sky temperature (T_sky) will vary significantly across the 400 MHz bandwidth. Therefore, in the estimation of the expected S/N, we separately consider three subbands and sum the three S/N values in quadrature. We take the flux density values at 400, 600, and 800 MHz (S₄₀₀, S₆₀₀, and S₈₀₀, respectively) from the ATNF Pulsar Catalogue. Where flux densities are not available, we extrapolate the subband flux density from the published spectral index. For each of the 209 sight lines, we obtain T_sky at 400, 600, and 800 MHz from the Haslam survey (Haslam et al. 1982). For the system temperature (T_sys), we consider both the optimistic nominal value of 50 K and a pessimistic worst-case scenario of 100 K to determine a range of expected S/N values.

Figure 8 summarizes the results of this analysis at 600 MHz. From the top left panel, it can be seen that most of our redetections are close to the 1:1 diagonal line. The lowest published flux density of a pulsar we have redetected is 0.6 mJy. The CHIME primary beam response is strongly dependent on frequency and zenith angle, where at large zenith angles (low declinations; blue points in Figure 8), there is a significant decrease in sensitivity. Using a preliminary model, we have taken the primary beam attenuation into account when computing the expected S/N for pulsars, equalizing the response as a function of decl.

Zoom In Zoom Out Reset image size

Precision pulsar timing requires exquisite polarization calibration. In particular, the stability of the observed polarization properties of any given pulsar over different epochs is critical (e.g., van Straten 2006; Foster et al. 2015). A general polarization calibration scheme for CHIME/Pulsar is under investigation and is a high-priority item for all CHIME projects. Currently, however, we cannot transform from instrumental to intrinsic Stokes parameters. Nevertheless, Stokes profile stability can nominally be ensured by using arbitrary Mueller (transfer) matrices to effectively absorb any telescope miscalibration and depolarization effects, including those introduced by averaging over frequency and time, when creating times of arrival (TOAs; e.g., Archibald et al. 2018). ³² This approach is powerful and flexible, but it is not easy to translate the elements of the resulting transfer matrix to physical telescope properties.

To examine the stability of the Stokes parameters over time, we inspected the polarization profiles of several bright targets, including B1937+21 (see Figure 9). A high-S/N smoothed frequency- and time-averaged template for each pulsar was created and used as a reference to compare observations of the same pulsar at different epochs. We compared the Stokes profiles and the resulting transfer matrix required to correct each observation to the reference profile, where the matrix elements should not differ substantially over time if the polarization response of the system is stable. The absolute difference between the template and observed Stokes profiles (without any correction) relative to the Stokes-I maximum on different days is typically ∼10%. Applying the arbitrary inverse transfer matrix corrections results in a residual difference of ∼1% or less. For individual pulsars, especially those that scintillate or exhibit mode changing, there are instances where the Stokes parameters change significantly on daily timescales relative to a given template.

Zoom In Zoom Out Reset image size

Ultimately, polarization stability and calibration are an ongoing challenge for CHIME/Pulsar. Applying the proposed arbitrary transfer matrix approach to stabilize the Stokes parameters from day to day will be a necessary step when conducting high-precision timing data analysis. Further effort will also be made to adapt this approach to a wideband regime, where the frequency dependence of the polarization response is also considered.

The PFB used by the CHIME FX correlator applies a sinc-Hann windowing function in the Fourier domain to the raw digitized sky signal when forming 1024 frequency channels across the CHIME band (Bandura et al. 2014). While chosen to maximize channel sensitivity, the imperfect response in each synthesized channel produces a leakage of signal between adjacent channels. Coherent dedispersion by the CHIME/Pulsar back end will ultimately produce aliasing of the detected pulsar signal, with a frequency-dependent lag between the original pulse and its aliased counterpart across the CHIME band.

Figure 10 shows examples of spectral leakage in pulse-averaged spectra for PSRs B1937+21 and J0740+6620. Artifacts due to leakage are apparent within a single observation of PSR B1937+21 with CHIME/Pulsar, though similar features are far less prominent in slower, fainter pulsars like J0740+6620, even after integrating data taken over 100 epochs, as shown in Figure 10.

