Interferometric Trigger Mechanisms

• Interferometric trigger is a real-time control mechanism that uses phase differences or beam power from multiple channels to maintain fringe coherence and resolve ambiguities.
• It is applied in optical systems like PHASECam for multi-wavelength fringe-jump detection and correction, enabling rapid and precise OPD adjustments.
• In radio systems, coherent beamforming in arrays such as NuPhase and ARA A5 improves SNR by summing correlated signals, lowering detection thresholds for weak impulses.

Interferometric trigger denotes a trigger or control-decision mechanism derived from interferometric observables rather than from single-channel amplitudes alone. In fringe tracking and phase control, it is any real-time, autonomous decision event based on interferometric phase measurements that initiates a control action, such as an OPD adjustment, to restore or maintain the desired fringe coherence. In phased radio arrays, the same idea appears as a low-threshold trigger formed from coherent delay-and-sum beamforming and thresholding of beam power. Across these settings, the trigger acts on quantities that preserve phase information across apertures, wavelengths, or antennas, thereby resolving ambiguities, reducing thresholds, or automating recovery and follow-up (Maier et al., 2020, Allison et al., 2018).

1. Core concept and trigger observables

The cited literature suggests two principal interferometric-trigger regimes. The first is a phase-error trigger, in which the monitored observable is an interferometric phase or a multi-wavelength phase combination, and the trigger commands a pathlength correction when the loop has slipped by an integer number of fringes. The second is a beam-power trigger, in which multiple antenna voltages are coherently summed into one or more trial arrival directions and the trigger fires when the power in any formed beam exceeds threshold.

In optical fringe tracking, the relevant ambiguity is the cyclic nature of phase: only phase modulo $360^\circ$ is directly measured. PHASECam therefore constructs a multi-wavelength scalar metric from simultaneous $H$- and $K$-band phase telemetry to break the $2\pi$ ambiguity over several $K$-band fringes. Its basic observable is

$$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$

with the modulo operation yielding values in $[0^\circ,360^\circ)$ (Maier et al., 2020).

In phased radio triggering, the basic observable is the coherent sum

$V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$

where $v_n(t)$ is the digitized voltage from antenna $n$, $H$0 are amplitude weights, and $H$1 are geometric delays for a plane wave from zenith angle $H$2. The trigger is then formed from beam power integrated over a short window, effectively testing whether a weak impulsive signal adds coherently across the array (Allison et al., 2018).

The formal SNR advantage is the same in both the 2016 development study and the later NuPhase implementation: a true coherent signal grows proportionally to $H$3, while uncorrelated noise grows as $H$4. In the ideal uncorrelated limit, this yields the familiar $H$5 array gain (Avva et al., 2016). A common simplification is to treat interferometric triggering as ordinary thresholding with many inputs; the beamforming formalism shows instead that the trigger statistic is explicitly directional and phase-aligned.

2. Multi-wavelength fringe-jump triggering in PHASECam

PHASECam is the fringe tracker for the Large Binocular Telescope Interferometer. It is a near-infrared camera used to measure both tip/tilt and fringe phase variations between the two AO-corrected apertures of the Large Binocular Telescope. Tip/tilt and phase sensing are performed in the $H$6 ($H$7) and $H$8 ($H$9) bands at $K$0, but only the $K$1-band phase telemetry is used to send corrections to maintain fringe coherence and visibility. Because the phase is cyclic, PHASECam’s phase-unwrapping algorithm can fail during fast, large phase variations or at low SNR, producing a fringe jump in which the OPD correction is incorrect by a wavelength (Maier et al., 2020).

The multi-wavelength fringe-jump capture and correction algorithm uses the difference-modulo metric $K$2 defined above. Over an OPD span of $K$3—the first common multiple of $K$4 and $K$5, corresponding to $K$6—this metric increases or decreases monotonically across $K$7–$K$8. To suppress measurement noise and obtain a continuous estimate of fringe displacement relative to a reference fringe, PHASECam forms the phasor average

$K$9

with $2\pi$0 at $2\pi$1, implying $2\pi$2. The phasor is thus mapped back into $2\pi$3 (Maier et al., 2020).

