Matched Filtering for the Canadian Hydrogen Observatory and Radio-Transient Detector Galaxy Search

This paper presents a matched filtering strategy and a prediction tool for the CHORD galaxy survey that demonstrates how combining data over time and using offset declination scans can effectively resolve spatial aliasing caused by the telescope's highly redundant dish layout.

The Gist

Imagine you are trying to take a clear photo of a single firefly in a dark field using a very special, high-tech camera. This camera, called CHORD, isn't just one lens; it's a massive grid of 512 small dishes (like a giant honeycomb of satellite dishes) working together to listen to the "hum" of hydrogen gas in distant galaxies.

The problem? Because this camera is so perfectly symmetrical and redundant (it has many dishes doing the exact same job), it suffers from a weird optical illusion called spatial aliasing.

The "Hall of Mirrors" Problem

Think of CHORD's layout like a hall of mirrors. If you stand in the middle and look at a firefly, you don't just see the real firefly; you see dozens of reflections (aliases) scattered around the room.

In a normal telescope, the different angles of the lenses help you figure out which reflection is the real one. But because CHORD's dishes are arranged in a perfect, repeating grid, all the dishes agree on where the fake reflections are. If you take a single snapshot, the telescope can't tell the difference between the real galaxy and its "ghost" images. They all look equally bright and real.

The Solution: Time and Movement

The paper explains how to solve this "hall of mirrors" problem without building a new telescope. They use two clever tricks:

1. The "Moving Spotlight" Trick (Time Integration)

Imagine the firefly isn't standing still; it's flying across the sky as the Earth rotates.

• The Real Galaxy: As it flies across the sky, it passes right through the center of the telescope's "spotlight" (the primary beam). It gets very bright, then fades.
• The Ghost (Alias): Because of the mirror effect, the ghost image follows a slightly different path. It might pass through the spotlight at the wrong time, or it might skim the edge of the beam instead of hitting the center.

By watching the galaxy for a whole night (integrating the data over time), the telescope can see that the real galaxy has a perfect "bright-then-dim" curve, while the ghosts have a messy, mismatched curve. The math (called a "matched filter") acts like a detective that says, "This pattern matches the real firefly; those other patterns are just echoes."

2. The "Step-Aside" Trick (Offset Scanning)

Sometimes, even with time, the ghosts are still too similar to the real thing, especially if the telescope is looking near the celestial equator (the sky's "equator").

The paper suggests a simple fix: Move the telescope.
Imagine you are trying to find a specific person in a crowd. If you stand still, you might confuse them with someone standing behind a pillar. But if you take two steps to the left, the person behind the pillar is now in a different spot relative to you, while the real person stays the same.

The authors found that if CHORD shifts its aim by about 2 degrees (roughly the width of four full moons) and scans a second strip of sky right next to the first one, the "ghosts" move to completely different, confusing locations. When you combine the data from the first scan and the second scan, the ghosts cancel each other out, leaving only the real galaxy.

Why This Matters

If they don't solve this, the final catalog of galaxies CHORD produces will be full of errors. They might think a galaxy is in one part of the sky when it's actually 2 degrees away, or they might miss faint galaxies because a bright "ghost" is drowning them out.

The Bottom Line:
This paper is a blueprint for how to clean up the "echoes" in CHORD's data. It proves that by simply waiting (watching the sky rotate) and moving (scanning adjacent strips of sky), the telescope can distinguish the real universe from its own confusing reflections. This ensures that when CHORD finishes its survey, the map of the universe it creates will be accurate, not a hall of mirrors.

Technical Summary

Here is a detailed technical summary of the paper "Matched Filtering for the Canadian Hydrogen Observatory and Radio-Transient Detector Galaxy Search."

1. Problem Statement

The Canadian Hydrogen Observatory and Radio-transient Detector (CHORD) is a drift-scanning interferometric radio telescope consisting of 512 six-meter dishes arranged in a highly redundant $22 \times 24$ rectangular grid. While this redundancy aids in noise reduction and calibration, it creates a severe spatial aliasing problem.

• The Core Issue: In a redundant array, the baselines (vectors between dish pairs) are all integer linear combinations of the two shortest baselines. Consequently, a single point source on the sky produces identical phase differences for multiple distinct sky locations.
• Consequence: A single source generates "aliases"—multiple bright spots in the resulting map that are indistinguishable from the true source in a single snapshot.
• Specific Challenge for CHORD: Unlike Fast Radio Bursts (FRBs) or pulsars, which have broad spectral signatures that can help distinguish aliases, galaxies are spectrally narrow (HI 21 cm emission). Therefore, frequency-domain techniques cannot be used to resolve the spatial degeneracy.
• Risks:
  1. Mislocalization: Algorithms may identify an alias as the true source, reporting incorrect coordinates and flux.
  2. Confusion: Faint sources near bright aliases may be drowned out.
  3. False Positives: Aliases from sources outside the scanned strip may appear as real sources within the strip.

2. Methodology

The authors propose a search strategy based on Matched Filtering in the visibility plane, specifically analyzing the Correlation Coefficient ($R$) between a template and the data to quantify alias severity.