Zoom In Zoom Out Reset image size

The MeerTime pulsar-timing back end mitigates spectral leakage in the MeerKAT observing system by using a modified sinc-Hann windowing function in their F-Engine system; the modified windowing function used by MeerTime suppresses the response of the synthesized-channel boundaries, which minimizes leakage but simultaneously lowers the effective sensitivity in each channel (Bailes et al. 2016). The commensal nature of pulsar/FRB observations with CHIME requires that the F-Engine use the same PFB configuration for all back ends when generating channelized data streams. Therefore, spectral leakage will be present to varying degrees in CHIME/Pulsar observations, and its impact on timing will be assessed during offline processing.

In order to establish the timing capabilities of the CHIME/Pulsar system, we collected near-daily observations of MSPs observed by the North American Nanohertz Observatory for Gravitational Waves (NANOGrav) during a several-month commissioning period when telescope sensitivity was generally stable. We used timing solutions for online folding that are also used by NANOGrav at the 305 m Arecibo Observatory and the 100 m Green Bank Telescope (GBT) for pulsar-timing data acquisition. Initial “template” pulse profiles of CHIME/Pulsar data, necessary for cross-correlation and arrival-time estimation, were generated by excising channels containing RFI, coherently adding timing data taken across the commissioning period, and fully averaging the stacked set in time and frequency. The time- and frequency-averaged profile was then denoised using a wavelet transform.

The TOAs were then computed via cross-correlation between the template profiles and a downsampled form of the CHIME/Pulsar fold-mode timing data. For analysis of NANOGrav pulsars, we fully integrated available fold-mode data in time and downsampled in frequency from the native resolution to 32 channels prior to TOA generation, yielding a maximum of 32 channelized TOAs per epoch. This level of downsampling is similar to the reduction methods used by NANOGrav in order to evaluate DM variations over time (e.g., Levin et al. 2016).

Best-fit timing residuals for PSR J0645+5158, computed using the TEMPO pulsar-timing analysis package ³³ (Nice et al. 2015) when modeling NANOGrav and CHIME/Pulsar TOA data simultaneously, are shown in Figure 11. We used the timing solution generated for the 12.5 yr NANOGrav data release (Alam et al. 2021a, 2021b) to model this combined data set, fitting for physical and instrumental parameters such as pulsar-spin frequency and its first time derivative, astrometry (ecliptic coordinates, proper motion, and timing parallax), DM values estimated within 1 day bins, and an arbitrary offset between the NANOGrav and CHIME data that reflects instrumental differences. As shown in Figure 11, we found that a numerically adequate timing model can be achieved when combining the CHIME and NANOGrav data sets, yielding best-fit parameters and fit statistics consistent with those previously reported from NANOGrav. The rms residual for the CHIME data set on J0645+5158 is 0.83 μs, which compares well with the rms values for the GBT data sets collected using the 800 MHz receiver (0.71 μs) and the 1400 MHz receiver (1.38 μs). A detailed timing study of NANOGrav sources observed with CHIME will be presented in a forthcoming study.

Zoom In Zoom Out Reset image size

The long-term timing of pulsars, with a focus on PTA targets, is the primary science driver for CHIME/Pulsar. Detection of nanohertz-frequency GWs using a PTA directly depends on a regular, high-cadence observing program of a large MSP sample over many years. The CHIME/Pulsar system is naturally producing such a data set with a near-daily cadence. As such, CHIME/Pulsar timing data of NANOGrav sources (e.g., Arzoumanian et al. 2018) will be combined in future NANOGrav data releases in order to improve detection of the stochastic GW background at nanohertz frequencies.