Immediately after closing the fringe-tracking loop, the system averages $2\pi$4 over approximately $2\pi$5 to form a reference value $2\pi$6. When the manual or automated pathlength setpoint $2\pi$7 is adjusted without a fringe jump, the reference is updated by

$2\pi$8

because a pure $2\pi$9 setpoint shift in $K$0 band corresponds to a $K$1 shift in diffmod space (Maier et al., 2020).

Fringe-jump detection is evaluated every $K$2 by comparing $K$3 to $K$4 and assigning a relative fringe index $K$5 through thresholds at $K$6 and the wrap regions near $K$7:

$K$8

Each fringe in diffmod space is $K$9 wide, so the $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$0 bounds separate in-fringe from out-of-fringe states. To suppress spurious triggers, PHASECam uses a signed hysteresis counter $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$1: each detected nonzero fringe value increments $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$2 by $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$3, each zero-fringe sample decrements it by $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$4, and a fringe jump is permanently declared only when $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$5 reaches a user-set minimum, currently five consecutive out-of-bounds detections (Maier et al., 2020).

Once a fringe jump with $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$6 is latched, the next OPD correction sent to the Fast/Slow pathlength correctors is offset by $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$7, that is, by $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$8 in $$d_i = (\phi_{H,\mathrm{raw},i} - \phi_{K,\mathrm{raw},i}) \bmod 360^\circ,$$9 band. After this bump is applied, the counter drains back to zero, the relative fringe index returns to zero, and the diffmod loop resumes monitoring. The operational sequence is therefore: wait for fringe-tracker loop closure, compute the reference over $[0^\circ,360^\circ)$0, compute rolling $[0^\circ,360^\circ)$1 every $[0^\circ,360^\circ)$2, compare to $[0^\circ,360^\circ)$3 and update $[0^\circ,360^\circ)$4, and apply a $[0^\circ,360^\circ)$5 OPD bump if $[0^\circ,360^\circ)$6 and $[0^\circ,360^\circ)$7 (Maier et al., 2020).

An archival $[0^\circ,360^\circ)$8 minute nulling fringe-tracking sequence on a $[0^\circ,360^\circ)$9 star from UT 03/28/2018 contained $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$0 fringe-jump events in $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$1, of which $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$2 were primary single-fringe slips or the first in a cluster. Manual operator correction typically required $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$3 per jump, or $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$4 total lost time. The diffmod algorithm detected all $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$5 primary jumps, declared each within $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$6 of occurrence, and applied correction within $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$7 total latency. By those metrics, the automated trigger could have recovered $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$8 of on-fringe time in that sequence; overall, $V(\theta,t) = \sum_n w_n v_n(t-\tau_n(\theta)),$9, or $v_n(t)$0 of the $v_n(t)$1, were lost to fringe jumps. The implementation is described as mode-independent because it uses only raw $v_n(t)$2 phase streams and can therefore be deployed in nulling, multi-axial Fizeau imaging, and non-redundant aperture masking without modification to the optical train (Maier et al., 2020).

3. Interferometric phased-array triggering for radio neutrino detection

For in-ice radio neutrino detection, the motivation for interferometric triggering is the weak, nanosecond-scale Askaryan impulse produced by a high-energy neutrino interaction in ice. Thermal noise in the ice of about $v_n(t)$3 and system noise of about $v_n(t)$4 set a noise-voltage RMS $v_n(t)$5, limiting trigger sensitivity. Coherent summation lowers the effective threshold by allowing a true impulsive signal to add linearly across channels while uncorrelated noise adds only in quadrature (Allison et al., 2018).