A. The Matched Filter Framework

The paper defines three key quantities derived from visibility data ($d$) and a point-source template ($t$):

1. Likelihood Change Map (LCM): A "dirty map" showing log-likelihood increase per unit brightness.
2. Maximum Likelihood Map (MLM): Optimizes brightness distribution but diverges in low-response regions.
3. Matched Filter ($M$): The optimal linear filter that maximizes the Signal-to-Noise Ratio (SNR). It is defined as:
   $$M(n; d) = \text{Re} \left( \frac{t(n)^\dagger N^{-1} d}{\sqrt{t(n)^\dagger N^{-1} t(n)}} \right)$$
   Where $N$ is the noise covariance matrix. The matched filter is preferred because it provides correct normalization for detection significance.

B. The Correlation Coefficient ($R$)

To quantify the risk of mislocalization, the authors define the correlation coefficient between a true source location ($n_s$) and a potential alias location ($n'$):
$$R(n', n_s) = \frac{t(n')^\dagger N^{-1} t(n_s)}{\sqrt{t(n')^\dagger N^{-1} t(n')}\sqrt{t(n_s)^\dagger N^{-1} t(n_s)}}$$

• $R=1$ implies perfect degeneracy (indistinguishable).
• $R<1$ implies the locations can be distinguished.

C. Three Analytical Scenarios

The paper derives $R$ for three increasingly complex observing scenarios:

1. Instantaneous Snapshot: A single integration period.
2. Time-Integrated: Combining data over a full sidereal day (drift scan).
3. Offset-Adjacent-Strip Scanning: Combining data from multiple pointings (re-pointing the array in declination).

3. Key Results

A. Instantaneous Case

• Result: Aliases are perfectly degenerate ($R=1$).
• Pattern: Aliases form a rectangular grid in tangent-plane coordinates, separated by approximately $1.5^\circ$to$2^\circ$ depending on wavelength and dish spacing.
• Implication: A single snapshot cannot distinguish the true source from its aliases.

B. Time-Integrated Case (Single Pointing)

By integrating data over a night, two effects break the degeneracy:

1. Primary Beam Crossing Effect: As the Earth rotates, the true source and its aliases pass through the primary beam at different times. The matched filter expects the signal to peak when the source is at the beam center. Aliases peak at the "wrong" time relative to the template, reducing their correlation.
   • East/West Aliases: Highly suppressed ($R < 0.3$) because their beam-crossing times differ significantly from the true source.
   • North/South Aliases: Less suppressed because they cross the beam at similar times, though the beam shape differs slightly.
2. Sky Curvature Effect: Real sources follow paths of constant declination. Aliases, however, follow paths of constant offset in tangent-plane coordinates. Over large hour angles, these paths diverge.
   • Latitude Dependence: This effect is strongest near the poles and negligible near the equator (flat sky approximation).
   • Equatorial Limit: If CHORD points near the celestial equator, the curvature effect vanishes. If a source is positioned such that its alias is symmetric across the equator, $R$ can approach 1 even with time integration.

C. Offset-Adjacent-Strip Scanning (Re-pointing)

The paper investigates re-pointing the array in declination by an offset angle $\delta\theta_o$ to observe adjacent strips.

• Mechanism: An alias that is degenerate for one pointing will generally not be degenerate for a different pointing because the relative geometry between the source, the alias, and the primary beam changes.
• Optimal Offset: The authors determine that an offset of $\approx 2^\circ$ (roughly the primary beam Full Width at Half Maximum) is optimal.
  • Offsets smaller than this provide redundant information.
  • Offsets larger than this may attenuate the signal too much in the second strip.
• Outcome: Combining two strips separated by $2^\circ$ significantly reduces the correlation coefficient of North/South aliases, breaking the degeneracy even at low declinations where time integration alone fails.

D. Probabilistic Interpretation

The authors translate correlation coefficients into mislocalization probabilities.

• Using Monte Carlo simulations, they show that for a $6\sigma$ detection threshold, the probability of an alias being brighter than the true source drops significantly when using the adjacent-strip strategy.
• Without re-pointing, low-declination observations have a high risk of mislocalization due to North/South aliases.
• With the $2^\circ$ offset strategy, mislocalization probability is minimized across the survey area.

4. Significance and Contributions

• Algorithmic Strategy: The paper establishes a rigorous mathematical framework for handling spatial aliasing in highly redundant interferometric arrays specifically for narrow-band galaxy surveys.
• Observing Strategy Optimization: It provides a concrete, actionable recommendation for CHORD operations: Re-point the array by $\approx 2^\circ$ in declination between survey strips. This simple operational change drastically reduces the risk of catalog errors without requiring complex hardware modifications.
• Predictive Tool: The authors present a tool to quickly estimate matched filter responses and alias locations, allowing for the prediction of survey completeness and reliability.
• Future Work: This paper lays the spatial groundwork for a future study (Bij et al., in prep.) that will combine these spatial findings with noise estimates and galaxy population statistics to forecast the total number of galaxies CHORD will detect.

Conclusion

The paper demonstrates that while the CHORD array's redundancy creates severe spatial aliasing, these aliases are not permanent barriers. By leveraging time integration (drift scanning) and strategic re-pointing (adjacent strip scanning with a $\sim 2^\circ$ offset), the telescope can effectively distinguish true galaxy positions from aliases, ensuring the production of a reliable and accurate HI galaxy catalogue.