The autonomous design of the CHIME/Pulsar system, combined with the wide field of view of the telescope, allows for regular, priority-based timing observations of sufficiently bright pulsars, with 400–500 pulsars significantly detected within a single day of operation. This capability is unprecedented in the northern hemisphere and only recently achieved for southern pulsars with the development of the MeerKAT pulsar-timing program (MeerTime; Bailes et al. 2016) and the UTMOST pulsar instrument (Bailes et al. 2017; Jankowski et al. 2019). The CHIME/Pulsar system actively monitors observable sources that are reported in the ATNF pulsar catalog with a range of cadences. CHIME/Pulsar also observes newly discovered pulsars found by the Green Bank North Celestial Cap (Stovall et al. 2014) survey in order to aid in the confirmation and follow-up of their discovered sources.

Long-term timing of binary radio pulsars often yields secular and/or periodic variations from purely Keplerian motion. The most famous examples of secular variations are those associated with general-relativistic orbital decay and precession in compact orbits (see Stairs 2003, for a review), though many nonrelativistic pulsar orbits have been observed to vary over time due to evolving sky orientations induced by proper motion (e.g., Kopeikin 1995, 1996). By contrast, the relativistic Shapiro time delay (Shapiro 1964) is a periodic effect observed in sufficiently inclined binary systems of any size where the pulsed signal traverses varying amounts of spacetime curvature induced in the companion star over the course of the orbit. Measurements of such effects allow for direct constraints on the masses (e.g., PSR J0740+6622; Cromartie et al. 2020) and geometry of the systems in question (e.g., PSR J0437−4715; van Straten et al. 2001) and thus stand to yield high-impact information that is otherwise inaccessible from purely Keplerian dynamics.

The CHIME/Pulsar system is expected to produce a rich data set for probing binary astrophysics by monitoring all visible binary pulsars with near-daily observing cadences over the course of telescope operation. As an example, Figure 12 shows the simulated best-fit estimates of a timing parameter quantifying relativistic time dilation and gravitational redshift in the PSR J0509+3801 double neutron star system (Lynch et al. 2018). While typical programs observe such sources on monthly or bimonthly cadences, the simulated estimates in Figure 12 suggest that such effects can be better constrained with high-cadence observations—like those achievable with CHIME/Pulsar—due to quickened evaluations of orbital parameters and their variations. Moreover, we anticipate the daily cadence to yield detections of the Shapiro delay on timescales much shorter than those typically seen in the current pulsar-timing literature, due to the faster rate of achieving dense orbital coverage.

Zoom In Zoom Out Reset image size

Pulsars are sensitive probes of the ISM and its structural variations across many lines of sight. By processing and recording data at low radio frequencies, the CHIME/Pulsar system is producing a rich and growing data set for autonomously monitoring frequency-dependent features in pulsar data. Examples of such effects include temporal variation of DM (e.g., Lam et al. 2016; Lentati et al. 2017; Lam et al. 2018), frequency-dependent DM (Cordes et al. 2016; Donner et al. 2019), “echoes” in pulsar spectra (e.g., Graham Smith et al. 2011; Michilli et al. 2018; Driessen et al. 2019; Bansal et al. 2020), “extreme scattering events” in flux density modulations (e.g., Coles et al. 2015; Kerr et al. 2018), scintillation (e.g., Bhat et al. 1999a, 1999b, 1999c; Wang et al. 2005, 2008), or multipath scattering/pulse broadening (e.g., Bhat et al. 2004; McKee et al. 2018) caused by small-scale plasma structures. Understanding these effects is critical in achieving a GW background detection using PTAs (e.g., Cordes & Shannon 2010; Levin et al. 2016), but they are also informative when considering pulsar astrometry and system dynamics (e.g., Lyne 1984; Pen et al. 2014; Reardon et al. 2019).