The 2016 development study gives the beamforming formalism in terms of time-series voltages $v_n(t)$6:

$v_n(t)$7

with equal weighting often sufficient in practice once channels are gain-matched to about $v_n(t)$8. If each antenna sees a deterministic Askaryan pulse of amplitude $v_n(t)$9 and additive zero-mean noise with variance $n$0 and cross-correlations $n$1, then the beamformed signal amplitude is $n$2 and the beam noise variance is $n$3. The resulting beamformed SNR is

$n$4

which reduces to the usual $n$5 improvement in the uncorrelated limit $n$6 (Avva et al., 2016).

A central empirical question is therefore whether nearby antennas suffer enough thermal-noise correlation to compromise the gain. The same study defines the thermal-noise covariance matrix by $n$7 and normalized coefficients $n$8. In an anechoic-chamber measurement using about $n$9 unbiased noise-only waveforms per channel pair, the peak cross-correlation coefficient was computed over the causal window $H$00. For feed-feed spacings down to about $H$01, the measured $H$02 remained consistent with zero within the $H$03 measurement noise floor, in agreement with a simple $H$04 black-body simulation. The paper therefore concludes that a close-packed in-ice borehole array may safely assume nearly uncorrelated thermal noise between adjacent antennas and recover essentially the full $H$05 SNR gain (Avva et al., 2016).

The trigger statistic is a short-window beam power. In the 2016 formulation,

$H$06

while in NuPhase the beamformer computes a power sum over a programmable window of $H$07 samples, about $H$08, and the trigger fires when the power in any beam exceeds threshold (Allison et al., 2018). The 2016 bench measurements with a three-antenna test array showed the $H$09 trigger-efficiency point shifting from $H$10 for a single channel to $H$11 for two-antenna beamforming and $H$12 for three-antenna beamforming at a per-beam rate of $H$13. Extrapolated simulation for a single beam at the same rate gave $H$14 efficiency at $H$15 for $H$16, $H$17 for $H$18, and $H$19 for $H$20 (Avva et al., 2016).

These results clarify a recurring misconception: the threshold improvement is not obtained merely by adding more antennas, but by delaying and summing them in a geometry matched to a putative plane-wave arrival direction. The trigger is therefore an interferometric matched filter in real time, implemented either in analog or in FPGA-based digital beamforming.

4. NuPhase and ARA A5 as mature in-ice implementations

NuPhase operationalized this trigger concept in the Askaryan Radio Array environment. The detector is a compact receiving array deployed about $H$21 deep in glacial ice near the South Pole, centered on an ARA station. It comprises $H$22 antennas, including $H$23 vertically polarized birdcages at $H$24 spacing and $H$25 horizontal quad slots, although three VPol channels were inoperable after deployment, leaving seven active VPol channels for triggering. Surface electronics digitize the signals at $H$26 with $H$27 bit resolution, and an Intel Arria V FPGA performs real-time beamforming using the lower five bits. The system forms $H$28 simultaneous beams over $H$29 elevation from seven-antenna and five-antenna sub-arrays, with a typical per-beam trigger rate of about $H$30, overall RF-trigger rate of about $H$31, and livetime at least $H$32 (Allison et al., 2018).

In situ calibration with an impulsive near-field pulser and a deep IceCube pulser established timing performance and trigger efficiency. Cross-correlation of deep-pulser waveforms on top and bottom VPol antennas gave a two-channel timing resolution below $H$33. Systematic channel offsets of $H$34–$H$35 were measured and corrected offline. For the near-field pulser, NuPhase achieved $H$36 at $H$37 per beam, $H$38 at $H$39 per beam, and $H$40 at $H$41 per beam. The standard ARA combinatoric trigger had $H$42 at a comparable total rate, so NuPhase lowered the voltage threshold by a factor of about $H$43 relative to ARA. Hardware-level simulation, validated against the calibration data, predicted $H$44 for far-field on-beam plane waves and about $H$45 for off-beam events between beams (Allison et al., 2018).