As a demonstration of the suitability of CHIME/Pulsar for studying small-scale dispersive variations, we measured the DM time series for PSR B1937+21, one of the fastest-known MSPs and one that is known to exhibit large and rapid variations in DM (e.g., Jones et al. 2017), following the process detailed in Donner et al. (2019). The 10 highest-S/N observations were selected, which were phase-aligned and summed to create a high-S/N frequency-resolved reference profile, with very little correlated noise. The original 1024 frequency channels were integrated to 16 channels, and a wavelet-smoothing algorithm was applied, resulting in a noise-free standard template. The remaining observations in our data set were also integrated to 16 frequency channels, and TOAs were measured for each via cross-correlation with the template. The TOAs were analyzed using TEMPO2 ³⁴ (Hobbs et al. 2006), with a timing model based on the one presented in Perera et al. (2019), with the time-varying DM and noise models removed. We fit for DM on each observing epoch while keeping the other timing-model parameters fixed to obtain a time series with a mean cadence of 1.2 days and a median DM precision of 2.9 × 10⁻⁵ pc cm⁻³, which we present in Figure 13. As a comparison, the DM time series for B1937+21 obtained from NANOGrav observations with the Arecibo and Green Bank observatories yield a similar mean DM precision of ∼2 × 10⁻⁵ pc cm⁻³, though evaluated over a monthly cadence (Alam et al. 2021a, 2021b). The measured DM time series from CHIME/Pulsar data displays an overall linear trend, with additional short-duration excesses lasting a few tens of days, which would not be easily resolved in low-cadence data sets (such as those normally employed by PTA experiments).

Zoom In Zoom Out Reset image size

Another important PTA science consideration that will benefit from the CHIME/Pulsar observing campaign is the study of the solar wind, which contributes an annual variation in DM to pulsar-timing data. This effect varies in time throughout the course of the 11 yr solar cycle, and it has been proposed (e.g., Madison et al. 2019; Tiburzi et al. 2019) that long-term, high-cadence radio monitoring of pulsars close to the ecliptic plane will provide valuable information about this phenomenon. Recent work has shown that in high-cadence data sets with high DM precision, typical simple solar wind models do not adequately describe the excess dispersive delay (Tiburzi et al. 2019), and that its robust mitigation will become increasingly important in searches for low-frequency GWs as PTA sensitivity continues to improve (Madison et al. 2019).

The polarization properties of radio pulsars have provided a wealth of information about their emission mechanism and geometry and the magnetized Galactic medium (e.g., Lorimer & Kramer 2004). The CHIME/Pulsar back end records and stores full Stokes information, allowing polarization studies to be performed for all detected sources. However, polarization calibration, necessary for many pulsar studies, is difficult to obtain for a transit telescope such as CHIME, where the interplay between the two orthogonal sets of feeds changes during the observation and is strongly frequency dependent. Obtaining a beam model of the telescope accurate enough to recover the intrinsic polarization of the signal is a work in progress involving measurements from various CHIME back ends, including CHIME/Pulsar (see also Section 4.4). Nevertheless, even before being able to calibrate the instrument, some interesting properties of polarized sources can be measured, in particular, the effect of Faraday rotation. CHIME/Pulsar, operating at low frequencies and with a large fractional bandwidth, is able to provide very precise measurements of rotation measure (RM).

The capability of CHIME/Pulsar to measure precise RM values has already been demonstrated in a recent work by Ng et al. (2020a), where we present RM values for 80 pulsars, 55 of which are measured for the first time. We plan to measure the RM of most detectable northern pulsars, many of which either do not have a reported RM value, or the cataloged value was measured decades ago. By virtue of observing a large fraction of northern pulsars regularly, CHIME/Pulsar will also significantly contribute to efforts to map the Galactic magnetic field, complementing similar ongoing studies at low frequencies (e.g., Lenc et al. 2018; Sobey et al. 2019). In addition, frequent observations present an exciting opportunity to investigate temporal RM variations of every observed source. Measuring simultaneous DM variations, as described in Section 5.3, will enable the measurement of the Galactic magnetic field gradient.

Approximately 200 pulsars to date ³⁵ have been observed to exhibit “glitches,” where their rotation period abruptly changes and can remain altered for several weeks or months subsequently (e.g., Radhakrishnan & Manchester 1969; Weltevrede et al. 2011; McKee et al. 2016; Lower et al. 2020). Studying glitches and their aftermath provides insight into the interior structure of neutron stars and the nuclear equation of state (e.g., Haskell & Melatos 2015; Haskell & Sedrakian 2018). The modular design of the CHIME/Pulsar system lends itself to immediate offline analysis of timing variations consistent with pulsar glitches from many different sources and potentially enables near-real-time alerting of such events to the transient astronomical community. Such analyses are especially interesting in the era of LIGO/VIRGO detection of GWs, where correlations of glitch and GW transient events are expected to yield multimessenger constraints on the interior structure of neutron stars. Furthermore, the observing cadence of the pulsar with a detected glitch can be increased in order to better model the postglitch recovery properties of the pulsar.