In detector-sensitivity studies with ARASim, inclusion of the already-achieved NuPhase trigger performance increased the trigger-level effective detector volume by a factor of $H$46 at neutrino energies between $H$47 and $H$48 relative to the currently used ARA combinatoric trigger. The same study identified a near-term path toward $H$49 through a $H$50-antenna VPol string, real-time fractional-delay filters or upsampling, and doubling the beam count to $H$51 beams; lowering $H$52 to $H$53 would yield more than a factor of $H$54 increase in single-station effective volume over the same energy range (Allison et al., 2018).

The ARA A5 phased-array trigger, described in the 2024 search contribution, places a dedicated phased-array string at the center of the fifth ARA station. The string contains $H$55 closely spaced antennas—seven VPol dipoles and two HPol dipoles—at approximately $H$56 depth with inter-element spacing of order $H$57. Around it, at about $H$58 radius, lies the standard outer ARA sub-array of four measurement strings with $H$59 channels total. The trigger FPGA forms $H$60 pre-defined zenith-angle beams from the seven VPol channels and computes power in a sliding $H$61 window,

$H$62

issuing a global trigger whenever any beam exceeds $H$63. The threshold is set so that the total phased-array trigger rate is about $H$64, including calibration triggers (Dasgupta, 2024).

That system reports a $H$65 trigger-efficiency point at an Askaryan peak field strength of about $H$66 in ice for the phased-array trigger, versus about $H$67 for the conventional eight-fold coincidence trigger. For simulated on-cone neutrino pulses, the phased-array trigger reaches $H$68 efficiency at electric-field $H$69 per channel, whereas the single-antenna trigger needs $H$70. The paper states that the phased-array trigger lowers the neutrino energy threshold by about $H$71 and, for $H$72, improves effective volumetric acceptance by roughly a factor of $H$73 compared to the traditional trigger. When the beam direction is used as a seed for offline reconstruction, the hybrid A5/PA approach achieves about $H$74 RMS in azimuth versus about $H$75 for the phased array alone, about $H$76 in zenith versus about $H$77, and reduces the $H$78D vertex-position uncertainty from about $H$79 to about $H$80 for high-SNR calibration pulses (Dasgupta, 2024).

5. Trigger orchestration for an interferometric observatory: the MWA case

A distinct but related use of the term appears at the observatory-control layer. The Murchison Widefield Array is an electronically steered low-frequency radio interferometer with a slew time less than $H$81, and its automatic trigger system responds to external VOEvent notices rather than to internally formed beam-power or fringe-phase metrics. This suggests a broader operational meaning of interferometric trigger: automated trigger handling for an interferometric instrument whose schedule and observing mode can be reconfigured dynamically (Hancock et al., 2019).

The MWA system is built as a two-layer architecture. A front-end VOEvent handling service, using COMET and the $H$82 Sky broker, receives, queues, and parses incoming XML VOEvent packets. Each event is placed on a serial processing queue and handled by a Python daemon, which passes it to one or more user-supplied plugins. Those plugins extract fields such as sky position, classification probability, dispersion measure, duration parameters, observability, and prior detections, and if the logic decides to follow the event, they invoke wrapper routines that call a back-end RESTful scheduling API integrated into the MWA’s PostgreSQL-backed schedule manager. The back-end exposes calls including obslist, busy, triggerobs, triggervcs, and triggerbuffer (Hancock et al., 2019).

Latency is governed jointly by software and scheduling cadence. Upon arrival of a transient VOEvent, COMET invokes the front-end within at most about $H$83, the packet is enqueued, and if a trigger call results, an HTTP request is sent to the back-end with about a $H$84 round-trip. However, the scheduler can insert observations only on an $H$85-aligned GPS-second boundary, and each subsystem requires configuration $H$86 in advance. The built-in latency is therefore $H$87–$H$88 before the new pointing and mode can begin. Additional delays come from VOEvent parsing and decision logic, Sun-avoidance beam selection, automatic calibrator lookup, and correlator-to-VCS mode changes. The total time from external alert reception to correlated or VCS data capture is therefore at most about $H$89–$H$90 (Hancock et al., 2019).