Daily observations, flexible recording configurations, and a wide observing band are ideal for long-term monitoring of the emission from objects such as repeating FRBs (e.g., Spitler et al. 2016; CHIME/FRB Collaboration et al. 2019a, 2019b; Fonseca et al. 2020), RRATs (McLaughlin et al. 2006), and intermittent (Kramer et al. 2006), nulling (Backer 1970a), and mode-changing pulsars (Backer 1970b; Lyne 1971; Naidu et al. 2017). This will allow us to probe any short- and long-term variation in the rotational properties of the sources and their potential link with the variability in emission properties (e.g., Lyne et al. 2010; Perera et al. 2016; Naidu et al. 2018; Stairs et al. 2019).

The high-cadence observations also allow us to search for previously unknown emission variability in known pulsars. As a testament to this, Ng et al. (2020b) identified new nulling and mode-changing pulsars from the commissioning data set (2018 July–2019 March).

The filterbank mode of the CHIME/Pulsar back end is being actively used to follow up both unknown single pulses of Galactic origin (Good et al., submitted) and repeating FRB sources discovered by CHIME/FRB. The large instantaneous field of view and the transit nature of CHIME/FRB has allowed for the blind detection of these sources. A number of these sources are then followed up with the more sensitive CHIME/Pulsar back end as the data are coherently dedispersed to the DM of the sources, allowing detection of faint pulses that will otherwise be missed by CHIME/FRB. Moreover, the tracking beams generated for CHIME/Pulsar remain centered on the positions of the sources during their transit at the CHIME site, allowing for robust detection of the intrinsic dynamic spectrum. Furthermore, CHIME/Pulsar has a time resolution of 327.68 μs, three times higher than CHIME/FRB, that could resolve smaller-scale structures of the single pulses otherwise undetectable by CHIME/FRB.

Follow-up tracking observations of the unknown Galactic sources with CHIME/Pulsar are also used to look for periodicity akin to pulsars and RRATs. These are done by either measuring the gaps between successive pulses or conducting periodicity searches including both FFT and “fast folding algorithms” (e.g., Parent et al. 2018). The new candidates detected by the CHIME/FRB system and confirmed by CHIME/Pulsar are subsequently monitored for long-term timing. ³⁶ New Galactic discoveries from the use of both CHIME/FRB and CHIME/Pulsar are reported in Good et al. (2020).

While individual observations of a source with CHIME/Pulsar are limited by the primary beam size of CHIME, which restricts transit time, the capability to perform multiple simultaneous observations means that we can afford to repeatedly observe a number of sources daily to stack and search the data for pulsations. This will allow for the potential detection of faint sources by leveraging long integration times. Such a strategy has been employed before in targeted searches of globular clusters (e.g., Cadelano et al. 2018).

CHIME/Pulsar is also able to rapidly follow up on any transient astronomical events that might emit radio pulses. The regular schedule can be overwritten quickly to point a pulsar beam toward the source of interest during the forthcoming transit. We expect to apply the capability to follow up on events such as magnetars undergoing X-ray outburst (e.g., Camilo et al. 2006; Rea et al. 2013; Esposito et al. 2020), GW events (Abbott et al. 2016, 2017), and new transitional MSPs (Archibald et al. 2009).

As with fold-mode observations, filterbank scans of known or candidate sources are assigned priorities and included in the automatic scheduling of acquisition with CHIME/Pulsar. However, it is possible to instead program CHIME/Pulsar to operate as a dedicated blind-search machine. In this case, the FX correlator and CHIME/Pulsar back end can beamform and process, respectively, up to ∼3000 distinct lines of sight per day for 5 minute acquisition times.