The observing modes include standard correlator mode with integration times of at least $H$91, selectable frequency resolution of $H$92, $H$93, or $H$94, and $H$95 coarse channels per snapshot, as well as the Voltage Capture System with full tile-voltage recording at $H$96 resolution. A buffered VCS mode stores all $H$97 tile voltages in a $H$98 ring buffer, so that on receipt of a triggerbuffer() call the system writes out the past $H$99 while continuing to record in real time, with no additional latency penalty beyond VOEvent handling (Hancock et al., 2019).

The MWA implementation illustrates another important point: trigger robustness often depends on queue discipline, priority control, and explicit exclusion logic as much as on the trigger criterion itself. The front-end guarantees serial, in-order processing; the back-end busy check prevents conflicting correlator-mode triggers; ongoing VCS observations cannot currently be pre-empted by a correlator trigger; and buffered VCS observations are exclusive until the scheduled buffer-to-disk operation completes (Hancock et al., 2019).

6. Cross-cutting design principles, limitations, and extensions

Across the optical and radio implementations, several common design principles recur. First, the trigger statistic is deliberately chosen to preserve coherence information that would be lost in single-channel thresholding: PHASECam uses a two-band phase-difference metric, whereas NuPhase and ARA A5 use coherent delay-and-sum voltages and beam powers. Second, all systems rely on explicit latency management. PHASECam declares primary jumps within about $K$00 and corrects within about $K$01 (Maier et al., 2020); NuPhase forms beams continuously in FPGA hardware at trigger rates of order hertz per beam (Allison et al., 2018); the MWA constrains latency through queueing and $K$02 schedule quantization (Hancock et al., 2019).

Third, robustness requires more than a threshold. PHASECam adds a signed hysteresis counter $K$03 to avoid spurious fringe-jump declarations (Maier et al., 2020). NuPhase identifies firmware-level limitations that degrade ideal performance, including the absence of real-time sub-sample delay corrections, a finite number of beams, and timing mismatches uncorrected in firmware, which create beam-pattern gaps and off-beam threshold penalties (Allison et al., 2018). The ARA A5 analysis emphasizes that the outer sparse array is crucial for breaking the intrinsic azimuthal degeneracy of a single string and that real-time FPGA beamforming must be calibrated in situ for antenna positions and cable delays at the sub-nanosecond level (Dasgupta, 2024).

Fourth, scalability is intrinsic to the interferometric approach. The PHASECam study states that the same concept extends to more wavelengths or to synthetic wavelengths, allowing unambiguous tracking over larger OPD excursions (Maier et al., 2020). In radio, the 2016 and 2018 studies both show the progression toward lower thresholds with larger phased arrays, culminating in the forecast that a $K$04-antenna VPol string together with fractional-delay correction and more beams could push $K$05 to $K$06 and increase single-station effective volume by more than a factor of $K$07 at $K$08–$K$09 (Avva et al., 2016, Allison et al., 2018).

A final conceptual distinction is that interferometric triggering is not a single technology but a family of trigger logics operating at different layers of the instrument. In PHASECam it is a control trigger that restores lock by applying an exact $K$10 OPD correction. In NuPhase and ARA A5 it is a detection trigger that promotes weak impulsive signals above thermal noise through coherent summation. In the MWA it is an observatory trigger that turns external event metadata into rapid reconfiguration of an electronically steered interferometer. The cited systems therefore locate the defining feature not in a particular hardware platform, but in the use of interferometric structure—phase differences, coherent sums, or interferometer steering and schedule control—to make a real-time decision.